Debian Codegraph @

tools/lib/bpf/libbpf.c

raw
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2
3 /*
4 * Common eBPF ELF object loading operations.
5 *
6 * Copyright (C) 2013-2015 Alexei Starovoitov <[email protected]>
7 * Copyright (C) 2015 Wang Nan <[email protected]>
8 * Copyright (C) 2015 Huawei Inc.
9 * Copyright (C) 2017 Nicira, Inc.
10 * Copyright (C) 2019 Isovalent, Inc.
11 */
12
13 #ifndef _GNU_SOURCE
14 #define _GNU_SOURCE
15 #endif
16 #include <stdlib.h>
17 #include <stdio.h>
18 #include <stdarg.h>
19 #include <libgen.h>
20 #include <inttypes.h>
21 #include <limits.h>
22 #include <string.h>
23 #include <unistd.h>
24 #include <endian.h>
25 #include <fcntl.h>
26 #include <errno.h>
27 #include <ctype.h>
28 #include <asm/unistd.h>
29 #include <linux/err.h>
30 #include <linux/kernel.h>
31 #include <linux/bpf.h>
32 #include <linux/btf.h>
33 #include <linux/filter.h>
34 #include <linux/limits.h>
35 #include <linux/perf_event.h>
36 #include <linux/bpf_perf_event.h>
37 #include <linux/ring_buffer.h>
38 #include <sys/epoll.h>
39 #include <sys/ioctl.h>
40 #include <sys/mman.h>
41 #include <sys/stat.h>
42 #include <sys/types.h>
43 #include <sys/vfs.h>
44 #include <sys/utsname.h>
45 #include <sys/resource.h>
46 #include <libelf.h>
47 #include <gelf.h>
48 #include <zlib.h>
49
50 #include "libbpf.h"
51 #include "bpf.h"
52 #include "btf.h"
53 #include "libbpf_internal.h"
54 #include "hashmap.h"
55 #include "bpf_gen_internal.h"
56 #include "zip.h"
57
58 #ifndef BPF_FS_MAGIC
59 #define BPF_FS_MAGIC 0xcafe4a11
60 #endif
61
62 #define MAX_EVENT_NAME_LEN 64
63
64 #define BPF_FS_DEFAULT_PATH "/sys/fs/bpf"
65
66 #define BPF_INSN_SZ (sizeof(struct bpf_insn))
67
68 /* vsprintf() in __base_pr() uses nonliteral format string. It may break
69 * compilation if user enables corresponding warning. Disable it explicitly.
70 */
71 #pragma GCC diagnostic ignored "-Wformat-nonliteral"
72
73 #define __printf(a, b) __attribute__((format(printf, a, b)))
74
75 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj);
76 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog);
77 static int map_set_def_max_entries(struct bpf_map *map);
78
79 static const char * const attach_type_name[] = {
80 [BPF_CGROUP_INET_INGRESS] = "cgroup_inet_ingress",
81 [BPF_CGROUP_INET_EGRESS] = "cgroup_inet_egress",
82 [BPF_CGROUP_INET_SOCK_CREATE] = "cgroup_inet_sock_create",
83 [BPF_CGROUP_INET_SOCK_RELEASE] = "cgroup_inet_sock_release",
84 [BPF_CGROUP_SOCK_OPS] = "cgroup_sock_ops",
85 [BPF_CGROUP_DEVICE] = "cgroup_device",
86 [BPF_CGROUP_INET4_BIND] = "cgroup_inet4_bind",
87 [BPF_CGROUP_INET6_BIND] = "cgroup_inet6_bind",
88 [BPF_CGROUP_INET4_CONNECT] = "cgroup_inet4_connect",
89 [BPF_CGROUP_INET6_CONNECT] = "cgroup_inet6_connect",
90 [BPF_CGROUP_UNIX_CONNECT] = "cgroup_unix_connect",
91 [BPF_CGROUP_INET4_POST_BIND] = "cgroup_inet4_post_bind",
92 [BPF_CGROUP_INET6_POST_BIND] = "cgroup_inet6_post_bind",
93 [BPF_CGROUP_INET4_GETPEERNAME] = "cgroup_inet4_getpeername",
94 [BPF_CGROUP_INET6_GETPEERNAME] = "cgroup_inet6_getpeername",
95 [BPF_CGROUP_UNIX_GETPEERNAME] = "cgroup_unix_getpeername",
96 [BPF_CGROUP_INET4_GETSOCKNAME] = "cgroup_inet4_getsockname",
97 [BPF_CGROUP_INET6_GETSOCKNAME] = "cgroup_inet6_getsockname",
98 [BPF_CGROUP_UNIX_GETSOCKNAME] = "cgroup_unix_getsockname",
99 [BPF_CGROUP_UDP4_SENDMSG] = "cgroup_udp4_sendmsg",
100 [BPF_CGROUP_UDP6_SENDMSG] = "cgroup_udp6_sendmsg",
101 [BPF_CGROUP_UNIX_SENDMSG] = "cgroup_unix_sendmsg",
102 [BPF_CGROUP_SYSCTL] = "cgroup_sysctl",
103 [BPF_CGROUP_UDP4_RECVMSG] = "cgroup_udp4_recvmsg",
104 [BPF_CGROUP_UDP6_RECVMSG] = "cgroup_udp6_recvmsg",
105 [BPF_CGROUP_UNIX_RECVMSG] = "cgroup_unix_recvmsg",
106 [BPF_CGROUP_GETSOCKOPT] = "cgroup_getsockopt",
107 [BPF_CGROUP_SETSOCKOPT] = "cgroup_setsockopt",
108 [BPF_SK_SKB_STREAM_PARSER] = "sk_skb_stream_parser",
109 [BPF_SK_SKB_STREAM_VERDICT] = "sk_skb_stream_verdict",
110 [BPF_SK_SKB_VERDICT] = "sk_skb_verdict",
111 [BPF_SK_MSG_VERDICT] = "sk_msg_verdict",
112 [BPF_LIRC_MODE2] = "lirc_mode2",
113 [BPF_FLOW_DISSECTOR] = "flow_dissector",
114 [BPF_TRACE_RAW_TP] = "trace_raw_tp",
115 [BPF_TRACE_FENTRY] = "trace_fentry",
116 [BPF_TRACE_FEXIT] = "trace_fexit",
117 [BPF_MODIFY_RETURN] = "modify_return",
118 [BPF_TRACE_FSESSION] = "trace_fsession",
119 [BPF_LSM_MAC] = "lsm_mac",
120 [BPF_LSM_CGROUP] = "lsm_cgroup",
121 [BPF_SK_LOOKUP] = "sk_lookup",
122 [BPF_TRACE_ITER] = "trace_iter",
123 [BPF_XDP_DEVMAP] = "xdp_devmap",
124 [BPF_XDP_CPUMAP] = "xdp_cpumap",
125 [BPF_XDP] = "xdp",
126 [BPF_SK_REUSEPORT_SELECT] = "sk_reuseport_select",
127 [BPF_SK_REUSEPORT_SELECT_OR_MIGRATE] = "sk_reuseport_select_or_migrate",
128 [BPF_PERF_EVENT] = "perf_event",
129 [BPF_TRACE_KPROBE_MULTI] = "trace_kprobe_multi",
130 [BPF_STRUCT_OPS] = "struct_ops",
131 [BPF_NETFILTER] = "netfilter",
132 [BPF_TCX_INGRESS] = "tcx_ingress",
133 [BPF_TCX_EGRESS] = "tcx_egress",
134 [BPF_TRACE_UPROBE_MULTI] = "trace_uprobe_multi",
135 [BPF_NETKIT_PRIMARY] = "netkit_primary",
136 [BPF_NETKIT_PEER] = "netkit_peer",
137 [BPF_TRACE_KPROBE_SESSION] = "trace_kprobe_session",
138 [BPF_TRACE_UPROBE_SESSION] = "trace_uprobe_session",
139 };
140
141 static const char * const link_type_name[] = {
142 [BPF_LINK_TYPE_UNSPEC] = "unspec",
143 [BPF_LINK_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
144 [BPF_LINK_TYPE_TRACING] = "tracing",
145 [BPF_LINK_TYPE_CGROUP] = "cgroup",
146 [BPF_LINK_TYPE_ITER] = "iter",
147 [BPF_LINK_TYPE_NETNS] = "netns",
148 [BPF_LINK_TYPE_XDP] = "xdp",
149 [BPF_LINK_TYPE_PERF_EVENT] = "perf_event",
150 [BPF_LINK_TYPE_KPROBE_MULTI] = "kprobe_multi",
151 [BPF_LINK_TYPE_STRUCT_OPS] = "struct_ops",
152 [BPF_LINK_TYPE_NETFILTER] = "netfilter",
153 [BPF_LINK_TYPE_TCX] = "tcx",
154 [BPF_LINK_TYPE_UPROBE_MULTI] = "uprobe_multi",
155 [BPF_LINK_TYPE_NETKIT] = "netkit",
156 [BPF_LINK_TYPE_SOCKMAP] = "sockmap",
157 };
158
159 static const char * const map_type_name[] = {
160 [BPF_MAP_TYPE_UNSPEC] = "unspec",
161 [BPF_MAP_TYPE_HASH] = "hash",
162 [BPF_MAP_TYPE_ARRAY] = "array",
163 [BPF_MAP_TYPE_PROG_ARRAY] = "prog_array",
164 [BPF_MAP_TYPE_PERF_EVENT_ARRAY] = "perf_event_array",
165 [BPF_MAP_TYPE_PERCPU_HASH] = "percpu_hash",
166 [BPF_MAP_TYPE_PERCPU_ARRAY] = "percpu_array",
167 [BPF_MAP_TYPE_STACK_TRACE] = "stack_trace",
168 [BPF_MAP_TYPE_CGROUP_ARRAY] = "cgroup_array",
169 [BPF_MAP_TYPE_LRU_HASH] = "lru_hash",
170 [BPF_MAP_TYPE_LRU_PERCPU_HASH] = "lru_percpu_hash",
171 [BPF_MAP_TYPE_LPM_TRIE] = "lpm_trie",
172 [BPF_MAP_TYPE_ARRAY_OF_MAPS] = "array_of_maps",
173 [BPF_MAP_TYPE_HASH_OF_MAPS] = "hash_of_maps",
174 [BPF_MAP_TYPE_DEVMAP] = "devmap",
175 [BPF_MAP_TYPE_DEVMAP_HASH] = "devmap_hash",
176 [BPF_MAP_TYPE_SOCKMAP] = "sockmap",
177 [BPF_MAP_TYPE_CPUMAP] = "cpumap",
178 [BPF_MAP_TYPE_XSKMAP] = "xskmap",
179 [BPF_MAP_TYPE_SOCKHASH] = "sockhash",
180 [BPF_MAP_TYPE_CGROUP_STORAGE] = "cgroup_storage",
181 [BPF_MAP_TYPE_REUSEPORT_SOCKARRAY] = "reuseport_sockarray",
182 [BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE] = "percpu_cgroup_storage",
183 [BPF_MAP_TYPE_QUEUE] = "queue",
184 [BPF_MAP_TYPE_STACK] = "stack",
185 [BPF_MAP_TYPE_SK_STORAGE] = "sk_storage",
186 [BPF_MAP_TYPE_STRUCT_OPS] = "struct_ops",
187 [BPF_MAP_TYPE_RINGBUF] = "ringbuf",
188 [BPF_MAP_TYPE_INODE_STORAGE] = "inode_storage",
189 [BPF_MAP_TYPE_TASK_STORAGE] = "task_storage",
190 [BPF_MAP_TYPE_BLOOM_FILTER] = "bloom_filter",
191 [BPF_MAP_TYPE_USER_RINGBUF] = "user_ringbuf",
192 [BPF_MAP_TYPE_CGRP_STORAGE] = "cgrp_storage",
193 [BPF_MAP_TYPE_ARENA] = "arena",
194 [BPF_MAP_TYPE_INSN_ARRAY] = "insn_array",
195 };
196
197 static const char * const prog_type_name[] = {
198 [BPF_PROG_TYPE_UNSPEC] = "unspec",
199 [BPF_PROG_TYPE_SOCKET_FILTER] = "socket_filter",
200 [BPF_PROG_TYPE_KPROBE] = "kprobe",
201 [BPF_PROG_TYPE_SCHED_CLS] = "sched_cls",
202 [BPF_PROG_TYPE_SCHED_ACT] = "sched_act",
203 [BPF_PROG_TYPE_TRACEPOINT] = "tracepoint",
204 [BPF_PROG_TYPE_XDP] = "xdp",
205 [BPF_PROG_TYPE_PERF_EVENT] = "perf_event",
206 [BPF_PROG_TYPE_CGROUP_SKB] = "cgroup_skb",
207 [BPF_PROG_TYPE_CGROUP_SOCK] = "cgroup_sock",
208 [BPF_PROG_TYPE_LWT_IN] = "lwt_in",
209 [BPF_PROG_TYPE_LWT_OUT] = "lwt_out",
210 [BPF_PROG_TYPE_LWT_XMIT] = "lwt_xmit",
211 [BPF_PROG_TYPE_SOCK_OPS] = "sock_ops",
212 [BPF_PROG_TYPE_SK_SKB] = "sk_skb",
213 [BPF_PROG_TYPE_CGROUP_DEVICE] = "cgroup_device",
214 [BPF_PROG_TYPE_SK_MSG] = "sk_msg",
215 [BPF_PROG_TYPE_RAW_TRACEPOINT] = "raw_tracepoint",
216 [BPF_PROG_TYPE_CGROUP_SOCK_ADDR] = "cgroup_sock_addr",
217 [BPF_PROG_TYPE_LWT_SEG6LOCAL] = "lwt_seg6local",
218 [BPF_PROG_TYPE_LIRC_MODE2] = "lirc_mode2",
219 [BPF_PROG_TYPE_SK_REUSEPORT] = "sk_reuseport",
220 [BPF_PROG_TYPE_FLOW_DISSECTOR] = "flow_dissector",
221 [BPF_PROG_TYPE_CGROUP_SYSCTL] = "cgroup_sysctl",
222 [BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE] = "raw_tracepoint_writable",
223 [BPF_PROG_TYPE_CGROUP_SOCKOPT] = "cgroup_sockopt",
224 [BPF_PROG_TYPE_TRACING] = "tracing",
225 [BPF_PROG_TYPE_STRUCT_OPS] = "struct_ops",
226 [BPF_PROG_TYPE_EXT] = "ext",
227 [BPF_PROG_TYPE_LSM] = "lsm",
228 [BPF_PROG_TYPE_SK_LOOKUP] = "sk_lookup",
229 [BPF_PROG_TYPE_SYSCALL] = "syscall",
230 [BPF_PROG_TYPE_NETFILTER] = "netfilter",
231 };
232
233 static int __base_pr(enum libbpf_print_level level, const char *format,
234 va_list args)
235 {
236 const char *env_var = "LIBBPF_LOG_LEVEL";
237 static enum libbpf_print_level min_level = LIBBPF_INFO;
238 static bool initialized;
239
240 if (!initialized) {
241 char *verbosity;
242
243 initialized = true;
244 verbosity = getenv(env_var);
245 if (verbosity) {
246 if (strcasecmp(verbosity, "warn") == 0)
247 min_level = LIBBPF_WARN;
248 else if (strcasecmp(verbosity, "debug") == 0)
249 min_level = LIBBPF_DEBUG;
250 else if (strcasecmp(verbosity, "info") == 0)
251 min_level = LIBBPF_INFO;
252 else
253 fprintf(stderr, "libbpf: unrecognized '%s' envvar value: '%s', should be one of 'warn', 'debug', or 'info'.\n",
254 env_var, verbosity);
255 }
256 }
257
258 /* if too verbose, skip logging */
259 if (level > min_level)
260 return 0;
261
262 return vfprintf(stderr, format, args);
263 }
264
265 static libbpf_print_fn_t __libbpf_pr = __base_pr;
266
267 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
268 {
269 libbpf_print_fn_t old_print_fn;
270
271 old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
272
273 return old_print_fn;
274 }
275
276 __printf(2, 3)
277 void libbpf_print(enum libbpf_print_level level, const char *format, ...)
278 {
279 va_list args;
280 int old_errno;
281 libbpf_print_fn_t print_fn;
282
283 print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
284 if (!print_fn)
285 return;
286
287 old_errno = errno;
288
289 va_start(args, format);
290 print_fn(level, format, args);
291 va_end(args);
292
293 errno = old_errno;
294 }
295
296 static void pr_perm_msg(int err)
297 {
298 struct rlimit limit;
299 char buf[100];
300
301 if (err != -EPERM || geteuid() != 0)
302 return;
303
304 err = getrlimit(RLIMIT_MEMLOCK, &limit);
305 if (err)
306 return;
307
308 if (limit.rlim_cur == RLIM_INFINITY)
309 return;
310
311 if (limit.rlim_cur < 1024)
312 snprintf(buf, sizeof(buf), "%zu bytes", (size_t)limit.rlim_cur);
313 else if (limit.rlim_cur < 1024*1024)
314 snprintf(buf, sizeof(buf), "%.1f KiB", (double)limit.rlim_cur / 1024);
315 else
316 snprintf(buf, sizeof(buf), "%.1f MiB", (double)limit.rlim_cur / (1024*1024));
317
318 pr_warn("permission error while running as root; try raising 'ulimit -l'? current value: %s\n",
319 buf);
320 }
321
322 /* Copied from tools/perf/util/util.h */
323 #ifndef zfree
324 # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
325 #endif
326
327 #ifndef zclose
328 # define zclose(fd) ({ \
329 int ___err = 0; \
330 if ((fd) >= 0) \
331 ___err = close((fd)); \
332 fd = -1; \
333 ___err; })
334 #endif
335
336 static inline __u64 ptr_to_u64(const void *ptr)
337 {
338 return (__u64) (unsigned long) ptr;
339 }
340
341 int libbpf_set_strict_mode(enum libbpf_strict_mode mode)
342 {
343 /* as of v1.0 libbpf_set_strict_mode() is a no-op */
344 return 0;
345 }
346
347 __u32 libbpf_major_version(void)
348 {
349 return LIBBPF_MAJOR_VERSION;
350 }
351
352 __u32 libbpf_minor_version(void)
353 {
354 return LIBBPF_MINOR_VERSION;
355 }
356
357 const char *libbpf_version_string(void)
358 {
359 #define __S(X) #X
360 #define _S(X) __S(X)
361 return "v" _S(LIBBPF_MAJOR_VERSION) "." _S(LIBBPF_MINOR_VERSION);
362 #undef _S
363 #undef __S
364 }
365
366 enum reloc_type {
367 RELO_LD64,
368 RELO_CALL,
369 RELO_DATA,
370 RELO_EXTERN_LD64,
371 RELO_EXTERN_CALL,
372 RELO_SUBPROG_ADDR,
373 RELO_CORE,
374 RELO_INSN_ARRAY,
375 };
376
377 struct reloc_desc {
378 enum reloc_type type;
379 int insn_idx;
380 union {
381 const struct bpf_core_relo *core_relo; /* used when type == RELO_CORE */
382 struct {
383 int map_idx;
384 unsigned int sym_off;
385 /*
386 * The following two fields can be unionized, as the
387 * ext_idx field is used for extern symbols, and the
388 * sym_size is used for jump tables, which are never
389 * extern
390 */
391 union {
392 int ext_idx;
393 int sym_size;
394 };
395 };
396 };
397 };
398
399 /* stored as sec_def->cookie for all libbpf-supported SEC()s */
400 enum sec_def_flags {
401 SEC_NONE = 0,
402 /* expected_attach_type is optional, if kernel doesn't support that */
403 SEC_EXP_ATTACH_OPT = 1,
404 /* legacy, only used by libbpf_get_type_names() and
405 * libbpf_attach_type_by_name(), not used by libbpf itself at all.
406 * This used to be associated with cgroup (and few other) BPF programs
407 * that were attachable through BPF_PROG_ATTACH command. Pretty
408 * meaningless nowadays, though.
409 */
410 SEC_ATTACHABLE = 2,
411 SEC_ATTACHABLE_OPT = SEC_ATTACHABLE | SEC_EXP_ATTACH_OPT,
412 /* attachment target is specified through BTF ID in either kernel or
413 * other BPF program's BTF object
414 */
415 SEC_ATTACH_BTF = 4,
416 /* BPF program type allows sleeping/blocking in kernel */
417 SEC_SLEEPABLE = 8,
418 /* BPF program support non-linear XDP buffer */
419 SEC_XDP_FRAGS = 16,
420 /* Setup proper attach type for usdt probes. */
421 SEC_USDT = 32,
422 };
423
424 struct bpf_sec_def {
425 char *sec;
426 enum bpf_prog_type prog_type;
427 enum bpf_attach_type expected_attach_type;
428 long cookie;
429 int handler_id;
430
431 libbpf_prog_setup_fn_t prog_setup_fn;
432 libbpf_prog_prepare_load_fn_t prog_prepare_load_fn;
433 libbpf_prog_attach_fn_t prog_attach_fn;
434 };
435
436 struct bpf_light_subprog {
437 __u32 sec_insn_off;
438 __u32 sub_insn_off;
439 };
440
441 /*
442 * bpf_prog should be a better name but it has been used in
443 * linux/filter.h.
444 */
445 struct bpf_program {
446 char *name;
447 char *sec_name;
448 size_t sec_idx;
449 const struct bpf_sec_def *sec_def;
450 /* this program's instruction offset (in number of instructions)
451 * within its containing ELF section
452 */
453 size_t sec_insn_off;
454 /* number of original instructions in ELF section belonging to this
455 * program, not taking into account subprogram instructions possible
456 * appended later during relocation
457 */
458 size_t sec_insn_cnt;
459 /* Offset (in number of instructions) of the start of instruction
460 * belonging to this BPF program within its containing main BPF
461 * program. For the entry-point (main) BPF program, this is always
462 * zero. For a sub-program, this gets reset before each of main BPF
463 * programs are processed and relocated and is used to determined
464 * whether sub-program was already appended to the main program, and
465 * if yes, at which instruction offset.
466 */
467 size_t sub_insn_off;
468
469 /* instructions that belong to BPF program; insns[0] is located at
470 * sec_insn_off instruction within its ELF section in ELF file, so
471 * when mapping ELF file instruction index to the local instruction,
472 * one needs to subtract sec_insn_off; and vice versa.
473 */
474 struct bpf_insn *insns;
475 /* actual number of instruction in this BPF program's image; for
476 * entry-point BPF programs this includes the size of main program
477 * itself plus all the used sub-programs, appended at the end
478 */
479 size_t insns_cnt;
480
481 struct reloc_desc *reloc_desc;
482 int nr_reloc;
483
484 /* BPF verifier log settings */
485 char *log_buf;
486 size_t log_size;
487 __u32 log_level;
488
489 struct bpf_object *obj;
490
491 int fd;
492 bool autoload;
493 bool autoattach;
494 bool sym_global;
495 bool mark_btf_static;
496 enum bpf_prog_type type;
497 enum bpf_attach_type expected_attach_type;
498 int exception_cb_idx;
499
500 int prog_ifindex;
501 __u32 attach_btf_obj_fd;
502 __u32 attach_btf_id;
503 __u32 attach_prog_fd;
504
505 void *func_info;
506 __u32 func_info_rec_size;
507 __u32 func_info_cnt;
508
509 void *line_info;
510 __u32 line_info_rec_size;
511 __u32 line_info_cnt;
512 __u32 prog_flags;
513 __u8 hash[SHA256_DIGEST_LENGTH];
514
515 struct bpf_light_subprog *subprogs;
516 __u32 subprog_cnt;
517 };
518
519 struct bpf_struct_ops {
520 struct bpf_program **progs;
521 __u32 *kern_func_off;
522 /* e.g. struct tcp_congestion_ops in bpf_prog's btf format */
523 void *data;
524 /* e.g. struct bpf_struct_ops_tcp_congestion_ops in
525 * btf_vmlinux's format.
526 * struct bpf_struct_ops_tcp_congestion_ops {
527 * [... some other kernel fields ...]
528 * struct tcp_congestion_ops data;
529 * }
530 * kern_vdata-size == sizeof(struct bpf_struct_ops_tcp_congestion_ops)
531 * bpf_map__init_kern_struct_ops() will populate the "kern_vdata"
532 * from "data".
533 */
534 void *kern_vdata;
535 __u32 type_id;
536 };
537
538 #define DATA_SEC ".data"
539 #define BSS_SEC ".bss"
540 #define RODATA_SEC ".rodata"
541 #define KCONFIG_SEC ".kconfig"
542 #define KSYMS_SEC ".ksyms"
543 #define STRUCT_OPS_SEC ".struct_ops"
544 #define STRUCT_OPS_LINK_SEC ".struct_ops.link"
545 #define ARENA_SEC ".addr_space.1"
546
547 enum libbpf_map_type {
548 LIBBPF_MAP_UNSPEC,
549 LIBBPF_MAP_DATA,
550 LIBBPF_MAP_BSS,
551 LIBBPF_MAP_RODATA,
552 LIBBPF_MAP_KCONFIG,
553 };
554
555 struct bpf_map_def {
556 unsigned int type;
557 unsigned int key_size;
558 unsigned int value_size;
559 unsigned int max_entries;
560 unsigned int map_flags;
561 };
562
563 struct bpf_map {
564 struct bpf_object *obj;
565 char *name;
566 /* real_name is defined for special internal maps (.rodata*,
567 * .data*, .bss, .kconfig) and preserves their original ELF section
568 * name. This is important to be able to find corresponding BTF
569 * DATASEC information.
570 */
571 char *real_name;
572 int fd;
573 int sec_idx;
574 size_t sec_offset;
575 int map_ifindex;
576 int inner_map_fd;
577 struct bpf_map_def def;
578 __u32 numa_node;
579 __u32 btf_var_idx;
580 int mod_btf_fd;
581 __u32 btf_key_type_id;
582 __u32 btf_value_type_id;
583 __u32 btf_vmlinux_value_type_id;
584 enum libbpf_map_type libbpf_type;
585 void *mmaped;
586 struct bpf_struct_ops *st_ops;
587 struct bpf_map *inner_map;
588 void **init_slots;
589 int init_slots_sz;
590 char *pin_path;
591 bool pinned;
592 bool reused;
593 bool autocreate;
594 bool autoattach;
595 __u64 map_extra;
596 struct bpf_program *excl_prog;
597 };
598
599 enum extern_type {
600 EXT_UNKNOWN,
601 EXT_KCFG,
602 EXT_KSYM,
603 };
604
605 enum kcfg_type {
606 KCFG_UNKNOWN,
607 KCFG_CHAR,
608 KCFG_BOOL,
609 KCFG_INT,
610 KCFG_TRISTATE,
611 KCFG_CHAR_ARR,
612 };
613
614 struct extern_desc {
615 enum extern_type type;
616 int sym_idx;
617 int btf_id;
618 int sec_btf_id;
619 char *name;
620 char *essent_name;
621 bool is_set;
622 bool is_weak;
623 union {
624 struct {
625 enum kcfg_type type;
626 int sz;
627 int align;
628 int data_off;
629 bool is_signed;
630 } kcfg;
631 struct {
632 unsigned long long addr;
633
634 /* target btf_id of the corresponding kernel var. */
635 int kernel_btf_obj_fd;
636 int kernel_btf_id;
637
638 /* local btf_id of the ksym extern's type. */
639 __u32 type_id;
640 /* BTF fd index to be patched in for insn->off, this is
641 * 0 for vmlinux BTF, index in obj->fd_array for module
642 * BTF
643 */
644 __s16 btf_fd_idx;
645 } ksym;
646 };
647 };
648
649 struct module_btf {
650 struct btf *btf;
651 char *name;
652 __u32 id;
653 int fd;
654 int fd_array_idx;
655 };
656
657 enum sec_type {
658 SEC_UNUSED = 0,
659 SEC_RELO,
660 SEC_BSS,
661 SEC_DATA,
662 SEC_RODATA,
663 SEC_ST_OPS,
664 };
665
666 struct elf_sec_desc {
667 enum sec_type sec_type;
668 Elf64_Shdr *shdr;
669 Elf_Data *data;
670 };
671
672 struct elf_state {
673 int fd;
674 const void *obj_buf;
675 size_t obj_buf_sz;
676 Elf *elf;
677 Elf64_Ehdr *ehdr;
678 Elf_Data *symbols;
679 Elf_Data *arena_data;
680 size_t shstrndx; /* section index for section name strings */
681 size_t strtabidx;
682 struct elf_sec_desc *secs;
683 size_t sec_cnt;
684 int btf_maps_shndx;
685 __u32 btf_maps_sec_btf_id;
686 int text_shndx;
687 int symbols_shndx;
688 bool has_st_ops;
689 int arena_data_shndx;
690 int jumptables_data_shndx;
691 };
692
693 struct usdt_manager;
694
695 enum bpf_object_state {
696 OBJ_OPEN,
697 OBJ_PREPARED,
698 OBJ_LOADED,
699 };
700
701 struct bpf_object {
702 char name[BPF_OBJ_NAME_LEN];
703 char license[64];
704 __u32 kern_version;
705
706 enum bpf_object_state state;
707 struct bpf_program *programs;
708 size_t nr_programs;
709 struct bpf_map *maps;
710 size_t nr_maps;
711 size_t maps_cap;
712
713 char *kconfig;
714 struct extern_desc *externs;
715 int nr_extern;
716 int kconfig_map_idx;
717
718 bool has_subcalls;
719 bool has_rodata;
720
721 struct bpf_gen *gen_loader;
722
723 /* Information when doing ELF related work. Only valid if efile.elf is not NULL */
724 struct elf_state efile;
725
726 unsigned char byteorder;
727
728 struct btf *btf;
729 struct btf_ext *btf_ext;
730
731 /* Parse and load BTF vmlinux if any of the programs in the object need
732 * it at load time.
733 */
734 struct btf *btf_vmlinux;
735 /* Path to the custom BTF to be used for BPF CO-RE relocations as an
736 * override for vmlinux BTF.
737 */
738 char *btf_custom_path;
739 /* vmlinux BTF override for CO-RE relocations */
740 struct btf *btf_vmlinux_override;
741 /* Lazily initialized kernel module BTFs */
742 struct module_btf *btf_modules;
743 bool btf_modules_loaded;
744 size_t btf_module_cnt;
745 size_t btf_module_cap;
746
747 /* optional log settings passed to BPF_BTF_LOAD and BPF_PROG_LOAD commands */
748 char *log_buf;
749 size_t log_size;
750 __u32 log_level;
751
752 int *fd_array;
753 size_t fd_array_cap;
754 size_t fd_array_cnt;
755
756 struct usdt_manager *usdt_man;
757
758 int arena_map_idx;
759 void *arena_data;
760 size_t arena_data_sz;
761 size_t arena_data_off;
762
763 void *jumptables_data;
764 size_t jumptables_data_sz;
765
766 struct {
767 struct bpf_program *prog;
768 unsigned int sym_off;
769 int fd;
770 } *jumptable_maps;
771 size_t jumptable_map_cnt;
772
773 struct kern_feature_cache *feat_cache;
774 char *token_path;
775 int token_fd;
776
777 char path[];
778 };
779
780 static const char *elf_sym_str(const struct bpf_object *obj, size_t off);
781 static const char *elf_sec_str(const struct bpf_object *obj, size_t off);
782 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx);
783 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name);
784 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn);
785 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn);
786 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn);
787 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx);
788 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx);
789
790 void bpf_program__unload(struct bpf_program *prog)
791 {
792 if (!prog)
793 return;
794
795 zclose(prog->fd);
796
797 zfree(&prog->func_info);
798 zfree(&prog->line_info);
799 zfree(&prog->subprogs);
800 }
801
802 static void bpf_program__exit(struct bpf_program *prog)
803 {
804 if (!prog)
805 return;
806
807 bpf_program__unload(prog);
808 zfree(&prog->name);
809 zfree(&prog->sec_name);
810 zfree(&prog->insns);
811 zfree(&prog->reloc_desc);
812
813 prog->nr_reloc = 0;
814 prog->insns_cnt = 0;
815 prog->sec_idx = -1;
816 }
817
818 static bool insn_is_subprog_call(const struct bpf_insn *insn)
819 {
820 return BPF_CLASS(insn->code) == BPF_JMP &&
821 BPF_OP(insn->code) == BPF_CALL &&
822 BPF_SRC(insn->code) == BPF_K &&
823 insn->src_reg == BPF_PSEUDO_CALL &&
824 insn->dst_reg == 0 &&
825 insn->off == 0;
826 }
827
828 static bool is_call_insn(const struct bpf_insn *insn)
829 {
830 return insn->code == (BPF_JMP | BPF_CALL);
831 }
832
833 static bool insn_is_pseudo_func(struct bpf_insn *insn)
834 {
835 return is_ldimm64_insn(insn) && insn->src_reg == BPF_PSEUDO_FUNC;
836 }
837
838 static int
839 bpf_object__init_prog(struct bpf_object *obj, struct bpf_program *prog,
840 const char *name, size_t sec_idx, const char *sec_name,
841 size_t sec_off, void *insn_data, size_t insn_data_sz)
842 {
843 if (insn_data_sz == 0 || insn_data_sz % BPF_INSN_SZ || sec_off % BPF_INSN_SZ) {
844 pr_warn("sec '%s': corrupted program '%s', offset %zu, size %zu\n",
845 sec_name, name, sec_off, insn_data_sz);
846 return -EINVAL;
847 }
848
849 memset(prog, 0, sizeof(*prog));
850 prog->obj = obj;
851
852 prog->sec_idx = sec_idx;
853 prog->sec_insn_off = sec_off / BPF_INSN_SZ;
854 prog->sec_insn_cnt = insn_data_sz / BPF_INSN_SZ;
855 /* insns_cnt can later be increased by appending used subprograms */
856 prog->insns_cnt = prog->sec_insn_cnt;
857
858 prog->type = BPF_PROG_TYPE_UNSPEC;
859 prog->fd = -1;
860 prog->exception_cb_idx = -1;
861
862 /* libbpf's convention for SEC("?abc...") is that it's just like
863 * SEC("abc...") but the corresponding bpf_program starts out with
864 * autoload set to false.
865 */
866 if (sec_name[0] == '?') {
867 prog->autoload = false;
868 /* from now on forget there was ? in section name */
869 sec_name++;
870 } else {
871 prog->autoload = true;
872 }
873
874 prog->autoattach = true;
875
876 /* inherit object's log_level */
877 prog->log_level = obj->log_level;
878
879 prog->sec_name = strdup(sec_name);
880 if (!prog->sec_name)
881 goto errout;
882
883 prog->name = strdup(name);
884 if (!prog->name)
885 goto errout;
886
887 prog->insns = malloc(insn_data_sz);
888 if (!prog->insns)
889 goto errout;
890 memcpy(prog->insns, insn_data, insn_data_sz);
891
892 return 0;
893 errout:
894 pr_warn("sec '%s': failed to allocate memory for prog '%s'\n", sec_name, name);
895 bpf_program__exit(prog);
896 return -ENOMEM;
897 }
898
899 static int
900 bpf_object__add_programs(struct bpf_object *obj, Elf_Data *sec_data,
901 const char *sec_name, int sec_idx)
902 {
903 Elf_Data *symbols = obj->efile.symbols;
904 struct bpf_program *prog, *progs;
905 void *data = sec_data->d_buf;
906 size_t sec_sz = sec_data->d_size, sec_off, prog_sz, nr_syms;
907 int nr_progs, err, i;
908 const char *name;
909 Elf64_Sym *sym;
910
911 progs = obj->programs;
912 nr_progs = obj->nr_programs;
913 nr_syms = symbols->d_size / sizeof(Elf64_Sym);
914
915 for (i = 0; i < nr_syms; i++) {
916 sym = elf_sym_by_idx(obj, i);
917
918 if (sym->st_shndx != sec_idx)
919 continue;
920 if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
921 continue;
922
923 prog_sz = sym->st_size;
924 sec_off = sym->st_value;
925
926 name = elf_sym_str(obj, sym->st_name);
927 if (!name) {
928 pr_warn("sec '%s': failed to get symbol name for offset %zu\n",
929 sec_name, sec_off);
930 return -LIBBPF_ERRNO__FORMAT;
931 }
932
933 if (sec_off + prog_sz > sec_sz || sec_off + prog_sz < sec_off) {
934 pr_warn("sec '%s': program at offset %zu crosses section boundary\n",
935 sec_name, sec_off);
936 return -LIBBPF_ERRNO__FORMAT;
937 }
938
939 if (sec_idx != obj->efile.text_shndx && ELF64_ST_BIND(sym->st_info) == STB_LOCAL) {
940 pr_warn("sec '%s': program '%s' is static and not supported\n", sec_name, name);
941 return -ENOTSUP;
942 }
943
944 pr_debug("sec '%s': found program '%s' at insn offset %zu (%zu bytes), code size %zu insns (%zu bytes)\n",
945 sec_name, name, sec_off / BPF_INSN_SZ, sec_off, prog_sz / BPF_INSN_SZ, prog_sz);
946
947 progs = libbpf_reallocarray(progs, nr_progs + 1, sizeof(*progs));
948 if (!progs) {
949 /*
950 * In this case the original obj->programs
951 * is still valid, so don't need special treat for
952 * bpf_close_object().
953 */
954 pr_warn("sec '%s': failed to alloc memory for new program '%s'\n",
955 sec_name, name);
956 return -ENOMEM;
957 }
958 obj->programs = progs;
959
960 prog = &progs[nr_progs];
961
962 err = bpf_object__init_prog(obj, prog, name, sec_idx, sec_name,
963 sec_off, data + sec_off, prog_sz);
964 if (err)
965 return err;
966
967 if (ELF64_ST_BIND(sym->st_info) != STB_LOCAL)
968 prog->sym_global = true;
969
970 /* if function is a global/weak symbol, but has restricted
971 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF FUNC
972 * as static to enable more permissive BPF verification mode
973 * with more outside context available to BPF verifier
974 */
975 if (prog->sym_global && (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
976 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL))
977 prog->mark_btf_static = true;
978
979 nr_progs++;
980 obj->nr_programs = nr_progs;
981 }
982
983 return 0;
984 }
985
986 static void bpf_object_bswap_progs(struct bpf_object *obj)
987 {
988 struct bpf_program *prog = obj->programs;
989 struct bpf_insn *insn;
990 int p, i;
991
992 for (p = 0; p < obj->nr_programs; p++, prog++) {
993 insn = prog->insns;
994 for (i = 0; i < prog->insns_cnt; i++, insn++)
995 bpf_insn_bswap(insn);
996 }
997 pr_debug("converted %zu BPF programs to native byte order\n", obj->nr_programs);
998 }
999
1000 static const struct btf_member *
1001 find_member_by_offset(const struct btf_type *t, __u32 bit_offset)
1002 {
1003 struct btf_member *m;
1004 int i;
1005
1006 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
1007 if (btf_member_bit_offset(t, i) == bit_offset)
1008 return m;
1009 }
1010
1011 return NULL;
1012 }
1013
1014 static const struct btf_member *
1015 find_member_by_name(const struct btf *btf, const struct btf_type *t,
1016 const char *name)
1017 {
1018 struct btf_member *m;
1019 int i;
1020
1021 for (i = 0, m = btf_members(t); i < btf_vlen(t); i++, m++) {
1022 if (!strcmp(btf__name_by_offset(btf, m->name_off), name))
1023 return m;
1024 }
1025
1026 return NULL;
1027 }
1028
1029 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
1030 __u16 kind, struct btf **res_btf,
1031 struct module_btf **res_mod_btf);
1032
1033 #define STRUCT_OPS_VALUE_PREFIX "bpf_struct_ops_"
1034 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
1035 const char *name, __u32 kind);
1036
1037 static int
1038 find_struct_ops_kern_types(struct bpf_object *obj, const char *tname_raw,
1039 struct module_btf **mod_btf,
1040 const struct btf_type **type, __u32 *type_id,
1041 const struct btf_type **vtype, __u32 *vtype_id,
1042 const struct btf_member **data_member)
1043 {
1044 const struct btf_type *kern_type, *kern_vtype;
1045 const struct btf_member *kern_data_member;
1046 struct btf *btf = NULL;
1047 __s32 kern_vtype_id, kern_type_id;
1048 char tname[192], stname[256];
1049 __u32 i;
1050
1051 snprintf(tname, sizeof(tname), "%.*s",
1052 (int)bpf_core_essential_name_len(tname_raw), tname_raw);
1053
1054 snprintf(stname, sizeof(stname), "%s%s", STRUCT_OPS_VALUE_PREFIX, tname);
1055
1056 /* Look for the corresponding "map_value" type that will be used
1057 * in map_update(BPF_MAP_TYPE_STRUCT_OPS) first, figure out the btf
1058 * and the mod_btf.
1059 * For example, find "struct bpf_struct_ops_tcp_congestion_ops".
1060 */
1061 kern_vtype_id = find_ksym_btf_id(obj, stname, BTF_KIND_STRUCT, &btf, mod_btf);
1062 if (kern_vtype_id < 0) {
1063 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", stname);
1064 return kern_vtype_id;
1065 }
1066 kern_vtype = btf__type_by_id(btf, kern_vtype_id);
1067
1068 kern_type_id = btf__find_by_name_kind(btf, tname, BTF_KIND_STRUCT);
1069 if (kern_type_id < 0) {
1070 pr_warn("struct_ops init_kern: struct %s is not found in kernel BTF\n", tname);
1071 return kern_type_id;
1072 }
1073 kern_type = btf__type_by_id(btf, kern_type_id);
1074
1075 /* Find "struct tcp_congestion_ops" from
1076 * struct bpf_struct_ops_tcp_congestion_ops {
1077 * [ ... ]
1078 * struct tcp_congestion_ops data;
1079 * }
1080 */
1081 kern_data_member = btf_members(kern_vtype);
1082 for (i = 0; i < btf_vlen(kern_vtype); i++, kern_data_member++) {
1083 if (kern_data_member->type == kern_type_id)
1084 break;
1085 }
1086 if (i == btf_vlen(kern_vtype)) {
1087 pr_warn("struct_ops init_kern: struct %s data is not found in struct %s\n",
1088 tname, stname);
1089 return -EINVAL;
1090 }
1091
1092 *type = kern_type;
1093 *type_id = kern_type_id;
1094 *vtype = kern_vtype;
1095 *vtype_id = kern_vtype_id;
1096 *data_member = kern_data_member;
1097
1098 return 0;
1099 }
1100
1101 static bool bpf_map__is_struct_ops(const struct bpf_map *map)
1102 {
1103 return map->def.type == BPF_MAP_TYPE_STRUCT_OPS;
1104 }
1105
1106 static bool is_valid_st_ops_program(struct bpf_object *obj,
1107 const struct bpf_program *prog)
1108 {
1109 int i;
1110
1111 for (i = 0; i < obj->nr_programs; i++) {
1112 if (&obj->programs[i] == prog)
1113 return prog->type == BPF_PROG_TYPE_STRUCT_OPS;
1114 }
1115
1116 return false;
1117 }
1118
1119 /* For each struct_ops program P, referenced from some struct_ops map M,
1120 * enable P.autoload if there are Ms for which M.autocreate is true,
1121 * disable P.autoload if for all Ms M.autocreate is false.
1122 * Don't change P.autoload for programs that are not referenced from any maps.
1123 */
1124 static int bpf_object_adjust_struct_ops_autoload(struct bpf_object *obj)
1125 {
1126 struct bpf_program *prog, *slot_prog;
1127 struct bpf_map *map;
1128 int i, j, k, vlen;
1129
1130 for (i = 0; i < obj->nr_programs; ++i) {
1131 int should_load = false;
1132 int use_cnt = 0;
1133
1134 prog = &obj->programs[i];
1135 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS)
1136 continue;
1137
1138 for (j = 0; j < obj->nr_maps; ++j) {
1139 const struct btf_type *type;
1140
1141 map = &obj->maps[j];
1142 if (!bpf_map__is_struct_ops(map))
1143 continue;
1144
1145 type = btf__type_by_id(obj->btf, map->st_ops->type_id);
1146 vlen = btf_vlen(type);
1147 for (k = 0; k < vlen; ++k) {
1148 slot_prog = map->st_ops->progs[k];
1149 if (prog != slot_prog)
1150 continue;
1151
1152 use_cnt++;
1153 if (map->autocreate)
1154 should_load = true;
1155 }
1156 }
1157 if (use_cnt)
1158 prog->autoload = should_load;
1159 }
1160
1161 return 0;
1162 }
1163
1164 /* Init the map's fields that depend on kern_btf */
1165 static int bpf_map__init_kern_struct_ops(struct bpf_map *map)
1166 {
1167 const struct btf_member *member, *kern_member, *kern_data_member;
1168 const struct btf_type *type, *kern_type, *kern_vtype;
1169 __u32 i, kern_type_id, kern_vtype_id, kern_data_off;
1170 struct bpf_object *obj = map->obj;
1171 const struct btf *btf = obj->btf;
1172 struct bpf_struct_ops *st_ops;
1173 const struct btf *kern_btf;
1174 struct module_btf *mod_btf = NULL;
1175 void *data, *kern_data;
1176 const char *tname;
1177 int err;
1178
1179 st_ops = map->st_ops;
1180 type = btf__type_by_id(btf, st_ops->type_id);
1181 tname = btf__name_by_offset(btf, type->name_off);
1182 err = find_struct_ops_kern_types(obj, tname, &mod_btf,
1183 &kern_type, &kern_type_id,
1184 &kern_vtype, &kern_vtype_id,
1185 &kern_data_member);
1186 if (err)
1187 return err;
1188
1189 kern_btf = mod_btf ? mod_btf->btf : obj->btf_vmlinux;
1190
1191 pr_debug("struct_ops init_kern %s: type_id:%u kern_type_id:%u kern_vtype_id:%u\n",
1192 map->name, st_ops->type_id, kern_type_id, kern_vtype_id);
1193
1194 map->mod_btf_fd = mod_btf ? mod_btf->fd : -1;
1195 map->def.value_size = kern_vtype->size;
1196 map->btf_vmlinux_value_type_id = kern_vtype_id;
1197
1198 st_ops->kern_vdata = calloc(1, kern_vtype->size);
1199 if (!st_ops->kern_vdata)
1200 return -ENOMEM;
1201
1202 data = st_ops->data;
1203 kern_data_off = kern_data_member->offset / 8;
1204 kern_data = st_ops->kern_vdata + kern_data_off;
1205
1206 member = btf_members(type);
1207 for (i = 0; i < btf_vlen(type); i++, member++) {
1208 const struct btf_type *mtype, *kern_mtype;
1209 __u32 mtype_id, kern_mtype_id;
1210 void *mdata, *kern_mdata;
1211 struct bpf_program *prog;
1212 __s64 msize, kern_msize;
1213 __u32 moff, kern_moff;
1214 __u32 kern_member_idx;
1215 const char *mname;
1216
1217 mname = btf__name_by_offset(btf, member->name_off);
1218 moff = member->offset / 8;
1219 mdata = data + moff;
1220 msize = btf__resolve_size(btf, member->type);
1221 if (msize < 0) {
1222 pr_warn("struct_ops init_kern %s: failed to resolve the size of member %s\n",
1223 map->name, mname);
1224 return msize;
1225 }
1226
1227 kern_member = find_member_by_name(kern_btf, kern_type, mname);
1228 if (!kern_member) {
1229 if (!libbpf_is_mem_zeroed(mdata, msize)) {
1230 pr_warn("struct_ops init_kern %s: Cannot find member %s in kernel BTF\n",
1231 map->name, mname);
1232 return -ENOTSUP;
1233 }
1234
1235 if (st_ops->progs[i]) {
1236 /* If we had declaratively set struct_ops callback, we need to
1237 * force its autoload to false, because it doesn't have
1238 * a chance of succeeding from POV of the current struct_ops map.
1239 * If this program is still referenced somewhere else, though,
1240 * then bpf_object_adjust_struct_ops_autoload() will update its
1241 * autoload accordingly.
1242 */
1243 st_ops->progs[i]->autoload = false;
1244 st_ops->progs[i] = NULL;
1245 }
1246
1247 /* Skip all-zero/NULL fields if they are not present in the kernel BTF */
1248 pr_info("struct_ops %s: member %s not found in kernel, skipping it as it's set to zero\n",
1249 map->name, mname);
1250 continue;
1251 }
1252
1253 kern_member_idx = kern_member - btf_members(kern_type);
1254 if (btf_member_bitfield_size(type, i) ||
1255 btf_member_bitfield_size(kern_type, kern_member_idx)) {
1256 pr_warn("struct_ops init_kern %s: bitfield %s is not supported\n",
1257 map->name, mname);
1258 return -ENOTSUP;
1259 }
1260
1261 kern_moff = kern_member->offset / 8;
1262 kern_mdata = kern_data + kern_moff;
1263
1264 mtype = skip_mods_and_typedefs(btf, member->type, &mtype_id);
1265 kern_mtype = skip_mods_and_typedefs(kern_btf, kern_member->type,
1266 &kern_mtype_id);
1267 if (BTF_INFO_KIND(mtype->info) !=
1268 BTF_INFO_KIND(kern_mtype->info)) {
1269 pr_warn("struct_ops init_kern %s: Unmatched member type %s %u != %u(kernel)\n",
1270 map->name, mname, BTF_INFO_KIND(mtype->info),
1271 BTF_INFO_KIND(kern_mtype->info));
1272 return -ENOTSUP;
1273 }
1274
1275 if (btf_is_ptr(mtype)) {
1276 prog = *(void **)mdata;
1277 /* just like for !kern_member case above, reset declaratively
1278 * set (at compile time) program's autload to false,
1279 * if user replaced it with another program or NULL
1280 */
1281 if (st_ops->progs[i] && st_ops->progs[i] != prog)
1282 st_ops->progs[i]->autoload = false;
1283
1284 /* Update the value from the shadow type */
1285 st_ops->progs[i] = prog;
1286 if (!prog)
1287 continue;
1288
1289 if (!is_valid_st_ops_program(obj, prog)) {
1290 pr_warn("struct_ops init_kern %s: member %s is not a struct_ops program\n",
1291 map->name, mname);
1292 return -ENOTSUP;
1293 }
1294
1295 kern_mtype = skip_mods_and_typedefs(kern_btf,
1296 kern_mtype->type,
1297 &kern_mtype_id);
1298
1299 /* mtype->type must be a func_proto which was
1300 * guaranteed in bpf_object__collect_st_ops_relos(),
1301 * so only check kern_mtype for func_proto here.
1302 */
1303 if (!btf_is_func_proto(kern_mtype)) {
1304 pr_warn("struct_ops init_kern %s: kernel member %s is not a func ptr\n",
1305 map->name, mname);
1306 return -ENOTSUP;
1307 }
1308
1309 if (mod_btf)
1310 prog->attach_btf_obj_fd = mod_btf->fd;
1311
1312 /* if we haven't yet processed this BPF program, record proper
1313 * attach_btf_id and member_idx
1314 */
1315 if (!prog->attach_btf_id) {
1316 prog->attach_btf_id = kern_type_id;
1317 prog->expected_attach_type = kern_member_idx;
1318 }
1319
1320 /* struct_ops BPF prog can be re-used between multiple
1321 * .struct_ops & .struct_ops.link as long as it's the
1322 * same struct_ops struct definition and the same
1323 * function pointer field
1324 */
1325 if (prog->attach_btf_id != kern_type_id) {
1326 pr_warn("struct_ops init_kern %s func ptr %s: invalid reuse of prog %s in sec %s with type %u: attach_btf_id %u != kern_type_id %u\n",
1327 map->name, mname, prog->name, prog->sec_name, prog->type,
1328 prog->attach_btf_id, kern_type_id);
1329 return -EINVAL;
1330 }
1331 if (prog->expected_attach_type != kern_member_idx) {
1332 pr_warn("struct_ops init_kern %s func ptr %s: invalid reuse of prog %s in sec %s with type %u: expected_attach_type %u != kern_member_idx %u\n",
1333 map->name, mname, prog->name, prog->sec_name, prog->type,
1334 prog->expected_attach_type, kern_member_idx);
1335 return -EINVAL;
1336 }
1337
1338 st_ops->kern_func_off[i] = kern_data_off + kern_moff;
1339
1340 pr_debug("struct_ops init_kern %s: func ptr %s is set to prog %s from data(+%u) to kern_data(+%u)\n",
1341 map->name, mname, prog->name, moff,
1342 kern_moff);
1343
1344 continue;
1345 }
1346
1347 kern_msize = btf__resolve_size(kern_btf, kern_mtype_id);
1348 if (kern_msize < 0 || msize != kern_msize) {
1349 pr_warn("struct_ops init_kern %s: Error in size of member %s: %zd != %zd(kernel)\n",
1350 map->name, mname, (ssize_t)msize,
1351 (ssize_t)kern_msize);
1352 return -ENOTSUP;
1353 }
1354
1355 pr_debug("struct_ops init_kern %s: copy %s %u bytes from data(+%u) to kern_data(+%u)\n",
1356 map->name, mname, (unsigned int)msize,
1357 moff, kern_moff);
1358 memcpy(kern_mdata, mdata, msize);
1359 }
1360
1361 return 0;
1362 }
1363
1364 static int bpf_object__init_kern_struct_ops_maps(struct bpf_object *obj)
1365 {
1366 struct bpf_map *map;
1367 size_t i;
1368 int err;
1369
1370 for (i = 0; i < obj->nr_maps; i++) {
1371 map = &obj->maps[i];
1372
1373 if (!bpf_map__is_struct_ops(map))
1374 continue;
1375
1376 if (!map->autocreate)
1377 continue;
1378
1379 err = bpf_map__init_kern_struct_ops(map);
1380 if (err)
1381 return err;
1382 }
1383
1384 return 0;
1385 }
1386
1387 static int init_struct_ops_maps(struct bpf_object *obj, const char *sec_name,
1388 int shndx, Elf_Data *data)
1389 {
1390 const struct btf_type *type, *datasec;
1391 const struct btf_var_secinfo *vsi;
1392 struct bpf_struct_ops *st_ops;
1393 const char *tname, *var_name;
1394 __s32 type_id, datasec_id;
1395 const struct btf *btf;
1396 struct bpf_map *map;
1397 __u32 i;
1398
1399 if (shndx == -1)
1400 return 0;
1401
1402 btf = obj->btf;
1403 datasec_id = btf__find_by_name_kind(btf, sec_name,
1404 BTF_KIND_DATASEC);
1405 if (datasec_id < 0) {
1406 pr_warn("struct_ops init: DATASEC %s not found\n",
1407 sec_name);
1408 return -EINVAL;
1409 }
1410
1411 datasec = btf__type_by_id(btf, datasec_id);
1412 vsi = btf_var_secinfos(datasec);
1413 for (i = 0; i < btf_vlen(datasec); i++, vsi++) {
1414 type = btf__type_by_id(obj->btf, vsi->type);
1415 var_name = btf__name_by_offset(obj->btf, type->name_off);
1416
1417 type_id = btf__resolve_type(obj->btf, vsi->type);
1418 if (type_id < 0) {
1419 pr_warn("struct_ops init: Cannot resolve var type_id %u in DATASEC %s\n",
1420 vsi->type, sec_name);
1421 return -EINVAL;
1422 }
1423
1424 type = btf__type_by_id(obj->btf, type_id);
1425 tname = btf__name_by_offset(obj->btf, type->name_off);
1426 if (!tname[0]) {
1427 pr_warn("struct_ops init: anonymous type is not supported\n");
1428 return -ENOTSUP;
1429 }
1430 if (!btf_is_struct(type)) {
1431 pr_warn("struct_ops init: %s is not a struct\n", tname);
1432 return -EINVAL;
1433 }
1434
1435 map = bpf_object__add_map(obj);
1436 if (IS_ERR(map))
1437 return PTR_ERR(map);
1438
1439 map->sec_idx = shndx;
1440 map->sec_offset = vsi->offset;
1441 map->name = strdup(var_name);
1442 if (!map->name)
1443 return -ENOMEM;
1444 map->btf_value_type_id = type_id;
1445
1446 /* Follow same convention as for programs autoload:
1447 * SEC("?.struct_ops") means map is not created by default.
1448 */
1449 if (sec_name[0] == '?') {
1450 map->autocreate = false;
1451 /* from now on forget there was ? in section name */
1452 sec_name++;
1453 }
1454
1455 map->def.type = BPF_MAP_TYPE_STRUCT_OPS;
1456 map->def.key_size = sizeof(int);
1457 map->def.value_size = type->size;
1458 map->def.max_entries = 1;
1459 map->def.map_flags = strcmp(sec_name, STRUCT_OPS_LINK_SEC) == 0 ? BPF_F_LINK : 0;
1460 map->autoattach = true;
1461
1462 map->st_ops = calloc(1, sizeof(*map->st_ops));
1463 if (!map->st_ops)
1464 return -ENOMEM;
1465 st_ops = map->st_ops;
1466 st_ops->data = malloc(type->size);
1467 st_ops->progs = calloc(btf_vlen(type), sizeof(*st_ops->progs));
1468 st_ops->kern_func_off = malloc(btf_vlen(type) *
1469 sizeof(*st_ops->kern_func_off));
1470 if (!st_ops->data || !st_ops->progs || !st_ops->kern_func_off)
1471 return -ENOMEM;
1472
1473 if (vsi->offset + type->size > data->d_size) {
1474 pr_warn("struct_ops init: var %s is beyond the end of DATASEC %s\n",
1475 var_name, sec_name);
1476 return -EINVAL;
1477 }
1478
1479 memcpy(st_ops->data,
1480 data->d_buf + vsi->offset,
1481 type->size);
1482 st_ops->type_id = type_id;
1483
1484 pr_debug("struct_ops init: struct %s(type_id=%u) %s found at offset %u\n",
1485 tname, type_id, var_name, vsi->offset);
1486 }
1487
1488 return 0;
1489 }
1490
1491 static int bpf_object_init_struct_ops(struct bpf_object *obj)
1492 {
1493 const char *sec_name;
1494 int sec_idx, err;
1495
1496 for (sec_idx = 0; sec_idx < obj->efile.sec_cnt; ++sec_idx) {
1497 struct elf_sec_desc *desc = &obj->efile.secs[sec_idx];
1498
1499 if (desc->sec_type != SEC_ST_OPS)
1500 continue;
1501
1502 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
1503 if (!sec_name)
1504 return -LIBBPF_ERRNO__FORMAT;
1505
1506 err = init_struct_ops_maps(obj, sec_name, sec_idx, desc->data);
1507 if (err)
1508 return err;
1509 }
1510
1511 return 0;
1512 }
1513
1514 static struct bpf_object *bpf_object__new(const char *path,
1515 const void *obj_buf,
1516 size_t obj_buf_sz,
1517 const char *obj_name)
1518 {
1519 struct bpf_object *obj;
1520 char *end;
1521
1522 obj = calloc(1, sizeof(struct bpf_object) + strlen(path) + 1);
1523 if (!obj) {
1524 pr_warn("alloc memory failed for %s\n", path);
1525 return ERR_PTR(-ENOMEM);
1526 }
1527
1528 strcpy(obj->path, path);
1529 if (obj_name) {
1530 libbpf_strlcpy(obj->name, obj_name, sizeof(obj->name));
1531 } else {
1532 /* Using basename() GNU version which doesn't modify arg. */
1533 libbpf_strlcpy(obj->name, basename((void *)path), sizeof(obj->name));
1534 end = strchr(obj->name, '.');
1535 if (end)
1536 *end = 0;
1537 }
1538
1539 obj->efile.fd = -1;
1540 /*
1541 * Caller of this function should also call
1542 * bpf_object__elf_finish() after data collection to return
1543 * obj_buf to user. If not, we should duplicate the buffer to
1544 * avoid user freeing them before elf finish.
1545 */
1546 obj->efile.obj_buf = obj_buf;
1547 obj->efile.obj_buf_sz = obj_buf_sz;
1548 obj->efile.btf_maps_shndx = -1;
1549 obj->kconfig_map_idx = -1;
1550 obj->arena_map_idx = -1;
1551
1552 obj->kern_version = get_kernel_version();
1553 obj->state = OBJ_OPEN;
1554
1555 return obj;
1556 }
1557
1558 static void bpf_object__elf_finish(struct bpf_object *obj)
1559 {
1560 if (!obj->efile.elf)
1561 return;
1562
1563 elf_end(obj->efile.elf);
1564 obj->efile.elf = NULL;
1565 obj->efile.ehdr = NULL;
1566 obj->efile.symbols = NULL;
1567 obj->efile.arena_data = NULL;
1568
1569 zfree(&obj->efile.secs);
1570 obj->efile.sec_cnt = 0;
1571 zclose(obj->efile.fd);
1572 obj->efile.obj_buf = NULL;
1573 obj->efile.obj_buf_sz = 0;
1574 }
1575
1576 static int bpf_object__elf_init(struct bpf_object *obj)
1577 {
1578 Elf64_Ehdr *ehdr;
1579 int err = 0;
1580 Elf *elf;
1581
1582 if (obj->efile.elf) {
1583 pr_warn("elf: init internal error\n");
1584 return -LIBBPF_ERRNO__LIBELF;
1585 }
1586
1587 if (obj->efile.obj_buf_sz > 0) {
1588 /* obj_buf should have been validated by bpf_object__open_mem(). */
1589 elf = elf_memory((char *)obj->efile.obj_buf, obj->efile.obj_buf_sz);
1590 } else {
1591 obj->efile.fd = open(obj->path, O_RDONLY | O_CLOEXEC);
1592 if (obj->efile.fd < 0) {
1593 err = -errno;
1594 pr_warn("elf: failed to open %s: %s\n", obj->path, errstr(err));
1595 return err;
1596 }
1597
1598 elf = elf_begin(obj->efile.fd, ELF_C_READ_MMAP, NULL);
1599 }
1600
1601 if (!elf) {
1602 pr_warn("elf: failed to open %s as ELF file: %s\n", obj->path, elf_errmsg(-1));
1603 err = -LIBBPF_ERRNO__LIBELF;
1604 goto errout;
1605 }
1606
1607 obj->efile.elf = elf;
1608
1609 if (elf_kind(elf) != ELF_K_ELF) {
1610 err = -LIBBPF_ERRNO__FORMAT;
1611 pr_warn("elf: '%s' is not a proper ELF object\n", obj->path);
1612 goto errout;
1613 }
1614
1615 if (gelf_getclass(elf) != ELFCLASS64) {
1616 err = -LIBBPF_ERRNO__FORMAT;
1617 pr_warn("elf: '%s' is not a 64-bit ELF object\n", obj->path);
1618 goto errout;
1619 }
1620
1621 obj->efile.ehdr = ehdr = elf64_getehdr(elf);
1622 if (!obj->efile.ehdr) {
1623 pr_warn("elf: failed to get ELF header from %s: %s\n", obj->path, elf_errmsg(-1));
1624 err = -LIBBPF_ERRNO__FORMAT;
1625 goto errout;
1626 }
1627
1628 /* Validate ELF object endianness... */
1629 if (ehdr->e_ident[EI_DATA] != ELFDATA2LSB &&
1630 ehdr->e_ident[EI_DATA] != ELFDATA2MSB) {
1631 err = -LIBBPF_ERRNO__ENDIAN;
1632 pr_warn("elf: '%s' has unknown byte order\n", obj->path);
1633 goto errout;
1634 }
1635 /* and save after bpf_object_open() frees ELF data */
1636 obj->byteorder = ehdr->e_ident[EI_DATA];
1637
1638 if (elf_getshdrstrndx(elf, &obj->efile.shstrndx)) {
1639 pr_warn("elf: failed to get section names section index for %s: %s\n",
1640 obj->path, elf_errmsg(-1));
1641 err = -LIBBPF_ERRNO__FORMAT;
1642 goto errout;
1643 }
1644
1645 /* ELF is corrupted/truncated, avoid calling elf_strptr. */
1646 if (!elf_rawdata(elf_getscn(elf, obj->efile.shstrndx), NULL)) {
1647 pr_warn("elf: failed to get section names strings from %s: %s\n",
1648 obj->path, elf_errmsg(-1));
1649 err = -LIBBPF_ERRNO__FORMAT;
1650 goto errout;
1651 }
1652
1653 /* Old LLVM set e_machine to EM_NONE */
1654 if (ehdr->e_type != ET_REL || (ehdr->e_machine && ehdr->e_machine != EM_BPF)) {
1655 pr_warn("elf: %s is not a valid eBPF object file\n", obj->path);
1656 err = -LIBBPF_ERRNO__FORMAT;
1657 goto errout;
1658 }
1659
1660 return 0;
1661 errout:
1662 bpf_object__elf_finish(obj);
1663 return err;
1664 }
1665
1666 static bool is_native_endianness(struct bpf_object *obj)
1667 {
1668 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
1669 return obj->byteorder == ELFDATA2LSB;
1670 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
1671 return obj->byteorder == ELFDATA2MSB;
1672 #else
1673 # error "Unrecognized __BYTE_ORDER__"
1674 #endif
1675 }
1676
1677 static int
1678 bpf_object__init_license(struct bpf_object *obj, void *data, size_t size)
1679 {
1680 if (!data) {
1681 pr_warn("invalid license section in %s\n", obj->path);
1682 return -LIBBPF_ERRNO__FORMAT;
1683 }
1684 /* libbpf_strlcpy() only copies first N - 1 bytes, so size + 1 won't
1685 * go over allowed ELF data section buffer
1686 */
1687 libbpf_strlcpy(obj->license, data, min(size + 1, sizeof(obj->license)));
1688 pr_debug("license of %s is %s\n", obj->path, obj->license);
1689 return 0;
1690 }
1691
1692 static int
1693 bpf_object__init_kversion(struct bpf_object *obj, void *data, size_t size)
1694 {
1695 __u32 kver;
1696
1697 if (!data || size != sizeof(kver)) {
1698 pr_warn("invalid kver section in %s\n", obj->path);
1699 return -LIBBPF_ERRNO__FORMAT;
1700 }
1701 memcpy(&kver, data, sizeof(kver));
1702 obj->kern_version = kver;
1703 pr_debug("kernel version of %s is %x\n", obj->path, obj->kern_version);
1704 return 0;
1705 }
1706
1707 static bool bpf_map_type__is_map_in_map(enum bpf_map_type type)
1708 {
1709 if (type == BPF_MAP_TYPE_ARRAY_OF_MAPS ||
1710 type == BPF_MAP_TYPE_HASH_OF_MAPS)
1711 return true;
1712 return false;
1713 }
1714
1715 static int find_elf_sec_sz(const struct bpf_object *obj, const char *name, __u32 *size)
1716 {
1717 Elf_Data *data;
1718 Elf_Scn *scn;
1719
1720 if (!name)
1721 return -EINVAL;
1722
1723 scn = elf_sec_by_name(obj, name);
1724 data = elf_sec_data(obj, scn);
1725 if (data) {
1726 *size = data->d_size;
1727 return 0; /* found it */
1728 }
1729
1730 return -ENOENT;
1731 }
1732
1733 static Elf64_Sym *find_elf_var_sym(const struct bpf_object *obj, const char *name)
1734 {
1735 Elf_Data *symbols = obj->efile.symbols;
1736 const char *sname;
1737 size_t si;
1738
1739 for (si = 0; si < symbols->d_size / sizeof(Elf64_Sym); si++) {
1740 Elf64_Sym *sym = elf_sym_by_idx(obj, si);
1741
1742 if (ELF64_ST_TYPE(sym->st_info) != STT_OBJECT)
1743 continue;
1744
1745 if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL &&
1746 ELF64_ST_BIND(sym->st_info) != STB_WEAK)
1747 continue;
1748
1749 sname = elf_sym_str(obj, sym->st_name);
1750 if (!sname) {
1751 pr_warn("failed to get sym name string for var %s\n", name);
1752 return ERR_PTR(-EIO);
1753 }
1754 if (strcmp(name, sname) == 0)
1755 return sym;
1756 }
1757
1758 return ERR_PTR(-ENOENT);
1759 }
1760
1761 #ifndef MFD_CLOEXEC
1762 #define MFD_CLOEXEC 0x0001U
1763 #endif
1764 #ifndef MFD_NOEXEC_SEAL
1765 #define MFD_NOEXEC_SEAL 0x0008U
1766 #endif
1767
1768 static int create_placeholder_fd(void)
1769 {
1770 unsigned int flags = MFD_CLOEXEC | MFD_NOEXEC_SEAL;
1771 const char *name = "libbpf-placeholder-fd";
1772 int fd;
1773
1774 fd = ensure_good_fd(sys_memfd_create(name, flags));
1775 if (fd >= 0)
1776 return fd;
1777 else if (errno != EINVAL)
1778 return -errno;
1779
1780 /* Possibly running on kernel without MFD_NOEXEC_SEAL */
1781 fd = ensure_good_fd(sys_memfd_create(name, flags & ~MFD_NOEXEC_SEAL));
1782 if (fd < 0)
1783 return -errno;
1784 return fd;
1785 }
1786
1787 static struct bpf_map *bpf_object__add_map(struct bpf_object *obj)
1788 {
1789 struct bpf_map *map;
1790 int err;
1791
1792 err = libbpf_ensure_mem((void **)&obj->maps, &obj->maps_cap,
1793 sizeof(*obj->maps), obj->nr_maps + 1);
1794 if (err)
1795 return ERR_PTR(err);
1796
1797 map = &obj->maps[obj->nr_maps++];
1798 map->obj = obj;
1799 /* Preallocate map FD without actually creating BPF map just yet.
1800 * These map FD "placeholders" will be reused later without changing
1801 * FD value when map is actually created in the kernel.
1802 *
1803 * This is useful to be able to perform BPF program relocations
1804 * without having to create BPF maps before that step. This allows us
1805 * to finalize and load BTF very late in BPF object's loading phase,
1806 * right before BPF maps have to be created and BPF programs have to
1807 * be loaded. By having these map FD placeholders we can perform all
1808 * the sanitizations, relocations, and any other adjustments before we
1809 * start creating actual BPF kernel objects (BTF, maps, progs).
1810 */
1811 map->fd = create_placeholder_fd();
1812 if (map->fd < 0)
1813 return ERR_PTR(map->fd);
1814 map->inner_map_fd = -1;
1815 map->autocreate = true;
1816
1817 return map;
1818 }
1819
1820 static size_t array_map_mmap_sz(unsigned int value_sz, unsigned int max_entries)
1821 {
1822 const long page_sz = sysconf(_SC_PAGE_SIZE);
1823 size_t map_sz;
1824
1825 map_sz = (size_t)roundup(value_sz, 8) * max_entries;
1826 map_sz = roundup(map_sz, page_sz);
1827 return map_sz;
1828 }
1829
1830 static size_t bpf_map_mmap_sz(const struct bpf_map *map)
1831 {
1832 const long page_sz = sysconf(_SC_PAGE_SIZE);
1833
1834 switch (map->def.type) {
1835 case BPF_MAP_TYPE_ARRAY:
1836 return array_map_mmap_sz(map->def.value_size, map->def.max_entries);
1837 case BPF_MAP_TYPE_ARENA:
1838 return page_sz * map->def.max_entries;
1839 default:
1840 return 0; /* not supported */
1841 }
1842 }
1843
1844 static int bpf_map_mmap_resize(struct bpf_map *map, size_t old_sz, size_t new_sz)
1845 {
1846 void *mmaped;
1847
1848 if (!map->mmaped)
1849 return -EINVAL;
1850
1851 if (old_sz == new_sz)
1852 return 0;
1853
1854 mmaped = mmap(NULL, new_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0);
1855 if (mmaped == MAP_FAILED)
1856 return -errno;
1857
1858 memcpy(mmaped, map->mmaped, min(old_sz, new_sz));
1859 munmap(map->mmaped, old_sz);
1860 map->mmaped = mmaped;
1861 return 0;
1862 }
1863
1864 static char *internal_map_name(struct bpf_object *obj, const char *real_name)
1865 {
1866 char map_name[BPF_OBJ_NAME_LEN], *p;
1867 int pfx_len, sfx_len = max((size_t)7, strlen(real_name));
1868
1869 /* This is one of the more confusing parts of libbpf for various
1870 * reasons, some of which are historical. The original idea for naming
1871 * internal names was to include as much of BPF object name prefix as
1872 * possible, so that it can be distinguished from similar internal
1873 * maps of a different BPF object.
1874 * As an example, let's say we have bpf_object named 'my_object_name'
1875 * and internal map corresponding to '.rodata' ELF section. The final
1876 * map name advertised to user and to the kernel will be
1877 * 'my_objec.rodata', taking first 8 characters of object name and
1878 * entire 7 characters of '.rodata'.
1879 * Somewhat confusingly, if internal map ELF section name is shorter
1880 * than 7 characters, e.g., '.bss', we still reserve 7 characters
1881 * for the suffix, even though we only have 4 actual characters, and
1882 * resulting map will be called 'my_objec.bss', not even using all 15
1883 * characters allowed by the kernel. Oh well, at least the truncated
1884 * object name is somewhat consistent in this case. But if the map
1885 * name is '.kconfig', we'll still have entirety of '.kconfig' added
1886 * (8 chars) and thus will be left with only first 7 characters of the
1887 * object name ('my_obje'). Happy guessing, user, that the final map
1888 * name will be "my_obje.kconfig".
1889 * Now, with libbpf starting to support arbitrarily named .rodata.*
1890 * and .data.* data sections, it's possible that ELF section name is
1891 * longer than allowed 15 chars, so we now need to be careful to take
1892 * only up to 15 first characters of ELF name, taking no BPF object
1893 * name characters at all. So '.rodata.abracadabra' will result in
1894 * '.rodata.abracad' kernel and user-visible name.
1895 * We need to keep this convoluted logic intact for .data, .bss and
1896 * .rodata maps, but for new custom .data.custom and .rodata.custom
1897 * maps we use their ELF names as is, not prepending bpf_object name
1898 * in front. We still need to truncate them to 15 characters for the
1899 * kernel. Full name can be recovered for such maps by using DATASEC
1900 * BTF type associated with such map's value type, though.
1901 */
1902 if (sfx_len >= BPF_OBJ_NAME_LEN)
1903 sfx_len = BPF_OBJ_NAME_LEN - 1;
1904
1905 /* if there are two or more dots in map name, it's a custom dot map */
1906 if (strchr(real_name + 1, '.') != NULL)
1907 pfx_len = 0;
1908 else
1909 pfx_len = min((size_t)BPF_OBJ_NAME_LEN - sfx_len - 1, strlen(obj->name));
1910
1911 snprintf(map_name, sizeof(map_name), "%.*s%.*s", pfx_len, obj->name,
1912 sfx_len, real_name);
1913
1914 /* sanities map name to characters allowed by kernel */
1915 for (p = map_name; *p && p < map_name + sizeof(map_name); p++)
1916 if (!isalnum(*p) && *p != '_' && *p != '.')
1917 *p = '_';
1918
1919 return strdup(map_name);
1920 }
1921
1922 static int
1923 map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map);
1924
1925 /* Internal BPF map is mmap()'able only if at least one of corresponding
1926 * DATASEC's VARs are to be exposed through BPF skeleton. I.e., it's a GLOBAL
1927 * variable and it's not marked as __hidden (which turns it into, effectively,
1928 * a STATIC variable).
1929 */
1930 static bool map_is_mmapable(struct bpf_object *obj, struct bpf_map *map)
1931 {
1932 const struct btf_type *t, *vt;
1933 struct btf_var_secinfo *vsi;
1934 int i, n;
1935
1936 if (!map->btf_value_type_id)
1937 return false;
1938
1939 t = btf__type_by_id(obj->btf, map->btf_value_type_id);
1940 if (!btf_is_datasec(t))
1941 return false;
1942
1943 vsi = btf_var_secinfos(t);
1944 for (i = 0, n = btf_vlen(t); i < n; i++, vsi++) {
1945 vt = btf__type_by_id(obj->btf, vsi->type);
1946 if (!btf_is_var(vt))
1947 continue;
1948
1949 if (btf_var(vt)->linkage != BTF_VAR_STATIC)
1950 return true;
1951 }
1952
1953 return false;
1954 }
1955
1956 static int
1957 bpf_object__init_internal_map(struct bpf_object *obj, enum libbpf_map_type type,
1958 const char *real_name, int sec_idx, void *data, size_t data_sz)
1959 {
1960 struct bpf_map_def *def;
1961 struct bpf_map *map;
1962 size_t mmap_sz;
1963 int err;
1964
1965 map = bpf_object__add_map(obj);
1966 if (IS_ERR(map))
1967 return PTR_ERR(map);
1968
1969 map->libbpf_type = type;
1970 map->sec_idx = sec_idx;
1971 map->sec_offset = 0;
1972 map->real_name = strdup(real_name);
1973 map->name = internal_map_name(obj, real_name);
1974 if (!map->real_name || !map->name) {
1975 zfree(&map->real_name);
1976 zfree(&map->name);
1977 return -ENOMEM;
1978 }
1979
1980 def = &map->def;
1981 def->type = BPF_MAP_TYPE_ARRAY;
1982 def->key_size = sizeof(int);
1983 def->value_size = data_sz;
1984 def->max_entries = 1;
1985 def->map_flags = type == LIBBPF_MAP_RODATA || type == LIBBPF_MAP_KCONFIG
1986 ? BPF_F_RDONLY_PROG : 0;
1987
1988 /* failures are fine because of maps like .rodata.str1.1 */
1989 (void) map_fill_btf_type_info(obj, map);
1990
1991 if (map_is_mmapable(obj, map))
1992 def->map_flags |= BPF_F_MMAPABLE;
1993
1994 pr_debug("map '%s' (global data): at sec_idx %d, offset %zu, flags %x.\n",
1995 map->name, map->sec_idx, map->sec_offset, def->map_flags);
1996
1997 mmap_sz = bpf_map_mmap_sz(map);
1998 map->mmaped = mmap(NULL, mmap_sz, PROT_READ | PROT_WRITE,
1999 MAP_SHARED | MAP_ANONYMOUS, -1, 0);
2000 if (map->mmaped == MAP_FAILED) {
2001 err = -errno;
2002 map->mmaped = NULL;
2003 pr_warn("failed to alloc map '%s' content buffer: %s\n", map->name, errstr(err));
2004 zfree(&map->real_name);
2005 zfree(&map->name);
2006 return err;
2007 }
2008
2009 if (data)
2010 memcpy(map->mmaped, data, data_sz);
2011
2012 pr_debug("map %td is \"%s\"\n", map - obj->maps, map->name);
2013 return 0;
2014 }
2015
2016 static int bpf_object__init_global_data_maps(struct bpf_object *obj)
2017 {
2018 struct elf_sec_desc *sec_desc;
2019 const char *sec_name;
2020 int err = 0, sec_idx;
2021
2022 /*
2023 * Populate obj->maps with libbpf internal maps.
2024 */
2025 for (sec_idx = 1; sec_idx < obj->efile.sec_cnt; sec_idx++) {
2026 sec_desc = &obj->efile.secs[sec_idx];
2027
2028 /* Skip recognized sections with size 0. */
2029 if (!sec_desc->data || sec_desc->data->d_size == 0)
2030 continue;
2031
2032 switch (sec_desc->sec_type) {
2033 case SEC_DATA:
2034 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
2035 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_DATA,
2036 sec_name, sec_idx,
2037 sec_desc->data->d_buf,
2038 sec_desc->data->d_size);
2039 break;
2040 case SEC_RODATA:
2041 obj->has_rodata = true;
2042 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
2043 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_RODATA,
2044 sec_name, sec_idx,
2045 sec_desc->data->d_buf,
2046 sec_desc->data->d_size);
2047 break;
2048 case SEC_BSS:
2049 sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, sec_idx));
2050 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_BSS,
2051 sec_name, sec_idx,
2052 NULL,
2053 sec_desc->data->d_size);
2054 break;
2055 default:
2056 /* skip */
2057 break;
2058 }
2059 if (err)
2060 return err;
2061 }
2062 return 0;
2063 }
2064
2065
2066 static struct extern_desc *find_extern_by_name(const struct bpf_object *obj,
2067 const void *name)
2068 {
2069 int i;
2070
2071 for (i = 0; i < obj->nr_extern; i++) {
2072 if (strcmp(obj->externs[i].name, name) == 0)
2073 return &obj->externs[i];
2074 }
2075 return NULL;
2076 }
2077
2078 static struct extern_desc *find_extern_by_name_with_len(const struct bpf_object *obj,
2079 const void *name, int len)
2080 {
2081 const char *ext_name;
2082 int i;
2083
2084 for (i = 0; i < obj->nr_extern; i++) {
2085 ext_name = obj->externs[i].name;
2086 if (strlen(ext_name) == len && strncmp(ext_name, name, len) == 0)
2087 return &obj->externs[i];
2088 }
2089 return NULL;
2090 }
2091
2092 static int set_kcfg_value_tri(struct extern_desc *ext, void *ext_val,
2093 char value)
2094 {
2095 switch (ext->kcfg.type) {
2096 case KCFG_BOOL:
2097 if (value == 'm') {
2098 pr_warn("extern (kcfg) '%s': value '%c' implies tristate or char type\n",
2099 ext->name, value);
2100 return -EINVAL;
2101 }
2102 *(bool *)ext_val = value == 'y' ? true : false;
2103 break;
2104 case KCFG_TRISTATE:
2105 if (value == 'y')
2106 *(enum libbpf_tristate *)ext_val = TRI_YES;
2107 else if (value == 'm')
2108 *(enum libbpf_tristate *)ext_val = TRI_MODULE;
2109 else /* value == 'n' */
2110 *(enum libbpf_tristate *)ext_val = TRI_NO;
2111 break;
2112 case KCFG_CHAR:
2113 *(char *)ext_val = value;
2114 break;
2115 case KCFG_UNKNOWN:
2116 case KCFG_INT:
2117 case KCFG_CHAR_ARR:
2118 default:
2119 pr_warn("extern (kcfg) '%s': value '%c' implies bool, tristate, or char type\n",
2120 ext->name, value);
2121 return -EINVAL;
2122 }
2123 ext->is_set = true;
2124 return 0;
2125 }
2126
2127 static int set_kcfg_value_str(struct extern_desc *ext, char *ext_val,
2128 const char *value)
2129 {
2130 size_t len;
2131
2132 if (ext->kcfg.type != KCFG_CHAR_ARR) {
2133 pr_warn("extern (kcfg) '%s': value '%s' implies char array type\n",
2134 ext->name, value);
2135 return -EINVAL;
2136 }
2137
2138 len = strlen(value);
2139 if (len < 2 || value[len - 1] != '"') {
2140 pr_warn("extern (kcfg) '%s': invalid string config '%s'\n",
2141 ext->name, value);
2142 return -EINVAL;
2143 }
2144
2145 /* strip quotes */
2146 len -= 2;
2147 if (len >= ext->kcfg.sz) {
2148 pr_warn("extern (kcfg) '%s': long string '%s' of (%zu bytes) truncated to %d bytes\n",
2149 ext->name, value, len, ext->kcfg.sz - 1);
2150 len = ext->kcfg.sz - 1;
2151 }
2152 memcpy(ext_val, value + 1, len);
2153 ext_val[len] = '\0';
2154 ext->is_set = true;
2155 return 0;
2156 }
2157
2158 static int parse_u64(const char *value, __u64 *res)
2159 {
2160 char *value_end;
2161 int err;
2162
2163 errno = 0;
2164 *res = strtoull(value, &value_end, 0);
2165 if (errno) {
2166 err = -errno;
2167 pr_warn("failed to parse '%s': %s\n", value, errstr(err));
2168 return err;
2169 }
2170 if (*value_end) {
2171 pr_warn("failed to parse '%s' as integer completely\n", value);
2172 return -EINVAL;
2173 }
2174 return 0;
2175 }
2176
2177 static bool is_kcfg_value_in_range(const struct extern_desc *ext, __u64 v)
2178 {
2179 int bit_sz = ext->kcfg.sz * 8;
2180
2181 if (ext->kcfg.sz == 8)
2182 return true;
2183
2184 /* Validate that value stored in u64 fits in integer of `ext->sz`
2185 * bytes size without any loss of information. If the target integer
2186 * is signed, we rely on the following limits of integer type of
2187 * Y bits and subsequent transformation:
2188 *
2189 * -2^(Y-1) <= X <= 2^(Y-1) - 1
2190 * 0 <= X + 2^(Y-1) <= 2^Y - 1
2191 * 0 <= X + 2^(Y-1) < 2^Y
2192 *
2193 * For unsigned target integer, check that all the (64 - Y) bits are
2194 * zero.
2195 */
2196 if (ext->kcfg.is_signed)
2197 return v + (1ULL << (bit_sz - 1)) < (1ULL << bit_sz);
2198 else
2199 return (v >> bit_sz) == 0;
2200 }
2201
2202 static int set_kcfg_value_num(struct extern_desc *ext, void *ext_val,
2203 __u64 value)
2204 {
2205 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR &&
2206 ext->kcfg.type != KCFG_BOOL) {
2207 pr_warn("extern (kcfg) '%s': value '%llu' implies integer, char, or boolean type\n",
2208 ext->name, (unsigned long long)value);
2209 return -EINVAL;
2210 }
2211 if (ext->kcfg.type == KCFG_BOOL && value > 1) {
2212 pr_warn("extern (kcfg) '%s': value '%llu' isn't boolean compatible\n",
2213 ext->name, (unsigned long long)value);
2214 return -EINVAL;
2215
2216 }
2217 if (!is_kcfg_value_in_range(ext, value)) {
2218 pr_warn("extern (kcfg) '%s': value '%llu' doesn't fit in %d bytes\n",
2219 ext->name, (unsigned long long)value, ext->kcfg.sz);
2220 return -ERANGE;
2221 }
2222 switch (ext->kcfg.sz) {
2223 case 1:
2224 *(__u8 *)ext_val = value;
2225 break;
2226 case 2:
2227 *(__u16 *)ext_val = value;
2228 break;
2229 case 4:
2230 *(__u32 *)ext_val = value;
2231 break;
2232 case 8:
2233 *(__u64 *)ext_val = value;
2234 break;
2235 default:
2236 return -EINVAL;
2237 }
2238 ext->is_set = true;
2239 return 0;
2240 }
2241
2242 static int bpf_object__process_kconfig_line(struct bpf_object *obj,
2243 char *buf, void *data)
2244 {
2245 struct extern_desc *ext;
2246 char *sep, *value;
2247 int len, err = 0;
2248 void *ext_val;
2249 __u64 num;
2250
2251 if (!str_has_pfx(buf, "CONFIG_"))
2252 return 0;
2253
2254 sep = strchr(buf, '=');
2255 if (!sep) {
2256 pr_warn("failed to parse '%s': no separator\n", buf);
2257 return -EINVAL;
2258 }
2259
2260 /* Trim ending '\n' */
2261 len = strlen(buf);
2262 if (buf[len - 1] == '\n')
2263 buf[len - 1] = '\0';
2264 /* Split on '=' and ensure that a value is present. */
2265 *sep = '\0';
2266 if (!sep[1]) {
2267 *sep = '=';
2268 pr_warn("failed to parse '%s': no value\n", buf);
2269 return -EINVAL;
2270 }
2271
2272 ext = find_extern_by_name(obj, buf);
2273 if (!ext || ext->is_set)
2274 return 0;
2275
2276 ext_val = data + ext->kcfg.data_off;
2277 value = sep + 1;
2278
2279 switch (*value) {
2280 case 'y': case 'n': case 'm':
2281 err = set_kcfg_value_tri(ext, ext_val, *value);
2282 break;
2283 case '"':
2284 err = set_kcfg_value_str(ext, ext_val, value);
2285 break;
2286 default:
2287 /* assume integer */
2288 err = parse_u64(value, &num);
2289 if (err) {
2290 pr_warn("extern (kcfg) '%s': value '%s' isn't a valid integer\n", ext->name, value);
2291 return err;
2292 }
2293 if (ext->kcfg.type != KCFG_INT && ext->kcfg.type != KCFG_CHAR) {
2294 pr_warn("extern (kcfg) '%s': value '%s' implies integer type\n", ext->name, value);
2295 return -EINVAL;
2296 }
2297 err = set_kcfg_value_num(ext, ext_val, num);
2298 break;
2299 }
2300 if (err)
2301 return err;
2302 pr_debug("extern (kcfg) '%s': set to %s\n", ext->name, value);
2303 return 0;
2304 }
2305
2306 static int bpf_object__read_kconfig_file(struct bpf_object *obj, void *data)
2307 {
2308 char buf[PATH_MAX];
2309 struct utsname uts;
2310 int len, err = 0;
2311 gzFile file;
2312
2313 uname(&uts);
2314 len = snprintf(buf, PATH_MAX, "/boot/config-%s", uts.release);
2315 if (len < 0)
2316 return -EINVAL;
2317 else if (len >= PATH_MAX)
2318 return -ENAMETOOLONG;
2319
2320 /* gzopen also accepts uncompressed files. */
2321 file = gzopen(buf, "re");
2322 if (!file)
2323 file = gzopen("/proc/config.gz", "re");
2324
2325 if (!file) {
2326 pr_warn("failed to open system Kconfig\n");
2327 return -ENOENT;
2328 }
2329
2330 while (gzgets(file, buf, sizeof(buf))) {
2331 err = bpf_object__process_kconfig_line(obj, buf, data);
2332 if (err) {
2333 pr_warn("error parsing system Kconfig line '%s': %s\n",
2334 buf, errstr(err));
2335 goto out;
2336 }
2337 }
2338
2339 out:
2340 gzclose(file);
2341 return err;
2342 }
2343
2344 static int bpf_object__read_kconfig_mem(struct bpf_object *obj,
2345 const char *config, void *data)
2346 {
2347 char buf[PATH_MAX];
2348 int err = 0;
2349 FILE *file;
2350
2351 file = fmemopen((void *)config, strlen(config), "r");
2352 if (!file) {
2353 err = -errno;
2354 pr_warn("failed to open in-memory Kconfig: %s\n", errstr(err));
2355 return err;
2356 }
2357
2358 while (fgets(buf, sizeof(buf), file)) {
2359 err = bpf_object__process_kconfig_line(obj, buf, data);
2360 if (err) {
2361 pr_warn("error parsing in-memory Kconfig line '%s': %s\n",
2362 buf, errstr(err));
2363 break;
2364 }
2365 }
2366
2367 fclose(file);
2368 return err;
2369 }
2370
2371 static int bpf_object__init_kconfig_map(struct bpf_object *obj)
2372 {
2373 struct extern_desc *last_ext = NULL, *ext;
2374 size_t map_sz;
2375 int i, err;
2376
2377 for (i = 0; i < obj->nr_extern; i++) {
2378 ext = &obj->externs[i];
2379 if (ext->type == EXT_KCFG)
2380 last_ext = ext;
2381 }
2382
2383 if (!last_ext)
2384 return 0;
2385
2386 map_sz = last_ext->kcfg.data_off + last_ext->kcfg.sz;
2387 err = bpf_object__init_internal_map(obj, LIBBPF_MAP_KCONFIG,
2388 ".kconfig", obj->efile.symbols_shndx,
2389 NULL, map_sz);
2390 if (err)
2391 return err;
2392
2393 obj->kconfig_map_idx = obj->nr_maps - 1;
2394
2395 return 0;
2396 }
2397
2398 const struct btf_type *
2399 skip_mods_and_typedefs(const struct btf *btf, __u32 id, __u32 *res_id)
2400 {
2401 const struct btf_type *t = btf__type_by_id(btf, id);
2402
2403 if (res_id)
2404 *res_id = id;
2405
2406 while (btf_is_mod(t) || btf_is_typedef(t)) {
2407 if (res_id)
2408 *res_id = t->type;
2409 t = btf__type_by_id(btf, t->type);
2410 }
2411
2412 return t;
2413 }
2414
2415 static const struct btf_type *
2416 resolve_func_ptr(const struct btf *btf, __u32 id, __u32 *res_id)
2417 {
2418 const struct btf_type *t;
2419
2420 t = skip_mods_and_typedefs(btf, id, NULL);
2421 if (!btf_is_ptr(t))
2422 return NULL;
2423
2424 t = skip_mods_and_typedefs(btf, t->type, res_id);
2425
2426 return btf_is_func_proto(t) ? t : NULL;
2427 }
2428
2429 static const char *__btf_kind_str(__u16 kind)
2430 {
2431 switch (kind) {
2432 case BTF_KIND_UNKN: return "void";
2433 case BTF_KIND_INT: return "int";
2434 case BTF_KIND_PTR: return "ptr";
2435 case BTF_KIND_ARRAY: return "array";
2436 case BTF_KIND_STRUCT: return "struct";
2437 case BTF_KIND_UNION: return "union";
2438 case BTF_KIND_ENUM: return "enum";
2439 case BTF_KIND_FWD: return "fwd";
2440 case BTF_KIND_TYPEDEF: return "typedef";
2441 case BTF_KIND_VOLATILE: return "volatile";
2442 case BTF_KIND_CONST: return "const";
2443 case BTF_KIND_RESTRICT: return "restrict";
2444 case BTF_KIND_FUNC: return "func";
2445 case BTF_KIND_FUNC_PROTO: return "func_proto";
2446 case BTF_KIND_VAR: return "var";
2447 case BTF_KIND_DATASEC: return "datasec";
2448 case BTF_KIND_FLOAT: return "float";
2449 case BTF_KIND_DECL_TAG: return "decl_tag";
2450 case BTF_KIND_TYPE_TAG: return "type_tag";
2451 case BTF_KIND_ENUM64: return "enum64";
2452 default: return "unknown";
2453 }
2454 }
2455
2456 const char *btf_kind_str(const struct btf_type *t)
2457 {
2458 return __btf_kind_str(btf_kind(t));
2459 }
2460
2461 /*
2462 * Fetch integer attribute of BTF map definition. Such attributes are
2463 * represented using a pointer to an array, in which dimensionality of array
2464 * encodes specified integer value. E.g., int (*type)[BPF_MAP_TYPE_ARRAY];
2465 * encodes `type => BPF_MAP_TYPE_ARRAY` key/value pair completely using BTF
2466 * type definition, while using only sizeof(void *) space in ELF data section.
2467 */
2468 static bool get_map_field_int(const char *map_name, const struct btf *btf,
2469 const struct btf_member *m, __u32 *res)
2470 {
2471 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2472 const char *name = btf__name_by_offset(btf, m->name_off);
2473 const struct btf_array *arr_info;
2474 const struct btf_type *arr_t;
2475
2476 if (!btf_is_ptr(t)) {
2477 pr_warn("map '%s': attr '%s': expected PTR, got %s.\n",
2478 map_name, name, btf_kind_str(t));
2479 return false;
2480 }
2481
2482 arr_t = btf__type_by_id(btf, t->type);
2483 if (!arr_t) {
2484 pr_warn("map '%s': attr '%s': type [%u] not found.\n",
2485 map_name, name, t->type);
2486 return false;
2487 }
2488 if (!btf_is_array(arr_t)) {
2489 pr_warn("map '%s': attr '%s': expected ARRAY, got %s.\n",
2490 map_name, name, btf_kind_str(arr_t));
2491 return false;
2492 }
2493 arr_info = btf_array(arr_t);
2494 *res = arr_info->nelems;
2495 return true;
2496 }
2497
2498 static bool get_map_field_long(const char *map_name, const struct btf *btf,
2499 const struct btf_member *m, __u64 *res)
2500 {
2501 const struct btf_type *t = skip_mods_and_typedefs(btf, m->type, NULL);
2502 const char *name = btf__name_by_offset(btf, m->name_off);
2503
2504 if (btf_is_ptr(t)) {
2505 __u32 res32;
2506 bool ret;
2507
2508 ret = get_map_field_int(map_name, btf, m, &res32);
2509 if (ret)
2510 *res = (__u64)res32;
2511 return ret;
2512 }
2513
2514 if (!btf_is_enum(t) && !btf_is_enum64(t)) {
2515 pr_warn("map '%s': attr '%s': expected ENUM or ENUM64, got %s.\n",
2516 map_name, name, btf_kind_str(t));
2517 return false;
2518 }
2519
2520 if (btf_vlen(t) != 1) {
2521 pr_warn("map '%s': attr '%s': invalid __ulong\n",
2522 map_name, name);
2523 return false;
2524 }
2525
2526 if (btf_is_enum(t)) {
2527 const struct btf_enum *e = btf_enum(t);
2528
2529 *res = e->val;
2530 } else {
2531 const struct btf_enum64 *e = btf_enum64(t);
2532
2533 *res = btf_enum64_value(e);
2534 }
2535 return true;
2536 }
2537
2538 static int pathname_concat(char *buf, size_t buf_sz, const char *path, const char *name)
2539 {
2540 int len;
2541
2542 len = snprintf(buf, buf_sz, "%s/%s", path, name);
2543 if (len < 0)
2544 return -EINVAL;
2545 if (len >= buf_sz)
2546 return -ENAMETOOLONG;
2547
2548 return 0;
2549 }
2550
2551 static int build_map_pin_path(struct bpf_map *map, const char *path)
2552 {
2553 char buf[PATH_MAX];
2554 int err;
2555
2556 if (!path)
2557 path = BPF_FS_DEFAULT_PATH;
2558
2559 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
2560 if (err)
2561 return err;
2562
2563 return bpf_map__set_pin_path(map, buf);
2564 }
2565
2566 /* should match definition in bpf_helpers.h */
2567 enum libbpf_pin_type {
2568 LIBBPF_PIN_NONE,
2569 /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */
2570 LIBBPF_PIN_BY_NAME,
2571 };
2572
2573 int parse_btf_map_def(const char *map_name, struct btf *btf,
2574 const struct btf_type *def_t, bool strict,
2575 struct btf_map_def *map_def, struct btf_map_def *inner_def)
2576 {
2577 const struct btf_type *t;
2578 const struct btf_member *m;
2579 bool is_inner = inner_def == NULL;
2580 int vlen, i;
2581
2582 vlen = btf_vlen(def_t);
2583 m = btf_members(def_t);
2584 for (i = 0; i < vlen; i++, m++) {
2585 const char *name = btf__name_by_offset(btf, m->name_off);
2586
2587 if (!name) {
2588 pr_warn("map '%s': invalid field #%d.\n", map_name, i);
2589 return -EINVAL;
2590 }
2591 if (strcmp(name, "type") == 0) {
2592 if (!get_map_field_int(map_name, btf, m, &map_def->map_type))
2593 return -EINVAL;
2594 map_def->parts |= MAP_DEF_MAP_TYPE;
2595 } else if (strcmp(name, "max_entries") == 0) {
2596 if (!get_map_field_int(map_name, btf, m, &map_def->max_entries))
2597 return -EINVAL;
2598 map_def->parts |= MAP_DEF_MAX_ENTRIES;
2599 } else if (strcmp(name, "map_flags") == 0) {
2600 if (!get_map_field_int(map_name, btf, m, &map_def->map_flags))
2601 return -EINVAL;
2602 map_def->parts |= MAP_DEF_MAP_FLAGS;
2603 } else if (strcmp(name, "numa_node") == 0) {
2604 if (!get_map_field_int(map_name, btf, m, &map_def->numa_node))
2605 return -EINVAL;
2606 map_def->parts |= MAP_DEF_NUMA_NODE;
2607 } else if (strcmp(name, "key_size") == 0) {
2608 __u32 sz;
2609
2610 if (!get_map_field_int(map_name, btf, m, &sz))
2611 return -EINVAL;
2612 if (map_def->key_size && map_def->key_size != sz) {
2613 pr_warn("map '%s': conflicting key size %u != %u.\n",
2614 map_name, map_def->key_size, sz);
2615 return -EINVAL;
2616 }
2617 map_def->key_size = sz;
2618 map_def->parts |= MAP_DEF_KEY_SIZE;
2619 } else if (strcmp(name, "key") == 0) {
2620 __s64 sz;
2621
2622 t = btf__type_by_id(btf, m->type);
2623 if (!t) {
2624 pr_warn("map '%s': key type [%d] not found.\n",
2625 map_name, m->type);
2626 return -EINVAL;
2627 }
2628 if (!btf_is_ptr(t)) {
2629 pr_warn("map '%s': key spec is not PTR: %s.\n",
2630 map_name, btf_kind_str(t));
2631 return -EINVAL;
2632 }
2633 sz = btf__resolve_size(btf, t->type);
2634 if (sz < 0) {
2635 pr_warn("map '%s': can't determine key size for type [%u]: %zd.\n",
2636 map_name, t->type, (ssize_t)sz);
2637 return sz;
2638 }
2639 if (map_def->key_size && map_def->key_size != sz) {
2640 pr_warn("map '%s': conflicting key size %u != %zd.\n",
2641 map_name, map_def->key_size, (ssize_t)sz);
2642 return -EINVAL;
2643 }
2644 map_def->key_size = sz;
2645 map_def->key_type_id = t->type;
2646 map_def->parts |= MAP_DEF_KEY_SIZE | MAP_DEF_KEY_TYPE;
2647 } else if (strcmp(name, "value_size") == 0) {
2648 __u32 sz;
2649
2650 if (!get_map_field_int(map_name, btf, m, &sz))
2651 return -EINVAL;
2652 if (map_def->value_size && map_def->value_size != sz) {
2653 pr_warn("map '%s': conflicting value size %u != %u.\n",
2654 map_name, map_def->value_size, sz);
2655 return -EINVAL;
2656 }
2657 map_def->value_size = sz;
2658 map_def->parts |= MAP_DEF_VALUE_SIZE;
2659 } else if (strcmp(name, "value") == 0) {
2660 __s64 sz;
2661
2662 t = btf__type_by_id(btf, m->type);
2663 if (!t) {
2664 pr_warn("map '%s': value type [%d] not found.\n",
2665 map_name, m->type);
2666 return -EINVAL;
2667 }
2668 if (!btf_is_ptr(t)) {
2669 pr_warn("map '%s': value spec is not PTR: %s.\n",
2670 map_name, btf_kind_str(t));
2671 return -EINVAL;
2672 }
2673 sz = btf__resolve_size(btf, t->type);
2674 if (sz < 0) {
2675 pr_warn("map '%s': can't determine value size for type [%u]: %zd.\n",
2676 map_name, t->type, (ssize_t)sz);
2677 return sz;
2678 }
2679 if (map_def->value_size && map_def->value_size != sz) {
2680 pr_warn("map '%s': conflicting value size %u != %zd.\n",
2681 map_name, map_def->value_size, (ssize_t)sz);
2682 return -EINVAL;
2683 }
2684 map_def->value_size = sz;
2685 map_def->value_type_id = t->type;
2686 map_def->parts |= MAP_DEF_VALUE_SIZE | MAP_DEF_VALUE_TYPE;
2687 }
2688 else if (strcmp(name, "values") == 0) {
2689 bool is_map_in_map = bpf_map_type__is_map_in_map(map_def->map_type);
2690 bool is_prog_array = map_def->map_type == BPF_MAP_TYPE_PROG_ARRAY;
2691 const char *desc = is_map_in_map ? "map-in-map inner" : "prog-array value";
2692 char inner_map_name[128];
2693 int err;
2694
2695 if (is_inner) {
2696 pr_warn("map '%s': multi-level inner maps not supported.\n",
2697 map_name);
2698 return -ENOTSUP;
2699 }
2700 if (i != vlen - 1) {
2701 pr_warn("map '%s': '%s' member should be last.\n",
2702 map_name, name);
2703 return -EINVAL;
2704 }
2705 if (!is_map_in_map && !is_prog_array) {
2706 pr_warn("map '%s': should be map-in-map or prog-array.\n",
2707 map_name);
2708 return -ENOTSUP;
2709 }
2710 if (map_def->value_size && map_def->value_size != 4) {
2711 pr_warn("map '%s': conflicting value size %u != 4.\n",
2712 map_name, map_def->value_size);
2713 return -EINVAL;
2714 }
2715 map_def->value_size = 4;
2716 t = btf__type_by_id(btf, m->type);
2717 if (!t) {
2718 pr_warn("map '%s': %s type [%d] not found.\n",
2719 map_name, desc, m->type);
2720 return -EINVAL;
2721 }
2722 if (!btf_is_array(t) || btf_array(t)->nelems) {
2723 pr_warn("map '%s': %s spec is not a zero-sized array.\n",
2724 map_name, desc);
2725 return -EINVAL;
2726 }
2727 t = skip_mods_and_typedefs(btf, btf_array(t)->type, NULL);
2728 if (!btf_is_ptr(t)) {
2729 pr_warn("map '%s': %s def is of unexpected kind %s.\n",
2730 map_name, desc, btf_kind_str(t));
2731 return -EINVAL;
2732 }
2733 t = skip_mods_and_typedefs(btf, t->type, NULL);
2734 if (is_prog_array) {
2735 if (!btf_is_func_proto(t)) {
2736 pr_warn("map '%s': prog-array value def is of unexpected kind %s.\n",
2737 map_name, btf_kind_str(t));
2738 return -EINVAL;
2739 }
2740 continue;
2741 }
2742 if (!btf_is_struct(t)) {
2743 pr_warn("map '%s': map-in-map inner def is of unexpected kind %s.\n",
2744 map_name, btf_kind_str(t));
2745 return -EINVAL;
2746 }
2747
2748 snprintf(inner_map_name, sizeof(inner_map_name), "%s.inner", map_name);
2749 err = parse_btf_map_def(inner_map_name, btf, t, strict, inner_def, NULL);
2750 if (err)
2751 return err;
2752
2753 map_def->parts |= MAP_DEF_INNER_MAP;
2754 } else if (strcmp(name, "pinning") == 0) {
2755 __u32 val;
2756
2757 if (is_inner) {
2758 pr_warn("map '%s': inner def can't be pinned.\n", map_name);
2759 return -EINVAL;
2760 }
2761 if (!get_map_field_int(map_name, btf, m, &val))
2762 return -EINVAL;
2763 if (val != LIBBPF_PIN_NONE && val != LIBBPF_PIN_BY_NAME) {
2764 pr_warn("map '%s': invalid pinning value %u.\n",
2765 map_name, val);
2766 return -EINVAL;
2767 }
2768 map_def->pinning = val;
2769 map_def->parts |= MAP_DEF_PINNING;
2770 } else if (strcmp(name, "map_extra") == 0) {
2771 __u64 map_extra;
2772
2773 if (!get_map_field_long(map_name, btf, m, &map_extra))
2774 return -EINVAL;
2775 map_def->map_extra = map_extra;
2776 map_def->parts |= MAP_DEF_MAP_EXTRA;
2777 } else {
2778 if (strict) {
2779 pr_warn("map '%s': unknown field '%s'.\n", map_name, name);
2780 return -ENOTSUP;
2781 }
2782 pr_debug("map '%s': ignoring unknown field '%s'.\n", map_name, name);
2783 }
2784 }
2785
2786 if (map_def->map_type == BPF_MAP_TYPE_UNSPEC) {
2787 pr_warn("map '%s': map type isn't specified.\n", map_name);
2788 return -EINVAL;
2789 }
2790
2791 return 0;
2792 }
2793
2794 static size_t adjust_ringbuf_sz(size_t sz)
2795 {
2796 __u32 page_sz = sysconf(_SC_PAGE_SIZE);
2797 __u32 mul;
2798
2799 /* if user forgot to set any size, make sure they see error */
2800 if (sz == 0)
2801 return 0;
2802 /* Kernel expects BPF_MAP_TYPE_RINGBUF's max_entries to be
2803 * a power-of-2 multiple of kernel's page size. If user diligently
2804 * satisified these conditions, pass the size through.
2805 */
2806 if ((sz % page_sz) == 0 && is_pow_of_2(sz / page_sz))
2807 return sz;
2808
2809 /* Otherwise find closest (page_sz * power_of_2) product bigger than
2810 * user-set size to satisfy both user size request and kernel
2811 * requirements and substitute correct max_entries for map creation.
2812 */
2813 for (mul = 1; mul <= UINT_MAX / page_sz; mul <<= 1) {
2814 if (mul * page_sz > sz)
2815 return mul * page_sz;
2816 }
2817
2818 /* if it's impossible to satisfy the conditions (i.e., user size is
2819 * very close to UINT_MAX but is not a power-of-2 multiple of
2820 * page_size) then just return original size and let kernel reject it
2821 */
2822 return sz;
2823 }
2824
2825 static bool map_is_ringbuf(const struct bpf_map *map)
2826 {
2827 return map->def.type == BPF_MAP_TYPE_RINGBUF ||
2828 map->def.type == BPF_MAP_TYPE_USER_RINGBUF;
2829 }
2830
2831 static void fill_map_from_def(struct bpf_map *map, const struct btf_map_def *def)
2832 {
2833 map->def.type = def->map_type;
2834 map->def.key_size = def->key_size;
2835 map->def.value_size = def->value_size;
2836 map->def.max_entries = def->max_entries;
2837 map->def.map_flags = def->map_flags;
2838 map->map_extra = def->map_extra;
2839
2840 map->numa_node = def->numa_node;
2841 map->btf_key_type_id = def->key_type_id;
2842 map->btf_value_type_id = def->value_type_id;
2843
2844 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
2845 if (map_is_ringbuf(map))
2846 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
2847
2848 if (def->parts & MAP_DEF_MAP_TYPE)
2849 pr_debug("map '%s': found type = %u.\n", map->name, def->map_type);
2850
2851 if (def->parts & MAP_DEF_KEY_TYPE)
2852 pr_debug("map '%s': found key [%u], sz = %u.\n",
2853 map->name, def->key_type_id, def->key_size);
2854 else if (def->parts & MAP_DEF_KEY_SIZE)
2855 pr_debug("map '%s': found key_size = %u.\n", map->name, def->key_size);
2856
2857 if (def->parts & MAP_DEF_VALUE_TYPE)
2858 pr_debug("map '%s': found value [%u], sz = %u.\n",
2859 map->name, def->value_type_id, def->value_size);
2860 else if (def->parts & MAP_DEF_VALUE_SIZE)
2861 pr_debug("map '%s': found value_size = %u.\n", map->name, def->value_size);
2862
2863 if (def->parts & MAP_DEF_MAX_ENTRIES)
2864 pr_debug("map '%s': found max_entries = %u.\n", map->name, def->max_entries);
2865 if (def->parts & MAP_DEF_MAP_FLAGS)
2866 pr_debug("map '%s': found map_flags = 0x%x.\n", map->name, def->map_flags);
2867 if (def->parts & MAP_DEF_MAP_EXTRA)
2868 pr_debug("map '%s': found map_extra = 0x%llx.\n", map->name,
2869 (unsigned long long)def->map_extra);
2870 if (def->parts & MAP_DEF_PINNING)
2871 pr_debug("map '%s': found pinning = %u.\n", map->name, def->pinning);
2872 if (def->parts & MAP_DEF_NUMA_NODE)
2873 pr_debug("map '%s': found numa_node = %u.\n", map->name, def->numa_node);
2874
2875 if (def->parts & MAP_DEF_INNER_MAP)
2876 pr_debug("map '%s': found inner map definition.\n", map->name);
2877 }
2878
2879 static const char *btf_var_linkage_str(__u32 linkage)
2880 {
2881 switch (linkage) {
2882 case BTF_VAR_STATIC: return "static";
2883 case BTF_VAR_GLOBAL_ALLOCATED: return "global";
2884 case BTF_VAR_GLOBAL_EXTERN: return "extern";
2885 default: return "unknown";
2886 }
2887 }
2888
2889 static int bpf_object__init_user_btf_map(struct bpf_object *obj,
2890 const struct btf_type *sec,
2891 int var_idx, int sec_idx,
2892 const Elf_Data *data, bool strict,
2893 const char *pin_root_path)
2894 {
2895 struct btf_map_def map_def = {}, inner_def = {};
2896 const struct btf_type *var, *def;
2897 const struct btf_var_secinfo *vi;
2898 const struct btf_var *var_extra;
2899 const char *map_name;
2900 struct bpf_map *map;
2901 int err;
2902
2903 vi = btf_var_secinfos(sec) + var_idx;
2904 var = btf__type_by_id(obj->btf, vi->type);
2905 var_extra = btf_var(var);
2906 map_name = btf__name_by_offset(obj->btf, var->name_off);
2907
2908 if (str_is_empty(map_name)) {
2909 pr_warn("map #%d: empty name.\n", var_idx);
2910 return -EINVAL;
2911 }
2912 if ((__u64)vi->offset + vi->size > data->d_size) {
2913 pr_warn("map '%s' BTF data is corrupted.\n", map_name);
2914 return -EINVAL;
2915 }
2916 if (!btf_is_var(var)) {
2917 pr_warn("map '%s': unexpected var kind %s.\n",
2918 map_name, btf_kind_str(var));
2919 return -EINVAL;
2920 }
2921 if (var_extra->linkage != BTF_VAR_GLOBAL_ALLOCATED) {
2922 pr_warn("map '%s': unsupported map linkage %s.\n",
2923 map_name, btf_var_linkage_str(var_extra->linkage));
2924 return -EOPNOTSUPP;
2925 }
2926
2927 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
2928 if (!btf_is_struct(def)) {
2929 pr_warn("map '%s': unexpected def kind %s.\n",
2930 map_name, btf_kind_str(var));
2931 return -EINVAL;
2932 }
2933 if (def->size > vi->size) {
2934 pr_warn("map '%s': invalid def size.\n", map_name);
2935 return -EINVAL;
2936 }
2937
2938 map = bpf_object__add_map(obj);
2939 if (IS_ERR(map))
2940 return PTR_ERR(map);
2941 map->name = strdup(map_name);
2942 if (!map->name) {
2943 pr_warn("map '%s': failed to alloc map name.\n", map_name);
2944 return -ENOMEM;
2945 }
2946 map->libbpf_type = LIBBPF_MAP_UNSPEC;
2947 map->def.type = BPF_MAP_TYPE_UNSPEC;
2948 map->sec_idx = sec_idx;
2949 map->sec_offset = vi->offset;
2950 map->btf_var_idx = var_idx;
2951 pr_debug("map '%s': at sec_idx %d, offset %zu.\n",
2952 map_name, map->sec_idx, map->sec_offset);
2953
2954 err = parse_btf_map_def(map->name, obj->btf, def, strict, &map_def, &inner_def);
2955 if (err)
2956 return err;
2957
2958 fill_map_from_def(map, &map_def);
2959
2960 if (map_def.pinning == LIBBPF_PIN_BY_NAME) {
2961 err = build_map_pin_path(map, pin_root_path);
2962 if (err) {
2963 pr_warn("map '%s': couldn't build pin path.\n", map->name);
2964 return err;
2965 }
2966 }
2967
2968 if (map_def.parts & MAP_DEF_INNER_MAP) {
2969 map->inner_map = calloc(1, sizeof(*map->inner_map));
2970 if (!map->inner_map)
2971 return -ENOMEM;
2972 map->inner_map->fd = create_placeholder_fd();
2973 if (map->inner_map->fd < 0)
2974 return map->inner_map->fd;
2975 map->inner_map->sec_idx = sec_idx;
2976 map->inner_map->name = malloc(strlen(map_name) + sizeof(".inner") + 1);
2977 if (!map->inner_map->name)
2978 return -ENOMEM;
2979 sprintf(map->inner_map->name, "%s.inner", map_name);
2980
2981 fill_map_from_def(map->inner_map, &inner_def);
2982 }
2983
2984 err = map_fill_btf_type_info(obj, map);
2985 if (err)
2986 return err;
2987
2988 return 0;
2989 }
2990
2991 static int init_arena_map_data(struct bpf_object *obj, struct bpf_map *map,
2992 const char *sec_name, int sec_idx,
2993 void *data, size_t data_sz)
2994 {
2995 const long page_sz = sysconf(_SC_PAGE_SIZE);
2996 const size_t data_alloc_sz = roundup(data_sz, page_sz);
2997 size_t mmap_sz;
2998
2999 mmap_sz = bpf_map_mmap_sz(map);
3000 if (data_alloc_sz > mmap_sz) {
3001 pr_warn("elf: sec '%s': declared ARENA map size (%zu) is too small to hold global __arena variables of size %zu\n",
3002 sec_name, mmap_sz, data_sz);
3003 return -E2BIG;
3004 }
3005
3006 obj->arena_data = malloc(data_sz);
3007 if (!obj->arena_data)
3008 return -ENOMEM;
3009 memcpy(obj->arena_data, data, data_sz);
3010 obj->arena_data_sz = data_sz;
3011
3012 /* make bpf_map__init_value() work for ARENA maps */
3013 map->mmaped = obj->arena_data;
3014
3015 return 0;
3016 }
3017
3018 static int bpf_object__init_user_btf_maps(struct bpf_object *obj, bool strict,
3019 const char *pin_root_path)
3020 {
3021 const struct btf_type *sec = NULL;
3022 int nr_types, i, vlen, err;
3023 const struct btf_type *t;
3024 const char *name;
3025 Elf_Data *data;
3026 Elf_Scn *scn;
3027
3028 if (obj->efile.btf_maps_shndx < 0)
3029 return 0;
3030
3031 scn = elf_sec_by_idx(obj, obj->efile.btf_maps_shndx);
3032 data = elf_sec_data(obj, scn);
3033 if (!data) {
3034 pr_warn("elf: failed to get %s map definitions for %s\n",
3035 MAPS_ELF_SEC, obj->path);
3036 return -EINVAL;
3037 }
3038
3039 nr_types = btf__type_cnt(obj->btf);
3040 for (i = 1; i < nr_types; i++) {
3041 t = btf__type_by_id(obj->btf, i);
3042 if (!btf_is_datasec(t))
3043 continue;
3044 name = btf__name_by_offset(obj->btf, t->name_off);
3045 if (strcmp(name, MAPS_ELF_SEC) == 0) {
3046 sec = t;
3047 obj->efile.btf_maps_sec_btf_id = i;
3048 break;
3049 }
3050 }
3051
3052 if (!sec) {
3053 pr_warn("DATASEC '%s' not found.\n", MAPS_ELF_SEC);
3054 return -ENOENT;
3055 }
3056
3057 vlen = btf_vlen(sec);
3058 for (i = 0; i < vlen; i++) {
3059 err = bpf_object__init_user_btf_map(obj, sec, i,
3060 obj->efile.btf_maps_shndx,
3061 data, strict,
3062 pin_root_path);
3063 if (err)
3064 return err;
3065 }
3066
3067 for (i = 0; i < obj->nr_maps; i++) {
3068 struct bpf_map *map = &obj->maps[i];
3069
3070 if (map->def.type != BPF_MAP_TYPE_ARENA)
3071 continue;
3072
3073 if (obj->arena_map_idx >= 0) {
3074 pr_warn("map '%s': only single ARENA map is supported (map '%s' is also ARENA)\n",
3075 map->name, obj->maps[obj->arena_map_idx].name);
3076 return -EINVAL;
3077 }
3078 obj->arena_map_idx = i;
3079
3080 if (obj->efile.arena_data) {
3081 err = init_arena_map_data(obj, map, ARENA_SEC, obj->efile.arena_data_shndx,
3082 obj->efile.arena_data->d_buf,
3083 obj->efile.arena_data->d_size);
3084 if (err)
3085 return err;
3086 }
3087 }
3088 if (obj->efile.arena_data && obj->arena_map_idx < 0) {
3089 pr_warn("elf: sec '%s': to use global __arena variables the ARENA map should be explicitly declared in SEC(\".maps\")\n",
3090 ARENA_SEC);
3091 return -ENOENT;
3092 }
3093
3094 return 0;
3095 }
3096
3097 static int bpf_object__init_maps(struct bpf_object *obj,
3098 const struct bpf_object_open_opts *opts)
3099 {
3100 const char *pin_root_path;
3101 bool strict;
3102 int err = 0;
3103
3104 strict = !OPTS_GET(opts, relaxed_maps, false);
3105 pin_root_path = OPTS_GET(opts, pin_root_path, NULL);
3106
3107 err = bpf_object__init_user_btf_maps(obj, strict, pin_root_path);
3108 err = err ?: bpf_object__init_global_data_maps(obj);
3109 err = err ?: bpf_object__init_kconfig_map(obj);
3110 err = err ?: bpf_object_init_struct_ops(obj);
3111
3112 return err;
3113 }
3114
3115 static bool section_have_execinstr(struct bpf_object *obj, int idx)
3116 {
3117 Elf64_Shdr *sh;
3118
3119 sh = elf_sec_hdr(obj, elf_sec_by_idx(obj, idx));
3120 if (!sh)
3121 return false;
3122
3123 return sh->sh_flags & SHF_EXECINSTR;
3124 }
3125
3126 static bool starts_with_qmark(const char *s)
3127 {
3128 return s && s[0] == '?';
3129 }
3130
3131 static bool btf_needs_sanitization(struct bpf_object *obj)
3132 {
3133 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
3134 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
3135 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
3136 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
3137 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
3138 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
3139 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
3140 bool has_qmark_datasec = kernel_supports(obj, FEAT_BTF_QMARK_DATASEC);
3141
3142 return !has_func || !has_datasec || !has_func_global || !has_float ||
3143 !has_decl_tag || !has_type_tag || !has_enum64 || !has_qmark_datasec;
3144 }
3145
3146 static int bpf_object__sanitize_btf(struct bpf_object *obj, struct btf *btf)
3147 {
3148 bool has_func_global = kernel_supports(obj, FEAT_BTF_GLOBAL_FUNC);
3149 bool has_datasec = kernel_supports(obj, FEAT_BTF_DATASEC);
3150 bool has_float = kernel_supports(obj, FEAT_BTF_FLOAT);
3151 bool has_func = kernel_supports(obj, FEAT_BTF_FUNC);
3152 bool has_decl_tag = kernel_supports(obj, FEAT_BTF_DECL_TAG);
3153 bool has_type_tag = kernel_supports(obj, FEAT_BTF_TYPE_TAG);
3154 bool has_enum64 = kernel_supports(obj, FEAT_BTF_ENUM64);
3155 bool has_qmark_datasec = kernel_supports(obj, FEAT_BTF_QMARK_DATASEC);
3156 int enum64_placeholder_id = 0;
3157 struct btf_type *t;
3158 int i, j, vlen;
3159
3160 for (i = 1; i < btf__type_cnt(btf); i++) {
3161 t = (struct btf_type *)btf__type_by_id(btf, i);
3162
3163 if ((!has_datasec && btf_is_var(t)) || (!has_decl_tag && btf_is_decl_tag(t))) {
3164 /* replace VAR/DECL_TAG with INT */
3165 t->info = BTF_INFO_ENC(BTF_KIND_INT, 0, 0);
3166 /*
3167 * using size = 1 is the safest choice, 4 will be too
3168 * big and cause kernel BTF validation failure if
3169 * original variable took less than 4 bytes
3170 */
3171 t->size = 1;
3172 *(int *)(t + 1) = BTF_INT_ENC(0, 0, 8);
3173 } else if (!has_datasec && btf_is_datasec(t)) {
3174 /* replace DATASEC with STRUCT */
3175 const struct btf_var_secinfo *v = btf_var_secinfos(t);
3176 struct btf_member *m = btf_members(t);
3177 struct btf_type *vt;
3178 char *name;
3179
3180 name = (char *)btf__name_by_offset(btf, t->name_off);
3181 while (*name) {
3182 if (*name == '.' || *name == '?')
3183 *name = '_';
3184 name++;
3185 }
3186
3187 vlen = btf_vlen(t);
3188 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, vlen);
3189 for (j = 0; j < vlen; j++, v++, m++) {
3190 /* order of field assignments is important */
3191 m->offset = v->offset * 8;
3192 m->type = v->type;
3193 /* preserve variable name as member name */
3194 vt = (void *)btf__type_by_id(btf, v->type);
3195 m->name_off = vt->name_off;
3196 }
3197 } else if (!has_qmark_datasec && btf_is_datasec(t) &&
3198 starts_with_qmark(btf__name_by_offset(btf, t->name_off))) {
3199 /* replace '?' prefix with '_' for DATASEC names */
3200 char *name;
3201
3202 name = (char *)btf__name_by_offset(btf, t->name_off);
3203 if (name[0] == '?')
3204 name[0] = '_';
3205 } else if (!has_func && btf_is_func_proto(t)) {
3206 /* replace FUNC_PROTO with ENUM */
3207 vlen = btf_vlen(t);
3208 t->info = BTF_INFO_ENC(BTF_KIND_ENUM, 0, vlen);
3209 t->size = sizeof(__u32); /* kernel enforced */
3210 } else if (!has_func && btf_is_func(t)) {
3211 /* replace FUNC with TYPEDEF */
3212 t->info = BTF_INFO_ENC(BTF_KIND_TYPEDEF, 0, 0);
3213 } else if (!has_func_global && btf_is_func(t)) {
3214 /* replace BTF_FUNC_GLOBAL with BTF_FUNC_STATIC */
3215 t->info = BTF_INFO_ENC(BTF_KIND_FUNC, 0, 0);
3216 } else if (!has_float && btf_is_float(t)) {
3217 /* replace FLOAT with an equally-sized empty STRUCT;
3218 * since C compilers do not accept e.g. "float" as a
3219 * valid struct name, make it anonymous
3220 */
3221 t->name_off = 0;
3222 t->info = BTF_INFO_ENC(BTF_KIND_STRUCT, 0, 0);
3223 } else if (!has_type_tag && btf_is_type_tag(t)) {
3224 /* replace TYPE_TAG with a CONST */
3225 t->name_off = 0;
3226 t->info = BTF_INFO_ENC(BTF_KIND_CONST, 0, 0);
3227 } else if (!has_enum64 && btf_is_enum(t)) {
3228 /* clear the kflag */
3229 t->info = btf_type_info(btf_kind(t), btf_vlen(t), false);
3230 } else if (!has_enum64 && btf_is_enum64(t)) {
3231 /* replace ENUM64 with a union */
3232 struct btf_member *m;
3233
3234 if (enum64_placeholder_id == 0) {
3235 enum64_placeholder_id = btf__add_int(btf, "enum64_placeholder", 1, 0);
3236 if (enum64_placeholder_id < 0)
3237 return enum64_placeholder_id;
3238
3239 t = (struct btf_type *)btf__type_by_id(btf, i);
3240 }
3241
3242 m = btf_members(t);
3243 vlen = btf_vlen(t);
3244 t->info = BTF_INFO_ENC(BTF_KIND_UNION, 0, vlen);
3245 for (j = 0; j < vlen; j++, m++) {
3246 m->type = enum64_placeholder_id;
3247 m->offset = 0;
3248 }
3249 }
3250 }
3251
3252 return 0;
3253 }
3254
3255 static bool libbpf_needs_btf(const struct bpf_object *obj)
3256 {
3257 return obj->efile.btf_maps_shndx >= 0 ||
3258 obj->efile.has_st_ops ||
3259 obj->nr_extern > 0;
3260 }
3261
3262 static bool kernel_needs_btf(const struct bpf_object *obj)
3263 {
3264 return obj->efile.has_st_ops;
3265 }
3266
3267 static int bpf_object__init_btf(struct bpf_object *obj,
3268 Elf_Data *btf_data,
3269 Elf_Data *btf_ext_data)
3270 {
3271 int err = -ENOENT;
3272
3273 if (btf_data) {
3274 obj->btf = btf__new(btf_data->d_buf, btf_data->d_size);
3275 err = libbpf_get_error(obj->btf);
3276 if (err) {
3277 obj->btf = NULL;
3278 pr_warn("Error loading ELF section %s: %s.\n", BTF_ELF_SEC, errstr(err));
3279 goto out;
3280 }
3281 /* enforce 8-byte pointers for BPF-targeted BTFs */
3282 btf__set_pointer_size(obj->btf, 8);
3283 }
3284 if (btf_ext_data) {
3285 struct btf_ext_info *ext_segs[3];
3286 int seg_num, sec_num;
3287
3288 if (!obj->btf) {
3289 pr_debug("Ignore ELF section %s because its depending ELF section %s is not found.\n",
3290 BTF_EXT_ELF_SEC, BTF_ELF_SEC);
3291 goto out;
3292 }
3293 obj->btf_ext = btf_ext__new(btf_ext_data->d_buf, btf_ext_data->d_size);
3294 err = libbpf_get_error(obj->btf_ext);
3295 if (err) {
3296 pr_warn("Error loading ELF section %s: %s. Ignored and continue.\n",
3297 BTF_EXT_ELF_SEC, errstr(err));
3298 obj->btf_ext = NULL;
3299 goto out;
3300 }
3301
3302 /* setup .BTF.ext to ELF section mapping */
3303 ext_segs[0] = &obj->btf_ext->func_info;
3304 ext_segs[1] = &obj->btf_ext->line_info;
3305 ext_segs[2] = &obj->btf_ext->core_relo_info;
3306 for (seg_num = 0; seg_num < ARRAY_SIZE(ext_segs); seg_num++) {
3307 struct btf_ext_info *seg = ext_segs[seg_num];
3308 const struct btf_ext_info_sec *sec;
3309 const char *sec_name;
3310 Elf_Scn *scn;
3311
3312 if (seg->sec_cnt == 0)
3313 continue;
3314
3315 seg->sec_idxs = calloc(seg->sec_cnt, sizeof(*seg->sec_idxs));
3316 if (!seg->sec_idxs) {
3317 err = -ENOMEM;
3318 goto out;
3319 }
3320
3321 sec_num = 0;
3322 for_each_btf_ext_sec(seg, sec) {
3323 /* preventively increment index to avoid doing
3324 * this before every continue below
3325 */
3326 sec_num++;
3327
3328 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
3329 if (str_is_empty(sec_name))
3330 continue;
3331 scn = elf_sec_by_name(obj, sec_name);
3332 if (!scn)
3333 continue;
3334
3335 seg->sec_idxs[sec_num - 1] = elf_ndxscn(scn);
3336 }
3337 }
3338 }
3339 out:
3340 if (err && libbpf_needs_btf(obj)) {
3341 pr_warn("BTF is required, but is missing or corrupted.\n");
3342 return err;
3343 }
3344 return 0;
3345 }
3346
3347 static int compare_vsi_off(const void *_a, const void *_b)
3348 {
3349 const struct btf_var_secinfo *a = _a;
3350 const struct btf_var_secinfo *b = _b;
3351
3352 return a->offset - b->offset;
3353 }
3354
3355 static int btf_fixup_datasec(struct bpf_object *obj, struct btf *btf,
3356 struct btf_type *t)
3357 {
3358 __u32 size = 0, i, vars = btf_vlen(t);
3359 const char *sec_name = btf__name_by_offset(btf, t->name_off);
3360 struct btf_var_secinfo *vsi;
3361 bool fixup_offsets = false;
3362 int err;
3363
3364 if (!sec_name) {
3365 pr_debug("No name found in string section for DATASEC kind.\n");
3366 return -ENOENT;
3367 }
3368
3369 /* Extern-backing datasecs (.ksyms, .kconfig) have their size and
3370 * variable offsets set at the previous step. Further, not every
3371 * extern BTF VAR has corresponding ELF symbol preserved, so we skip
3372 * all fixups altogether for such sections and go straight to sorting
3373 * VARs within their DATASEC.
3374 */
3375 if (strcmp(sec_name, KCONFIG_SEC) == 0 || strcmp(sec_name, KSYMS_SEC) == 0)
3376 goto sort_vars;
3377
3378 /* Clang leaves DATASEC size and VAR offsets as zeroes, so we need to
3379 * fix this up. But BPF static linker already fixes this up and fills
3380 * all the sizes and offsets during static linking. So this step has
3381 * to be optional. But the STV_HIDDEN handling is non-optional for any
3382 * non-extern DATASEC, so the variable fixup loop below handles both
3383 * functions at the same time, paying the cost of BTF VAR <-> ELF
3384 * symbol matching just once.
3385 */
3386 if (t->size == 0) {
3387 err = find_elf_sec_sz(obj, sec_name, &size);
3388 if (err || !size) {
3389 pr_debug("sec '%s': failed to determine size from ELF: size %u, err %s\n",
3390 sec_name, size, errstr(err));
3391 return -ENOENT;
3392 }
3393
3394 t->size = size;
3395 fixup_offsets = true;
3396 }
3397
3398 for (i = 0, vsi = btf_var_secinfos(t); i < vars; i++, vsi++) {
3399 const struct btf_type *t_var;
3400 struct btf_var *var;
3401 const char *var_name;
3402 Elf64_Sym *sym;
3403
3404 t_var = btf__type_by_id(btf, vsi->type);
3405 if (!t_var || !btf_is_var(t_var)) {
3406 pr_debug("sec '%s': unexpected non-VAR type found\n", sec_name);
3407 return -EINVAL;
3408 }
3409
3410 var = btf_var(t_var);
3411 if (var->linkage == BTF_VAR_STATIC || var->linkage == BTF_VAR_GLOBAL_EXTERN)
3412 continue;
3413
3414 var_name = btf__name_by_offset(btf, t_var->name_off);
3415 if (!var_name) {
3416 pr_debug("sec '%s': failed to find name of DATASEC's member #%d\n",
3417 sec_name, i);
3418 return -ENOENT;
3419 }
3420
3421 sym = find_elf_var_sym(obj, var_name);
3422 if (IS_ERR(sym)) {
3423 pr_debug("sec '%s': failed to find ELF symbol for VAR '%s'\n",
3424 sec_name, var_name);
3425 return -ENOENT;
3426 }
3427
3428 if (fixup_offsets)
3429 vsi->offset = sym->st_value;
3430
3431 /* if variable is a global/weak symbol, but has restricted
3432 * (STV_HIDDEN or STV_INTERNAL) visibility, mark its BTF VAR
3433 * as static. This follows similar logic for functions (BPF
3434 * subprogs) and influences libbpf's further decisions about
3435 * whether to make global data BPF array maps as
3436 * BPF_F_MMAPABLE.
3437 */
3438 if (ELF64_ST_VISIBILITY(sym->st_other) == STV_HIDDEN
3439 || ELF64_ST_VISIBILITY(sym->st_other) == STV_INTERNAL)
3440 var->linkage = BTF_VAR_STATIC;
3441 }
3442
3443 sort_vars:
3444 qsort(btf_var_secinfos(t), vars, sizeof(*vsi), compare_vsi_off);
3445 return 0;
3446 }
3447
3448 static int bpf_object_fixup_btf(struct bpf_object *obj)
3449 {
3450 int i, n, err = 0;
3451
3452 if (!obj->btf)
3453 return 0;
3454
3455 n = btf__type_cnt(obj->btf);
3456 for (i = 1; i < n; i++) {
3457 struct btf_type *t = btf_type_by_id(obj->btf, i);
3458
3459 /* Loader needs to fix up some of the things compiler
3460 * couldn't get its hands on while emitting BTF. This
3461 * is section size and global variable offset. We use
3462 * the info from the ELF itself for this purpose.
3463 */
3464 if (btf_is_datasec(t)) {
3465 err = btf_fixup_datasec(obj, obj->btf, t);
3466 if (err)
3467 return err;
3468 }
3469 }
3470
3471 return 0;
3472 }
3473
3474 static bool prog_needs_vmlinux_btf(struct bpf_program *prog)
3475 {
3476 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS ||
3477 prog->type == BPF_PROG_TYPE_LSM)
3478 return true;
3479
3480 /* BPF_PROG_TYPE_TRACING programs which do not attach to other programs
3481 * also need vmlinux BTF
3482 */
3483 if (prog->type == BPF_PROG_TYPE_TRACING && !prog->attach_prog_fd)
3484 return true;
3485
3486 return false;
3487 }
3488
3489 static bool map_needs_vmlinux_btf(struct bpf_map *map)
3490 {
3491 return bpf_map__is_struct_ops(map);
3492 }
3493
3494 static bool obj_needs_vmlinux_btf(const struct bpf_object *obj)
3495 {
3496 struct bpf_program *prog;
3497 struct bpf_map *map;
3498 int i;
3499
3500 /* CO-RE relocations need kernel BTF, only when btf_custom_path
3501 * is not specified
3502 */
3503 if (obj->btf_ext && obj->btf_ext->core_relo_info.len && !obj->btf_custom_path)
3504 return true;
3505
3506 /* Support for typed ksyms needs kernel BTF */
3507 for (i = 0; i < obj->nr_extern; i++) {
3508 const struct extern_desc *ext;
3509
3510 ext = &obj->externs[i];
3511 if (ext->type == EXT_KSYM && ext->ksym.type_id)
3512 return true;
3513 }
3514
3515 bpf_object__for_each_program(prog, obj) {
3516 if (!prog->autoload)
3517 continue;
3518 if (prog_needs_vmlinux_btf(prog))
3519 return true;
3520 }
3521
3522 bpf_object__for_each_map(map, obj) {
3523 if (map_needs_vmlinux_btf(map))
3524 return true;
3525 }
3526
3527 return false;
3528 }
3529
3530 static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
3531 {
3532 int err;
3533
3534 /* btf_vmlinux could be loaded earlier */
3535 if (obj->btf_vmlinux || obj->gen_loader)
3536 return 0;
3537
3538 if (!force && !obj_needs_vmlinux_btf(obj))
3539 return 0;
3540
3541 obj->btf_vmlinux = btf__load_vmlinux_btf();
3542 err = libbpf_get_error(obj->btf_vmlinux);
3543 if (err) {
3544 pr_warn("Error loading vmlinux BTF: %s\n", errstr(err));
3545 obj->btf_vmlinux = NULL;
3546 return err;
3547 }
3548 return 0;
3549 }
3550
3551 static int bpf_object__sanitize_and_load_btf(struct bpf_object *obj)
3552 {
3553 struct btf *kern_btf = obj->btf;
3554 bool btf_mandatory, sanitize;
3555 int i, err = 0;
3556
3557 if (!obj->btf)
3558 return 0;
3559
3560 if (!kernel_supports(obj, FEAT_BTF)) {
3561 if (kernel_needs_btf(obj)) {
3562 err = -EOPNOTSUPP;
3563 goto report;
3564 }
3565 pr_debug("Kernel doesn't support BTF, skipping uploading it.\n");
3566 return 0;
3567 }
3568
3569 /* Even though some subprogs are global/weak, user might prefer more
3570 * permissive BPF verification process that BPF verifier performs for
3571 * static functions, taking into account more context from the caller
3572 * functions. In such case, they need to mark such subprogs with
3573 * __attribute__((visibility("hidden"))) and libbpf will adjust
3574 * corresponding FUNC BTF type to be marked as static and trigger more
3575 * involved BPF verification process.
3576 */
3577 for (i = 0; i < obj->nr_programs; i++) {
3578 struct bpf_program *prog = &obj->programs[i];
3579 struct btf_type *t;
3580 const char *name;
3581 int j, n;
3582
3583 if (!prog->mark_btf_static || !prog_is_subprog(obj, prog))
3584 continue;
3585
3586 n = btf__type_cnt(obj->btf);
3587 for (j = 1; j < n; j++) {
3588 t = btf_type_by_id(obj->btf, j);
3589 if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL)
3590 continue;
3591
3592 name = btf__str_by_offset(obj->btf, t->name_off);
3593 if (strcmp(name, prog->name) != 0)
3594 continue;
3595
3596 t->info = btf_type_info(BTF_KIND_FUNC, BTF_FUNC_STATIC, 0);
3597 break;
3598 }
3599 }
3600
3601 sanitize = btf_needs_sanitization(obj);
3602 if (sanitize) {
3603 const void *raw_data;
3604 __u32 sz;
3605
3606 /* clone BTF to sanitize a copy and leave the original intact */
3607 raw_data = btf__raw_data(obj->btf, &sz);
3608 kern_btf = btf__new(raw_data, sz);
3609 err = libbpf_get_error(kern_btf);
3610 if (err)
3611 return err;
3612
3613 /* enforce 8-byte pointers for BPF-targeted BTFs */
3614 btf__set_pointer_size(obj->btf, 8);
3615 err = bpf_object__sanitize_btf(obj, kern_btf);
3616 if (err)
3617 return err;
3618 }
3619
3620 if (obj->gen_loader) {
3621 __u32 raw_size = 0;
3622 const void *raw_data = btf__raw_data(kern_btf, &raw_size);
3623
3624 if (!raw_data)
3625 return -ENOMEM;
3626 bpf_gen__load_btf(obj->gen_loader, raw_data, raw_size);
3627 /* Pretend to have valid FD to pass various fd >= 0 checks.
3628 * This fd == 0 will not be used with any syscall and will be reset to -1 eventually.
3629 */
3630 btf__set_fd(kern_btf, 0);
3631 } else {
3632 /* currently BPF_BTF_LOAD only supports log_level 1 */
3633 err = btf_load_into_kernel(kern_btf, obj->log_buf, obj->log_size,
3634 obj->log_level ? 1 : 0, obj->token_fd);
3635 }
3636 if (sanitize) {
3637 if (!err) {
3638 /* move fd to libbpf's BTF */
3639 btf__set_fd(obj->btf, btf__fd(kern_btf));
3640 btf__set_fd(kern_btf, -1);
3641 }
3642 btf__free(kern_btf);
3643 }
3644 report:
3645 if (err) {
3646 btf_mandatory = kernel_needs_btf(obj);
3647 if (btf_mandatory) {
3648 pr_warn("Error loading .BTF into kernel: %s. BTF is mandatory, can't proceed.\n",
3649 errstr(err));
3650 } else {
3651 pr_info("Error loading .BTF into kernel: %s. BTF is optional, ignoring.\n",
3652 errstr(err));
3653 err = 0;
3654 }
3655 }
3656 return err;
3657 }
3658
3659 static const char *elf_sym_str(const struct bpf_object *obj, size_t off)
3660 {
3661 const char *name;
3662
3663 name = elf_strptr(obj->efile.elf, obj->efile.strtabidx, off);
3664 if (!name) {
3665 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3666 off, obj->path, elf_errmsg(-1));
3667 return NULL;
3668 }
3669
3670 return name;
3671 }
3672
3673 static const char *elf_sec_str(const struct bpf_object *obj, size_t off)
3674 {
3675 const char *name;
3676
3677 name = elf_strptr(obj->efile.elf, obj->efile.shstrndx, off);
3678 if (!name) {
3679 pr_warn("elf: failed to get section name string at offset %zu from %s: %s\n",
3680 off, obj->path, elf_errmsg(-1));
3681 return NULL;
3682 }
3683
3684 return name;
3685 }
3686
3687 static Elf_Scn *elf_sec_by_idx(const struct bpf_object *obj, size_t idx)
3688 {
3689 Elf_Scn *scn;
3690
3691 scn = elf_getscn(obj->efile.elf, idx);
3692 if (!scn) {
3693 pr_warn("elf: failed to get section(%zu) from %s: %s\n",
3694 idx, obj->path, elf_errmsg(-1));
3695 return NULL;
3696 }
3697 return scn;
3698 }
3699
3700 static Elf_Scn *elf_sec_by_name(const struct bpf_object *obj, const char *name)
3701 {
3702 Elf_Scn *scn = NULL;
3703 Elf *elf = obj->efile.elf;
3704 const char *sec_name;
3705
3706 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3707 sec_name = elf_sec_name(obj, scn);
3708 if (!sec_name)
3709 return NULL;
3710
3711 if (strcmp(sec_name, name) != 0)
3712 continue;
3713
3714 return scn;
3715 }
3716 return NULL;
3717 }
3718
3719 static Elf64_Shdr *elf_sec_hdr(const struct bpf_object *obj, Elf_Scn *scn)
3720 {
3721 Elf64_Shdr *shdr;
3722
3723 if (!scn)
3724 return NULL;
3725
3726 shdr = elf64_getshdr(scn);
3727 if (!shdr) {
3728 pr_warn("elf: failed to get section(%zu) header from %s: %s\n",
3729 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3730 return NULL;
3731 }
3732
3733 return shdr;
3734 }
3735
3736 static const char *elf_sec_name(const struct bpf_object *obj, Elf_Scn *scn)
3737 {
3738 const char *name;
3739 Elf64_Shdr *sh;
3740
3741 if (!scn)
3742 return NULL;
3743
3744 sh = elf_sec_hdr(obj, scn);
3745 if (!sh)
3746 return NULL;
3747
3748 name = elf_sec_str(obj, sh->sh_name);
3749 if (!name) {
3750 pr_warn("elf: failed to get section(%zu) name from %s: %s\n",
3751 elf_ndxscn(scn), obj->path, elf_errmsg(-1));
3752 return NULL;
3753 }
3754
3755 return name;
3756 }
3757
3758 static Elf_Data *elf_sec_data(const struct bpf_object *obj, Elf_Scn *scn)
3759 {
3760 Elf_Data *data;
3761
3762 if (!scn)
3763 return NULL;
3764
3765 data = elf_getdata(scn, 0);
3766 if (!data) {
3767 pr_warn("elf: failed to get section(%zu) %s data from %s: %s\n",
3768 elf_ndxscn(scn), elf_sec_name(obj, scn) ?: "<?>",
3769 obj->path, elf_errmsg(-1));
3770 return NULL;
3771 }
3772
3773 return data;
3774 }
3775
3776 static Elf64_Sym *elf_sym_by_idx(const struct bpf_object *obj, size_t idx)
3777 {
3778 if (idx >= obj->efile.symbols->d_size / sizeof(Elf64_Sym))
3779 return NULL;
3780
3781 return (Elf64_Sym *)obj->efile.symbols->d_buf + idx;
3782 }
3783
3784 static Elf64_Rel *elf_rel_by_idx(Elf_Data *data, size_t idx)
3785 {
3786 if (idx >= data->d_size / sizeof(Elf64_Rel))
3787 return NULL;
3788
3789 return (Elf64_Rel *)data->d_buf + idx;
3790 }
3791
3792 static bool is_sec_name_dwarf(const char *name)
3793 {
3794 /* approximation, but the actual list is too long */
3795 return str_has_pfx(name, ".debug_");
3796 }
3797
3798 static bool ignore_elf_section(Elf64_Shdr *hdr, const char *name)
3799 {
3800 /* no special handling of .strtab */
3801 if (hdr->sh_type == SHT_STRTAB)
3802 return true;
3803
3804 /* ignore .llvm_addrsig section as well */
3805 if (hdr->sh_type == SHT_LLVM_ADDRSIG)
3806 return true;
3807
3808 /* no subprograms will lead to an empty .text section, ignore it */
3809 if (hdr->sh_type == SHT_PROGBITS && hdr->sh_size == 0 &&
3810 strcmp(name, ".text") == 0)
3811 return true;
3812
3813 /* DWARF sections */
3814 if (is_sec_name_dwarf(name))
3815 return true;
3816
3817 if (str_has_pfx(name, ".rel")) {
3818 name += sizeof(".rel") - 1;
3819 /* DWARF section relocations */
3820 if (is_sec_name_dwarf(name))
3821 return true;
3822
3823 /* .BTF and .BTF.ext don't need relocations */
3824 if (strcmp(name, BTF_ELF_SEC) == 0 ||
3825 strcmp(name, BTF_EXT_ELF_SEC) == 0)
3826 return true;
3827 }
3828
3829 return false;
3830 }
3831
3832 static int cmp_progs(const void *_a, const void *_b)
3833 {
3834 const struct bpf_program *a = _a;
3835 const struct bpf_program *b = _b;
3836
3837 if (a->sec_idx != b->sec_idx)
3838 return a->sec_idx < b->sec_idx ? -1 : 1;
3839
3840 /* sec_insn_off can't be the same within the section */
3841 return a->sec_insn_off < b->sec_insn_off ? -1 : 1;
3842 }
3843
3844 static int bpf_object__elf_collect(struct bpf_object *obj)
3845 {
3846 struct elf_sec_desc *sec_desc;
3847 Elf *elf = obj->efile.elf;
3848 Elf_Data *btf_ext_data = NULL;
3849 Elf_Data *btf_data = NULL;
3850 int idx = 0, err = 0;
3851 const char *name;
3852 Elf_Data *data;
3853 Elf_Scn *scn;
3854 Elf64_Shdr *sh;
3855
3856 /* ELF section indices are 0-based, but sec #0 is special "invalid"
3857 * section. Since section count retrieved by elf_getshdrnum() does
3858 * include sec #0, it is already the necessary size of an array to keep
3859 * all the sections.
3860 */
3861 if (elf_getshdrnum(obj->efile.elf, &obj->efile.sec_cnt)) {
3862 pr_warn("elf: failed to get the number of sections for %s: %s\n",
3863 obj->path, elf_errmsg(-1));
3864 return -LIBBPF_ERRNO__FORMAT;
3865 }
3866 obj->efile.secs = calloc(obj->efile.sec_cnt, sizeof(*obj->efile.secs));
3867 if (!obj->efile.secs)
3868 return -ENOMEM;
3869
3870 /* a bunch of ELF parsing functionality depends on processing symbols,
3871 * so do the first pass and find the symbol table
3872 */
3873 scn = NULL;
3874 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3875 sh = elf_sec_hdr(obj, scn);
3876 if (!sh)
3877 return -LIBBPF_ERRNO__FORMAT;
3878
3879 if (sh->sh_type == SHT_SYMTAB) {
3880 if (obj->efile.symbols) {
3881 pr_warn("elf: multiple symbol tables in %s\n", obj->path);
3882 return -LIBBPF_ERRNO__FORMAT;
3883 }
3884
3885 data = elf_sec_data(obj, scn);
3886 if (!data)
3887 return -LIBBPF_ERRNO__FORMAT;
3888
3889 idx = elf_ndxscn(scn);
3890
3891 obj->efile.symbols = data;
3892 obj->efile.symbols_shndx = idx;
3893 obj->efile.strtabidx = sh->sh_link;
3894 }
3895 }
3896
3897 if (!obj->efile.symbols) {
3898 pr_warn("elf: couldn't find symbol table in %s, stripped object file?\n",
3899 obj->path);
3900 return -ENOENT;
3901 }
3902
3903 scn = NULL;
3904 while ((scn = elf_nextscn(elf, scn)) != NULL) {
3905 idx = elf_ndxscn(scn);
3906 sec_desc = &obj->efile.secs[idx];
3907
3908 sh = elf_sec_hdr(obj, scn);
3909 if (!sh)
3910 return -LIBBPF_ERRNO__FORMAT;
3911
3912 name = elf_sec_str(obj, sh->sh_name);
3913 if (!name)
3914 return -LIBBPF_ERRNO__FORMAT;
3915
3916 if (ignore_elf_section(sh, name))
3917 continue;
3918
3919 data = elf_sec_data(obj, scn);
3920 if (!data)
3921 return -LIBBPF_ERRNO__FORMAT;
3922
3923 pr_debug("elf: section(%d) %s, size %ld, link %d, flags %lx, type=%d\n",
3924 idx, name, (unsigned long)data->d_size,
3925 (int)sh->sh_link, (unsigned long)sh->sh_flags,
3926 (int)sh->sh_type);
3927
3928 if (strcmp(name, "license") == 0) {
3929 err = bpf_object__init_license(obj, data->d_buf, data->d_size);
3930 if (err)
3931 return err;
3932 } else if (strcmp(name, "version") == 0) {
3933 err = bpf_object__init_kversion(obj, data->d_buf, data->d_size);
3934 if (err)
3935 return err;
3936 } else if (strcmp(name, "maps") == 0) {
3937 pr_warn("elf: legacy map definitions in 'maps' section are not supported by libbpf v1.0+\n");
3938 return -ENOTSUP;
3939 } else if (strcmp(name, MAPS_ELF_SEC) == 0) {
3940 obj->efile.btf_maps_shndx = idx;
3941 } else if (strcmp(name, BTF_ELF_SEC) == 0) {
3942 if (sh->sh_type != SHT_PROGBITS)
3943 return -LIBBPF_ERRNO__FORMAT;
3944 btf_data = data;
3945 } else if (strcmp(name, BTF_EXT_ELF_SEC) == 0) {
3946 if (sh->sh_type != SHT_PROGBITS)
3947 return -LIBBPF_ERRNO__FORMAT;
3948 btf_ext_data = data;
3949 } else if (sh->sh_type == SHT_SYMTAB) {
3950 /* already processed during the first pass above */
3951 } else if (sh->sh_type == SHT_PROGBITS && data->d_size > 0) {
3952 if (sh->sh_flags & SHF_EXECINSTR) {
3953 if (strcmp(name, ".text") == 0)
3954 obj->efile.text_shndx = idx;
3955 err = bpf_object__add_programs(obj, data, name, idx);
3956 if (err)
3957 return err;
3958 } else if (strcmp(name, DATA_SEC) == 0 ||
3959 str_has_pfx(name, DATA_SEC ".")) {
3960 sec_desc->sec_type = SEC_DATA;
3961 sec_desc->shdr = sh;
3962 sec_desc->data = data;
3963 } else if (strcmp(name, RODATA_SEC) == 0 ||
3964 str_has_pfx(name, RODATA_SEC ".")) {
3965 sec_desc->sec_type = SEC_RODATA;
3966 sec_desc->shdr = sh;
3967 sec_desc->data = data;
3968 } else if (strcmp(name, STRUCT_OPS_SEC) == 0 ||
3969 strcmp(name, STRUCT_OPS_LINK_SEC) == 0 ||
3970 strcmp(name, "?" STRUCT_OPS_SEC) == 0 ||
3971 strcmp(name, "?" STRUCT_OPS_LINK_SEC) == 0) {
3972 sec_desc->sec_type = SEC_ST_OPS;
3973 sec_desc->shdr = sh;
3974 sec_desc->data = data;
3975 obj->efile.has_st_ops = true;
3976 } else if (strcmp(name, ARENA_SEC) == 0) {
3977 obj->efile.arena_data = data;
3978 obj->efile.arena_data_shndx = idx;
3979 } else if (strcmp(name, JUMPTABLES_SEC) == 0) {
3980 obj->jumptables_data = malloc(data->d_size);
3981 if (!obj->jumptables_data)
3982 return -ENOMEM;
3983 memcpy(obj->jumptables_data, data->d_buf, data->d_size);
3984 obj->jumptables_data_sz = data->d_size;
3985 obj->efile.jumptables_data_shndx = idx;
3986 } else {
3987 pr_info("elf: skipping unrecognized data section(%d) %s\n",
3988 idx, name);
3989 }
3990 } else if (sh->sh_type == SHT_REL) {
3991 int targ_sec_idx = sh->sh_info; /* points to other section */
3992
3993 if (sh->sh_entsize != sizeof(Elf64_Rel) ||
3994 targ_sec_idx >= obj->efile.sec_cnt)
3995 return -LIBBPF_ERRNO__FORMAT;
3996
3997 /* Only do relo for section with exec instructions */
3998 if (!section_have_execinstr(obj, targ_sec_idx) &&
3999 strcmp(name, ".rel" STRUCT_OPS_SEC) &&
4000 strcmp(name, ".rel" STRUCT_OPS_LINK_SEC) &&
4001 strcmp(name, ".rel?" STRUCT_OPS_SEC) &&
4002 strcmp(name, ".rel?" STRUCT_OPS_LINK_SEC) &&
4003 strcmp(name, ".rel" MAPS_ELF_SEC)) {
4004 pr_info("elf: skipping relo section(%d) %s for section(%d) %s\n",
4005 idx, name, targ_sec_idx,
4006 elf_sec_name(obj, elf_sec_by_idx(obj, targ_sec_idx)) ?: "<?>");
4007 continue;
4008 }
4009
4010 sec_desc->sec_type = SEC_RELO;
4011 sec_desc->shdr = sh;
4012 sec_desc->data = data;
4013 } else if (sh->sh_type == SHT_NOBITS && (strcmp(name, BSS_SEC) == 0 ||
4014 str_has_pfx(name, BSS_SEC "."))) {
4015 sec_desc->sec_type = SEC_BSS;
4016 sec_desc->shdr = sh;
4017 sec_desc->data = data;
4018 } else {
4019 pr_info("elf: skipping section(%d) %s (size %zu)\n", idx, name,
4020 (size_t)sh->sh_size);
4021 }
4022 }
4023
4024 if (!obj->efile.strtabidx || obj->efile.strtabidx > idx) {
4025 pr_warn("elf: symbol strings section missing or invalid in %s\n", obj->path);
4026 return -LIBBPF_ERRNO__FORMAT;
4027 }
4028
4029 /* change BPF program insns to native endianness for introspection */
4030 if (!is_native_endianness(obj))
4031 bpf_object_bswap_progs(obj);
4032
4033 /* sort BPF programs by section name and in-section instruction offset
4034 * for faster search
4035 */
4036 if (obj->nr_programs)
4037 qsort(obj->programs, obj->nr_programs, sizeof(*obj->programs), cmp_progs);
4038
4039 return bpf_object__init_btf(obj, btf_data, btf_ext_data);
4040 }
4041
4042 static bool sym_is_extern(const Elf64_Sym *sym)
4043 {
4044 int bind = ELF64_ST_BIND(sym->st_info);
4045 /* externs are symbols w/ type=NOTYPE, bind=GLOBAL|WEAK, section=UND */
4046 return sym->st_shndx == SHN_UNDEF &&
4047 (bind == STB_GLOBAL || bind == STB_WEAK) &&
4048 ELF64_ST_TYPE(sym->st_info) == STT_NOTYPE;
4049 }
4050
4051 static bool sym_is_subprog(const Elf64_Sym *sym, int text_shndx)
4052 {
4053 int bind = ELF64_ST_BIND(sym->st_info);
4054 int type = ELF64_ST_TYPE(sym->st_info);
4055
4056 /* in .text section */
4057 if (sym->st_shndx != text_shndx)
4058 return false;
4059
4060 /* local function */
4061 if (bind == STB_LOCAL && type == STT_SECTION)
4062 return true;
4063
4064 /* global function */
4065 return (bind == STB_GLOBAL || bind == STB_WEAK) && type == STT_FUNC;
4066 }
4067
4068 static int find_extern_btf_id(const struct btf *btf, const char *ext_name)
4069 {
4070 const struct btf_type *t;
4071 const char *tname;
4072 int i, n;
4073
4074 if (!btf)
4075 return -ESRCH;
4076
4077 n = btf__type_cnt(btf);
4078 for (i = 1; i < n; i++) {
4079 t = btf__type_by_id(btf, i);
4080
4081 if (!btf_is_var(t) && !btf_is_func(t))
4082 continue;
4083
4084 tname = btf__name_by_offset(btf, t->name_off);
4085 if (strcmp(tname, ext_name))
4086 continue;
4087
4088 if (btf_is_var(t) &&
4089 btf_var(t)->linkage != BTF_VAR_GLOBAL_EXTERN)
4090 return -EINVAL;
4091
4092 if (btf_is_func(t) && btf_func_linkage(t) != BTF_FUNC_EXTERN)
4093 return -EINVAL;
4094
4095 return i;
4096 }
4097
4098 return -ENOENT;
4099 }
4100
4101 static int find_extern_sec_btf_id(struct btf *btf, int ext_btf_id) {
4102 const struct btf_var_secinfo *vs;
4103 const struct btf_type *t;
4104 int i, j, n;
4105
4106 if (!btf)
4107 return -ESRCH;
4108
4109 n = btf__type_cnt(btf);
4110 for (i = 1; i < n; i++) {
4111 t = btf__type_by_id(btf, i);
4112
4113 if (!btf_is_datasec(t))
4114 continue;
4115
4116 vs = btf_var_secinfos(t);
4117 for (j = 0; j < btf_vlen(t); j++, vs++) {
4118 if (vs->type == ext_btf_id)
4119 return i;
4120 }
4121 }
4122
4123 return -ENOENT;
4124 }
4125
4126 static enum kcfg_type find_kcfg_type(const struct btf *btf, int id,
4127 bool *is_signed)
4128 {
4129 const struct btf_type *t;
4130 const char *name;
4131
4132 t = skip_mods_and_typedefs(btf, id, NULL);
4133 name = btf__name_by_offset(btf, t->name_off);
4134
4135 if (is_signed)
4136 *is_signed = false;
4137 switch (btf_kind(t)) {
4138 case BTF_KIND_INT: {
4139 int enc = btf_int_encoding(t);
4140
4141 if (enc & BTF_INT_BOOL)
4142 return t->size == 1 ? KCFG_BOOL : KCFG_UNKNOWN;
4143 if (is_signed)
4144 *is_signed = enc & BTF_INT_SIGNED;
4145 if (t->size == 1)
4146 return KCFG_CHAR;
4147 if (t->size < 1 || t->size > 8 || (t->size & (t->size - 1)))
4148 return KCFG_UNKNOWN;
4149 return KCFG_INT;
4150 }
4151 case BTF_KIND_ENUM:
4152 if (t->size != 4)
4153 return KCFG_UNKNOWN;
4154 if (strcmp(name, "libbpf_tristate"))
4155 return KCFG_UNKNOWN;
4156 return KCFG_TRISTATE;
4157 case BTF_KIND_ENUM64:
4158 if (strcmp(name, "libbpf_tristate"))
4159 return KCFG_UNKNOWN;
4160 return KCFG_TRISTATE;
4161 case BTF_KIND_ARRAY:
4162 if (btf_array(t)->nelems == 0)
4163 return KCFG_UNKNOWN;
4164 if (find_kcfg_type(btf, btf_array(t)->type, NULL) != KCFG_CHAR)
4165 return KCFG_UNKNOWN;
4166 return KCFG_CHAR_ARR;
4167 default:
4168 return KCFG_UNKNOWN;
4169 }
4170 }
4171
4172 static int cmp_externs(const void *_a, const void *_b)
4173 {
4174 const struct extern_desc *a = _a;
4175 const struct extern_desc *b = _b;
4176
4177 if (a->type != b->type)
4178 return a->type < b->type ? -1 : 1;
4179
4180 if (a->type == EXT_KCFG) {
4181 /* descending order by alignment requirements */
4182 if (a->kcfg.align != b->kcfg.align)
4183 return a->kcfg.align > b->kcfg.align ? -1 : 1;
4184 /* ascending order by size, within same alignment class */
4185 if (a->kcfg.sz != b->kcfg.sz)
4186 return a->kcfg.sz < b->kcfg.sz ? -1 : 1;
4187 }
4188
4189 /* resolve ties by name */
4190 return strcmp(a->name, b->name);
4191 }
4192
4193 static int find_int_btf_id(const struct btf *btf)
4194 {
4195 const struct btf_type *t;
4196 int i, n;
4197
4198 n = btf__type_cnt(btf);
4199 for (i = 1; i < n; i++) {
4200 t = btf__type_by_id(btf, i);
4201
4202 if (btf_is_int(t) && btf_int_bits(t) == 32)
4203 return i;
4204 }
4205
4206 return 0;
4207 }
4208
4209 static int add_dummy_ksym_var(struct btf *btf)
4210 {
4211 int i, int_btf_id, sec_btf_id, dummy_var_btf_id;
4212 const struct btf_var_secinfo *vs;
4213 const struct btf_type *sec;
4214
4215 if (!btf)
4216 return 0;
4217
4218 sec_btf_id = btf__find_by_name_kind(btf, KSYMS_SEC,
4219 BTF_KIND_DATASEC);
4220 if (sec_btf_id < 0)
4221 return 0;
4222
4223 sec = btf__type_by_id(btf, sec_btf_id);
4224 vs = btf_var_secinfos(sec);
4225 for (i = 0; i < btf_vlen(sec); i++, vs++) {
4226 const struct btf_type *vt;
4227
4228 vt = btf__type_by_id(btf, vs->type);
4229 if (btf_is_func(vt))
4230 break;
4231 }
4232
4233 /* No func in ksyms sec. No need to add dummy var. */
4234 if (i == btf_vlen(sec))
4235 return 0;
4236
4237 int_btf_id = find_int_btf_id(btf);
4238 dummy_var_btf_id = btf__add_var(btf,
4239 "dummy_ksym",
4240 BTF_VAR_GLOBAL_ALLOCATED,
4241 int_btf_id);
4242 if (dummy_var_btf_id < 0)
4243 pr_warn("cannot create a dummy_ksym var\n");
4244
4245 return dummy_var_btf_id;
4246 }
4247
4248 static int bpf_object__collect_externs(struct bpf_object *obj)
4249 {
4250 struct btf_type *sec, *kcfg_sec = NULL, *ksym_sec = NULL;
4251 const struct btf_type *t;
4252 struct extern_desc *ext;
4253 int i, n, off, dummy_var_btf_id;
4254 const char *ext_name, *sec_name;
4255 size_t ext_essent_len;
4256 Elf_Scn *scn;
4257 Elf64_Shdr *sh;
4258
4259 if (!obj->efile.symbols)
4260 return 0;
4261
4262 scn = elf_sec_by_idx(obj, obj->efile.symbols_shndx);
4263 sh = elf_sec_hdr(obj, scn);
4264 if (!sh || sh->sh_entsize != sizeof(Elf64_Sym))
4265 return -LIBBPF_ERRNO__FORMAT;
4266
4267 dummy_var_btf_id = add_dummy_ksym_var(obj->btf);
4268 if (dummy_var_btf_id < 0)
4269 return dummy_var_btf_id;
4270
4271 n = sh->sh_size / sh->sh_entsize;
4272 pr_debug("looking for externs among %d symbols...\n", n);
4273
4274 for (i = 0; i < n; i++) {
4275 Elf64_Sym *sym = elf_sym_by_idx(obj, i);
4276
4277 if (!sym)
4278 return -LIBBPF_ERRNO__FORMAT;
4279 if (!sym_is_extern(sym))
4280 continue;
4281 ext_name = elf_sym_str(obj, sym->st_name);
4282 if (str_is_empty(ext_name))
4283 continue;
4284
4285 ext = obj->externs;
4286 ext = libbpf_reallocarray(ext, obj->nr_extern + 1, sizeof(*ext));
4287 if (!ext)
4288 return -ENOMEM;
4289 obj->externs = ext;
4290 ext = &ext[obj->nr_extern];
4291 memset(ext, 0, sizeof(*ext));
4292 obj->nr_extern++;
4293
4294 ext->btf_id = find_extern_btf_id(obj->btf, ext_name);
4295 if (ext->btf_id <= 0) {
4296 pr_warn("failed to find BTF for extern '%s': %d\n",
4297 ext_name, ext->btf_id);
4298 return ext->btf_id;
4299 }
4300 t = btf__type_by_id(obj->btf, ext->btf_id);
4301 ext->name = strdup(btf__name_by_offset(obj->btf, t->name_off));
4302 if (!ext->name)
4303 return -ENOMEM;
4304 ext->sym_idx = i;
4305 ext->is_weak = ELF64_ST_BIND(sym->st_info) == STB_WEAK;
4306
4307 ext_essent_len = bpf_core_essential_name_len(ext->name);
4308 ext->essent_name = NULL;
4309 if (ext_essent_len != strlen(ext->name)) {
4310 ext->essent_name = strndup(ext->name, ext_essent_len);
4311 if (!ext->essent_name)
4312 return -ENOMEM;
4313 }
4314
4315 ext->sec_btf_id = find_extern_sec_btf_id(obj->btf, ext->btf_id);
4316 if (ext->sec_btf_id <= 0) {
4317 pr_warn("failed to find BTF for extern '%s' [%d] section: %d\n",
4318 ext_name, ext->btf_id, ext->sec_btf_id);
4319 return ext->sec_btf_id;
4320 }
4321 sec = (void *)btf__type_by_id(obj->btf, ext->sec_btf_id);
4322 sec_name = btf__name_by_offset(obj->btf, sec->name_off);
4323
4324 if (strcmp(sec_name, KCONFIG_SEC) == 0) {
4325 if (btf_is_func(t)) {
4326 pr_warn("extern function %s is unsupported under %s section\n",
4327 ext->name, KCONFIG_SEC);
4328 return -ENOTSUP;
4329 }
4330 kcfg_sec = sec;
4331 ext->type = EXT_KCFG;
4332 ext->kcfg.sz = btf__resolve_size(obj->btf, t->type);
4333 if (ext->kcfg.sz <= 0) {
4334 pr_warn("failed to resolve size of extern (kcfg) '%s': %d\n",
4335 ext_name, ext->kcfg.sz);
4336 return ext->kcfg.sz;
4337 }
4338 ext->kcfg.align = btf__align_of(obj->btf, t->type);
4339 if (ext->kcfg.align <= 0) {
4340 pr_warn("failed to determine alignment of extern (kcfg) '%s': %d\n",
4341 ext_name, ext->kcfg.align);
4342 return -EINVAL;
4343 }
4344 ext->kcfg.type = find_kcfg_type(obj->btf, t->type,
4345 &ext->kcfg.is_signed);
4346 if (ext->kcfg.type == KCFG_UNKNOWN) {
4347 pr_warn("extern (kcfg) '%s': type is unsupported\n", ext_name);
4348 return -ENOTSUP;
4349 }
4350 } else if (strcmp(sec_name, KSYMS_SEC) == 0) {
4351 ksym_sec = sec;
4352 ext->type = EXT_KSYM;
4353 skip_mods_and_typedefs(obj->btf, t->type,
4354 &ext->ksym.type_id);
4355 } else {
4356 pr_warn("unrecognized extern section '%s'\n", sec_name);
4357 return -ENOTSUP;
4358 }
4359 }
4360 pr_debug("collected %d externs total\n", obj->nr_extern);
4361
4362 if (!obj->nr_extern)
4363 return 0;
4364
4365 /* sort externs by type, for kcfg ones also by (align, size, name) */
4366 qsort(obj->externs, obj->nr_extern, sizeof(*ext), cmp_externs);
4367
4368 /* for .ksyms section, we need to turn all externs into allocated
4369 * variables in BTF to pass kernel verification; we do this by
4370 * pretending that each extern is a 8-byte variable
4371 */
4372 if (ksym_sec) {
4373 /* find existing 4-byte integer type in BTF to use for fake
4374 * extern variables in DATASEC
4375 */
4376 int int_btf_id = find_int_btf_id(obj->btf);
4377 /* For extern function, a dummy_var added earlier
4378 * will be used to replace the vs->type and
4379 * its name string will be used to refill
4380 * the missing param's name.
4381 */
4382 const struct btf_type *dummy_var;
4383
4384 dummy_var = btf__type_by_id(obj->btf, dummy_var_btf_id);
4385 for (i = 0; i < obj->nr_extern; i++) {
4386 ext = &obj->externs[i];
4387 if (ext->type != EXT_KSYM)
4388 continue;
4389 pr_debug("extern (ksym) #%d: symbol %d, name %s\n",
4390 i, ext->sym_idx, ext->name);
4391 }
4392
4393 sec = ksym_sec;
4394 n = btf_vlen(sec);
4395 for (i = 0, off = 0; i < n; i++, off += sizeof(int)) {
4396 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4397 struct btf_type *vt;
4398
4399 vt = (void *)btf__type_by_id(obj->btf, vs->type);
4400 ext_name = btf__name_by_offset(obj->btf, vt->name_off);
4401 ext = find_extern_by_name(obj, ext_name);
4402 if (!ext) {
4403 pr_warn("failed to find extern definition for BTF %s '%s'\n",
4404 btf_kind_str(vt), ext_name);
4405 return -ESRCH;
4406 }
4407 if (btf_is_func(vt)) {
4408 const struct btf_type *func_proto;
4409 struct btf_param *param;
4410 int j;
4411
4412 func_proto = btf__type_by_id(obj->btf,
4413 vt->type);
4414 param = btf_params(func_proto);
4415 /* Reuse the dummy_var string if the
4416 * func proto does not have param name.
4417 */
4418 for (j = 0; j < btf_vlen(func_proto); j++)
4419 if (param[j].type && !param[j].name_off)
4420 param[j].name_off =
4421 dummy_var->name_off;
4422 vs->type = dummy_var_btf_id;
4423 vt->info &= ~0xffff;
4424 vt->info |= BTF_FUNC_GLOBAL;
4425 } else {
4426 btf_var(vt)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4427 vt->type = int_btf_id;
4428 }
4429 vs->offset = off;
4430 vs->size = sizeof(int);
4431 }
4432 sec->size = off;
4433 }
4434
4435 if (kcfg_sec) {
4436 sec = kcfg_sec;
4437 /* for kcfg externs calculate their offsets within a .kconfig map */
4438 off = 0;
4439 for (i = 0; i < obj->nr_extern; i++) {
4440 ext = &obj->externs[i];
4441 if (ext->type != EXT_KCFG)
4442 continue;
4443
4444 ext->kcfg.data_off = roundup(off, ext->kcfg.align);
4445 off = ext->kcfg.data_off + ext->kcfg.sz;
4446 pr_debug("extern (kcfg) #%d: symbol %d, off %u, name %s\n",
4447 i, ext->sym_idx, ext->kcfg.data_off, ext->name);
4448 }
4449 sec->size = off;
4450 n = btf_vlen(sec);
4451 for (i = 0; i < n; i++) {
4452 struct btf_var_secinfo *vs = btf_var_secinfos(sec) + i;
4453
4454 t = btf__type_by_id(obj->btf, vs->type);
4455 ext_name = btf__name_by_offset(obj->btf, t->name_off);
4456 ext = find_extern_by_name(obj, ext_name);
4457 if (!ext) {
4458 pr_warn("failed to find extern definition for BTF var '%s'\n",
4459 ext_name);
4460 return -ESRCH;
4461 }
4462 btf_var(t)->linkage = BTF_VAR_GLOBAL_ALLOCATED;
4463 vs->offset = ext->kcfg.data_off;
4464 }
4465 }
4466 return 0;
4467 }
4468
4469 static bool prog_is_subprog(const struct bpf_object *obj, const struct bpf_program *prog)
4470 {
4471 return prog->sec_idx == obj->efile.text_shndx;
4472 }
4473
4474 struct bpf_program *
4475 bpf_object__find_program_by_name(const struct bpf_object *obj,
4476 const char *name)
4477 {
4478 struct bpf_program *prog;
4479
4480 bpf_object__for_each_program(prog, obj) {
4481 if (prog_is_subprog(obj, prog))
4482 continue;
4483 if (!strcmp(prog->name, name))
4484 return prog;
4485 }
4486 return errno = ENOENT, NULL;
4487 }
4488
4489 static bool bpf_object__shndx_is_data(const struct bpf_object *obj,
4490 int shndx)
4491 {
4492 switch (obj->efile.secs[shndx].sec_type) {
4493 case SEC_BSS:
4494 case SEC_DATA:
4495 case SEC_RODATA:
4496 return true;
4497 default:
4498 return false;
4499 }
4500 }
4501
4502 static bool bpf_object__shndx_is_maps(const struct bpf_object *obj,
4503 int shndx)
4504 {
4505 return shndx == obj->efile.btf_maps_shndx;
4506 }
4507
4508 static enum libbpf_map_type
4509 bpf_object__section_to_libbpf_map_type(const struct bpf_object *obj, int shndx)
4510 {
4511 if (shndx == obj->efile.symbols_shndx)
4512 return LIBBPF_MAP_KCONFIG;
4513
4514 switch (obj->efile.secs[shndx].sec_type) {
4515 case SEC_BSS:
4516 return LIBBPF_MAP_BSS;
4517 case SEC_DATA:
4518 return LIBBPF_MAP_DATA;
4519 case SEC_RODATA:
4520 return LIBBPF_MAP_RODATA;
4521 default:
4522 return LIBBPF_MAP_UNSPEC;
4523 }
4524 }
4525
4526 static int bpf_prog_compute_hash(struct bpf_program *prog)
4527 {
4528 struct bpf_insn *purged;
4529 int i, err = 0;
4530
4531 purged = calloc(prog->insns_cnt, BPF_INSN_SZ);
4532 if (!purged)
4533 return -ENOMEM;
4534
4535 /* If relocations have been done, the map_fd needs to be
4536 * discarded for the digest calculation.
4537 */
4538 for (i = 0; i < prog->insns_cnt; i++) {
4539 purged[i] = prog->insns[i];
4540 if (purged[i].code == (BPF_LD | BPF_IMM | BPF_DW) &&
4541 (purged[i].src_reg == BPF_PSEUDO_MAP_FD ||
4542 purged[i].src_reg == BPF_PSEUDO_MAP_VALUE)) {
4543 purged[i].imm = 0;
4544 i++;
4545 if (i >= prog->insns_cnt ||
4546 prog->insns[i].code != 0 ||
4547 prog->insns[i].dst_reg != 0 ||
4548 prog->insns[i].src_reg != 0 ||
4549 prog->insns[i].off != 0) {
4550 err = -EINVAL;
4551 goto out;
4552 }
4553 purged[i] = prog->insns[i];
4554 purged[i].imm = 0;
4555 }
4556 }
4557 libbpf_sha256(purged, prog->insns_cnt * sizeof(struct bpf_insn),
4558 prog->hash);
4559 out:
4560 free(purged);
4561 return err;
4562 }
4563
4564 static int bpf_program__record_reloc(struct bpf_program *prog,
4565 struct reloc_desc *reloc_desc,
4566 __u32 insn_idx, const char *sym_name,
4567 const Elf64_Sym *sym, const Elf64_Rel *rel)
4568 {
4569 struct bpf_insn *insn = &prog->insns[insn_idx];
4570 size_t map_idx, nr_maps = prog->obj->nr_maps;
4571 struct bpf_object *obj = prog->obj;
4572 __u32 shdr_idx = sym->st_shndx;
4573 enum libbpf_map_type type;
4574 const char *sym_sec_name;
4575 struct bpf_map *map;
4576
4577 if (!is_call_insn(insn) && !is_ldimm64_insn(insn)) {
4578 pr_warn("prog '%s': invalid relo against '%s' for insns[%d].code 0x%x\n",
4579 prog->name, sym_name, insn_idx, insn->code);
4580 return -LIBBPF_ERRNO__RELOC;
4581 }
4582
4583 if (sym_is_extern(sym)) {
4584 int sym_idx = ELF64_R_SYM(rel->r_info);
4585 int i, n = obj->nr_extern;
4586 struct extern_desc *ext;
4587
4588 for (i = 0; i < n; i++) {
4589 ext = &obj->externs[i];
4590 if (ext->sym_idx == sym_idx)
4591 break;
4592 }
4593 if (i >= n) {
4594 pr_warn("prog '%s': extern relo failed to find extern for '%s' (%d)\n",
4595 prog->name, sym_name, sym_idx);
4596 return -LIBBPF_ERRNO__RELOC;
4597 }
4598 pr_debug("prog '%s': found extern #%d '%s' (sym %d) for insn #%u\n",
4599 prog->name, i, ext->name, ext->sym_idx, insn_idx);
4600 if (insn->code == (BPF_JMP | BPF_CALL))
4601 reloc_desc->type = RELO_EXTERN_CALL;
4602 else
4603 reloc_desc->type = RELO_EXTERN_LD64;
4604 reloc_desc->insn_idx = insn_idx;
4605 reloc_desc->ext_idx = i;
4606 return 0;
4607 }
4608
4609 /* sub-program call relocation */
4610 if (is_call_insn(insn)) {
4611 if (insn->src_reg != BPF_PSEUDO_CALL) {
4612 pr_warn("prog '%s': incorrect bpf_call opcode\n", prog->name);
4613 return -LIBBPF_ERRNO__RELOC;
4614 }
4615 /* text_shndx can be 0, if no default "main" program exists */
4616 if (!shdr_idx || shdr_idx != obj->efile.text_shndx) {
4617 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4618 pr_warn("prog '%s': bad call relo against '%s' in section '%s'\n",
4619 prog->name, sym_name, sym_sec_name);
4620 return -LIBBPF_ERRNO__RELOC;
4621 }
4622 if (sym->st_value % BPF_INSN_SZ) {
4623 pr_warn("prog '%s': bad call relo against '%s' at offset %zu\n",
4624 prog->name, sym_name, (size_t)sym->st_value);
4625 return -LIBBPF_ERRNO__RELOC;
4626 }
4627 reloc_desc->type = RELO_CALL;
4628 reloc_desc->insn_idx = insn_idx;
4629 reloc_desc->sym_off = sym->st_value;
4630 return 0;
4631 }
4632
4633 if (!shdr_idx || shdr_idx >= SHN_LORESERVE) {
4634 pr_warn("prog '%s': invalid relo against '%s' in special section 0x%x; forgot to initialize global var?..\n",
4635 prog->name, sym_name, shdr_idx);
4636 return -LIBBPF_ERRNO__RELOC;
4637 }
4638
4639 /* loading subprog addresses */
4640 if (sym_is_subprog(sym, obj->efile.text_shndx)) {
4641 /* global_func: sym->st_value = offset in the section, insn->imm = 0.
4642 * local_func: sym->st_value = 0, insn->imm = offset in the section.
4643 */
4644 if ((sym->st_value % BPF_INSN_SZ) || (insn->imm % BPF_INSN_SZ)) {
4645 pr_warn("prog '%s': bad subprog addr relo against '%s' at offset %zu+%d\n",
4646 prog->name, sym_name, (size_t)sym->st_value, insn->imm);
4647 return -LIBBPF_ERRNO__RELOC;
4648 }
4649
4650 reloc_desc->type = RELO_SUBPROG_ADDR;
4651 reloc_desc->insn_idx = insn_idx;
4652 reloc_desc->sym_off = sym->st_value;
4653 return 0;
4654 }
4655
4656 type = bpf_object__section_to_libbpf_map_type(obj, shdr_idx);
4657 sym_sec_name = elf_sec_name(obj, elf_sec_by_idx(obj, shdr_idx));
4658
4659 /* arena data relocation */
4660 if (shdr_idx == obj->efile.arena_data_shndx) {
4661 if (obj->arena_map_idx < 0) {
4662 pr_warn("prog '%s': bad arena data relocation at insn %u, no arena maps defined\n",
4663 prog->name, insn_idx);
4664 return -LIBBPF_ERRNO__RELOC;
4665 }
4666 reloc_desc->type = RELO_DATA;
4667 reloc_desc->insn_idx = insn_idx;
4668 reloc_desc->map_idx = obj->arena_map_idx;
4669 reloc_desc->sym_off = sym->st_value;
4670
4671 map = &obj->maps[obj->arena_map_idx];
4672 pr_debug("prog '%s': found arena map %d (%s, sec %d, off %zu) for insn %u\n",
4673 prog->name, obj->arena_map_idx, map->name, map->sec_idx,
4674 map->sec_offset, insn_idx);
4675 return 0;
4676 }
4677
4678 /* jump table data relocation */
4679 if (shdr_idx == obj->efile.jumptables_data_shndx) {
4680 reloc_desc->type = RELO_INSN_ARRAY;
4681 reloc_desc->insn_idx = insn_idx;
4682 reloc_desc->map_idx = -1;
4683 reloc_desc->sym_off = sym->st_value;
4684 reloc_desc->sym_size = sym->st_size;
4685 return 0;
4686 }
4687
4688 /* generic map reference relocation */
4689 if (type == LIBBPF_MAP_UNSPEC) {
4690 if (!bpf_object__shndx_is_maps(obj, shdr_idx)) {
4691 pr_warn("prog '%s': bad map relo against '%s' in section '%s'\n",
4692 prog->name, sym_name, sym_sec_name);
4693 return -LIBBPF_ERRNO__RELOC;
4694 }
4695 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4696 map = &obj->maps[map_idx];
4697 if (map->libbpf_type != type ||
4698 map->sec_idx != sym->st_shndx ||
4699 map->sec_offset != sym->st_value)
4700 continue;
4701 pr_debug("prog '%s': found map %zd (%s, sec %d, off %zu) for insn #%u\n",
4702 prog->name, map_idx, map->name, map->sec_idx,
4703 map->sec_offset, insn_idx);
4704 break;
4705 }
4706 if (map_idx >= nr_maps) {
4707 pr_warn("prog '%s': map relo failed to find map for section '%s', off %zu\n",
4708 prog->name, sym_sec_name, (size_t)sym->st_value);
4709 return -LIBBPF_ERRNO__RELOC;
4710 }
4711 reloc_desc->type = RELO_LD64;
4712 reloc_desc->insn_idx = insn_idx;
4713 reloc_desc->map_idx = map_idx;
4714 reloc_desc->sym_off = 0; /* sym->st_value determines map_idx */
4715 return 0;
4716 }
4717
4718 /* global data map relocation */
4719 if (!bpf_object__shndx_is_data(obj, shdr_idx)) {
4720 pr_warn("prog '%s': bad data relo against section '%s'\n",
4721 prog->name, sym_sec_name);
4722 return -LIBBPF_ERRNO__RELOC;
4723 }
4724 for (map_idx = 0; map_idx < nr_maps; map_idx++) {
4725 map = &obj->maps[map_idx];
4726 if (map->libbpf_type != type || map->sec_idx != sym->st_shndx)
4727 continue;
4728 pr_debug("prog '%s': found data map %zd (%s, sec %d, off %zu) for insn %u\n",
4729 prog->name, map_idx, map->name, map->sec_idx,
4730 map->sec_offset, insn_idx);
4731 break;
4732 }
4733 if (map_idx >= nr_maps) {
4734 pr_warn("prog '%s': data relo failed to find map for section '%s'\n",
4735 prog->name, sym_sec_name);
4736 return -LIBBPF_ERRNO__RELOC;
4737 }
4738
4739 reloc_desc->type = RELO_DATA;
4740 reloc_desc->insn_idx = insn_idx;
4741 reloc_desc->map_idx = map_idx;
4742 reloc_desc->sym_off = sym->st_value;
4743 return 0;
4744 }
4745
4746 static bool prog_contains_insn(const struct bpf_program *prog, size_t insn_idx)
4747 {
4748 return insn_idx >= prog->sec_insn_off &&
4749 insn_idx < prog->sec_insn_off + prog->sec_insn_cnt;
4750 }
4751
4752 static struct bpf_program *find_prog_by_sec_insn(const struct bpf_object *obj,
4753 size_t sec_idx, size_t insn_idx)
4754 {
4755 int l = 0, r = obj->nr_programs - 1, m;
4756 struct bpf_program *prog;
4757
4758 if (!obj->nr_programs)
4759 return NULL;
4760
4761 while (l < r) {
4762 m = l + (r - l + 1) / 2;
4763 prog = &obj->programs[m];
4764
4765 if (prog->sec_idx < sec_idx ||
4766 (prog->sec_idx == sec_idx && prog->sec_insn_off <= insn_idx))
4767 l = m;
4768 else
4769 r = m - 1;
4770 }
4771 /* matching program could be at index l, but it still might be the
4772 * wrong one, so we need to double check conditions for the last time
4773 */
4774 prog = &obj->programs[l];
4775 if (prog->sec_idx == sec_idx && prog_contains_insn(prog, insn_idx))
4776 return prog;
4777 return NULL;
4778 }
4779
4780 static int
4781 bpf_object__collect_prog_relos(struct bpf_object *obj, Elf64_Shdr *shdr, Elf_Data *data)
4782 {
4783 const char *relo_sec_name, *sec_name;
4784 size_t sec_idx = shdr->sh_info, sym_idx;
4785 struct bpf_program *prog;
4786 struct reloc_desc *relos;
4787 int err, i, nrels;
4788 const char *sym_name;
4789 __u32 insn_idx;
4790 Elf_Scn *scn;
4791 Elf_Data *scn_data;
4792 Elf64_Sym *sym;
4793 Elf64_Rel *rel;
4794
4795 if (sec_idx >= obj->efile.sec_cnt)
4796 return -EINVAL;
4797
4798 scn = elf_sec_by_idx(obj, sec_idx);
4799 scn_data = elf_sec_data(obj, scn);
4800 if (!scn_data)
4801 return -LIBBPF_ERRNO__FORMAT;
4802
4803 relo_sec_name = elf_sec_str(obj, shdr->sh_name);
4804 sec_name = elf_sec_name(obj, scn);
4805 if (!relo_sec_name || !sec_name)
4806 return -EINVAL;
4807
4808 pr_debug("sec '%s': collecting relocation for section(%zu) '%s'\n",
4809 relo_sec_name, sec_idx, sec_name);
4810 nrels = shdr->sh_size / shdr->sh_entsize;
4811
4812 for (i = 0; i < nrels; i++) {
4813 rel = elf_rel_by_idx(data, i);
4814 if (!rel) {
4815 pr_warn("sec '%s': failed to get relo #%d\n", relo_sec_name, i);
4816 return -LIBBPF_ERRNO__FORMAT;
4817 }
4818
4819 sym_idx = ELF64_R_SYM(rel->r_info);
4820 sym = elf_sym_by_idx(obj, sym_idx);
4821 if (!sym) {
4822 pr_warn("sec '%s': symbol #%zu not found for relo #%d\n",
4823 relo_sec_name, sym_idx, i);
4824 return -LIBBPF_ERRNO__FORMAT;
4825 }
4826
4827 if (sym->st_shndx >= obj->efile.sec_cnt) {
4828 pr_warn("sec '%s': corrupted symbol #%zu pointing to invalid section #%zu for relo #%d\n",
4829 relo_sec_name, sym_idx, (size_t)sym->st_shndx, i);
4830 return -LIBBPF_ERRNO__FORMAT;
4831 }
4832
4833 if (rel->r_offset % BPF_INSN_SZ || rel->r_offset >= scn_data->d_size) {
4834 pr_warn("sec '%s': invalid offset 0x%zx for relo #%d\n",
4835 relo_sec_name, (size_t)rel->r_offset, i);
4836 return -LIBBPF_ERRNO__FORMAT;
4837 }
4838
4839 insn_idx = rel->r_offset / BPF_INSN_SZ;
4840 /* relocations against static functions are recorded as
4841 * relocations against the section that contains a function;
4842 * in such case, symbol will be STT_SECTION and sym.st_name
4843 * will point to empty string (0), so fetch section name
4844 * instead
4845 */
4846 if (ELF64_ST_TYPE(sym->st_info) == STT_SECTION && sym->st_name == 0)
4847 sym_name = elf_sec_name(obj, elf_sec_by_idx(obj, sym->st_shndx));
4848 else
4849 sym_name = elf_sym_str(obj, sym->st_name);
4850 sym_name = sym_name ?: "<?";
4851
4852 pr_debug("sec '%s': relo #%d: insn #%u against '%s'\n",
4853 relo_sec_name, i, insn_idx, sym_name);
4854
4855 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
4856 if (!prog) {
4857 pr_debug("sec '%s': relo #%d: couldn't find program in section '%s' for insn #%u, probably overridden weak function, skipping...\n",
4858 relo_sec_name, i, sec_name, insn_idx);
4859 continue;
4860 }
4861
4862 relos = libbpf_reallocarray(prog->reloc_desc,
4863 prog->nr_reloc + 1, sizeof(*relos));
4864 if (!relos)
4865 return -ENOMEM;
4866 prog->reloc_desc = relos;
4867
4868 /* adjust insn_idx to local BPF program frame of reference */
4869 insn_idx -= prog->sec_insn_off;
4870 err = bpf_program__record_reloc(prog, &relos[prog->nr_reloc],
4871 insn_idx, sym_name, sym, rel);
4872 if (err)
4873 return err;
4874
4875 prog->nr_reloc++;
4876 }
4877 return 0;
4878 }
4879
4880 static int map_fill_btf_type_info(struct bpf_object *obj, struct bpf_map *map)
4881 {
4882 int id;
4883
4884 if (!obj->btf)
4885 return -ENOENT;
4886
4887 /* if it's BTF-defined map, we don't need to search for type IDs.
4888 * For struct_ops map, it does not need btf_key_type_id and
4889 * btf_value_type_id.
4890 */
4891 if (map->sec_idx == obj->efile.btf_maps_shndx || bpf_map__is_struct_ops(map))
4892 return 0;
4893
4894 /*
4895 * LLVM annotates global data differently in BTF, that is,
4896 * only as '.data', '.bss' or '.rodata'.
4897 */
4898 if (!bpf_map__is_internal(map))
4899 return -ENOENT;
4900
4901 id = btf__find_by_name(obj->btf, map->real_name);
4902 if (id < 0)
4903 return id;
4904
4905 map->btf_key_type_id = 0;
4906 map->btf_value_type_id = id;
4907 return 0;
4908 }
4909
4910 static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info)
4911 {
4912 char file[PATH_MAX], buff[4096];
4913 FILE *fp;
4914 __u32 val;
4915 int err;
4916
4917 snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
4918 memset(info, 0, sizeof(*info));
4919
4920 fp = fopen(file, "re");
4921 if (!fp) {
4922 err = -errno;
4923 pr_warn("failed to open %s: %s. No procfs support?\n", file,
4924 errstr(err));
4925 return err;
4926 }
4927
4928 while (fgets(buff, sizeof(buff), fp)) {
4929 if (sscanf(buff, "map_type:\t%u", &val) == 1)
4930 info->type = val;
4931 else if (sscanf(buff, "key_size:\t%u", &val) == 1)
4932 info->key_size = val;
4933 else if (sscanf(buff, "value_size:\t%u", &val) == 1)
4934 info->value_size = val;
4935 else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
4936 info->max_entries = val;
4937 else if (sscanf(buff, "map_flags:\t%i", &val) == 1)
4938 info->map_flags = val;
4939 }
4940
4941 fclose(fp);
4942
4943 return 0;
4944 }
4945
4946 static bool map_is_created(const struct bpf_map *map)
4947 {
4948 return map->obj->state >= OBJ_PREPARED || map->reused;
4949 }
4950
4951 bool bpf_map__autocreate(const struct bpf_map *map)
4952 {
4953 return map->autocreate;
4954 }
4955
4956 int bpf_map__set_autocreate(struct bpf_map *map, bool autocreate)
4957 {
4958 if (map_is_created(map))
4959 return libbpf_err(-EBUSY);
4960
4961 map->autocreate = autocreate;
4962 return 0;
4963 }
4964
4965 int bpf_map__set_autoattach(struct bpf_map *map, bool autoattach)
4966 {
4967 if (!bpf_map__is_struct_ops(map))
4968 return libbpf_err(-EINVAL);
4969
4970 map->autoattach = autoattach;
4971 return 0;
4972 }
4973
4974 bool bpf_map__autoattach(const struct bpf_map *map)
4975 {
4976 return map->autoattach;
4977 }
4978
4979 int bpf_map__reuse_fd(struct bpf_map *map, int fd)
4980 {
4981 struct bpf_map_info info;
4982 __u32 len = sizeof(info), name_len;
4983 int new_fd, err;
4984 char *new_name;
4985
4986 memset(&info, 0, len);
4987 err = bpf_map_get_info_by_fd(fd, &info, &len);
4988 if (err && errno == EINVAL)
4989 err = bpf_get_map_info_from_fdinfo(fd, &info);
4990 if (err)
4991 return libbpf_err(err);
4992
4993 name_len = strlen(info.name);
4994 if (name_len == BPF_OBJ_NAME_LEN - 1 && strncmp(map->name, info.name, name_len) == 0)
4995 new_name = strdup(map->name);
4996 else
4997 new_name = strdup(info.name);
4998
4999 if (!new_name)
5000 return libbpf_err(-errno);
5001
5002 /*
5003 * Like dup(), but make sure new FD is >= 3 and has O_CLOEXEC set.
5004 * This is similar to what we do in ensure_good_fd(), but without
5005 * closing original FD.
5006 */
5007 new_fd = fcntl(fd, F_DUPFD_CLOEXEC, 3);
5008 if (new_fd < 0) {
5009 err = -errno;
5010 goto err_free_new_name;
5011 }
5012
5013 err = reuse_fd(map->fd, new_fd);
5014 if (err)
5015 goto err_free_new_name;
5016
5017 free(map->name);
5018
5019 map->name = new_name;
5020 map->def.type = info.type;
5021 map->def.key_size = info.key_size;
5022 map->def.value_size = info.value_size;
5023 map->def.max_entries = info.max_entries;
5024 map->def.map_flags = info.map_flags;
5025 map->btf_key_type_id = info.btf_key_type_id;
5026 map->btf_value_type_id = info.btf_value_type_id;
5027 map->reused = true;
5028 map->map_extra = info.map_extra;
5029
5030 return 0;
5031
5032 err_free_new_name:
5033 free(new_name);
5034 return libbpf_err(err);
5035 }
5036
5037 __u32 bpf_map__max_entries(const struct bpf_map *map)
5038 {
5039 return map->def.max_entries;
5040 }
5041
5042 struct bpf_map *bpf_map__inner_map(struct bpf_map *map)
5043 {
5044 if (!bpf_map_type__is_map_in_map(map->def.type))
5045 return errno = EINVAL, NULL;
5046
5047 return map->inner_map;
5048 }
5049
5050 int bpf_map__set_max_entries(struct bpf_map *map, __u32 max_entries)
5051 {
5052 if (map_is_created(map))
5053 return libbpf_err(-EBUSY);
5054
5055 map->def.max_entries = max_entries;
5056
5057 /* auto-adjust BPF ringbuf map max_entries to be a multiple of page size */
5058 if (map_is_ringbuf(map))
5059 map->def.max_entries = adjust_ringbuf_sz(map->def.max_entries);
5060
5061 return 0;
5062 }
5063
5064 static int bpf_object_prepare_token(struct bpf_object *obj)
5065 {
5066 const char *bpffs_path;
5067 int bpffs_fd = -1, token_fd, err;
5068 bool mandatory;
5069 enum libbpf_print_level level;
5070
5071 /* token is explicitly prevented */
5072 if (obj->token_path && obj->token_path[0] == '\0') {
5073 pr_debug("object '%s': token is prevented, skipping...\n", obj->name);
5074 return 0;
5075 }
5076
5077 mandatory = obj->token_path != NULL;
5078 level = mandatory ? LIBBPF_WARN : LIBBPF_DEBUG;
5079
5080 bpffs_path = obj->token_path ?: BPF_FS_DEFAULT_PATH;
5081 bpffs_fd = open(bpffs_path, O_DIRECTORY, O_RDWR);
5082 if (bpffs_fd < 0) {
5083 err = -errno;
5084 __pr(level, "object '%s': failed (%s) to open BPF FS mount at '%s'%s\n",
5085 obj->name, errstr(err), bpffs_path,
5086 mandatory ? "" : ", skipping optional step...");
5087 return mandatory ? err : 0;
5088 }
5089
5090 token_fd = bpf_token_create(bpffs_fd, 0);
5091 close(bpffs_fd);
5092 if (token_fd < 0) {
5093 if (!mandatory && token_fd == -ENOENT) {
5094 pr_debug("object '%s': BPF FS at '%s' doesn't have BPF token delegation set up, skipping...\n",
5095 obj->name, bpffs_path);
5096 return 0;
5097 }
5098 __pr(level, "object '%s': failed (%d) to create BPF token from '%s'%s\n",
5099 obj->name, token_fd, bpffs_path,
5100 mandatory ? "" : ", skipping optional step...");
5101 return mandatory ? token_fd : 0;
5102 }
5103
5104 obj->feat_cache = calloc(1, sizeof(*obj->feat_cache));
5105 if (!obj->feat_cache) {
5106 close(token_fd);
5107 return -ENOMEM;
5108 }
5109
5110 obj->token_fd = token_fd;
5111 obj->feat_cache->token_fd = token_fd;
5112
5113 return 0;
5114 }
5115
5116 static int
5117 bpf_object__probe_loading(struct bpf_object *obj)
5118 {
5119 struct bpf_insn insns[] = {
5120 BPF_MOV64_IMM(BPF_REG_0, 0),
5121 BPF_EXIT_INSN(),
5122 };
5123 int ret, insn_cnt = ARRAY_SIZE(insns);
5124 LIBBPF_OPTS(bpf_prog_load_opts, opts,
5125 .token_fd = obj->token_fd,
5126 .prog_flags = obj->token_fd ? BPF_F_TOKEN_FD : 0,
5127 );
5128
5129 if (obj->gen_loader)
5130 return 0;
5131
5132 ret = bump_rlimit_memlock();
5133 if (ret)
5134 pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %s), you might need to do it explicitly!\n",
5135 errstr(ret));
5136
5137 /* make sure basic loading works */
5138 ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, &opts);
5139 if (ret < 0)
5140 ret = bpf_prog_load(BPF_PROG_TYPE_TRACEPOINT, NULL, "GPL", insns, insn_cnt, &opts);
5141 if (ret < 0) {
5142 ret = errno;
5143 pr_warn("Error in %s(): %s. Couldn't load trivial BPF program. Make sure your kernel supports BPF (CONFIG_BPF_SYSCALL=y) and/or that RLIMIT_MEMLOCK is set to big enough value.\n",
5144 __func__, errstr(ret));
5145 return -ret;
5146 }
5147 close(ret);
5148
5149 return 0;
5150 }
5151
5152 bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
5153 {
5154 if (obj->gen_loader)
5155 /* To generate loader program assume the latest kernel
5156 * to avoid doing extra prog_load, map_create syscalls.
5157 */
5158 return true;
5159
5160 if (obj->token_fd)
5161 return feat_supported(obj->feat_cache, feat_id);
5162
5163 return feat_supported(NULL, feat_id);
5164 }
5165
5166 static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd)
5167 {
5168 struct bpf_map_info map_info;
5169 __u32 map_info_len = sizeof(map_info);
5170 int err;
5171
5172 memset(&map_info, 0, map_info_len);
5173 err = bpf_map_get_info_by_fd(map_fd, &map_info, &map_info_len);
5174 if (err && errno == EINVAL)
5175 err = bpf_get_map_info_from_fdinfo(map_fd, &map_info);
5176 if (err) {
5177 pr_warn("failed to get map info for map FD %d: %s\n", map_fd,
5178 errstr(err));
5179 return false;
5180 }
5181
5182 /*
5183 * bpf_get_map_info_by_fd() for DEVMAP will always return flags with
5184 * BPF_F_RDONLY_PROG set, but it generally is not set at map creation time.
5185 * Thus, ignore the BPF_F_RDONLY_PROG flag in the flags returned from
5186 * bpf_get_map_info_by_fd() when checking for compatibility with an
5187 * existing DEVMAP.
5188 */
5189 if (map->def.type == BPF_MAP_TYPE_DEVMAP || map->def.type == BPF_MAP_TYPE_DEVMAP_HASH)
5190 map_info.map_flags &= ~BPF_F_RDONLY_PROG;
5191
5192 return (map_info.type == map->def.type &&
5193 map_info.key_size == map->def.key_size &&
5194 map_info.value_size == map->def.value_size &&
5195 map_info.max_entries == map->def.max_entries &&
5196 map_info.map_flags == map->def.map_flags &&
5197 map_info.map_extra == map->map_extra);
5198 }
5199
5200 static int
5201 bpf_object__reuse_map(struct bpf_map *map)
5202 {
5203 int err, pin_fd;
5204
5205 pin_fd = bpf_obj_get(map->pin_path);
5206 if (pin_fd < 0) {
5207 err = -errno;
5208 if (err == -ENOENT) {
5209 pr_debug("found no pinned map to reuse at '%s'\n",
5210 map->pin_path);
5211 return 0;
5212 }
5213
5214 pr_warn("couldn't retrieve pinned map '%s': %s\n",
5215 map->pin_path, errstr(err));
5216 return err;
5217 }
5218
5219 if (!map_is_reuse_compat(map, pin_fd)) {
5220 pr_warn("couldn't reuse pinned map at '%s': parameter mismatch\n",
5221 map->pin_path);
5222 close(pin_fd);
5223 return -EINVAL;
5224 }
5225
5226 err = bpf_map__reuse_fd(map, pin_fd);
5227 close(pin_fd);
5228 if (err)
5229 return err;
5230
5231 map->pinned = true;
5232 pr_debug("reused pinned map at '%s'\n", map->pin_path);
5233
5234 return 0;
5235 }
5236
5237 static int
5238 bpf_object__populate_internal_map(struct bpf_object *obj, struct bpf_map *map)
5239 {
5240 enum libbpf_map_type map_type = map->libbpf_type;
5241 int err, zero = 0;
5242 size_t mmap_sz;
5243
5244 if (obj->gen_loader) {
5245 bpf_gen__map_update_elem(obj->gen_loader, map - obj->maps,
5246 map->mmaped, map->def.value_size);
5247 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG)
5248 bpf_gen__map_freeze(obj->gen_loader, map - obj->maps);
5249 return 0;
5250 }
5251
5252 err = bpf_map_update_elem(map->fd, &zero, map->mmaped, 0);
5253 if (err) {
5254 err = -errno;
5255 pr_warn("map '%s': failed to set initial contents: %s\n",
5256 bpf_map__name(map), errstr(err));
5257 return err;
5258 }
5259
5260 /* Freeze .rodata and .kconfig map as read-only from syscall side. */
5261 if (map_type == LIBBPF_MAP_RODATA || map_type == LIBBPF_MAP_KCONFIG) {
5262 err = bpf_map_freeze(map->fd);
5263 if (err) {
5264 err = -errno;
5265 pr_warn("map '%s': failed to freeze as read-only: %s\n",
5266 bpf_map__name(map), errstr(err));
5267 return err;
5268 }
5269 }
5270
5271 /* Remap anonymous mmap()-ed "map initialization image" as
5272 * a BPF map-backed mmap()-ed memory, but preserving the same
5273 * memory address. This will cause kernel to change process'
5274 * page table to point to a different piece of kernel memory,
5275 * but from userspace point of view memory address (and its
5276 * contents, being identical at this point) will stay the
5277 * same. This mapping will be released by bpf_object__close()
5278 * as per normal clean up procedure.
5279 */
5280 mmap_sz = bpf_map_mmap_sz(map);
5281 if (map->def.map_flags & BPF_F_MMAPABLE) {
5282 void *mmaped;
5283 int prot;
5284
5285 if (map->def.map_flags & BPF_F_RDONLY_PROG)
5286 prot = PROT_READ;
5287 else
5288 prot = PROT_READ | PROT_WRITE;
5289 mmaped = mmap(map->mmaped, mmap_sz, prot, MAP_SHARED | MAP_FIXED, map->fd, 0);
5290 if (mmaped == MAP_FAILED) {
5291 err = -errno;
5292 pr_warn("map '%s': failed to re-mmap() contents: %s\n",
5293 bpf_map__name(map), errstr(err));
5294 return err;
5295 }
5296 map->mmaped = mmaped;
5297 } else if (map->mmaped) {
5298 munmap(map->mmaped, mmap_sz);
5299 map->mmaped = NULL;
5300 }
5301
5302 return 0;
5303 }
5304
5305 static void bpf_map__destroy(struct bpf_map *map);
5306
5307 static int bpf_object__create_map(struct bpf_object *obj, struct bpf_map *map, bool is_inner)
5308 {
5309 LIBBPF_OPTS(bpf_map_create_opts, create_attr);
5310 struct bpf_map_def *def = &map->def;
5311 const char *map_name = NULL;
5312 int err = 0, map_fd;
5313
5314 if (kernel_supports(obj, FEAT_PROG_NAME))
5315 map_name = map->name;
5316 create_attr.map_ifindex = map->map_ifindex;
5317 create_attr.map_flags = def->map_flags;
5318 create_attr.numa_node = map->numa_node;
5319 create_attr.map_extra = map->map_extra;
5320 create_attr.token_fd = obj->token_fd;
5321 if (obj->token_fd)
5322 create_attr.map_flags |= BPF_F_TOKEN_FD;
5323 if (map->excl_prog) {
5324 err = bpf_prog_compute_hash(map->excl_prog);
5325 if (err)
5326 return err;
5327
5328 create_attr.excl_prog_hash = map->excl_prog->hash;
5329 create_attr.excl_prog_hash_size = SHA256_DIGEST_LENGTH;
5330 }
5331
5332 if (bpf_map__is_struct_ops(map)) {
5333 create_attr.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
5334 if (map->mod_btf_fd >= 0) {
5335 create_attr.value_type_btf_obj_fd = map->mod_btf_fd;
5336 create_attr.map_flags |= BPF_F_VTYPE_BTF_OBJ_FD;
5337 }
5338 }
5339
5340 if (obj->btf && btf__fd(obj->btf) >= 0) {
5341 create_attr.btf_fd = btf__fd(obj->btf);
5342 create_attr.btf_key_type_id = map->btf_key_type_id;
5343 create_attr.btf_value_type_id = map->btf_value_type_id;
5344 }
5345
5346 if (bpf_map_type__is_map_in_map(def->type)) {
5347 if (map->inner_map) {
5348 err = map_set_def_max_entries(map->inner_map);
5349 if (err)
5350 return err;
5351 err = bpf_object__create_map(obj, map->inner_map, true);
5352 if (err) {
5353 pr_warn("map '%s': failed to create inner map: %s\n",
5354 map->name, errstr(err));
5355 return err;
5356 }
5357 map->inner_map_fd = map->inner_map->fd;
5358 }
5359 if (map->inner_map_fd >= 0)
5360 create_attr.inner_map_fd = map->inner_map_fd;
5361 }
5362
5363 switch (def->type) {
5364 case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
5365 case BPF_MAP_TYPE_CGROUP_ARRAY:
5366 case BPF_MAP_TYPE_STACK_TRACE:
5367 case BPF_MAP_TYPE_ARRAY_OF_MAPS:
5368 case BPF_MAP_TYPE_HASH_OF_MAPS:
5369 case BPF_MAP_TYPE_DEVMAP:
5370 case BPF_MAP_TYPE_DEVMAP_HASH:
5371 case BPF_MAP_TYPE_CPUMAP:
5372 case BPF_MAP_TYPE_XSKMAP:
5373 case BPF_MAP_TYPE_SOCKMAP:
5374 case BPF_MAP_TYPE_SOCKHASH:
5375 case BPF_MAP_TYPE_QUEUE:
5376 case BPF_MAP_TYPE_STACK:
5377 case BPF_MAP_TYPE_ARENA:
5378 create_attr.btf_fd = 0;
5379 create_attr.btf_key_type_id = 0;
5380 create_attr.btf_value_type_id = 0;
5381 map->btf_key_type_id = 0;
5382 map->btf_value_type_id = 0;
5383 break;
5384 case BPF_MAP_TYPE_STRUCT_OPS:
5385 create_attr.btf_value_type_id = 0;
5386 break;
5387 default:
5388 break;
5389 }
5390
5391 if (obj->gen_loader) {
5392 bpf_gen__map_create(obj->gen_loader, def->type, map_name,
5393 def->key_size, def->value_size, def->max_entries,
5394 &create_attr, is_inner ? -1 : map - obj->maps);
5395 /* We keep pretenting we have valid FD to pass various fd >= 0
5396 * checks by just keeping original placeholder FDs in place.
5397 * See bpf_object__add_map() comment.
5398 * This placeholder fd will not be used with any syscall and
5399 * will be reset to -1 eventually.
5400 */
5401 map_fd = map->fd;
5402 } else {
5403 map_fd = bpf_map_create(def->type, map_name,
5404 def->key_size, def->value_size,
5405 def->max_entries, &create_attr);
5406 }
5407 if (map_fd < 0 && (create_attr.btf_key_type_id || create_attr.btf_value_type_id)) {
5408 err = -errno;
5409 pr_warn("Error in bpf_create_map_xattr(%s): %s. Retrying without BTF.\n",
5410 map->name, errstr(err));
5411 create_attr.btf_fd = 0;
5412 create_attr.btf_key_type_id = 0;
5413 create_attr.btf_value_type_id = 0;
5414 map->btf_key_type_id = 0;
5415 map->btf_value_type_id = 0;
5416 map_fd = bpf_map_create(def->type, map_name,
5417 def->key_size, def->value_size,
5418 def->max_entries, &create_attr);
5419 }
5420
5421 if (bpf_map_type__is_map_in_map(def->type) && map->inner_map) {
5422 if (obj->gen_loader)
5423 map->inner_map->fd = -1;
5424 bpf_map__destroy(map->inner_map);
5425 zfree(&map->inner_map);
5426 }
5427
5428 if (map_fd < 0)
5429 return map_fd;
5430
5431 /* obj->gen_loader case, prevent reuse_fd() from closing map_fd */
5432 if (map->fd == map_fd)
5433 return 0;
5434
5435 /* Keep placeholder FD value but now point it to the BPF map object.
5436 * This way everything that relied on this map's FD (e.g., relocated
5437 * ldimm64 instructions) will stay valid and won't need adjustments.
5438 * map->fd stays valid but now point to what map_fd points to.
5439 */
5440 return reuse_fd(map->fd, map_fd);
5441 }
5442
5443 static int init_map_in_map_slots(struct bpf_object *obj, struct bpf_map *map)
5444 {
5445 const struct bpf_map *targ_map;
5446 unsigned int i;
5447 int fd, err = 0;
5448
5449 for (i = 0; i < map->init_slots_sz; i++) {
5450 if (!map->init_slots[i])
5451 continue;
5452
5453 targ_map = map->init_slots[i];
5454 fd = targ_map->fd;
5455
5456 if (obj->gen_loader) {
5457 bpf_gen__populate_outer_map(obj->gen_loader,
5458 map - obj->maps, i,
5459 targ_map - obj->maps);
5460 } else {
5461 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5462 }
5463 if (err) {
5464 err = -errno;
5465 pr_warn("map '%s': failed to initialize slot [%d] to map '%s' fd=%d: %s\n",
5466 map->name, i, targ_map->name, fd, errstr(err));
5467 return err;
5468 }
5469 pr_debug("map '%s': slot [%d] set to map '%s' fd=%d\n",
5470 map->name, i, targ_map->name, fd);
5471 }
5472
5473 zfree(&map->init_slots);
5474 map->init_slots_sz = 0;
5475
5476 return 0;
5477 }
5478
5479 static int init_prog_array_slots(struct bpf_object *obj, struct bpf_map *map)
5480 {
5481 const struct bpf_program *targ_prog;
5482 unsigned int i;
5483 int fd, err;
5484
5485 if (obj->gen_loader)
5486 return -ENOTSUP;
5487
5488 for (i = 0; i < map->init_slots_sz; i++) {
5489 if (!map->init_slots[i])
5490 continue;
5491
5492 targ_prog = map->init_slots[i];
5493 fd = bpf_program__fd(targ_prog);
5494
5495 err = bpf_map_update_elem(map->fd, &i, &fd, 0);
5496 if (err) {
5497 err = -errno;
5498 pr_warn("map '%s': failed to initialize slot [%d] to prog '%s' fd=%d: %s\n",
5499 map->name, i, targ_prog->name, fd, errstr(err));
5500 return err;
5501 }
5502 pr_debug("map '%s': slot [%d] set to prog '%s' fd=%d\n",
5503 map->name, i, targ_prog->name, fd);
5504 }
5505
5506 zfree(&map->init_slots);
5507 map->init_slots_sz = 0;
5508
5509 return 0;
5510 }
5511
5512 static int bpf_object_init_prog_arrays(struct bpf_object *obj)
5513 {
5514 struct bpf_map *map;
5515 int i, err;
5516
5517 for (i = 0; i < obj->nr_maps; i++) {
5518 map = &obj->maps[i];
5519
5520 if (!map->init_slots_sz || map->def.type != BPF_MAP_TYPE_PROG_ARRAY)
5521 continue;
5522
5523 err = init_prog_array_slots(obj, map);
5524 if (err < 0)
5525 return err;
5526 }
5527 return 0;
5528 }
5529
5530 static int map_set_def_max_entries(struct bpf_map *map)
5531 {
5532 if (map->def.type == BPF_MAP_TYPE_PERF_EVENT_ARRAY && !map->def.max_entries) {
5533 int nr_cpus;
5534
5535 nr_cpus = libbpf_num_possible_cpus();
5536 if (nr_cpus < 0) {
5537 pr_warn("map '%s': failed to determine number of system CPUs: %d\n",
5538 map->name, nr_cpus);
5539 return nr_cpus;
5540 }
5541 pr_debug("map '%s': setting size to %d\n", map->name, nr_cpus);
5542 map->def.max_entries = nr_cpus;
5543 }
5544
5545 return 0;
5546 }
5547
5548 static int
5549 bpf_object__create_maps(struct bpf_object *obj)
5550 {
5551 struct bpf_map *map;
5552 unsigned int i, j;
5553 int err;
5554 bool retried;
5555
5556 for (i = 0; i < obj->nr_maps; i++) {
5557 map = &obj->maps[i];
5558
5559 /* To support old kernels, we skip creating global data maps
5560 * (.rodata, .data, .kconfig, etc); later on, during program
5561 * loading, if we detect that at least one of the to-be-loaded
5562 * programs is referencing any global data map, we'll error
5563 * out with program name and relocation index logged.
5564 * This approach allows to accommodate Clang emitting
5565 * unnecessary .rodata.str1.1 sections for string literals,
5566 * but also it allows to have CO-RE applications that use
5567 * global variables in some of BPF programs, but not others.
5568 * If those global variable-using programs are not loaded at
5569 * runtime due to bpf_program__set_autoload(prog, false),
5570 * bpf_object loading will succeed just fine even on old
5571 * kernels.
5572 */
5573 if (bpf_map__is_internal(map) && !kernel_supports(obj, FEAT_GLOBAL_DATA))
5574 map->autocreate = false;
5575
5576 if (!map->autocreate) {
5577 pr_debug("map '%s': skipped auto-creating...\n", map->name);
5578 continue;
5579 }
5580
5581 err = map_set_def_max_entries(map);
5582 if (err)
5583 goto err_out;
5584
5585 retried = false;
5586 retry:
5587 if (map->pin_path) {
5588 err = bpf_object__reuse_map(map);
5589 if (err) {
5590 pr_warn("map '%s': error reusing pinned map\n",
5591 map->name);
5592 goto err_out;
5593 }
5594 if (retried && map->fd < 0) {
5595 pr_warn("map '%s': cannot find pinned map\n",
5596 map->name);
5597 err = -ENOENT;
5598 goto err_out;
5599 }
5600 }
5601
5602 if (map->reused) {
5603 pr_debug("map '%s': skipping creation (preset fd=%d)\n",
5604 map->name, map->fd);
5605 } else {
5606 err = bpf_object__create_map(obj, map, false);
5607 if (err)
5608 goto err_out;
5609
5610 pr_debug("map '%s': created successfully, fd=%d\n",
5611 map->name, map->fd);
5612
5613 if (bpf_map__is_internal(map)) {
5614 err = bpf_object__populate_internal_map(obj, map);
5615 if (err < 0)
5616 goto err_out;
5617 } else if (map->def.type == BPF_MAP_TYPE_ARENA) {
5618 map->mmaped = mmap((void *)(long)map->map_extra,
5619 bpf_map_mmap_sz(map), PROT_READ | PROT_WRITE,
5620 map->map_extra ? MAP_SHARED | MAP_FIXED : MAP_SHARED,
5621 map->fd, 0);
5622 if (map->mmaped == MAP_FAILED) {
5623 err = -errno;
5624 map->mmaped = NULL;
5625 pr_warn("map '%s': failed to mmap arena: %s\n",
5626 map->name, errstr(err));
5627 return err;
5628 }
5629 if (obj->arena_data) {
5630 memcpy(map->mmaped + obj->arena_data_off, obj->arena_data,
5631 obj->arena_data_sz);
5632 zfree(&obj->arena_data);
5633 }
5634 }
5635 if (map->init_slots_sz && map->def.type != BPF_MAP_TYPE_PROG_ARRAY) {
5636 err = init_map_in_map_slots(obj, map);
5637 if (err < 0)
5638 goto err_out;
5639 }
5640 }
5641
5642 if (map->pin_path && !map->pinned) {
5643 err = bpf_map__pin(map, NULL);
5644 if (err) {
5645 if (!retried && err == -EEXIST) {
5646 retried = true;
5647 goto retry;
5648 }
5649 pr_warn("map '%s': failed to auto-pin at '%s': %s\n",
5650 map->name, map->pin_path, errstr(err));
5651 goto err_out;
5652 }
5653 }
5654 }
5655
5656 return 0;
5657
5658 err_out:
5659 pr_warn("map '%s': failed to create: %s\n", map->name, errstr(err));
5660 pr_perm_msg(err);
5661 for (j = 0; j < i; j++)
5662 zclose(obj->maps[j].fd);
5663 return err;
5664 }
5665
5666 static bool bpf_core_is_flavor_sep(const char *s)
5667 {
5668 /* check X___Y name pattern, where X and Y are not underscores */
5669 return s[0] != '_' && /* X */
5670 s[1] == '_' && s[2] == '_' && s[3] == '_' && /* ___ */
5671 s[4] != '_'; /* Y */
5672 }
5673
5674 /* Given 'some_struct_name___with_flavor' return the length of a name prefix
5675 * before last triple underscore. Struct name part after last triple
5676 * underscore is ignored by BPF CO-RE relocation during relocation matching.
5677 */
5678 size_t bpf_core_essential_name_len(const char *name)
5679 {
5680 size_t n = strlen(name);
5681 int i;
5682
5683 for (i = n - 5; i >= 0; i--) {
5684 if (bpf_core_is_flavor_sep(name + i))
5685 return i + 1;
5686 }
5687 return n;
5688 }
5689
5690 void bpf_core_free_cands(struct bpf_core_cand_list *cands)
5691 {
5692 if (!cands)
5693 return;
5694
5695 free(cands->cands);
5696 free(cands);
5697 }
5698
5699 int bpf_core_add_cands(struct bpf_core_cand *local_cand,
5700 size_t local_essent_len,
5701 const struct btf *targ_btf,
5702 const char *targ_btf_name,
5703 int targ_start_id,
5704 struct bpf_core_cand_list *cands)
5705 {
5706 struct bpf_core_cand *new_cands, *cand;
5707 const struct btf_type *t, *local_t;
5708 const char *targ_name, *local_name;
5709 size_t targ_essent_len;
5710 int n, i;
5711
5712 local_t = btf__type_by_id(local_cand->btf, local_cand->id);
5713 local_name = btf__str_by_offset(local_cand->btf, local_t->name_off);
5714
5715 n = btf__type_cnt(targ_btf);
5716 for (i = targ_start_id; i < n; i++) {
5717 t = btf__type_by_id(targ_btf, i);
5718 if (!btf_kind_core_compat(t, local_t))
5719 continue;
5720
5721 targ_name = btf__name_by_offset(targ_btf, t->name_off);
5722 if (str_is_empty(targ_name))
5723 continue;
5724
5725 targ_essent_len = bpf_core_essential_name_len(targ_name);
5726 if (targ_essent_len != local_essent_len)
5727 continue;
5728
5729 if (strncmp(local_name, targ_name, local_essent_len) != 0)
5730 continue;
5731
5732 pr_debug("CO-RE relocating [%d] %s %s: found target candidate [%d] %s %s in [%s]\n",
5733 local_cand->id, btf_kind_str(local_t),
5734 local_name, i, btf_kind_str(t), targ_name,
5735 targ_btf_name);
5736 new_cands = libbpf_reallocarray(cands->cands, cands->len + 1,
5737 sizeof(*cands->cands));
5738 if (!new_cands)
5739 return -ENOMEM;
5740
5741 cand = &new_cands[cands->len];
5742 cand->btf = targ_btf;
5743 cand->id = i;
5744
5745 cands->cands = new_cands;
5746 cands->len++;
5747 }
5748 return 0;
5749 }
5750
5751 static int load_module_btfs(struct bpf_object *obj)
5752 {
5753 struct bpf_btf_info info;
5754 struct module_btf *mod_btf;
5755 struct btf *btf;
5756 char name[64];
5757 __u32 id = 0, len;
5758 int err, fd;
5759
5760 if (obj->btf_modules_loaded)
5761 return 0;
5762
5763 if (obj->gen_loader)
5764 return 0;
5765
5766 /* don't do this again, even if we find no module BTFs */
5767 obj->btf_modules_loaded = true;
5768
5769 /* kernel too old to support module BTFs */
5770 if (!kernel_supports(obj, FEAT_MODULE_BTF))
5771 return 0;
5772
5773 while (true) {
5774 err = bpf_btf_get_next_id(id, &id);
5775 if (err && errno == ENOENT)
5776 return 0;
5777 if (err && errno == EPERM) {
5778 pr_debug("skipping module BTFs loading, missing privileges\n");
5779 return 0;
5780 }
5781 if (err) {
5782 err = -errno;
5783 pr_warn("failed to iterate BTF objects: %s\n", errstr(err));
5784 return err;
5785 }
5786
5787 fd = bpf_btf_get_fd_by_id(id);
5788 if (fd < 0) {
5789 if (errno == ENOENT)
5790 continue; /* expected race: BTF was unloaded */
5791 err = -errno;
5792 pr_warn("failed to get BTF object #%d FD: %s\n", id, errstr(err));
5793 return err;
5794 }
5795
5796 len = sizeof(info);
5797 memset(&info, 0, sizeof(info));
5798 info.name = ptr_to_u64(name);
5799 info.name_len = sizeof(name);
5800
5801 btf = NULL;
5802 err = bpf_btf_get_info_by_fd(fd, &info, &len);
5803 if (err) {
5804 err = -errno;
5805 pr_warn("failed to get BTF object #%d info: %s\n", id, errstr(err));
5806 break;
5807 }
5808
5809 /* ignore non-module BTFs */
5810 if (!info.kernel_btf || strcmp(name, "vmlinux") == 0) {
5811 close(fd);
5812 continue;
5813 }
5814
5815 btf = btf_get_from_fd(fd, obj->btf_vmlinux);
5816 err = libbpf_get_error(btf);
5817 if (err) {
5818 pr_warn("failed to load module [%s]'s BTF object #%d: %s\n",
5819 name, id, errstr(err));
5820 break;
5821 }
5822
5823 err = libbpf_ensure_mem((void **)&obj->btf_modules, &obj->btf_module_cap,
5824 sizeof(*obj->btf_modules), obj->btf_module_cnt + 1);
5825 if (err)
5826 break;
5827
5828 mod_btf = &obj->btf_modules[obj->btf_module_cnt];
5829
5830 mod_btf->btf = btf;
5831 mod_btf->id = id;
5832 mod_btf->fd = fd;
5833 mod_btf->name = strdup(name);
5834 if (!mod_btf->name) {
5835 err = -ENOMEM;
5836 break;
5837 }
5838 obj->btf_module_cnt++;
5839 }
5840
5841 if (err) {
5842 btf__free(btf);
5843 close(fd);
5844 }
5845 return err;
5846 }
5847
5848 static struct bpf_core_cand_list *
5849 bpf_core_find_cands(struct bpf_object *obj, const struct btf *local_btf, __u32 local_type_id)
5850 {
5851 struct bpf_core_cand local_cand = {};
5852 struct bpf_core_cand_list *cands;
5853 const struct btf *main_btf;
5854 const struct btf_type *local_t;
5855 const char *local_name;
5856 size_t local_essent_len;
5857 int err, i;
5858
5859 local_cand.btf = local_btf;
5860 local_cand.id = local_type_id;
5861 local_t = btf__type_by_id(local_btf, local_type_id);
5862 if (!local_t)
5863 return ERR_PTR(-EINVAL);
5864
5865 local_name = btf__name_by_offset(local_btf, local_t->name_off);
5866 if (str_is_empty(local_name))
5867 return ERR_PTR(-EINVAL);
5868 local_essent_len = bpf_core_essential_name_len(local_name);
5869
5870 cands = calloc(1, sizeof(*cands));
5871 if (!cands)
5872 return ERR_PTR(-ENOMEM);
5873
5874 /* Attempt to find target candidates in vmlinux BTF first */
5875 main_btf = obj->btf_vmlinux_override ?: obj->btf_vmlinux;
5876 err = bpf_core_add_cands(&local_cand, local_essent_len, main_btf, "vmlinux", 1, cands);
5877 if (err)
5878 goto err_out;
5879
5880 /* if vmlinux BTF has any candidate, don't got for module BTFs */
5881 if (cands->len)
5882 return cands;
5883
5884 /* if vmlinux BTF was overridden, don't attempt to load module BTFs */
5885 if (obj->btf_vmlinux_override)
5886 return cands;
5887
5888 /* now look through module BTFs, trying to still find candidates */
5889 err = load_module_btfs(obj);
5890 if (err)
5891 goto err_out;
5892
5893 for (i = 0; i < obj->btf_module_cnt; i++) {
5894 err = bpf_core_add_cands(&local_cand, local_essent_len,
5895 obj->btf_modules[i].btf,
5896 obj->btf_modules[i].name,
5897 btf__type_cnt(obj->btf_vmlinux),
5898 cands);
5899 if (err)
5900 goto err_out;
5901 }
5902
5903 return cands;
5904 err_out:
5905 bpf_core_free_cands(cands);
5906 return ERR_PTR(err);
5907 }
5908
5909 /* Check local and target types for compatibility. This check is used for
5910 * type-based CO-RE relocations and follow slightly different rules than
5911 * field-based relocations. This function assumes that root types were already
5912 * checked for name match. Beyond that initial root-level name check, names
5913 * are completely ignored. Compatibility rules are as follows:
5914 * - any two STRUCTs/UNIONs/FWDs/ENUMs/INTs are considered compatible, but
5915 * kind should match for local and target types (i.e., STRUCT is not
5916 * compatible with UNION);
5917 * - for ENUMs, the size is ignored;
5918 * - for INT, size and signedness are ignored;
5919 * - for ARRAY, dimensionality is ignored, element types are checked for
5920 * compatibility recursively;
5921 * - CONST/VOLATILE/RESTRICT modifiers are ignored;
5922 * - TYPEDEFs/PTRs are compatible if types they pointing to are compatible;
5923 * - FUNC_PROTOs are compatible if they have compatible signature: same
5924 * number of input args and compatible return and argument types.
5925 * These rules are not set in stone and probably will be adjusted as we get
5926 * more experience with using BPF CO-RE relocations.
5927 */
5928 int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id,
5929 const struct btf *targ_btf, __u32 targ_id)
5930 {
5931 return __bpf_core_types_are_compat(local_btf, local_id, targ_btf, targ_id, 32);
5932 }
5933
5934 int bpf_core_types_match(const struct btf *local_btf, __u32 local_id,
5935 const struct btf *targ_btf, __u32 targ_id)
5936 {
5937 return __bpf_core_types_match(local_btf, local_id, targ_btf, targ_id, false, 32);
5938 }
5939
5940 static size_t bpf_core_hash_fn(const long key, void *ctx)
5941 {
5942 return key;
5943 }
5944
5945 static bool bpf_core_equal_fn(const long k1, const long k2, void *ctx)
5946 {
5947 return k1 == k2;
5948 }
5949
5950 static int record_relo_core(struct bpf_program *prog,
5951 const struct bpf_core_relo *core_relo, int insn_idx)
5952 {
5953 struct reloc_desc *relos, *relo;
5954
5955 relos = libbpf_reallocarray(prog->reloc_desc,
5956 prog->nr_reloc + 1, sizeof(*relos));
5957 if (!relos)
5958 return -ENOMEM;
5959 relo = &relos[prog->nr_reloc];
5960 relo->type = RELO_CORE;
5961 relo->insn_idx = insn_idx;
5962 relo->core_relo = core_relo;
5963 prog->reloc_desc = relos;
5964 prog->nr_reloc++;
5965 return 0;
5966 }
5967
5968 static const struct bpf_core_relo *find_relo_core(struct bpf_program *prog, int insn_idx)
5969 {
5970 struct reloc_desc *relo;
5971 int i;
5972
5973 for (i = 0; i < prog->nr_reloc; i++) {
5974 relo = &prog->reloc_desc[i];
5975 if (relo->type != RELO_CORE || relo->insn_idx != insn_idx)
5976 continue;
5977
5978 return relo->core_relo;
5979 }
5980
5981 return NULL;
5982 }
5983
5984 static int bpf_core_resolve_relo(struct bpf_program *prog,
5985 const struct bpf_core_relo *relo,
5986 int relo_idx,
5987 const struct btf *local_btf,
5988 struct hashmap *cand_cache,
5989 struct bpf_core_relo_res *targ_res)
5990 {
5991 struct bpf_core_spec specs_scratch[3] = {};
5992 struct bpf_core_cand_list *cands = NULL;
5993 const char *prog_name = prog->name;
5994 const struct btf_type *local_type;
5995 const char *local_name;
5996 __u32 local_id = relo->type_id;
5997 int err;
5998
5999 local_type = btf__type_by_id(local_btf, local_id);
6000 if (!local_type)
6001 return -EINVAL;
6002
6003 local_name = btf__name_by_offset(local_btf, local_type->name_off);
6004 if (!local_name)
6005 return -EINVAL;
6006
6007 if (relo->kind != BPF_CORE_TYPE_ID_LOCAL &&
6008 !hashmap__find(cand_cache, local_id, &cands)) {
6009 cands = bpf_core_find_cands(prog->obj, local_btf, local_id);
6010 if (IS_ERR(cands)) {
6011 pr_warn("prog '%s': relo #%d: target candidate search failed for [%d] %s %s: %ld\n",
6012 prog_name, relo_idx, local_id, btf_kind_str(local_type),
6013 local_name, PTR_ERR(cands));
6014 return PTR_ERR(cands);
6015 }
6016 err = hashmap__set(cand_cache, local_id, cands, NULL, NULL);
6017 if (err) {
6018 bpf_core_free_cands(cands);
6019 return err;
6020 }
6021 }
6022
6023 return bpf_core_calc_relo_insn(prog_name, relo, relo_idx, local_btf, cands, specs_scratch,
6024 targ_res);
6025 }
6026
6027 static int
6028 bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path)
6029 {
6030 const struct btf_ext_info_sec *sec;
6031 struct bpf_core_relo_res targ_res;
6032 const struct bpf_core_relo *rec;
6033 const struct btf_ext_info *seg;
6034 struct hashmap_entry *entry;
6035 struct hashmap *cand_cache = NULL;
6036 struct bpf_program *prog;
6037 struct bpf_insn *insn;
6038 const char *sec_name;
6039 int i, err = 0, insn_idx, sec_idx, sec_num;
6040
6041 if (obj->btf_ext->core_relo_info.len == 0)
6042 return 0;
6043
6044 if (targ_btf_path) {
6045 obj->btf_vmlinux_override = btf__parse(targ_btf_path, NULL);
6046 err = libbpf_get_error(obj->btf_vmlinux_override);
6047 if (err) {
6048 pr_warn("failed to parse target BTF: %s\n", errstr(err));
6049 return err;
6050 }
6051 }
6052
6053 cand_cache = hashmap__new(bpf_core_hash_fn, bpf_core_equal_fn, NULL);
6054 if (IS_ERR(cand_cache)) {
6055 err = PTR_ERR(cand_cache);
6056 goto out;
6057 }
6058
6059 seg = &obj->btf_ext->core_relo_info;
6060 sec_num = 0;
6061 for_each_btf_ext_sec(seg, sec) {
6062 sec_idx = seg->sec_idxs[sec_num];
6063 sec_num++;
6064
6065 sec_name = btf__name_by_offset(obj->btf, sec->sec_name_off);
6066 if (str_is_empty(sec_name)) {
6067 err = -EINVAL;
6068 goto out;
6069 }
6070
6071 pr_debug("sec '%s': found %d CO-RE relocations\n", sec_name, sec->num_info);
6072
6073 for_each_btf_ext_rec(seg, sec, i, rec) {
6074 if (rec->insn_off % BPF_INSN_SZ)
6075 return -EINVAL;
6076 insn_idx = rec->insn_off / BPF_INSN_SZ;
6077 prog = find_prog_by_sec_insn(obj, sec_idx, insn_idx);
6078 if (!prog) {
6079 /* When __weak subprog is "overridden" by another instance
6080 * of the subprog from a different object file, linker still
6081 * appends all the .BTF.ext info that used to belong to that
6082 * eliminated subprogram.
6083 * This is similar to what x86-64 linker does for relocations.
6084 * So just ignore such relocations just like we ignore
6085 * subprog instructions when discovering subprograms.
6086 */
6087 pr_debug("sec '%s': skipping CO-RE relocation #%d for insn #%d belonging to eliminated weak subprogram\n",
6088 sec_name, i, insn_idx);
6089 continue;
6090 }
6091 /* no need to apply CO-RE relocation if the program is
6092 * not going to be loaded
6093 */
6094 if (!prog->autoload)
6095 continue;
6096
6097 /* adjust insn_idx from section frame of reference to the local
6098 * program's frame of reference; (sub-)program code is not yet
6099 * relocated, so it's enough to just subtract in-section offset
6100 */
6101 insn_idx = insn_idx - prog->sec_insn_off;
6102 if (insn_idx >= prog->insns_cnt)
6103 return -EINVAL;
6104 insn = &prog->insns[insn_idx];
6105
6106 err = record_relo_core(prog, rec, insn_idx);
6107 if (err) {
6108 pr_warn("prog '%s': relo #%d: failed to record relocation: %s\n",
6109 prog->name, i, errstr(err));
6110 goto out;
6111 }
6112
6113 if (prog->obj->gen_loader)
6114 continue;
6115
6116 err = bpf_core_resolve_relo(prog, rec, i, obj->btf, cand_cache, &targ_res);
6117 if (err) {
6118 pr_warn("prog '%s': relo #%d: failed to relocate: %s\n",
6119 prog->name, i, errstr(err));
6120 goto out;
6121 }
6122
6123 err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res);
6124 if (err) {
6125 pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %s\n",
6126 prog->name, i, insn_idx, errstr(err));
6127 goto out;
6128 }
6129 }
6130 }
6131
6132 out:
6133 /* obj->btf_vmlinux and module BTFs are freed after object load */
6134 btf__free(obj->btf_vmlinux_override);
6135 obj->btf_vmlinux_override = NULL;
6136
6137 if (!IS_ERR_OR_NULL(cand_cache)) {
6138 hashmap__for_each_entry(cand_cache, entry, i) {
6139 bpf_core_free_cands(entry->pvalue);
6140 }
6141 hashmap__free(cand_cache);
6142 }
6143 return err;
6144 }
6145
6146 /* base map load ldimm64 special constant, used also for log fixup logic */
6147 #define POISON_LDIMM64_MAP_BASE 2001000000
6148 #define POISON_LDIMM64_MAP_PFX "200100"
6149
6150 static void poison_map_ldimm64(struct bpf_program *prog, int relo_idx,
6151 int insn_idx, struct bpf_insn *insn,
6152 int map_idx, const struct bpf_map *map)
6153 {
6154 int i;
6155
6156 pr_debug("prog '%s': relo #%d: poisoning insn #%d that loads map #%d '%s'\n",
6157 prog->name, relo_idx, insn_idx, map_idx, map->name);
6158
6159 /* we turn single ldimm64 into two identical invalid calls */
6160 for (i = 0; i < 2; i++) {
6161 insn->code = BPF_JMP | BPF_CALL;
6162 insn->dst_reg = 0;
6163 insn->src_reg = 0;
6164 insn->off = 0;
6165 /* if this instruction is reachable (not a dead code),
6166 * verifier will complain with something like:
6167 * invalid func unknown#2001000123
6168 * where lower 123 is map index into obj->maps[] array
6169 */
6170 insn->imm = POISON_LDIMM64_MAP_BASE + map_idx;
6171
6172 insn++;
6173 }
6174 }
6175
6176 /* unresolved kfunc call special constant, used also for log fixup logic */
6177 #define POISON_CALL_KFUNC_BASE 2002000000
6178 #define POISON_CALL_KFUNC_PFX "2002"
6179
6180 static void poison_kfunc_call(struct bpf_program *prog, int relo_idx,
6181 int insn_idx, struct bpf_insn *insn,
6182 int ext_idx, const struct extern_desc *ext)
6183 {
6184 pr_debug("prog '%s': relo #%d: poisoning insn #%d that calls kfunc '%s'\n",
6185 prog->name, relo_idx, insn_idx, ext->name);
6186
6187 /* we turn kfunc call into invalid helper call with identifiable constant */
6188 insn->code = BPF_JMP | BPF_CALL;
6189 insn->dst_reg = 0;
6190 insn->src_reg = 0;
6191 insn->off = 0;
6192 /* if this instruction is reachable (not a dead code),
6193 * verifier will complain with something like:
6194 * invalid func unknown#2001000123
6195 * where lower 123 is extern index into obj->externs[] array
6196 */
6197 insn->imm = POISON_CALL_KFUNC_BASE + ext_idx;
6198 }
6199
6200 static int find_jt_map(struct bpf_object *obj, struct bpf_program *prog, unsigned int sym_off)
6201 {
6202 size_t i;
6203
6204 for (i = 0; i < obj->jumptable_map_cnt; i++) {
6205 /*
6206 * This might happen that same offset is used for two different
6207 * programs (as jump tables can be the same). However, for
6208 * different programs different maps should be created.
6209 */
6210 if (obj->jumptable_maps[i].sym_off == sym_off &&
6211 obj->jumptable_maps[i].prog == prog)
6212 return obj->jumptable_maps[i].fd;
6213 }
6214
6215 return -ENOENT;
6216 }
6217
6218 static int add_jt_map(struct bpf_object *obj, struct bpf_program *prog, unsigned int sym_off, int map_fd)
6219 {
6220 size_t cnt = obj->jumptable_map_cnt;
6221 size_t size = sizeof(obj->jumptable_maps[0]);
6222 void *tmp;
6223
6224 tmp = libbpf_reallocarray(obj->jumptable_maps, cnt + 1, size);
6225 if (!tmp)
6226 return -ENOMEM;
6227
6228 obj->jumptable_maps = tmp;
6229 obj->jumptable_maps[cnt].prog = prog;
6230 obj->jumptable_maps[cnt].sym_off = sym_off;
6231 obj->jumptable_maps[cnt].fd = map_fd;
6232 obj->jumptable_map_cnt++;
6233
6234 return 0;
6235 }
6236
6237 static int find_subprog_idx(struct bpf_program *prog, int insn_idx)
6238 {
6239 int i;
6240
6241 for (i = prog->subprog_cnt - 1; i >= 0; i--) {
6242 if (insn_idx >= prog->subprogs[i].sub_insn_off)
6243 return i;
6244 }
6245
6246 return -1;
6247 }
6248
6249 static int create_jt_map(struct bpf_object *obj, struct bpf_program *prog, struct reloc_desc *relo)
6250 {
6251 const __u32 jt_entry_size = 8;
6252 unsigned int sym_off = relo->sym_off;
6253 int jt_size = relo->sym_size;
6254 __u32 max_entries = jt_size / jt_entry_size;
6255 __u32 value_size = sizeof(struct bpf_insn_array_value);
6256 struct bpf_insn_array_value val = {};
6257 int subprog_idx;
6258 int map_fd, err;
6259 __u64 insn_off;
6260 __u64 *jt;
6261 __u32 i;
6262
6263 map_fd = find_jt_map(obj, prog, sym_off);
6264 if (map_fd >= 0)
6265 return map_fd;
6266
6267 if (sym_off % jt_entry_size) {
6268 pr_warn("map '.jumptables': jumptable start %u should be multiple of %u\n",
6269 sym_off, jt_entry_size);
6270 return -EINVAL;
6271 }
6272
6273 if (jt_size % jt_entry_size) {
6274 pr_warn("map '.jumptables': jumptable size %d should be multiple of %u\n",
6275 jt_size, jt_entry_size);
6276 return -EINVAL;
6277 }
6278
6279 map_fd = bpf_map_create(BPF_MAP_TYPE_INSN_ARRAY, ".jumptables",
6280 4, value_size, max_entries, NULL);
6281 if (map_fd < 0)
6282 return map_fd;
6283
6284 if (!obj->jumptables_data) {
6285 pr_warn("map '.jumptables': ELF file is missing jump table data\n");
6286 err = -EINVAL;
6287 goto err_close;
6288 }
6289 if (sym_off + jt_size > obj->jumptables_data_sz) {
6290 pr_warn("map '.jumptables': jumptables_data size is %zd, trying to access %d\n",
6291 obj->jumptables_data_sz, sym_off + jt_size);
6292 err = -EINVAL;
6293 goto err_close;
6294 }
6295
6296 subprog_idx = -1; /* main program */
6297 if (relo->insn_idx < 0 || relo->insn_idx >= prog->insns_cnt) {
6298 pr_warn("map '.jumptables': invalid instruction index %d\n", relo->insn_idx);
6299 err = -EINVAL;
6300 goto err_close;
6301 }
6302 if (prog->subprogs)
6303 subprog_idx = find_subprog_idx(prog, relo->insn_idx);
6304
6305 jt = (__u64 *)(obj->jumptables_data + sym_off);
6306 for (i = 0; i < max_entries; i++) {
6307 /*
6308 * The offset should be made to be relative to the beginning of
6309 * the main function, not the subfunction.
6310 */
6311 insn_off = jt[i]/sizeof(struct bpf_insn);
6312 if (subprog_idx >= 0) {
6313 insn_off -= prog->subprogs[subprog_idx].sec_insn_off;
6314 insn_off += prog->subprogs[subprog_idx].sub_insn_off;
6315 } else {
6316 insn_off -= prog->sec_insn_off;
6317 }
6318
6319 /*
6320 * LLVM-generated jump tables contain u64 records, however
6321 * should contain values that fit in u32.
6322 */
6323 if (insn_off > UINT32_MAX) {
6324 pr_warn("map '.jumptables': invalid jump table value 0x%llx at offset %u\n",
6325 (long long)jt[i], sym_off + i * jt_entry_size);
6326 err = -EINVAL;
6327 goto err_close;
6328 }
6329
6330 val.orig_off = insn_off;
6331 err = bpf_map_update_elem(map_fd, &i, &val, 0);
6332 if (err)
6333 goto err_close;
6334 }
6335
6336 err = bpf_map_freeze(map_fd);
6337 if (err)
6338 goto err_close;
6339
6340 err = add_jt_map(obj, prog, sym_off, map_fd);
6341 if (err)
6342 goto err_close;
6343
6344 return map_fd;
6345
6346 err_close:
6347 close(map_fd);
6348 return err;
6349 }
6350
6351 /* Relocate data references within program code:
6352 * - map references;
6353 * - global variable references;
6354 * - extern references.
6355 */
6356 static int
6357 bpf_object__relocate_data(struct bpf_object *obj, struct bpf_program *prog)
6358 {
6359 int i;
6360
6361 for (i = 0; i < prog->nr_reloc; i++) {
6362 struct reloc_desc *relo = &prog->reloc_desc[i];
6363 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
6364 const struct bpf_map *map;
6365 struct extern_desc *ext;
6366
6367 switch (relo->type) {
6368 case RELO_LD64:
6369 map = &obj->maps[relo->map_idx];
6370 if (obj->gen_loader) {
6371 insn[0].src_reg = BPF_PSEUDO_MAP_IDX;
6372 insn[0].imm = relo->map_idx;
6373 } else if (map->autocreate) {
6374 insn[0].src_reg = BPF_PSEUDO_MAP_FD;
6375 insn[0].imm = map->fd;
6376 } else {
6377 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6378 relo->map_idx, map);
6379 }
6380 break;
6381 case RELO_DATA:
6382 map = &obj->maps[relo->map_idx];
6383 insn[1].imm = insn[0].imm + relo->sym_off;
6384
6385 if (relo->map_idx == obj->arena_map_idx)
6386 insn[1].imm += obj->arena_data_off;
6387
6388 if (obj->gen_loader) {
6389 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6390 insn[0].imm = relo->map_idx;
6391 } else if (map->autocreate) {
6392 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6393 insn[0].imm = map->fd;
6394 } else {
6395 poison_map_ldimm64(prog, i, relo->insn_idx, insn,
6396 relo->map_idx, map);
6397 }
6398 break;
6399 case RELO_EXTERN_LD64:
6400 ext = &obj->externs[relo->ext_idx];
6401 if (ext->type == EXT_KCFG) {
6402 if (obj->gen_loader) {
6403 insn[0].src_reg = BPF_PSEUDO_MAP_IDX_VALUE;
6404 insn[0].imm = obj->kconfig_map_idx;
6405 } else {
6406 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6407 insn[0].imm = obj->maps[obj->kconfig_map_idx].fd;
6408 }
6409 insn[1].imm = ext->kcfg.data_off;
6410 } else /* EXT_KSYM */ {
6411 if (ext->ksym.type_id && ext->is_set) { /* typed ksyms */
6412 insn[0].src_reg = BPF_PSEUDO_BTF_ID;
6413 insn[0].imm = ext->ksym.kernel_btf_id;
6414 insn[1].imm = ext->ksym.kernel_btf_obj_fd;
6415 } else { /* typeless ksyms or unresolved typed ksyms */
6416 insn[0].imm = (__u32)ext->ksym.addr;
6417 insn[1].imm = ext->ksym.addr >> 32;
6418 }
6419 }
6420 break;
6421 case RELO_EXTERN_CALL:
6422 ext = &obj->externs[relo->ext_idx];
6423 insn[0].src_reg = BPF_PSEUDO_KFUNC_CALL;
6424 if (ext->is_set) {
6425 insn[0].imm = ext->ksym.kernel_btf_id;
6426 insn[0].off = ext->ksym.btf_fd_idx;
6427 } else { /* unresolved weak kfunc call */
6428 poison_kfunc_call(prog, i, relo->insn_idx, insn,
6429 relo->ext_idx, ext);
6430 }
6431 break;
6432 case RELO_SUBPROG_ADDR:
6433 if (insn[0].src_reg != BPF_PSEUDO_FUNC) {
6434 pr_warn("prog '%s': relo #%d: bad insn\n",
6435 prog->name, i);
6436 return -EINVAL;
6437 }
6438 /* handled already */
6439 break;
6440 case RELO_CALL:
6441 /* handled already */
6442 break;
6443 case RELO_CORE:
6444 /* will be handled by bpf_program_record_relos() */
6445 break;
6446 case RELO_INSN_ARRAY: {
6447 int map_fd;
6448
6449 map_fd = create_jt_map(obj, prog, relo);
6450 if (map_fd < 0) {
6451 pr_warn("prog '%s': relo #%d: can't create jump table: sym_off %u\n",
6452 prog->name, i, relo->sym_off);
6453 return map_fd;
6454 }
6455 insn[0].src_reg = BPF_PSEUDO_MAP_VALUE;
6456 insn->imm = map_fd;
6457 insn->off = 0;
6458 }
6459 break;
6460 default:
6461 pr_warn("prog '%s': relo #%d: bad relo type %d\n",
6462 prog->name, i, relo->type);
6463 return -EINVAL;
6464 }
6465 }
6466
6467 return 0;
6468 }
6469
6470 static int adjust_prog_btf_ext_info(const struct bpf_object *obj,
6471 const struct bpf_program *prog,
6472 const struct btf_ext_info *ext_info,
6473 void **prog_info, __u32 *prog_rec_cnt,
6474 __u32 *prog_rec_sz)
6475 {
6476 void *copy_start = NULL, *copy_end = NULL;
6477 void *rec, *rec_end, *new_prog_info;
6478 const struct btf_ext_info_sec *sec;
6479 size_t old_sz, new_sz;
6480 int i, sec_num, sec_idx, off_adj;
6481
6482 sec_num = 0;
6483 for_each_btf_ext_sec(ext_info, sec) {
6484 sec_idx = ext_info->sec_idxs[sec_num];
6485 sec_num++;
6486 if (prog->sec_idx != sec_idx)
6487 continue;
6488
6489 for_each_btf_ext_rec(ext_info, sec, i, rec) {
6490 __u32 insn_off = *(__u32 *)rec / BPF_INSN_SZ;
6491
6492 if (insn_off < prog->sec_insn_off)
6493 continue;
6494 if (insn_off >= prog->sec_insn_off + prog->sec_insn_cnt)
6495 break;
6496
6497 if (!copy_start)
6498 copy_start = rec;
6499 copy_end = rec + ext_info->rec_size;
6500 }
6501
6502 if (!copy_start)
6503 return -ENOENT;
6504
6505 /* append func/line info of a given (sub-)program to the main
6506 * program func/line info
6507 */
6508 old_sz = (size_t)(*prog_rec_cnt) * ext_info->rec_size;
6509 new_sz = old_sz + (copy_end - copy_start);
6510 new_prog_info = realloc(*prog_info, new_sz);
6511 if (!new_prog_info)
6512 return -ENOMEM;
6513 *prog_info = new_prog_info;
6514 *prog_rec_cnt = new_sz / ext_info->rec_size;
6515 memcpy(new_prog_info + old_sz, copy_start, copy_end - copy_start);
6516
6517 /* Kernel instruction offsets are in units of 8-byte
6518 * instructions, while .BTF.ext instruction offsets generated
6519 * by Clang are in units of bytes. So convert Clang offsets
6520 * into kernel offsets and adjust offset according to program
6521 * relocated position.
6522 */
6523 off_adj = prog->sub_insn_off - prog->sec_insn_off;
6524 rec = new_prog_info + old_sz;
6525 rec_end = new_prog_info + new_sz;
6526 for (; rec < rec_end; rec += ext_info->rec_size) {
6527 __u32 *insn_off = rec;
6528
6529 *insn_off = *insn_off / BPF_INSN_SZ + off_adj;
6530 }
6531 *prog_rec_sz = ext_info->rec_size;
6532 return 0;
6533 }
6534
6535 return -ENOENT;
6536 }
6537
6538 static int
6539 reloc_prog_func_and_line_info(const struct bpf_object *obj,
6540 struct bpf_program *main_prog,
6541 const struct bpf_program *prog)
6542 {
6543 int err;
6544
6545 /* no .BTF.ext relocation if .BTF.ext is missing or kernel doesn't
6546 * support func/line info
6547 */
6548 if (!obj->btf_ext || !kernel_supports(obj, FEAT_BTF_FUNC))
6549 return 0;
6550
6551 /* only attempt func info relocation if main program's func_info
6552 * relocation was successful
6553 */
6554 if (main_prog != prog && !main_prog->func_info)
6555 goto line_info;
6556
6557 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->func_info,
6558 &main_prog->func_info,
6559 &main_prog->func_info_cnt,
6560 &main_prog->func_info_rec_size);
6561 if (err) {
6562 if (err != -ENOENT) {
6563 pr_warn("prog '%s': error relocating .BTF.ext function info: %s\n",
6564 prog->name, errstr(err));
6565 return err;
6566 }
6567 if (main_prog->func_info) {
6568 /*
6569 * Some info has already been found but has problem
6570 * in the last btf_ext reloc. Must have to error out.
6571 */
6572 pr_warn("prog '%s': missing .BTF.ext function info.\n", prog->name);
6573 return err;
6574 }
6575 /* Have problem loading the very first info. Ignore the rest. */
6576 pr_warn("prog '%s': missing .BTF.ext function info for the main program, skipping all of .BTF.ext func info.\n",
6577 prog->name);
6578 }
6579
6580 line_info:
6581 /* don't relocate line info if main program's relocation failed */
6582 if (main_prog != prog && !main_prog->line_info)
6583 return 0;
6584
6585 err = adjust_prog_btf_ext_info(obj, prog, &obj->btf_ext->line_info,
6586 &main_prog->line_info,
6587 &main_prog->line_info_cnt,
6588 &main_prog->line_info_rec_size);
6589 if (err) {
6590 if (err != -ENOENT) {
6591 pr_warn("prog '%s': error relocating .BTF.ext line info: %s\n",
6592 prog->name, errstr(err));
6593 return err;
6594 }
6595 if (main_prog->line_info) {
6596 /*
6597 * Some info has already been found but has problem
6598 * in the last btf_ext reloc. Must have to error out.
6599 */
6600 pr_warn("prog '%s': missing .BTF.ext line info.\n", prog->name);
6601 return err;
6602 }
6603 /* Have problem loading the very first info. Ignore the rest. */
6604 pr_warn("prog '%s': missing .BTF.ext line info for the main program, skipping all of .BTF.ext line info.\n",
6605 prog->name);
6606 }
6607 return 0;
6608 }
6609
6610 static int cmp_relo_by_insn_idx(const void *key, const void *elem)
6611 {
6612 size_t insn_idx = *(const size_t *)key;
6613 const struct reloc_desc *relo = elem;
6614
6615 if (insn_idx == relo->insn_idx)
6616 return 0;
6617 return insn_idx < relo->insn_idx ? -1 : 1;
6618 }
6619
6620 static struct reloc_desc *find_prog_insn_relo(const struct bpf_program *prog, size_t insn_idx)
6621 {
6622 if (!prog->nr_reloc)
6623 return NULL;
6624 return bsearch(&insn_idx, prog->reloc_desc, prog->nr_reloc,
6625 sizeof(*prog->reloc_desc), cmp_relo_by_insn_idx);
6626 }
6627
6628 static int append_subprog_relos(struct bpf_program *main_prog, struct bpf_program *subprog)
6629 {
6630 int new_cnt = main_prog->nr_reloc + subprog->nr_reloc;
6631 struct reloc_desc *relos;
6632 int i;
6633
6634 if (main_prog == subprog)
6635 return 0;
6636 relos = libbpf_reallocarray(main_prog->reloc_desc, new_cnt, sizeof(*relos));
6637 /* if new count is zero, reallocarray can return a valid NULL result;
6638 * in this case the previous pointer will be freed, so we *have to*
6639 * reassign old pointer to the new value (even if it's NULL)
6640 */
6641 if (!relos && new_cnt)
6642 return -ENOMEM;
6643 if (subprog->nr_reloc)
6644 memcpy(relos + main_prog->nr_reloc, subprog->reloc_desc,
6645 sizeof(*relos) * subprog->nr_reloc);
6646
6647 for (i = main_prog->nr_reloc; i < new_cnt; i++)
6648 relos[i].insn_idx += subprog->sub_insn_off;
6649 /* After insn_idx adjustment the 'relos' array is still sorted
6650 * by insn_idx and doesn't break bsearch.
6651 */
6652 main_prog->reloc_desc = relos;
6653 main_prog->nr_reloc = new_cnt;
6654 return 0;
6655 }
6656
6657 static int save_subprog_offsets(struct bpf_program *main_prog, struct bpf_program *subprog)
6658 {
6659 size_t size = sizeof(main_prog->subprogs[0]);
6660 int cnt = main_prog->subprog_cnt;
6661 void *tmp;
6662
6663 tmp = libbpf_reallocarray(main_prog->subprogs, cnt + 1, size);
6664 if (!tmp)
6665 return -ENOMEM;
6666
6667 main_prog->subprogs = tmp;
6668 main_prog->subprogs[cnt].sec_insn_off = subprog->sec_insn_off;
6669 main_prog->subprogs[cnt].sub_insn_off = subprog->sub_insn_off;
6670 main_prog->subprog_cnt++;
6671
6672 return 0;
6673 }
6674
6675 static int
6676 bpf_object__append_subprog_code(struct bpf_object *obj, struct bpf_program *main_prog,
6677 struct bpf_program *subprog)
6678 {
6679 struct bpf_insn *insns;
6680 size_t new_cnt;
6681 int err;
6682
6683 subprog->sub_insn_off = main_prog->insns_cnt;
6684
6685 new_cnt = main_prog->insns_cnt + subprog->insns_cnt;
6686 insns = libbpf_reallocarray(main_prog->insns, new_cnt, sizeof(*insns));
6687 if (!insns) {
6688 pr_warn("prog '%s': failed to realloc prog code\n", main_prog->name);
6689 return -ENOMEM;
6690 }
6691 main_prog->insns = insns;
6692 main_prog->insns_cnt = new_cnt;
6693
6694 memcpy(main_prog->insns + subprog->sub_insn_off, subprog->insns,
6695 subprog->insns_cnt * sizeof(*insns));
6696
6697 pr_debug("prog '%s': added %zu insns from sub-prog '%s'\n",
6698 main_prog->name, subprog->insns_cnt, subprog->name);
6699
6700 /* The subprog insns are now appended. Append its relos too. */
6701 err = append_subprog_relos(main_prog, subprog);
6702 if (err)
6703 return err;
6704
6705 err = save_subprog_offsets(main_prog, subprog);
6706 if (err) {
6707 pr_warn("prog '%s': failed to add subprog offsets: %s\n",
6708 main_prog->name, errstr(err));
6709 return err;
6710 }
6711
6712 return 0;
6713 }
6714
6715 static int
6716 bpf_object__reloc_code(struct bpf_object *obj, struct bpf_program *main_prog,
6717 struct bpf_program *prog)
6718 {
6719 size_t sub_insn_idx, insn_idx;
6720 struct bpf_program *subprog;
6721 struct reloc_desc *relo;
6722 struct bpf_insn *insn;
6723 int err;
6724
6725 err = reloc_prog_func_and_line_info(obj, main_prog, prog);
6726 if (err)
6727 return err;
6728
6729 for (insn_idx = 0; insn_idx < prog->sec_insn_cnt; insn_idx++) {
6730 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6731 if (!insn_is_subprog_call(insn) && !insn_is_pseudo_func(insn))
6732 continue;
6733
6734 relo = find_prog_insn_relo(prog, insn_idx);
6735 if (relo && relo->type == RELO_EXTERN_CALL)
6736 /* kfunc relocations will be handled later
6737 * in bpf_object__relocate_data()
6738 */
6739 continue;
6740 if (relo && relo->type != RELO_CALL && relo->type != RELO_SUBPROG_ADDR) {
6741 pr_warn("prog '%s': unexpected relo for insn #%zu, type %d\n",
6742 prog->name, insn_idx, relo->type);
6743 return -LIBBPF_ERRNO__RELOC;
6744 }
6745 if (relo) {
6746 /* sub-program instruction index is a combination of
6747 * an offset of a symbol pointed to by relocation and
6748 * call instruction's imm field; for global functions,
6749 * call always has imm = -1, but for static functions
6750 * relocation is against STT_SECTION and insn->imm
6751 * points to a start of a static function
6752 *
6753 * for subprog addr relocation, the relo->sym_off + insn->imm is
6754 * the byte offset in the corresponding section.
6755 */
6756 if (relo->type == RELO_CALL)
6757 sub_insn_idx = relo->sym_off / BPF_INSN_SZ + insn->imm + 1;
6758 else
6759 sub_insn_idx = (relo->sym_off + insn->imm) / BPF_INSN_SZ;
6760 } else if (insn_is_pseudo_func(insn)) {
6761 /*
6762 * RELO_SUBPROG_ADDR relo is always emitted even if both
6763 * functions are in the same section, so it shouldn't reach here.
6764 */
6765 pr_warn("prog '%s': missing subprog addr relo for insn #%zu\n",
6766 prog->name, insn_idx);
6767 return -LIBBPF_ERRNO__RELOC;
6768 } else {
6769 /* if subprogram call is to a static function within
6770 * the same ELF section, there won't be any relocation
6771 * emitted, but it also means there is no additional
6772 * offset necessary, insns->imm is relative to
6773 * instruction's original position within the section
6774 */
6775 sub_insn_idx = prog->sec_insn_off + insn_idx + insn->imm + 1;
6776 }
6777
6778 /* we enforce that sub-programs should be in .text section */
6779 subprog = find_prog_by_sec_insn(obj, obj->efile.text_shndx, sub_insn_idx);
6780 if (!subprog) {
6781 pr_warn("prog '%s': no .text section found yet sub-program call exists\n",
6782 prog->name);
6783 return -LIBBPF_ERRNO__RELOC;
6784 }
6785
6786 /* if it's the first call instruction calling into this
6787 * subprogram (meaning this subprog hasn't been processed
6788 * yet) within the context of current main program:
6789 * - append it at the end of main program's instructions blog;
6790 * - process is recursively, while current program is put on hold;
6791 * - if that subprogram calls some other not yet processes
6792 * subprogram, same thing will happen recursively until
6793 * there are no more unprocesses subprograms left to append
6794 * and relocate.
6795 */
6796 if (subprog->sub_insn_off == 0) {
6797 err = bpf_object__append_subprog_code(obj, main_prog, subprog);
6798 if (err)
6799 return err;
6800 err = bpf_object__reloc_code(obj, main_prog, subprog);
6801 if (err)
6802 return err;
6803 }
6804
6805 /* main_prog->insns memory could have been re-allocated, so
6806 * calculate pointer again
6807 */
6808 insn = &main_prog->insns[prog->sub_insn_off + insn_idx];
6809 /* calculate correct instruction position within current main
6810 * prog; each main prog can have a different set of
6811 * subprograms appended (potentially in different order as
6812 * well), so position of any subprog can be different for
6813 * different main programs
6814 */
6815 insn->imm = subprog->sub_insn_off - (prog->sub_insn_off + insn_idx) - 1;
6816
6817 pr_debug("prog '%s': insn #%zu relocated, imm %d points to subprog '%s' (now at %zu offset)\n",
6818 prog->name, insn_idx, insn->imm, subprog->name, subprog->sub_insn_off);
6819 }
6820
6821 return 0;
6822 }
6823
6824 /*
6825 * Relocate sub-program calls.
6826 *
6827 * Algorithm operates as follows. Each entry-point BPF program (referred to as
6828 * main prog) is processed separately. For each subprog (non-entry functions,
6829 * that can be called from either entry progs or other subprogs) gets their
6830 * sub_insn_off reset to zero. This serves as indicator that this subprogram
6831 * hasn't been yet appended and relocated within current main prog. Once its
6832 * relocated, sub_insn_off will point at the position within current main prog
6833 * where given subprog was appended. This will further be used to relocate all
6834 * the call instructions jumping into this subprog.
6835 *
6836 * We start with main program and process all call instructions. If the call
6837 * is into a subprog that hasn't been processed (i.e., subprog->sub_insn_off
6838 * is zero), subprog instructions are appended at the end of main program's
6839 * instruction array. Then main program is "put on hold" while we recursively
6840 * process newly appended subprogram. If that subprogram calls into another
6841 * subprogram that hasn't been appended, new subprogram is appended again to
6842 * the *main* prog's instructions (subprog's instructions are always left
6843 * untouched, as they need to be in unmodified state for subsequent main progs
6844 * and subprog instructions are always sent only as part of a main prog) and
6845 * the process continues recursively. Once all the subprogs called from a main
6846 * prog or any of its subprogs are appended (and relocated), all their
6847 * positions within finalized instructions array are known, so it's easy to
6848 * rewrite call instructions with correct relative offsets, corresponding to
6849 * desired target subprog.
6850 *
6851 * Its important to realize that some subprogs might not be called from some
6852 * main prog and any of its called/used subprogs. Those will keep their
6853 * subprog->sub_insn_off as zero at all times and won't be appended to current
6854 * main prog and won't be relocated within the context of current main prog.
6855 * They might still be used from other main progs later.
6856 *
6857 * Visually this process can be shown as below. Suppose we have two main
6858 * programs mainA and mainB and BPF object contains three subprogs: subA,
6859 * subB, and subC. mainA calls only subA, mainB calls only subC, but subA and
6860 * subC both call subB:
6861 *
6862 * +--------+ +-------+
6863 * | v v |
6864 * +--+---+ +--+-+-+ +---+--+
6865 * | subA | | subB | | subC |
6866 * +--+---+ +------+ +---+--+
6867 * ^ ^
6868 * | |
6869 * +---+-------+ +------+----+
6870 * | mainA | | mainB |
6871 * +-----------+ +-----------+
6872 *
6873 * We'll start relocating mainA, will find subA, append it and start
6874 * processing sub A recursively:
6875 *
6876 * +-----------+------+
6877 * | mainA | subA |
6878 * +-----------+------+
6879 *
6880 * At this point we notice that subB is used from subA, so we append it and
6881 * relocate (there are no further subcalls from subB):
6882 *
6883 * +-----------+------+------+
6884 * | mainA | subA | subB |
6885 * +-----------+------+------+
6886 *
6887 * At this point, we relocate subA calls, then go one level up and finish with
6888 * relocatin mainA calls. mainA is done.
6889 *
6890 * For mainB process is similar but results in different order. We start with
6891 * mainB and skip subA and subB, as mainB never calls them (at least
6892 * directly), but we see subC is needed, so we append and start processing it:
6893 *
6894 * +-----------+------+
6895 * | mainB | subC |
6896 * +-----------+------+
6897 * Now we see subC needs subB, so we go back to it, append and relocate it:
6898 *
6899 * +-----------+------+------+
6900 * | mainB | subC | subB |
6901 * +-----------+------+------+
6902 *
6903 * At this point we unwind recursion, relocate calls in subC, then in mainB.
6904 */
6905 static int
6906 bpf_object__relocate_calls(struct bpf_object *obj, struct bpf_program *prog)
6907 {
6908 struct bpf_program *subprog;
6909 int i, err;
6910
6911 /* mark all subprogs as not relocated (yet) within the context of
6912 * current main program
6913 */
6914 for (i = 0; i < obj->nr_programs; i++) {
6915 subprog = &obj->programs[i];
6916 if (!prog_is_subprog(obj, subprog))
6917 continue;
6918
6919 subprog->sub_insn_off = 0;
6920 }
6921
6922 err = bpf_object__reloc_code(obj, prog, prog);
6923 if (err)
6924 return err;
6925
6926 return 0;
6927 }
6928
6929 static void
6930 bpf_object__free_relocs(struct bpf_object *obj)
6931 {
6932 struct bpf_program *prog;
6933 int i;
6934
6935 /* free up relocation descriptors */
6936 for (i = 0; i < obj->nr_programs; i++) {
6937 prog = &obj->programs[i];
6938 zfree(&prog->reloc_desc);
6939 prog->nr_reloc = 0;
6940 }
6941 }
6942
6943 static int cmp_relocs(const void *_a, const void *_b)
6944 {
6945 const struct reloc_desc *a = _a;
6946 const struct reloc_desc *b = _b;
6947
6948 if (a->insn_idx != b->insn_idx)
6949 return a->insn_idx < b->insn_idx ? -1 : 1;
6950
6951 /* no two relocations should have the same insn_idx, but ... */
6952 if (a->type != b->type)
6953 return a->type < b->type ? -1 : 1;
6954
6955 return 0;
6956 }
6957
6958 static void bpf_object__sort_relos(struct bpf_object *obj)
6959 {
6960 int i;
6961
6962 for (i = 0; i < obj->nr_programs; i++) {
6963 struct bpf_program *p = &obj->programs[i];
6964
6965 if (!p->nr_reloc)
6966 continue;
6967
6968 qsort(p->reloc_desc, p->nr_reloc, sizeof(*p->reloc_desc), cmp_relocs);
6969 }
6970 }
6971
6972 static int bpf_prog_assign_exc_cb(struct bpf_object *obj, struct bpf_program *prog)
6973 {
6974 const char *str = "exception_callback:";
6975 size_t pfx_len = strlen(str);
6976 int i, j, n;
6977
6978 if (!obj->btf || !kernel_supports(obj, FEAT_BTF_DECL_TAG))
6979 return 0;
6980
6981 n = btf__type_cnt(obj->btf);
6982 for (i = 1; i < n; i++) {
6983 const char *name;
6984 struct btf_type *t;
6985
6986 t = btf_type_by_id(obj->btf, i);
6987 if (!btf_is_decl_tag(t) || btf_decl_tag(t)->component_idx != -1)
6988 continue;
6989
6990 name = btf__str_by_offset(obj->btf, t->name_off);
6991 if (strncmp(name, str, pfx_len) != 0)
6992 continue;
6993
6994 t = btf_type_by_id(obj->btf, t->type);
6995 if (!btf_is_func(t) || btf_func_linkage(t) != BTF_FUNC_GLOBAL) {
6996 pr_warn("prog '%s': exception_callback:<value> decl tag not applied to the main program\n",
6997 prog->name);
6998 return -EINVAL;
6999 }
7000 if (strcmp(prog->name, btf__str_by_offset(obj->btf, t->name_off)) != 0)
7001 continue;
7002 /* Multiple callbacks are specified for the same prog,
7003 * the verifier will eventually return an error for this
7004 * case, hence simply skip appending a subprog.
7005 */
7006 if (prog->exception_cb_idx >= 0) {
7007 prog->exception_cb_idx = -1;
7008 break;
7009 }
7010
7011 name += pfx_len;
7012 if (str_is_empty(name)) {
7013 pr_warn("prog '%s': exception_callback:<value> decl tag contains empty value\n",
7014 prog->name);
7015 return -EINVAL;
7016 }
7017
7018 for (j = 0; j < obj->nr_programs; j++) {
7019 struct bpf_program *subprog = &obj->programs[j];
7020
7021 if (!prog_is_subprog(obj, subprog))
7022 continue;
7023 if (strcmp(name, subprog->name) != 0)
7024 continue;
7025 /* Enforce non-hidden, as from verifier point of
7026 * view it expects global functions, whereas the
7027 * mark_btf_static fixes up linkage as static.
7028 */
7029 if (!subprog->sym_global || subprog->mark_btf_static) {
7030 pr_warn("prog '%s': exception callback %s must be a global non-hidden function\n",
7031 prog->name, subprog->name);
7032 return -EINVAL;
7033 }
7034 /* Let's see if we already saw a static exception callback with the same name */
7035 if (prog->exception_cb_idx >= 0) {
7036 pr_warn("prog '%s': multiple subprogs with same name as exception callback '%s'\n",
7037 prog->name, subprog->name);
7038 return -EINVAL;
7039 }
7040 prog->exception_cb_idx = j;
7041 break;
7042 }
7043
7044 if (prog->exception_cb_idx >= 0)
7045 continue;
7046
7047 pr_warn("prog '%s': cannot find exception callback '%s'\n", prog->name, name);
7048 return -ENOENT;
7049 }
7050
7051 return 0;
7052 }
7053
7054 static struct {
7055 enum bpf_prog_type prog_type;
7056 const char *ctx_name;
7057 } global_ctx_map[] = {
7058 { BPF_PROG_TYPE_CGROUP_DEVICE, "bpf_cgroup_dev_ctx" },
7059 { BPF_PROG_TYPE_CGROUP_SKB, "__sk_buff" },
7060 { BPF_PROG_TYPE_CGROUP_SOCK, "bpf_sock" },
7061 { BPF_PROG_TYPE_CGROUP_SOCK_ADDR, "bpf_sock_addr" },
7062 { BPF_PROG_TYPE_CGROUP_SOCKOPT, "bpf_sockopt" },
7063 { BPF_PROG_TYPE_CGROUP_SYSCTL, "bpf_sysctl" },
7064 { BPF_PROG_TYPE_FLOW_DISSECTOR, "__sk_buff" },
7065 { BPF_PROG_TYPE_KPROBE, "bpf_user_pt_regs_t" },
7066 { BPF_PROG_TYPE_LWT_IN, "__sk_buff" },
7067 { BPF_PROG_TYPE_LWT_OUT, "__sk_buff" },
7068 { BPF_PROG_TYPE_LWT_SEG6LOCAL, "__sk_buff" },
7069 { BPF_PROG_TYPE_LWT_XMIT, "__sk_buff" },
7070 { BPF_PROG_TYPE_NETFILTER, "bpf_nf_ctx" },
7071 { BPF_PROG_TYPE_PERF_EVENT, "bpf_perf_event_data" },
7072 { BPF_PROG_TYPE_RAW_TRACEPOINT, "bpf_raw_tracepoint_args" },
7073 { BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE, "bpf_raw_tracepoint_args" },
7074 { BPF_PROG_TYPE_SCHED_ACT, "__sk_buff" },
7075 { BPF_PROG_TYPE_SCHED_CLS, "__sk_buff" },
7076 { BPF_PROG_TYPE_SK_LOOKUP, "bpf_sk_lookup" },
7077 { BPF_PROG_TYPE_SK_MSG, "sk_msg_md" },
7078 { BPF_PROG_TYPE_SK_REUSEPORT, "sk_reuseport_md" },
7079 { BPF_PROG_TYPE_SK_SKB, "__sk_buff" },
7080 { BPF_PROG_TYPE_SOCK_OPS, "bpf_sock_ops" },
7081 { BPF_PROG_TYPE_SOCKET_FILTER, "__sk_buff" },
7082 { BPF_PROG_TYPE_XDP, "xdp_md" },
7083 /* all other program types don't have "named" context structs */
7084 };
7085
7086 /* forward declarations for arch-specific underlying types of bpf_user_pt_regs_t typedef,
7087 * for below __builtin_types_compatible_p() checks;
7088 * with this approach we don't need any extra arch-specific #ifdef guards
7089 */
7090 struct pt_regs;
7091 struct user_pt_regs;
7092 struct user_regs_struct;
7093
7094 static bool need_func_arg_type_fixup(const struct btf *btf, const struct bpf_program *prog,
7095 const char *subprog_name, int arg_idx,
7096 int arg_type_id, const char *ctx_name)
7097 {
7098 const struct btf_type *t;
7099 const char *tname;
7100
7101 /* check if existing parameter already matches verifier expectations */
7102 t = skip_mods_and_typedefs(btf, arg_type_id, NULL);
7103 if (!btf_is_ptr(t))
7104 goto out_warn;
7105
7106 /* typedef bpf_user_pt_regs_t is a special PITA case, valid for kprobe
7107 * and perf_event programs, so check this case early on and forget
7108 * about it for subsequent checks
7109 */
7110 while (btf_is_mod(t))
7111 t = btf__type_by_id(btf, t->type);
7112 if (btf_is_typedef(t) &&
7113 (prog->type == BPF_PROG_TYPE_KPROBE || prog->type == BPF_PROG_TYPE_PERF_EVENT)) {
7114 tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>";
7115 if (strcmp(tname, "bpf_user_pt_regs_t") == 0)
7116 return false; /* canonical type for kprobe/perf_event */
7117 }
7118
7119 /* now we can ignore typedefs moving forward */
7120 t = skip_mods_and_typedefs(btf, t->type, NULL);
7121
7122 /* if it's `void *`, definitely fix up BTF info */
7123 if (btf_is_void(t))
7124 return true;
7125
7126 /* if it's already proper canonical type, no need to fix up */
7127 tname = btf__str_by_offset(btf, t->name_off) ?: "<anon>";
7128 if (btf_is_struct(t) && strcmp(tname, ctx_name) == 0)
7129 return false;
7130
7131 /* special cases */
7132 switch (prog->type) {
7133 case BPF_PROG_TYPE_KPROBE:
7134 /* `struct pt_regs *` is expected, but we need to fix up */
7135 if (btf_is_struct(t) && strcmp(tname, "pt_regs") == 0)
7136 return true;
7137 break;
7138 case BPF_PROG_TYPE_PERF_EVENT:
7139 if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct pt_regs) &&
7140 btf_is_struct(t) && strcmp(tname, "pt_regs") == 0)
7141 return true;
7142 if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_pt_regs) &&
7143 btf_is_struct(t) && strcmp(tname, "user_pt_regs") == 0)
7144 return true;
7145 if (__builtin_types_compatible_p(bpf_user_pt_regs_t, struct user_regs_struct) &&
7146 btf_is_struct(t) && strcmp(tname, "user_regs_struct") == 0)
7147 return true;
7148 break;
7149 case BPF_PROG_TYPE_RAW_TRACEPOINT:
7150 case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
7151 /* allow u64* as ctx */
7152 if (btf_is_int(t) && t->size == 8)
7153 return true;
7154 break;
7155 default:
7156 break;
7157 }
7158
7159 out_warn:
7160 pr_warn("prog '%s': subprog '%s' arg#%d is expected to be of `struct %s *` type\n",
7161 prog->name, subprog_name, arg_idx, ctx_name);
7162 return false;
7163 }
7164
7165 static int clone_func_btf_info(struct btf *btf, int orig_fn_id, struct bpf_program *prog)
7166 {
7167 int fn_id, fn_proto_id, ret_type_id, orig_proto_id;
7168 int i, err, arg_cnt, fn_name_off, linkage;
7169 struct btf_type *fn_t, *fn_proto_t, *t;
7170 struct btf_param *p;
7171
7172 /* caller already validated FUNC -> FUNC_PROTO validity */
7173 fn_t = btf_type_by_id(btf, orig_fn_id);
7174 fn_proto_t = btf_type_by_id(btf, fn_t->type);
7175
7176 /* Note that each btf__add_xxx() operation invalidates
7177 * all btf_type and string pointers, so we need to be
7178 * very careful when cloning BTF types. BTF type
7179 * pointers have to be always refetched. And to avoid
7180 * problems with invalidated string pointers, we
7181 * add empty strings initially, then just fix up
7182 * name_off offsets in place. Offsets are stable for
7183 * existing strings, so that works out.
7184 */
7185 fn_name_off = fn_t->name_off; /* we are about to invalidate fn_t */
7186 linkage = btf_func_linkage(fn_t);
7187 orig_proto_id = fn_t->type; /* original FUNC_PROTO ID */
7188 ret_type_id = fn_proto_t->type; /* fn_proto_t will be invalidated */
7189 arg_cnt = btf_vlen(fn_proto_t);
7190
7191 /* clone FUNC_PROTO and its params */
7192 fn_proto_id = btf__add_func_proto(btf, ret_type_id);
7193 if (fn_proto_id < 0)
7194 return -EINVAL;
7195
7196 for (i = 0; i < arg_cnt; i++) {
7197 int name_off;
7198
7199 /* copy original parameter data */
7200 t = btf_type_by_id(btf, orig_proto_id);
7201 p = &btf_params(t)[i];
7202 name_off = p->name_off;
7203
7204 err = btf__add_func_param(btf, "", p->type);
7205 if (err)
7206 return err;
7207
7208 fn_proto_t = btf_type_by_id(btf, fn_proto_id);
7209 p = &btf_params(fn_proto_t)[i];
7210 p->name_off = name_off; /* use remembered str offset */
7211 }
7212
7213 /* clone FUNC now, btf__add_func() enforces non-empty name, so use
7214 * entry program's name as a placeholder, which we replace immediately
7215 * with original name_off
7216 */
7217 fn_id = btf__add_func(btf, prog->name, linkage, fn_proto_id);
7218 if (fn_id < 0)
7219 return -EINVAL;
7220
7221 fn_t = btf_type_by_id(btf, fn_id);
7222 fn_t->name_off = fn_name_off; /* reuse original string */
7223
7224 return fn_id;
7225 }
7226
7227 /* Check if main program or global subprog's function prototype has `arg:ctx`
7228 * argument tags, and, if necessary, substitute correct type to match what BPF
7229 * verifier would expect, taking into account specific program type. This
7230 * allows to support __arg_ctx tag transparently on old kernels that don't yet
7231 * have a native support for it in the verifier, making user's life much
7232 * easier.
7233 */
7234 static int bpf_program_fixup_func_info(struct bpf_object *obj, struct bpf_program *prog)
7235 {
7236 const char *ctx_name = NULL, *ctx_tag = "arg:ctx", *fn_name;
7237 struct bpf_func_info_min *func_rec;
7238 struct btf_type *fn_t, *fn_proto_t;
7239 struct btf *btf = obj->btf;
7240 const struct btf_type *t;
7241 struct btf_param *p;
7242 int ptr_id = 0, struct_id, tag_id, orig_fn_id;
7243 int i, n, arg_idx, arg_cnt, err, rec_idx;
7244 int *orig_ids;
7245
7246 /* no .BTF.ext, no problem */
7247 if (!obj->btf_ext || !prog->func_info)
7248 return 0;
7249
7250 /* don't do any fix ups if kernel natively supports __arg_ctx */
7251 if (kernel_supports(obj, FEAT_ARG_CTX_TAG))
7252 return 0;
7253
7254 /* some BPF program types just don't have named context structs, so
7255 * this fallback mechanism doesn't work for them
7256 */
7257 for (i = 0; i < ARRAY_SIZE(global_ctx_map); i++) {
7258 if (global_ctx_map[i].prog_type != prog->type)
7259 continue;
7260 ctx_name = global_ctx_map[i].ctx_name;
7261 break;
7262 }
7263 if (!ctx_name)
7264 return 0;
7265
7266 /* remember original func BTF IDs to detect if we already cloned them */
7267 orig_ids = calloc(prog->func_info_cnt, sizeof(*orig_ids));
7268 if (!orig_ids)
7269 return -ENOMEM;
7270 for (i = 0; i < prog->func_info_cnt; i++) {
7271 func_rec = prog->func_info + prog->func_info_rec_size * i;
7272 orig_ids[i] = func_rec->type_id;
7273 }
7274
7275 /* go through each DECL_TAG with "arg:ctx" and see if it points to one
7276 * of our subprogs; if yes and subprog is global and needs adjustment,
7277 * clone and adjust FUNC -> FUNC_PROTO combo
7278 */
7279 for (i = 1, n = btf__type_cnt(btf); i < n; i++) {
7280 /* only DECL_TAG with "arg:ctx" value are interesting */
7281 t = btf__type_by_id(btf, i);
7282 if (!btf_is_decl_tag(t))
7283 continue;
7284 if (strcmp(btf__str_by_offset(btf, t->name_off), ctx_tag) != 0)
7285 continue;
7286
7287 /* only global funcs need adjustment, if at all */
7288 orig_fn_id = t->type;
7289 fn_t = btf_type_by_id(btf, orig_fn_id);
7290 if (!btf_is_func(fn_t) || btf_func_linkage(fn_t) != BTF_FUNC_GLOBAL)
7291 continue;
7292
7293 /* sanity check FUNC -> FUNC_PROTO chain, just in case */
7294 fn_proto_t = btf_type_by_id(btf, fn_t->type);
7295 if (!fn_proto_t || !btf_is_func_proto(fn_proto_t))
7296 continue;
7297
7298 /* find corresponding func_info record */
7299 func_rec = NULL;
7300 for (rec_idx = 0; rec_idx < prog->func_info_cnt; rec_idx++) {
7301 if (orig_ids[rec_idx] == t->type) {
7302 func_rec = prog->func_info + prog->func_info_rec_size * rec_idx;
7303 break;
7304 }
7305 }
7306 /* current main program doesn't call into this subprog */
7307 if (!func_rec)
7308 continue;
7309
7310 /* some more sanity checking of DECL_TAG */
7311 arg_cnt = btf_vlen(fn_proto_t);
7312 arg_idx = btf_decl_tag(t)->component_idx;
7313 if (arg_idx < 0 || arg_idx >= arg_cnt)
7314 continue;
7315
7316 /* check if we should fix up argument type */
7317 p = &btf_params(fn_proto_t)[arg_idx];
7318 fn_name = btf__str_by_offset(btf, fn_t->name_off) ?: "<anon>";
7319 if (!need_func_arg_type_fixup(btf, prog, fn_name, arg_idx, p->type, ctx_name))
7320 continue;
7321
7322 /* clone fn/fn_proto, unless we already did it for another arg */
7323 if (func_rec->type_id == orig_fn_id) {
7324 int fn_id;
7325
7326 fn_id = clone_func_btf_info(btf, orig_fn_id, prog);
7327 if (fn_id < 0) {
7328 err = fn_id;
7329 goto err_out;
7330 }
7331
7332 /* point func_info record to a cloned FUNC type */
7333 func_rec->type_id = fn_id;
7334 }
7335
7336 /* create PTR -> STRUCT type chain to mark PTR_TO_CTX argument;
7337 * we do it just once per main BPF program, as all global
7338 * funcs share the same program type, so need only PTR ->
7339 * STRUCT type chain
7340 */
7341 if (ptr_id == 0) {
7342 struct_id = btf__add_struct(btf, ctx_name, 0);
7343 ptr_id = btf__add_ptr(btf, struct_id);
7344 if (ptr_id < 0 || struct_id < 0) {
7345 err = -EINVAL;
7346 goto err_out;
7347 }
7348 }
7349
7350 /* for completeness, clone DECL_TAG and point it to cloned param */
7351 tag_id = btf__add_decl_tag(btf, ctx_tag, func_rec->type_id, arg_idx);
7352 if (tag_id < 0) {
7353 err = -EINVAL;
7354 goto err_out;
7355 }
7356
7357 /* all the BTF manipulations invalidated pointers, refetch them */
7358 fn_t = btf_type_by_id(btf, func_rec->type_id);
7359 fn_proto_t = btf_type_by_id(btf, fn_t->type);
7360
7361 /* fix up type ID pointed to by param */
7362 p = &btf_params(fn_proto_t)[arg_idx];
7363 p->type = ptr_id;
7364 }
7365
7366 free(orig_ids);
7367 return 0;
7368 err_out:
7369 free(orig_ids);
7370 return err;
7371 }
7372
7373 static int bpf_object__relocate(struct bpf_object *obj, const char *targ_btf_path)
7374 {
7375 struct bpf_program *prog;
7376 size_t i, j;
7377 int err;
7378
7379 if (obj->btf_ext) {
7380 err = bpf_object__relocate_core(obj, targ_btf_path);
7381 if (err) {
7382 pr_warn("failed to perform CO-RE relocations: %s\n",
7383 errstr(err));
7384 return err;
7385 }
7386 bpf_object__sort_relos(obj);
7387 }
7388
7389 /* place globals at the end of the arena (if supported) */
7390 if (obj->arena_map_idx >= 0 && kernel_supports(obj, FEAT_LDIMM64_FULL_RANGE_OFF)) {
7391 struct bpf_map *arena_map = &obj->maps[obj->arena_map_idx];
7392
7393 obj->arena_data_off = bpf_map_mmap_sz(arena_map) -
7394 roundup(obj->arena_data_sz, sysconf(_SC_PAGE_SIZE));
7395 }
7396
7397 /* Before relocating calls pre-process relocations and mark
7398 * few ld_imm64 instructions that points to subprogs.
7399 * Otherwise bpf_object__reloc_code() later would have to consider
7400 * all ld_imm64 insns as relocation candidates. That would
7401 * reduce relocation speed, since amount of find_prog_insn_relo()
7402 * would increase and most of them will fail to find a relo.
7403 */
7404 for (i = 0; i < obj->nr_programs; i++) {
7405 prog = &obj->programs[i];
7406 for (j = 0; j < prog->nr_reloc; j++) {
7407 struct reloc_desc *relo = &prog->reloc_desc[j];
7408 struct bpf_insn *insn = &prog->insns[relo->insn_idx];
7409
7410 /* mark the insn, so it's recognized by insn_is_pseudo_func() */
7411 if (relo->type == RELO_SUBPROG_ADDR)
7412 insn[0].src_reg = BPF_PSEUDO_FUNC;
7413 }
7414 }
7415
7416 /* relocate subprogram calls and append used subprograms to main
7417 * programs; each copy of subprogram code needs to be relocated
7418 * differently for each main program, because its code location might
7419 * have changed.
7420 * Append subprog relos to main programs to allow data relos to be
7421 * processed after text is completely relocated.
7422 */
7423 for (i = 0; i < obj->nr_programs; i++) {
7424 prog = &obj->programs[i];
7425 /* sub-program's sub-calls are relocated within the context of
7426 * its main program only
7427 */
7428 if (prog_is_subprog(obj, prog))
7429 continue;
7430 if (!prog->autoload)
7431 continue;
7432
7433 err = bpf_object__relocate_calls(obj, prog);
7434 if (err) {
7435 pr_warn("prog '%s': failed to relocate calls: %s\n",
7436 prog->name, errstr(err));
7437 return err;
7438 }
7439
7440 err = bpf_prog_assign_exc_cb(obj, prog);
7441 if (err)
7442 return err;
7443 /* Now, also append exception callback if it has not been done already. */
7444 if (prog->exception_cb_idx >= 0) {
7445 struct bpf_program *subprog = &obj->programs[prog->exception_cb_idx];
7446
7447 /* Calling exception callback directly is disallowed, which the
7448 * verifier will reject later. In case it was processed already,
7449 * we can skip this step, otherwise for all other valid cases we
7450 * have to append exception callback now.
7451 */
7452 if (subprog->sub_insn_off == 0) {
7453 err = bpf_object__append_subprog_code(obj, prog, subprog);
7454 if (err)
7455 return err;
7456 err = bpf_object__reloc_code(obj, prog, subprog);
7457 if (err)
7458 return err;
7459 }
7460 }
7461 }
7462 for (i = 0; i < obj->nr_programs; i++) {
7463 prog = &obj->programs[i];
7464 if (prog_is_subprog(obj, prog))
7465 continue;
7466 if (!prog->autoload)
7467 continue;
7468
7469 /* Process data relos for main programs */
7470 err = bpf_object__relocate_data(obj, prog);
7471 if (err) {
7472 pr_warn("prog '%s': failed to relocate data references: %s\n",
7473 prog->name, errstr(err));
7474 return err;
7475 }
7476
7477 /* Fix up .BTF.ext information, if necessary */
7478 err = bpf_program_fixup_func_info(obj, prog);
7479 if (err) {
7480 pr_warn("prog '%s': failed to perform .BTF.ext fix ups: %s\n",
7481 prog->name, errstr(err));
7482 return err;
7483 }
7484 }
7485
7486 return 0;
7487 }
7488
7489 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
7490 Elf64_Shdr *shdr, Elf_Data *data);
7491
7492 static int bpf_object__collect_map_relos(struct bpf_object *obj,
7493 Elf64_Shdr *shdr, Elf_Data *data)
7494 {
7495 const int bpf_ptr_sz = 8, host_ptr_sz = sizeof(void *);
7496 int i, j, nrels, new_sz;
7497 const struct btf_var_secinfo *vi = NULL;
7498 const struct btf_type *sec, *var, *def;
7499 struct bpf_map *map = NULL, *targ_map = NULL;
7500 struct bpf_program *targ_prog = NULL;
7501 bool is_prog_array, is_map_in_map;
7502 const struct btf_member *member;
7503 const char *name, *mname, *type;
7504 unsigned int moff;
7505 Elf64_Sym *sym;
7506 Elf64_Rel *rel;
7507 void *tmp;
7508
7509 if (!obj->efile.btf_maps_sec_btf_id || !obj->btf)
7510 return -EINVAL;
7511 sec = btf__type_by_id(obj->btf, obj->efile.btf_maps_sec_btf_id);
7512 if (!sec)
7513 return -EINVAL;
7514
7515 nrels = shdr->sh_size / shdr->sh_entsize;
7516 for (i = 0; i < nrels; i++) {
7517 rel = elf_rel_by_idx(data, i);
7518 if (!rel) {
7519 pr_warn(".maps relo #%d: failed to get ELF relo\n", i);
7520 return -LIBBPF_ERRNO__FORMAT;
7521 }
7522
7523 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
7524 if (!sym) {
7525 pr_warn(".maps relo #%d: symbol %zx not found\n",
7526 i, (size_t)ELF64_R_SYM(rel->r_info));
7527 return -LIBBPF_ERRNO__FORMAT;
7528 }
7529 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
7530
7531 pr_debug(".maps relo #%d: for %zd value %zd rel->r_offset %zu name %d ('%s')\n",
7532 i, (ssize_t)(rel->r_info >> 32), (size_t)sym->st_value,
7533 (size_t)rel->r_offset, sym->st_name, name);
7534
7535 for (j = 0; j < obj->nr_maps; j++) {
7536 map = &obj->maps[j];
7537 if (map->sec_idx != obj->efile.btf_maps_shndx)
7538 continue;
7539
7540 vi = btf_var_secinfos(sec) + map->btf_var_idx;
7541 if (vi->offset <= rel->r_offset &&
7542 rel->r_offset + bpf_ptr_sz <= vi->offset + vi->size)
7543 break;
7544 }
7545 if (j == obj->nr_maps) {
7546 pr_warn(".maps relo #%d: cannot find map '%s' at rel->r_offset %zu\n",
7547 i, name, (size_t)rel->r_offset);
7548 return -EINVAL;
7549 }
7550
7551 is_map_in_map = bpf_map_type__is_map_in_map(map->def.type);
7552 is_prog_array = map->def.type == BPF_MAP_TYPE_PROG_ARRAY;
7553 type = is_map_in_map ? "map" : "prog";
7554 if (is_map_in_map) {
7555 if (sym->st_shndx != obj->efile.btf_maps_shndx) {
7556 pr_warn(".maps relo #%d: '%s' isn't a BTF-defined map\n",
7557 i, name);
7558 return -LIBBPF_ERRNO__RELOC;
7559 }
7560 if (map->def.type == BPF_MAP_TYPE_HASH_OF_MAPS &&
7561 map->def.key_size != sizeof(int)) {
7562 pr_warn(".maps relo #%d: hash-of-maps '%s' should have key size %zu.\n",
7563 i, map->name, sizeof(int));
7564 return -EINVAL;
7565 }
7566 targ_map = bpf_object__find_map_by_name(obj, name);
7567 if (!targ_map) {
7568 pr_warn(".maps relo #%d: '%s' isn't a valid map reference\n",
7569 i, name);
7570 return -ESRCH;
7571 }
7572 } else if (is_prog_array) {
7573 targ_prog = bpf_object__find_program_by_name(obj, name);
7574 if (!targ_prog) {
7575 pr_warn(".maps relo #%d: '%s' isn't a valid program reference\n",
7576 i, name);
7577 return -ESRCH;
7578 }
7579 if (targ_prog->sec_idx != sym->st_shndx ||
7580 targ_prog->sec_insn_off * 8 != sym->st_value ||
7581 prog_is_subprog(obj, targ_prog)) {
7582 pr_warn(".maps relo #%d: '%s' isn't an entry-point program\n",
7583 i, name);
7584 return -LIBBPF_ERRNO__RELOC;
7585 }
7586 } else {
7587 return -EINVAL;
7588 }
7589
7590 var = btf__type_by_id(obj->btf, vi->type);
7591 def = skip_mods_and_typedefs(obj->btf, var->type, NULL);
7592 if (btf_vlen(def) == 0)
7593 return -EINVAL;
7594 member = btf_members(def) + btf_vlen(def) - 1;
7595 mname = btf__name_by_offset(obj->btf, member->name_off);
7596 if (strcmp(mname, "values"))
7597 return -EINVAL;
7598
7599 moff = btf_member_bit_offset(def, btf_vlen(def) - 1) / 8;
7600 if (rel->r_offset - vi->offset < moff)
7601 return -EINVAL;
7602
7603 moff = rel->r_offset - vi->offset - moff;
7604 /* here we use BPF pointer size, which is always 64 bit, as we
7605 * are parsing ELF that was built for BPF target
7606 */
7607 if (moff % bpf_ptr_sz)
7608 return -EINVAL;
7609 moff /= bpf_ptr_sz;
7610 if (moff >= map->init_slots_sz) {
7611 new_sz = moff + 1;
7612 tmp = libbpf_reallocarray(map->init_slots, new_sz, host_ptr_sz);
7613 if (!tmp)
7614 return -ENOMEM;
7615 map->init_slots = tmp;
7616 memset(map->init_slots + map->init_slots_sz, 0,
7617 (new_sz - map->init_slots_sz) * host_ptr_sz);
7618 map->init_slots_sz = new_sz;
7619 }
7620 map->init_slots[moff] = is_map_in_map ? (void *)targ_map : (void *)targ_prog;
7621
7622 pr_debug(".maps relo #%d: map '%s' slot [%d] points to %s '%s'\n",
7623 i, map->name, moff, type, name);
7624 }
7625
7626 return 0;
7627 }
7628
7629 static int bpf_object__collect_relos(struct bpf_object *obj)
7630 {
7631 int i, err;
7632
7633 for (i = 0; i < obj->efile.sec_cnt; i++) {
7634 struct elf_sec_desc *sec_desc = &obj->efile.secs[i];
7635 Elf64_Shdr *shdr;
7636 Elf_Data *data;
7637 int idx;
7638
7639 if (sec_desc->sec_type != SEC_RELO)
7640 continue;
7641
7642 shdr = sec_desc->shdr;
7643 data = sec_desc->data;
7644 idx = shdr->sh_info;
7645
7646 if (shdr->sh_type != SHT_REL || idx < 0 || idx >= obj->efile.sec_cnt) {
7647 pr_warn("internal error at %d\n", __LINE__);
7648 return -LIBBPF_ERRNO__INTERNAL;
7649 }
7650
7651 if (obj->efile.secs[idx].sec_type == SEC_ST_OPS)
7652 err = bpf_object__collect_st_ops_relos(obj, shdr, data);
7653 else if (idx == obj->efile.btf_maps_shndx)
7654 err = bpf_object__collect_map_relos(obj, shdr, data);
7655 else
7656 err = bpf_object__collect_prog_relos(obj, shdr, data);
7657 if (err)
7658 return err;
7659 }
7660
7661 bpf_object__sort_relos(obj);
7662 return 0;
7663 }
7664
7665 static bool insn_is_helper_call(struct bpf_insn *insn, enum bpf_func_id *func_id)
7666 {
7667 if (BPF_CLASS(insn->code) == BPF_JMP &&
7668 BPF_OP(insn->code) == BPF_CALL &&
7669 BPF_SRC(insn->code) == BPF_K &&
7670 insn->src_reg == 0 &&
7671 insn->dst_reg == 0) {
7672 *func_id = insn->imm;
7673 return true;
7674 }
7675 return false;
7676 }
7677
7678 static int bpf_object__sanitize_prog(struct bpf_object *obj, struct bpf_program *prog)
7679 {
7680 struct bpf_insn *insn = prog->insns;
7681 enum bpf_func_id func_id;
7682 int i;
7683
7684 if (obj->gen_loader)
7685 return 0;
7686
7687 for (i = 0; i < prog->insns_cnt; i++, insn++) {
7688 if (!insn_is_helper_call(insn, &func_id))
7689 continue;
7690
7691 /* on kernels that don't yet support
7692 * bpf_probe_read_{kernel,user}[_str] helpers, fall back
7693 * to bpf_probe_read() which works well for old kernels
7694 */
7695 switch (func_id) {
7696 case BPF_FUNC_probe_read_kernel:
7697 case BPF_FUNC_probe_read_user:
7698 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
7699 insn->imm = BPF_FUNC_probe_read;
7700 break;
7701 case BPF_FUNC_probe_read_kernel_str:
7702 case BPF_FUNC_probe_read_user_str:
7703 if (!kernel_supports(obj, FEAT_PROBE_READ_KERN))
7704 insn->imm = BPF_FUNC_probe_read_str;
7705 break;
7706 default:
7707 break;
7708 }
7709 }
7710 return 0;
7711 }
7712
7713 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
7714 int *btf_obj_fd, int *btf_type_id);
7715
7716 /* this is called as prog->sec_def->prog_prepare_load_fn for libbpf-supported sec_defs */
7717 static int libbpf_prepare_prog_load(struct bpf_program *prog,
7718 struct bpf_prog_load_opts *opts, long cookie)
7719 {
7720 enum sec_def_flags def = cookie;
7721
7722 /* old kernels might not support specifying expected_attach_type */
7723 if ((def & SEC_EXP_ATTACH_OPT) && !kernel_supports(prog->obj, FEAT_EXP_ATTACH_TYPE))
7724 opts->expected_attach_type = 0;
7725
7726 if (def & SEC_SLEEPABLE)
7727 opts->prog_flags |= BPF_F_SLEEPABLE;
7728
7729 if (prog->type == BPF_PROG_TYPE_XDP && (def & SEC_XDP_FRAGS))
7730 opts->prog_flags |= BPF_F_XDP_HAS_FRAGS;
7731
7732 /* special check for usdt to use uprobe_multi link */
7733 if ((def & SEC_USDT) && kernel_supports(prog->obj, FEAT_UPROBE_MULTI_LINK)) {
7734 /* for BPF_TRACE_UPROBE_MULTI, user might want to query expected_attach_type
7735 * in prog, and expected_attach_type we set in kernel is from opts, so we
7736 * update both.
7737 */
7738 prog->expected_attach_type = BPF_TRACE_UPROBE_MULTI;
7739 opts->expected_attach_type = BPF_TRACE_UPROBE_MULTI;
7740 }
7741
7742 if ((def & SEC_ATTACH_BTF) && !prog->attach_btf_id) {
7743 int btf_obj_fd = 0, btf_type_id = 0, err;
7744 const char *attach_name;
7745
7746 attach_name = strchr(prog->sec_name, '/');
7747 if (!attach_name) {
7748 /* if BPF program is annotated with just SEC("fentry")
7749 * (or similar) without declaratively specifying
7750 * target, then it is expected that target will be
7751 * specified with bpf_program__set_attach_target() at
7752 * runtime before BPF object load step. If not, then
7753 * there is nothing to load into the kernel as BPF
7754 * verifier won't be able to validate BPF program
7755 * correctness anyways.
7756 */
7757 pr_warn("prog '%s': no BTF-based attach target is specified, use bpf_program__set_attach_target()\n",
7758 prog->name);
7759 return -EINVAL;
7760 }
7761 attach_name++; /* skip over / */
7762
7763 err = libbpf_find_attach_btf_id(prog, attach_name, &btf_obj_fd, &btf_type_id);
7764 if (err)
7765 return err;
7766
7767 /* cache resolved BTF FD and BTF type ID in the prog */
7768 prog->attach_btf_obj_fd = btf_obj_fd;
7769 prog->attach_btf_id = btf_type_id;
7770
7771 /* but by now libbpf common logic is not utilizing
7772 * prog->atach_btf_obj_fd/prog->attach_btf_id anymore because
7773 * this callback is called after opts were populated by
7774 * libbpf, so this callback has to update opts explicitly here
7775 */
7776 opts->attach_btf_obj_fd = btf_obj_fd;
7777 opts->attach_btf_id = btf_type_id;
7778 }
7779 return 0;
7780 }
7781
7782 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz);
7783
7784 static int bpf_object_load_prog(struct bpf_object *obj, struct bpf_program *prog,
7785 struct bpf_insn *insns, int insns_cnt,
7786 const char *license, __u32 kern_version, int *prog_fd)
7787 {
7788 LIBBPF_OPTS(bpf_prog_load_opts, load_attr);
7789 const char *prog_name = NULL;
7790 size_t log_buf_size = 0;
7791 char *log_buf = NULL, *tmp;
7792 bool own_log_buf = true;
7793 __u32 log_level = prog->log_level;
7794 int ret, err;
7795
7796 /* Be more helpful by rejecting programs that can't be validated early
7797 * with more meaningful and actionable error message.
7798 */
7799 switch (prog->type) {
7800 case BPF_PROG_TYPE_UNSPEC:
7801 /*
7802 * The program type must be set. Most likely we couldn't find a proper
7803 * section definition at load time, and thus we didn't infer the type.
7804 */
7805 pr_warn("prog '%s': missing BPF prog type, check ELF section name '%s'\n",
7806 prog->name, prog->sec_name);
7807 return -EINVAL;
7808 case BPF_PROG_TYPE_STRUCT_OPS:
7809 if (prog->attach_btf_id == 0) {
7810 pr_warn("prog '%s': SEC(\"struct_ops\") program isn't referenced anywhere, did you forget to use it?\n",
7811 prog->name);
7812 return -EINVAL;
7813 }
7814 break;
7815 default:
7816 break;
7817 }
7818
7819 if (!insns || !insns_cnt)
7820 return -EINVAL;
7821
7822 if (kernel_supports(obj, FEAT_PROG_NAME))
7823 prog_name = prog->name;
7824 load_attr.attach_prog_fd = prog->attach_prog_fd;
7825 load_attr.attach_btf_obj_fd = prog->attach_btf_obj_fd;
7826 load_attr.attach_btf_id = prog->attach_btf_id;
7827 load_attr.kern_version = kern_version;
7828 load_attr.prog_ifindex = prog->prog_ifindex;
7829 load_attr.expected_attach_type = prog->expected_attach_type;
7830
7831 /* specify func_info/line_info only if kernel supports them */
7832 if (obj->btf && btf__fd(obj->btf) >= 0 && kernel_supports(obj, FEAT_BTF_FUNC)) {
7833 load_attr.prog_btf_fd = btf__fd(obj->btf);
7834 load_attr.func_info = prog->func_info;
7835 load_attr.func_info_rec_size = prog->func_info_rec_size;
7836 load_attr.func_info_cnt = prog->func_info_cnt;
7837 load_attr.line_info = prog->line_info;
7838 load_attr.line_info_rec_size = prog->line_info_rec_size;
7839 load_attr.line_info_cnt = prog->line_info_cnt;
7840 }
7841 load_attr.log_level = log_level;
7842 load_attr.prog_flags = prog->prog_flags;
7843 load_attr.fd_array = obj->fd_array;
7844
7845 load_attr.token_fd = obj->token_fd;
7846 if (obj->token_fd)
7847 load_attr.prog_flags |= BPF_F_TOKEN_FD;
7848
7849 /* adjust load_attr if sec_def provides custom preload callback */
7850 if (prog->sec_def && prog->sec_def->prog_prepare_load_fn) {
7851 err = prog->sec_def->prog_prepare_load_fn(prog, &load_attr, prog->sec_def->cookie);
7852 if (err < 0) {
7853 pr_warn("prog '%s': failed to prepare load attributes: %s\n",
7854 prog->name, errstr(err));
7855 return err;
7856 }
7857 insns = prog->insns;
7858 insns_cnt = prog->insns_cnt;
7859 }
7860
7861 if (obj->gen_loader) {
7862 bpf_gen__prog_load(obj->gen_loader, prog->type, prog->name,
7863 license, insns, insns_cnt, &load_attr,
7864 prog - obj->programs);
7865 *prog_fd = -1;
7866 return 0;
7867 }
7868
7869 retry_load:
7870 /* if log_level is zero, we don't request logs initially even if
7871 * custom log_buf is specified; if the program load fails, then we'll
7872 * bump log_level to 1 and use either custom log_buf or we'll allocate
7873 * our own and retry the load to get details on what failed
7874 */
7875 if (log_level) {
7876 if (prog->log_buf) {
7877 log_buf = prog->log_buf;
7878 log_buf_size = prog->log_size;
7879 own_log_buf = false;
7880 } else if (obj->log_buf) {
7881 log_buf = obj->log_buf;
7882 log_buf_size = obj->log_size;
7883 own_log_buf = false;
7884 } else {
7885 log_buf_size = max((size_t)BPF_LOG_BUF_SIZE, log_buf_size * 2);
7886 tmp = realloc(log_buf, log_buf_size);
7887 if (!tmp) {
7888 ret = -ENOMEM;
7889 goto out;
7890 }
7891 log_buf = tmp;
7892 log_buf[0] = '\0';
7893 own_log_buf = true;
7894 }
7895 }
7896
7897 load_attr.log_buf = log_buf;
7898 load_attr.log_size = log_buf_size;
7899 load_attr.log_level = log_level;
7900
7901 ret = bpf_prog_load(prog->type, prog_name, license, insns, insns_cnt, &load_attr);
7902 if (ret >= 0) {
7903 if (log_level && own_log_buf) {
7904 pr_debug("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7905 prog->name, log_buf);
7906 }
7907
7908 if (obj->has_rodata && kernel_supports(obj, FEAT_PROG_BIND_MAP)) {
7909 struct bpf_map *map;
7910 int i;
7911
7912 for (i = 0; i < obj->nr_maps; i++) {
7913 map = &prog->obj->maps[i];
7914 if (map->libbpf_type != LIBBPF_MAP_RODATA)
7915 continue;
7916
7917 if (bpf_prog_bind_map(ret, map->fd, NULL)) {
7918 pr_warn("prog '%s': failed to bind map '%s': %s\n",
7919 prog->name, map->real_name, errstr(errno));
7920 /* Don't fail hard if can't bind rodata. */
7921 }
7922 }
7923 }
7924
7925 *prog_fd = ret;
7926 ret = 0;
7927 goto out;
7928 }
7929
7930 if (log_level == 0) {
7931 log_level = 1;
7932 goto retry_load;
7933 }
7934 /* On ENOSPC, increase log buffer size and retry, unless custom
7935 * log_buf is specified.
7936 * Be careful to not overflow u32, though. Kernel's log buf size limit
7937 * isn't part of UAPI so it can always be bumped to full 4GB. So don't
7938 * multiply by 2 unless we are sure we'll fit within 32 bits.
7939 * Currently, we'll get -EINVAL when we reach (UINT_MAX >> 2).
7940 */
7941 if (own_log_buf && errno == ENOSPC && log_buf_size <= UINT_MAX / 2)
7942 goto retry_load;
7943
7944 ret = -errno;
7945
7946 /* post-process verifier log to improve error descriptions */
7947 fixup_verifier_log(prog, log_buf, log_buf_size);
7948
7949 pr_warn("prog '%s': BPF program load failed: %s\n", prog->name, errstr(errno));
7950 pr_perm_msg(ret);
7951
7952 if (own_log_buf && log_buf && log_buf[0] != '\0') {
7953 pr_warn("prog '%s': -- BEGIN PROG LOAD LOG --\n%s-- END PROG LOAD LOG --\n",
7954 prog->name, log_buf);
7955 }
7956
7957 out:
7958 if (own_log_buf)
7959 free(log_buf);
7960 return ret;
7961 }
7962
7963 static char *find_prev_line(char *buf, char *cur)
7964 {
7965 char *p;
7966
7967 if (cur == buf) /* end of a log buf */
7968 return NULL;
7969
7970 p = cur - 1;
7971 while (p - 1 >= buf && *(p - 1) != '\n')
7972 p--;
7973
7974 return p;
7975 }
7976
7977 static void patch_log(char *buf, size_t buf_sz, size_t log_sz,
7978 char *orig, size_t orig_sz, const char *patch)
7979 {
7980 /* size of the remaining log content to the right from the to-be-replaced part */
7981 size_t rem_sz = (buf + log_sz) - (orig + orig_sz);
7982 size_t patch_sz = strlen(patch);
7983
7984 if (patch_sz != orig_sz) {
7985 /* If patch line(s) are longer than original piece of verifier log,
7986 * shift log contents by (patch_sz - orig_sz) bytes to the right
7987 * starting from after to-be-replaced part of the log.
7988 *
7989 * If patch line(s) are shorter than original piece of verifier log,
7990 * shift log contents by (orig_sz - patch_sz) bytes to the left
7991 * starting from after to-be-replaced part of the log
7992 *
7993 * We need to be careful about not overflowing available
7994 * buf_sz capacity. If that's the case, we'll truncate the end
7995 * of the original log, as necessary.
7996 */
7997 if (patch_sz > orig_sz) {
7998 if (orig + patch_sz >= buf + buf_sz) {
7999 /* patch is big enough to cover remaining space completely */
8000 patch_sz -= (orig + patch_sz) - (buf + buf_sz) + 1;
8001 rem_sz = 0;
8002 } else if (patch_sz - orig_sz > buf_sz - log_sz) {
8003 /* patch causes part of remaining log to be truncated */
8004 rem_sz -= (patch_sz - orig_sz) - (buf_sz - log_sz);
8005 }
8006 }
8007 /* shift remaining log to the right by calculated amount */
8008 memmove(orig + patch_sz, orig + orig_sz, rem_sz);
8009 }
8010
8011 memcpy(orig, patch, patch_sz);
8012 }
8013
8014 static void fixup_log_failed_core_relo(struct bpf_program *prog,
8015 char *buf, size_t buf_sz, size_t log_sz,
8016 char *line1, char *line2, char *line3)
8017 {
8018 /* Expected log for failed and not properly guarded CO-RE relocation:
8019 * line1 -> 123: (85) call unknown#195896080
8020 * line2 -> invalid func unknown#195896080
8021 * line3 -> <anything else or end of buffer>
8022 *
8023 * "123" is the index of the instruction that was poisoned. We extract
8024 * instruction index to find corresponding CO-RE relocation and
8025 * replace this part of the log with more relevant information about
8026 * failed CO-RE relocation.
8027 */
8028 const struct bpf_core_relo *relo;
8029 struct bpf_core_spec spec;
8030 char patch[512], spec_buf[256];
8031 int insn_idx, err, spec_len;
8032
8033 if (sscanf(line1, "%d: (%*d) call unknown#195896080\n", &insn_idx) != 1)
8034 return;
8035
8036 relo = find_relo_core(prog, insn_idx);
8037 if (!relo)
8038 return;
8039
8040 err = bpf_core_parse_spec(prog->name, prog->obj->btf, relo, &spec);
8041 if (err)
8042 return;
8043
8044 spec_len = bpf_core_format_spec(spec_buf, sizeof(spec_buf), &spec);
8045 snprintf(patch, sizeof(patch),
8046 "%d: <invalid CO-RE relocation>\n"
8047 "failed to resolve CO-RE relocation %s%s\n",
8048 insn_idx, spec_buf, spec_len >= sizeof(spec_buf) ? "..." : "");
8049
8050 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
8051 }
8052
8053 static void fixup_log_missing_map_load(struct bpf_program *prog,
8054 char *buf, size_t buf_sz, size_t log_sz,
8055 char *line1, char *line2, char *line3)
8056 {
8057 /* Expected log for failed and not properly guarded map reference:
8058 * line1 -> 123: (85) call unknown#2001000345
8059 * line2 -> invalid func unknown#2001000345
8060 * line3 -> <anything else or end of buffer>
8061 *
8062 * "123" is the index of the instruction that was poisoned.
8063 * "345" in "2001000345" is a map index in obj->maps to fetch map name.
8064 */
8065 struct bpf_object *obj = prog->obj;
8066 const struct bpf_map *map;
8067 int insn_idx, map_idx;
8068 char patch[128];
8069
8070 if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &map_idx) != 2)
8071 return;
8072
8073 map_idx -= POISON_LDIMM64_MAP_BASE;
8074 if (map_idx < 0 || map_idx >= obj->nr_maps)
8075 return;
8076 map = &obj->maps[map_idx];
8077
8078 snprintf(patch, sizeof(patch),
8079 "%d: <invalid BPF map reference>\n"
8080 "BPF map '%s' is referenced but wasn't created\n",
8081 insn_idx, map->name);
8082
8083 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
8084 }
8085
8086 static void fixup_log_missing_kfunc_call(struct bpf_program *prog,
8087 char *buf, size_t buf_sz, size_t log_sz,
8088 char *line1, char *line2, char *line3)
8089 {
8090 /* Expected log for failed and not properly guarded kfunc call:
8091 * line1 -> 123: (85) call unknown#2002000345
8092 * line2 -> invalid func unknown#2002000345
8093 * line3 -> <anything else or end of buffer>
8094 *
8095 * "123" is the index of the instruction that was poisoned.
8096 * "345" in "2002000345" is an extern index in obj->externs to fetch kfunc name.
8097 */
8098 struct bpf_object *obj = prog->obj;
8099 const struct extern_desc *ext;
8100 int insn_idx, ext_idx;
8101 char patch[128];
8102
8103 if (sscanf(line1, "%d: (%*d) call unknown#%d\n", &insn_idx, &ext_idx) != 2)
8104 return;
8105
8106 ext_idx -= POISON_CALL_KFUNC_BASE;
8107 if (ext_idx < 0 || ext_idx >= obj->nr_extern)
8108 return;
8109 ext = &obj->externs[ext_idx];
8110
8111 snprintf(patch, sizeof(patch),
8112 "%d: <invalid kfunc call>\n"
8113 "kfunc '%s' is referenced but wasn't resolved\n",
8114 insn_idx, ext->name);
8115
8116 patch_log(buf, buf_sz, log_sz, line1, line3 - line1, patch);
8117 }
8118
8119 static void fixup_verifier_log(struct bpf_program *prog, char *buf, size_t buf_sz)
8120 {
8121 /* look for familiar error patterns in last N lines of the log */
8122 const size_t max_last_line_cnt = 10;
8123 char *prev_line, *cur_line, *next_line;
8124 size_t log_sz;
8125 int i;
8126
8127 if (!buf)
8128 return;
8129
8130 log_sz = strlen(buf) + 1;
8131 next_line = buf + log_sz - 1;
8132
8133 for (i = 0; i < max_last_line_cnt; i++, next_line = cur_line) {
8134 cur_line = find_prev_line(buf, next_line);
8135 if (!cur_line)
8136 return;
8137
8138 if (str_has_pfx(cur_line, "invalid func unknown#195896080\n")) {
8139 prev_line = find_prev_line(buf, cur_line);
8140 if (!prev_line)
8141 continue;
8142
8143 /* failed CO-RE relocation case */
8144 fixup_log_failed_core_relo(prog, buf, buf_sz, log_sz,
8145 prev_line, cur_line, next_line);
8146 return;
8147 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_LDIMM64_MAP_PFX)) {
8148 prev_line = find_prev_line(buf, cur_line);
8149 if (!prev_line)
8150 continue;
8151
8152 /* reference to uncreated BPF map */
8153 fixup_log_missing_map_load(prog, buf, buf_sz, log_sz,
8154 prev_line, cur_line, next_line);
8155 return;
8156 } else if (str_has_pfx(cur_line, "invalid func unknown#"POISON_CALL_KFUNC_PFX)) {
8157 prev_line = find_prev_line(buf, cur_line);
8158 if (!prev_line)
8159 continue;
8160
8161 /* reference to unresolved kfunc */
8162 fixup_log_missing_kfunc_call(prog, buf, buf_sz, log_sz,
8163 prev_line, cur_line, next_line);
8164 return;
8165 }
8166 }
8167 }
8168
8169 static int bpf_program_record_relos(struct bpf_program *prog)
8170 {
8171 struct bpf_object *obj = prog->obj;
8172 int i;
8173
8174 for (i = 0; i < prog->nr_reloc; i++) {
8175 struct reloc_desc *relo = &prog->reloc_desc[i];
8176 struct extern_desc *ext = &obj->externs[relo->ext_idx];
8177 int kind;
8178
8179 switch (relo->type) {
8180 case RELO_EXTERN_LD64:
8181 if (ext->type != EXT_KSYM)
8182 continue;
8183 kind = btf_is_var(btf__type_by_id(obj->btf, ext->btf_id)) ?
8184 BTF_KIND_VAR : BTF_KIND_FUNC;
8185 bpf_gen__record_extern(obj->gen_loader, ext->name,
8186 ext->is_weak, !ext->ksym.type_id,
8187 true, kind, relo->insn_idx);
8188 break;
8189 case RELO_EXTERN_CALL:
8190 bpf_gen__record_extern(obj->gen_loader, ext->name,
8191 ext->is_weak, false, false, BTF_KIND_FUNC,
8192 relo->insn_idx);
8193 break;
8194 case RELO_CORE: {
8195 struct bpf_core_relo cr = {
8196 .insn_off = relo->insn_idx * 8,
8197 .type_id = relo->core_relo->type_id,
8198 .access_str_off = relo->core_relo->access_str_off,
8199 .kind = relo->core_relo->kind,
8200 };
8201
8202 bpf_gen__record_relo_core(obj->gen_loader, &cr);
8203 break;
8204 }
8205 default:
8206 continue;
8207 }
8208 }
8209 return 0;
8210 }
8211
8212 static int
8213 bpf_object__load_progs(struct bpf_object *obj, int log_level)
8214 {
8215 struct bpf_program *prog;
8216 size_t i;
8217 int err;
8218
8219 for (i = 0; i < obj->nr_programs; i++) {
8220 prog = &obj->programs[i];
8221 if (prog_is_subprog(obj, prog))
8222 continue;
8223 if (!prog->autoload) {
8224 pr_debug("prog '%s': skipped loading\n", prog->name);
8225 continue;
8226 }
8227 prog->log_level |= log_level;
8228
8229 if (obj->gen_loader)
8230 bpf_program_record_relos(prog);
8231
8232 err = bpf_object_load_prog(obj, prog, prog->insns, prog->insns_cnt,
8233 obj->license, obj->kern_version, &prog->fd);
8234 if (err) {
8235 pr_warn("prog '%s': failed to load: %s\n", prog->name, errstr(err));
8236 return err;
8237 }
8238 }
8239
8240 bpf_object__free_relocs(obj);
8241 return 0;
8242 }
8243
8244 static int bpf_object_prepare_progs(struct bpf_object *obj)
8245 {
8246 struct bpf_program *prog;
8247 size_t i;
8248 int err;
8249
8250 for (i = 0; i < obj->nr_programs; i++) {
8251 prog = &obj->programs[i];
8252 err = bpf_object__sanitize_prog(obj, prog);
8253 if (err)
8254 return err;
8255 }
8256 return 0;
8257 }
8258
8259 static const struct bpf_sec_def *find_sec_def(const char *sec_name);
8260
8261 static int bpf_object_init_progs(struct bpf_object *obj, const struct bpf_object_open_opts *opts)
8262 {
8263 struct bpf_program *prog;
8264 int err;
8265
8266 bpf_object__for_each_program(prog, obj) {
8267 prog->sec_def = find_sec_def(prog->sec_name);
8268 if (!prog->sec_def) {
8269 /* couldn't guess, but user might manually specify */
8270 pr_debug("prog '%s': unrecognized ELF section name '%s'\n",
8271 prog->name, prog->sec_name);
8272 continue;
8273 }
8274
8275 prog->type = prog->sec_def->prog_type;
8276 prog->expected_attach_type = prog->sec_def->expected_attach_type;
8277
8278 /* sec_def can have custom callback which should be called
8279 * after bpf_program is initialized to adjust its properties
8280 */
8281 if (prog->sec_def->prog_setup_fn) {
8282 err = prog->sec_def->prog_setup_fn(prog, prog->sec_def->cookie);
8283 if (err < 0) {
8284 pr_warn("prog '%s': failed to initialize: %s\n",
8285 prog->name, errstr(err));
8286 return err;
8287 }
8288 }
8289 }
8290
8291 return 0;
8292 }
8293
8294 static struct bpf_object *bpf_object_open(const char *path, const void *obj_buf, size_t obj_buf_sz,
8295 const char *obj_name,
8296 const struct bpf_object_open_opts *opts)
8297 {
8298 const char *kconfig, *btf_tmp_path, *token_path;
8299 struct bpf_object *obj;
8300 int err;
8301 char *log_buf;
8302 size_t log_size;
8303 __u32 log_level;
8304
8305 if (obj_buf && !obj_name)
8306 return ERR_PTR(-EINVAL);
8307
8308 if (elf_version(EV_CURRENT) == EV_NONE) {
8309 pr_warn("failed to init libelf for %s\n",
8310 path ? : "(mem buf)");
8311 return ERR_PTR(-LIBBPF_ERRNO__LIBELF);
8312 }
8313
8314 if (!OPTS_VALID(opts, bpf_object_open_opts))
8315 return ERR_PTR(-EINVAL);
8316
8317 obj_name = OPTS_GET(opts, object_name, NULL) ?: obj_name;
8318 if (obj_buf) {
8319 path = obj_name;
8320 pr_debug("loading object '%s' from buffer\n", obj_name);
8321 } else {
8322 pr_debug("loading object from %s\n", path);
8323 }
8324
8325 log_buf = OPTS_GET(opts, kernel_log_buf, NULL);
8326 log_size = OPTS_GET(opts, kernel_log_size, 0);
8327 log_level = OPTS_GET(opts, kernel_log_level, 0);
8328 if (log_size > UINT_MAX)
8329 return ERR_PTR(-EINVAL);
8330 if (log_size && !log_buf)
8331 return ERR_PTR(-EINVAL);
8332
8333 token_path = OPTS_GET(opts, bpf_token_path, NULL);
8334 /* if user didn't specify bpf_token_path explicitly, check if
8335 * LIBBPF_BPF_TOKEN_PATH envvar was set and treat it as bpf_token_path
8336 * option
8337 */
8338 if (!token_path)
8339 token_path = getenv("LIBBPF_BPF_TOKEN_PATH");
8340 if (token_path && strlen(token_path) >= PATH_MAX)
8341 return ERR_PTR(-ENAMETOOLONG);
8342
8343 obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
8344 if (IS_ERR(obj))
8345 return obj;
8346
8347 obj->log_buf = log_buf;
8348 obj->log_size = log_size;
8349 obj->log_level = log_level;
8350
8351 if (token_path) {
8352 obj->token_path = strdup(token_path);
8353 if (!obj->token_path) {
8354 err = -ENOMEM;
8355 goto out;
8356 }
8357 }
8358
8359 btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
8360 if (btf_tmp_path) {
8361 if (strlen(btf_tmp_path) >= PATH_MAX) {
8362 err = -ENAMETOOLONG;
8363 goto out;
8364 }
8365 obj->btf_custom_path = strdup(btf_tmp_path);
8366 if (!obj->btf_custom_path) {
8367 err = -ENOMEM;
8368 goto out;
8369 }
8370 }
8371
8372 kconfig = OPTS_GET(opts, kconfig, NULL);
8373 if (kconfig) {
8374 obj->kconfig = strdup(kconfig);
8375 if (!obj->kconfig) {
8376 err = -ENOMEM;
8377 goto out;
8378 }
8379 }
8380
8381 err = bpf_object__elf_init(obj);
8382 err = err ? : bpf_object__elf_collect(obj);
8383 err = err ? : bpf_object__collect_externs(obj);
8384 err = err ? : bpf_object_fixup_btf(obj);
8385 err = err ? : bpf_object__init_maps(obj, opts);
8386 err = err ? : bpf_object_init_progs(obj, opts);
8387 err = err ? : bpf_object__collect_relos(obj);
8388 if (err)
8389 goto out;
8390
8391 bpf_object__elf_finish(obj);
8392
8393 return obj;
8394 out:
8395 bpf_object__close(obj);
8396 return ERR_PTR(err);
8397 }
8398
8399 struct bpf_object *
8400 bpf_object__open_file(const char *path, const struct bpf_object_open_opts *opts)
8401 {
8402 if (!path)
8403 return libbpf_err_ptr(-EINVAL);
8404
8405 return libbpf_ptr(bpf_object_open(path, NULL, 0, NULL, opts));
8406 }
8407
8408 struct bpf_object *bpf_object__open(const char *path)
8409 {
8410 return bpf_object__open_file(path, NULL);
8411 }
8412
8413 struct bpf_object *
8414 bpf_object__open_mem(const void *obj_buf, size_t obj_buf_sz,
8415 const struct bpf_object_open_opts *opts)
8416 {
8417 char tmp_name[64];
8418
8419 if (!obj_buf || obj_buf_sz == 0)
8420 return libbpf_err_ptr(-EINVAL);
8421
8422 /* create a (quite useless) default "name" for this memory buffer object */
8423 snprintf(tmp_name, sizeof(tmp_name), "%lx-%zx", (unsigned long)obj_buf, obj_buf_sz);
8424
8425 return libbpf_ptr(bpf_object_open(NULL, obj_buf, obj_buf_sz, tmp_name, opts));
8426 }
8427
8428 static int bpf_object_unload(struct bpf_object *obj)
8429 {
8430 size_t i;
8431
8432 if (!obj)
8433 return libbpf_err(-EINVAL);
8434
8435 for (i = 0; i < obj->nr_maps; i++) {
8436 zclose(obj->maps[i].fd);
8437 if (obj->maps[i].st_ops)
8438 zfree(&obj->maps[i].st_ops->kern_vdata);
8439 }
8440
8441 for (i = 0; i < obj->nr_programs; i++)
8442 bpf_program__unload(&obj->programs[i]);
8443
8444 return 0;
8445 }
8446
8447 static int bpf_object__sanitize_maps(struct bpf_object *obj)
8448 {
8449 struct bpf_map *m;
8450
8451 bpf_object__for_each_map(m, obj) {
8452 if (!bpf_map__is_internal(m))
8453 continue;
8454 if (!kernel_supports(obj, FEAT_ARRAY_MMAP))
8455 m->def.map_flags &= ~BPF_F_MMAPABLE;
8456 }
8457
8458 return 0;
8459 }
8460
8461 typedef int (*kallsyms_cb_t)(unsigned long long sym_addr, char sym_type,
8462 const char *sym_name, void *ctx);
8463
8464 static int libbpf_kallsyms_parse(kallsyms_cb_t cb, void *ctx)
8465 {
8466 char sym_type, sym_name[500];
8467 unsigned long long sym_addr;
8468 int ret, err = 0;
8469 FILE *f;
8470
8471 f = fopen("/proc/kallsyms", "re");
8472 if (!f) {
8473 err = -errno;
8474 pr_warn("failed to open /proc/kallsyms: %s\n", errstr(err));
8475 return err;
8476 }
8477
8478 while (true) {
8479 ret = fscanf(f, "%llx %c %499s%*[^\n]\n",
8480 &sym_addr, &sym_type, sym_name);
8481 if (ret == EOF && feof(f))
8482 break;
8483 if (ret != 3) {
8484 pr_warn("failed to read kallsyms entry: %d\n", ret);
8485 err = -EINVAL;
8486 break;
8487 }
8488
8489 err = cb(sym_addr, sym_type, sym_name, ctx);
8490 if (err)
8491 break;
8492 }
8493
8494 fclose(f);
8495 return err;
8496 }
8497
8498 static int kallsyms_cb(unsigned long long sym_addr, char sym_type,
8499 const char *sym_name, void *ctx)
8500 {
8501 struct bpf_object *obj = ctx;
8502 const struct btf_type *t;
8503 struct extern_desc *ext;
8504 const char *res;
8505
8506 res = strstr(sym_name, ".llvm.");
8507 if (sym_type == 'd' && res)
8508 ext = find_extern_by_name_with_len(obj, sym_name, res - sym_name);
8509 else
8510 ext = find_extern_by_name(obj, sym_name);
8511 if (!ext || ext->type != EXT_KSYM)
8512 return 0;
8513
8514 t = btf__type_by_id(obj->btf, ext->btf_id);
8515 if (!btf_is_var(t))
8516 return 0;
8517
8518 if (ext->is_set && ext->ksym.addr != sym_addr) {
8519 pr_warn("extern (ksym) '%s': resolution is ambiguous: 0x%llx or 0x%llx\n",
8520 sym_name, ext->ksym.addr, sym_addr);
8521 return -EINVAL;
8522 }
8523 if (!ext->is_set) {
8524 ext->is_set = true;
8525 ext->ksym.addr = sym_addr;
8526 pr_debug("extern (ksym) '%s': set to 0x%llx\n", sym_name, sym_addr);
8527 }
8528 return 0;
8529 }
8530
8531 static int bpf_object__read_kallsyms_file(struct bpf_object *obj)
8532 {
8533 return libbpf_kallsyms_parse(kallsyms_cb, obj);
8534 }
8535
8536 static int find_ksym_btf_id(struct bpf_object *obj, const char *ksym_name,
8537 __u16 kind, struct btf **res_btf,
8538 struct module_btf **res_mod_btf)
8539 {
8540 struct module_btf *mod_btf;
8541 struct btf *btf;
8542 int i, id, err;
8543
8544 btf = obj->btf_vmlinux;
8545 mod_btf = NULL;
8546 id = btf__find_by_name_kind(btf, ksym_name, kind);
8547
8548 if (id == -ENOENT) {
8549 err = load_module_btfs(obj);
8550 if (err)
8551 return err;
8552
8553 for (i = 0; i < obj->btf_module_cnt; i++) {
8554 /* we assume module_btf's BTF FD is always >0 */
8555 mod_btf = &obj->btf_modules[i];
8556 btf = mod_btf->btf;
8557 id = btf__find_by_name_kind_own(btf, ksym_name, kind);
8558 if (id != -ENOENT)
8559 break;
8560 }
8561 }
8562 if (id <= 0)
8563 return -ESRCH;
8564
8565 *res_btf = btf;
8566 *res_mod_btf = mod_btf;
8567 return id;
8568 }
8569
8570 static int bpf_object__resolve_ksym_var_btf_id(struct bpf_object *obj,
8571 struct extern_desc *ext)
8572 {
8573 const struct btf_type *targ_var, *targ_type;
8574 __u32 targ_type_id, local_type_id;
8575 struct module_btf *mod_btf = NULL;
8576 const char *targ_var_name;
8577 struct btf *btf = NULL;
8578 int id, err;
8579
8580 id = find_ksym_btf_id(obj, ext->name, BTF_KIND_VAR, &btf, &mod_btf);
8581 if (id < 0) {
8582 if (id == -ESRCH && ext->is_weak)
8583 return 0;
8584 pr_warn("extern (var ksym) '%s': not found in kernel BTF\n",
8585 ext->name);
8586 return id;
8587 }
8588
8589 /* find local type_id */
8590 local_type_id = ext->ksym.type_id;
8591
8592 /* find target type_id */
8593 targ_var = btf__type_by_id(btf, id);
8594 targ_var_name = btf__name_by_offset(btf, targ_var->name_off);
8595 targ_type = skip_mods_and_typedefs(btf, targ_var->type, &targ_type_id);
8596
8597 err = bpf_core_types_are_compat(obj->btf, local_type_id,
8598 btf, targ_type_id);
8599 if (err <= 0) {
8600 const struct btf_type *local_type;
8601 const char *targ_name, *local_name;
8602
8603 local_type = btf__type_by_id(obj->btf, local_type_id);
8604 local_name = btf__name_by_offset(obj->btf, local_type->name_off);
8605 targ_name = btf__name_by_offset(btf, targ_type->name_off);
8606
8607 pr_warn("extern (var ksym) '%s': incompatible types, expected [%d] %s %s, but kernel has [%d] %s %s\n",
8608 ext->name, local_type_id,
8609 btf_kind_str(local_type), local_name, targ_type_id,
8610 btf_kind_str(targ_type), targ_name);
8611 return -EINVAL;
8612 }
8613
8614 ext->is_set = true;
8615 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
8616 ext->ksym.kernel_btf_id = id;
8617 pr_debug("extern (var ksym) '%s': resolved to [%d] %s %s\n",
8618 ext->name, id, btf_kind_str(targ_var), targ_var_name);
8619
8620 return 0;
8621 }
8622
8623 static int bpf_object__resolve_ksym_func_btf_id(struct bpf_object *obj,
8624 struct extern_desc *ext)
8625 {
8626 int local_func_proto_id, kfunc_proto_id, kfunc_id;
8627 struct module_btf *mod_btf = NULL;
8628 const struct btf_type *kern_func;
8629 struct btf *kern_btf = NULL;
8630 int ret;
8631
8632 local_func_proto_id = ext->ksym.type_id;
8633
8634 kfunc_id = find_ksym_btf_id(obj, ext->essent_name ?: ext->name, BTF_KIND_FUNC, &kern_btf,
8635 &mod_btf);
8636 if (kfunc_id < 0) {
8637 if (kfunc_id == -ESRCH && ext->is_weak)
8638 return 0;
8639 pr_warn("extern (func ksym) '%s': not found in kernel or module BTFs\n",
8640 ext->name);
8641 return kfunc_id;
8642 }
8643
8644 kern_func = btf__type_by_id(kern_btf, kfunc_id);
8645 kfunc_proto_id = kern_func->type;
8646
8647 ret = bpf_core_types_are_compat(obj->btf, local_func_proto_id,
8648 kern_btf, kfunc_proto_id);
8649 if (ret <= 0) {
8650 if (ext->is_weak)
8651 return 0;
8652
8653 pr_warn("extern (func ksym) '%s': func_proto [%d] incompatible with %s [%d]\n",
8654 ext->name, local_func_proto_id,
8655 mod_btf ? mod_btf->name : "vmlinux", kfunc_proto_id);
8656 return -EINVAL;
8657 }
8658
8659 /* set index for module BTF fd in fd_array, if unset */
8660 if (mod_btf && !mod_btf->fd_array_idx) {
8661 /* insn->off is s16 */
8662 if (obj->fd_array_cnt == INT16_MAX) {
8663 pr_warn("extern (func ksym) '%s': module BTF fd index %d too big to fit in bpf_insn offset\n",
8664 ext->name, mod_btf->fd_array_idx);
8665 return -E2BIG;
8666 }
8667 /* Cannot use index 0 for module BTF fd */
8668 if (!obj->fd_array_cnt)
8669 obj->fd_array_cnt = 1;
8670
8671 ret = libbpf_ensure_mem((void **)&obj->fd_array, &obj->fd_array_cap, sizeof(int),
8672 obj->fd_array_cnt + 1);
8673 if (ret)
8674 return ret;
8675 mod_btf->fd_array_idx = obj->fd_array_cnt;
8676 /* we assume module BTF FD is always >0 */
8677 obj->fd_array[obj->fd_array_cnt++] = mod_btf->fd;
8678 }
8679
8680 ext->is_set = true;
8681 ext->ksym.kernel_btf_id = kfunc_id;
8682 ext->ksym.btf_fd_idx = mod_btf ? mod_btf->fd_array_idx : 0;
8683 /* Also set kernel_btf_obj_fd to make sure that bpf_object__relocate_data()
8684 * populates FD into ld_imm64 insn when it's used to point to kfunc.
8685 * {kernel_btf_id, btf_fd_idx} -> fixup bpf_call.
8686 * {kernel_btf_id, kernel_btf_obj_fd} -> fixup ld_imm64.
8687 */
8688 ext->ksym.kernel_btf_obj_fd = mod_btf ? mod_btf->fd : 0;
8689 pr_debug("extern (func ksym) '%s': resolved to %s [%d]\n",
8690 ext->name, mod_btf ? mod_btf->name : "vmlinux", kfunc_id);
8691
8692 return 0;
8693 }
8694
8695 static int bpf_object__resolve_ksyms_btf_id(struct bpf_object *obj)
8696 {
8697 const struct btf_type *t;
8698 struct extern_desc *ext;
8699 int i, err;
8700
8701 for (i = 0; i < obj->nr_extern; i++) {
8702 ext = &obj->externs[i];
8703 if (ext->type != EXT_KSYM || !ext->ksym.type_id)
8704 continue;
8705
8706 if (obj->gen_loader) {
8707 ext->is_set = true;
8708 ext->ksym.kernel_btf_obj_fd = 0;
8709 ext->ksym.kernel_btf_id = 0;
8710 continue;
8711 }
8712 t = btf__type_by_id(obj->btf, ext->btf_id);
8713 if (btf_is_var(t))
8714 err = bpf_object__resolve_ksym_var_btf_id(obj, ext);
8715 else
8716 err = bpf_object__resolve_ksym_func_btf_id(obj, ext);
8717 if (err)
8718 return err;
8719 }
8720 return 0;
8721 }
8722
8723 static int bpf_object__resolve_externs(struct bpf_object *obj,
8724 const char *extra_kconfig)
8725 {
8726 bool need_config = false, need_kallsyms = false;
8727 bool need_vmlinux_btf = false;
8728 struct extern_desc *ext;
8729 void *kcfg_data = NULL;
8730 int err, i;
8731
8732 if (obj->nr_extern == 0)
8733 return 0;
8734
8735 if (obj->kconfig_map_idx >= 0)
8736 kcfg_data = obj->maps[obj->kconfig_map_idx].mmaped;
8737
8738 for (i = 0; i < obj->nr_extern; i++) {
8739 ext = &obj->externs[i];
8740
8741 if (ext->type == EXT_KSYM) {
8742 if (ext->ksym.type_id)
8743 need_vmlinux_btf = true;
8744 else
8745 need_kallsyms = true;
8746 continue;
8747 } else if (ext->type == EXT_KCFG) {
8748 void *ext_ptr = kcfg_data + ext->kcfg.data_off;
8749 __u64 value = 0;
8750
8751 /* Kconfig externs need actual /proc/config.gz */
8752 if (str_has_pfx(ext->name, "CONFIG_")) {
8753 need_config = true;
8754 continue;
8755 }
8756
8757 /* Virtual kcfg externs are customly handled by libbpf */
8758 if (strcmp(ext->name, "LINUX_KERNEL_VERSION") == 0) {
8759 value = get_kernel_version();
8760 if (!value) {
8761 pr_warn("extern (kcfg) '%s': failed to get kernel version\n", ext->name);
8762 return -EINVAL;
8763 }
8764 } else if (strcmp(ext->name, "LINUX_HAS_BPF_COOKIE") == 0) {
8765 value = kernel_supports(obj, FEAT_BPF_COOKIE);
8766 } else if (strcmp(ext->name, "LINUX_HAS_SYSCALL_WRAPPER") == 0) {
8767 value = kernel_supports(obj, FEAT_SYSCALL_WRAPPER);
8768 } else if (!str_has_pfx(ext->name, "LINUX_") || !ext->is_weak) {
8769 /* Currently libbpf supports only CONFIG_ and LINUX_ prefixed
8770 * __kconfig externs, where LINUX_ ones are virtual and filled out
8771 * customly by libbpf (their values don't come from Kconfig).
8772 * If LINUX_xxx variable is not recognized by libbpf, but is marked
8773 * __weak, it defaults to zero value, just like for CONFIG_xxx
8774 * externs.
8775 */
8776 pr_warn("extern (kcfg) '%s': unrecognized virtual extern\n", ext->name);
8777 return -EINVAL;
8778 }
8779
8780 err = set_kcfg_value_num(ext, ext_ptr, value);
8781 if (err)
8782 return err;
8783 pr_debug("extern (kcfg) '%s': set to 0x%llx\n",
8784 ext->name, (long long)value);
8785 } else {
8786 pr_warn("extern '%s': unrecognized extern kind\n", ext->name);
8787 return -EINVAL;
8788 }
8789 }
8790 if (need_config && extra_kconfig) {
8791 err = bpf_object__read_kconfig_mem(obj, extra_kconfig, kcfg_data);
8792 if (err)
8793 return -EINVAL;
8794 need_config = false;
8795 for (i = 0; i < obj->nr_extern; i++) {
8796 ext = &obj->externs[i];
8797 if (ext->type == EXT_KCFG && !ext->is_set) {
8798 need_config = true;
8799 break;
8800 }
8801 }
8802 }
8803 if (need_config) {
8804 err = bpf_object__read_kconfig_file(obj, kcfg_data);
8805 if (err)
8806 return -EINVAL;
8807 }
8808 if (need_kallsyms) {
8809 err = bpf_object__read_kallsyms_file(obj);
8810 if (err)
8811 return -EINVAL;
8812 }
8813 if (need_vmlinux_btf) {
8814 err = bpf_object__resolve_ksyms_btf_id(obj);
8815 if (err)
8816 return -EINVAL;
8817 }
8818 for (i = 0; i < obj->nr_extern; i++) {
8819 ext = &obj->externs[i];
8820
8821 if (!ext->is_set && !ext->is_weak) {
8822 pr_warn("extern '%s' (strong): not resolved\n", ext->name);
8823 return -ESRCH;
8824 } else if (!ext->is_set) {
8825 pr_debug("extern '%s' (weak): not resolved, defaulting to zero\n",
8826 ext->name);
8827 }
8828 }
8829
8830 return 0;
8831 }
8832
8833 static void bpf_map_prepare_vdata(const struct bpf_map *map)
8834 {
8835 const struct btf_type *type;
8836 struct bpf_struct_ops *st_ops;
8837 __u32 i;
8838
8839 st_ops = map->st_ops;
8840 type = btf__type_by_id(map->obj->btf, st_ops->type_id);
8841 for (i = 0; i < btf_vlen(type); i++) {
8842 struct bpf_program *prog = st_ops->progs[i];
8843 void *kern_data;
8844 int prog_fd;
8845
8846 if (!prog)
8847 continue;
8848
8849 prog_fd = bpf_program__fd(prog);
8850 kern_data = st_ops->kern_vdata + st_ops->kern_func_off[i];
8851 *(unsigned long *)kern_data = prog_fd;
8852 }
8853 }
8854
8855 static int bpf_object_prepare_struct_ops(struct bpf_object *obj)
8856 {
8857 struct bpf_map *map;
8858 int i;
8859
8860 for (i = 0; i < obj->nr_maps; i++) {
8861 map = &obj->maps[i];
8862
8863 if (!bpf_map__is_struct_ops(map))
8864 continue;
8865
8866 if (!map->autocreate)
8867 continue;
8868
8869 bpf_map_prepare_vdata(map);
8870 }
8871
8872 return 0;
8873 }
8874
8875 static void bpf_object_unpin(struct bpf_object *obj)
8876 {
8877 int i;
8878
8879 /* unpin any maps that were auto-pinned during load */
8880 for (i = 0; i < obj->nr_maps; i++)
8881 if (obj->maps[i].pinned && !obj->maps[i].reused)
8882 bpf_map__unpin(&obj->maps[i], NULL);
8883 }
8884
8885 static void bpf_object_post_load_cleanup(struct bpf_object *obj)
8886 {
8887 int i;
8888
8889 /* clean up fd_array */
8890 zfree(&obj->fd_array);
8891
8892 /* clean up module BTFs */
8893 for (i = 0; i < obj->btf_module_cnt; i++) {
8894 close(obj->btf_modules[i].fd);
8895 btf__free(obj->btf_modules[i].btf);
8896 free(obj->btf_modules[i].name);
8897 }
8898 obj->btf_module_cnt = 0;
8899 zfree(&obj->btf_modules);
8900
8901 /* clean up vmlinux BTF */
8902 btf__free(obj->btf_vmlinux);
8903 obj->btf_vmlinux = NULL;
8904 }
8905
8906 static int bpf_object_prepare(struct bpf_object *obj, const char *target_btf_path)
8907 {
8908 int err;
8909
8910 if (obj->state >= OBJ_PREPARED) {
8911 pr_warn("object '%s': prepare loading can't be attempted twice\n", obj->name);
8912 return -EINVAL;
8913 }
8914
8915 err = bpf_object_prepare_token(obj);
8916 err = err ? : bpf_object__probe_loading(obj);
8917 err = err ? : bpf_object__load_vmlinux_btf(obj, false);
8918 err = err ? : bpf_object__resolve_externs(obj, obj->kconfig);
8919 err = err ? : bpf_object__sanitize_maps(obj);
8920 err = err ? : bpf_object__init_kern_struct_ops_maps(obj);
8921 err = err ? : bpf_object_adjust_struct_ops_autoload(obj);
8922 err = err ? : bpf_object__relocate(obj, obj->btf_custom_path ? : target_btf_path);
8923 err = err ? : bpf_object__sanitize_and_load_btf(obj);
8924 err = err ? : bpf_object__create_maps(obj);
8925 err = err ? : bpf_object_prepare_progs(obj);
8926
8927 if (err) {
8928 bpf_object_unpin(obj);
8929 bpf_object_unload(obj);
8930 obj->state = OBJ_LOADED;
8931 return err;
8932 }
8933
8934 obj->state = OBJ_PREPARED;
8935 return 0;
8936 }
8937
8938 static int bpf_object_load(struct bpf_object *obj, int extra_log_level, const char *target_btf_path)
8939 {
8940 int err;
8941
8942 if (!obj)
8943 return libbpf_err(-EINVAL);
8944
8945 if (obj->state >= OBJ_LOADED) {
8946 pr_warn("object '%s': load can't be attempted twice\n", obj->name);
8947 return libbpf_err(-EINVAL);
8948 }
8949
8950 /* Disallow kernel loading programs of non-native endianness but
8951 * permit cross-endian creation of "light skeleton".
8952 */
8953 if (obj->gen_loader) {
8954 bpf_gen__init(obj->gen_loader, extra_log_level, obj->nr_programs, obj->nr_maps);
8955 } else if (!is_native_endianness(obj)) {
8956 pr_warn("object '%s': loading non-native endianness is unsupported\n", obj->name);
8957 return libbpf_err(-LIBBPF_ERRNO__ENDIAN);
8958 }
8959
8960 if (obj->state < OBJ_PREPARED) {
8961 err = bpf_object_prepare(obj, target_btf_path);
8962 if (err)
8963 return libbpf_err(err);
8964 }
8965 err = bpf_object__load_progs(obj, extra_log_level);
8966 err = err ? : bpf_object_init_prog_arrays(obj);
8967 err = err ? : bpf_object_prepare_struct_ops(obj);
8968
8969 if (obj->gen_loader) {
8970 /* reset FDs */
8971 if (obj->btf)
8972 btf__set_fd(obj->btf, -1);
8973 if (!err)
8974 err = bpf_gen__finish(obj->gen_loader, obj->nr_programs, obj->nr_maps);
8975 }
8976
8977 bpf_object_post_load_cleanup(obj);
8978 obj->state = OBJ_LOADED; /* doesn't matter if successfully or not */
8979
8980 if (err) {
8981 bpf_object_unpin(obj);
8982 bpf_object_unload(obj);
8983 pr_warn("failed to load object '%s'\n", obj->path);
8984 return libbpf_err(err);
8985 }
8986
8987 return 0;
8988 }
8989
8990 int bpf_object__prepare(struct bpf_object *obj)
8991 {
8992 return libbpf_err(bpf_object_prepare(obj, NULL));
8993 }
8994
8995 int bpf_object__load(struct bpf_object *obj)
8996 {
8997 return bpf_object_load(obj, 0, NULL);
8998 }
8999
9000 static int make_parent_dir(const char *path)
9001 {
9002 char *dname, *dir;
9003 int err = 0;
9004
9005 dname = strdup(path);
9006 if (dname == NULL)
9007 return -ENOMEM;
9008
9009 dir = dirname(dname);
9010 if (mkdir(dir, 0700) && errno != EEXIST)
9011 err = -errno;
9012
9013 free(dname);
9014 if (err) {
9015 pr_warn("failed to mkdir %s: %s\n", path, errstr(err));
9016 }
9017 return err;
9018 }
9019
9020 static int check_path(const char *path)
9021 {
9022 struct statfs st_fs;
9023 char *dname, *dir;
9024 int err = 0;
9025
9026 if (path == NULL)
9027 return -EINVAL;
9028
9029 dname = strdup(path);
9030 if (dname == NULL)
9031 return -ENOMEM;
9032
9033 dir = dirname(dname);
9034 if (statfs(dir, &st_fs)) {
9035 pr_warn("failed to statfs %s: %s\n", dir, errstr(errno));
9036 err = -errno;
9037 }
9038 free(dname);
9039
9040 if (!err && st_fs.f_type != BPF_FS_MAGIC) {
9041 pr_warn("specified path %s is not on BPF FS\n", path);
9042 err = -EINVAL;
9043 }
9044
9045 return err;
9046 }
9047
9048 int bpf_program__pin(struct bpf_program *prog, const char *path)
9049 {
9050 int err;
9051
9052 if (prog->fd < 0) {
9053 pr_warn("prog '%s': can't pin program that wasn't loaded\n", prog->name);
9054 return libbpf_err(-EINVAL);
9055 }
9056
9057 err = make_parent_dir(path);
9058 if (err)
9059 return libbpf_err(err);
9060
9061 err = check_path(path);
9062 if (err)
9063 return libbpf_err(err);
9064
9065 if (bpf_obj_pin(prog->fd, path)) {
9066 err = -errno;
9067 pr_warn("prog '%s': failed to pin at '%s': %s\n", prog->name, path, errstr(err));
9068 return libbpf_err(err);
9069 }
9070
9071 pr_debug("prog '%s': pinned at '%s'\n", prog->name, path);
9072 return 0;
9073 }
9074
9075 int bpf_program__unpin(struct bpf_program *prog, const char *path)
9076 {
9077 int err;
9078
9079 if (prog->fd < 0) {
9080 pr_warn("prog '%s': can't unpin program that wasn't loaded\n", prog->name);
9081 return libbpf_err(-EINVAL);
9082 }
9083
9084 err = check_path(path);
9085 if (err)
9086 return libbpf_err(err);
9087
9088 err = unlink(path);
9089 if (err)
9090 return libbpf_err(-errno);
9091
9092 pr_debug("prog '%s': unpinned from '%s'\n", prog->name, path);
9093 return 0;
9094 }
9095
9096 int bpf_map__pin(struct bpf_map *map, const char *path)
9097 {
9098 int err;
9099
9100 if (map == NULL) {
9101 pr_warn("invalid map pointer\n");
9102 return libbpf_err(-EINVAL);
9103 }
9104
9105 if (map->fd < 0) {
9106 pr_warn("map '%s': can't pin BPF map without FD (was it created?)\n", map->name);
9107 return libbpf_err(-EINVAL);
9108 }
9109
9110 if (map->pin_path) {
9111 if (path && strcmp(path, map->pin_path)) {
9112 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
9113 bpf_map__name(map), map->pin_path, path);
9114 return libbpf_err(-EINVAL);
9115 } else if (map->pinned) {
9116 pr_debug("map '%s' already pinned at '%s'; not re-pinning\n",
9117 bpf_map__name(map), map->pin_path);
9118 return 0;
9119 }
9120 } else {
9121 if (!path) {
9122 pr_warn("missing a path to pin map '%s' at\n",
9123 bpf_map__name(map));
9124 return libbpf_err(-EINVAL);
9125 } else if (map->pinned) {
9126 pr_warn("map '%s' already pinned\n", bpf_map__name(map));
9127 return libbpf_err(-EEXIST);
9128 }
9129
9130 map->pin_path = strdup(path);
9131 if (!map->pin_path) {
9132 err = -errno;
9133 goto out_err;
9134 }
9135 }
9136
9137 err = make_parent_dir(map->pin_path);
9138 if (err)
9139 return libbpf_err(err);
9140
9141 err = check_path(map->pin_path);
9142 if (err)
9143 return libbpf_err(err);
9144
9145 if (bpf_obj_pin(map->fd, map->pin_path)) {
9146 err = -errno;
9147 goto out_err;
9148 }
9149
9150 map->pinned = true;
9151 pr_debug("pinned map '%s'\n", map->pin_path);
9152
9153 return 0;
9154
9155 out_err:
9156 pr_warn("failed to pin map: %s\n", errstr(err));
9157 return libbpf_err(err);
9158 }
9159
9160 int bpf_map__unpin(struct bpf_map *map, const char *path)
9161 {
9162 int err;
9163
9164 if (map == NULL) {
9165 pr_warn("invalid map pointer\n");
9166 return libbpf_err(-EINVAL);
9167 }
9168
9169 if (map->pin_path) {
9170 if (path && strcmp(path, map->pin_path)) {
9171 pr_warn("map '%s' already has pin path '%s' different from '%s'\n",
9172 bpf_map__name(map), map->pin_path, path);
9173 return libbpf_err(-EINVAL);
9174 }
9175 path = map->pin_path;
9176 } else if (!path) {
9177 pr_warn("no path to unpin map '%s' from\n",
9178 bpf_map__name(map));
9179 return libbpf_err(-EINVAL);
9180 }
9181
9182 err = check_path(path);
9183 if (err)
9184 return libbpf_err(err);
9185
9186 err = unlink(path);
9187 if (err != 0)
9188 return libbpf_err(-errno);
9189
9190 map->pinned = false;
9191 pr_debug("unpinned map '%s' from '%s'\n", bpf_map__name(map), path);
9192
9193 return 0;
9194 }
9195
9196 int bpf_map__set_pin_path(struct bpf_map *map, const char *path)
9197 {
9198 char *new = NULL;
9199
9200 if (path) {
9201 new = strdup(path);
9202 if (!new)
9203 return libbpf_err(-errno);
9204 }
9205
9206 free(map->pin_path);
9207 map->pin_path = new;
9208 return 0;
9209 }
9210
9211 __alias(bpf_map__pin_path)
9212 const char *bpf_map__get_pin_path(const struct bpf_map *map);
9213
9214 const char *bpf_map__pin_path(const struct bpf_map *map)
9215 {
9216 return map->pin_path;
9217 }
9218
9219 bool bpf_map__is_pinned(const struct bpf_map *map)
9220 {
9221 return map->pinned;
9222 }
9223
9224 static void sanitize_pin_path(char *s)
9225 {
9226 /* bpffs disallows periods in path names */
9227 while (*s) {
9228 if (*s == '.')
9229 *s = '_';
9230 s++;
9231 }
9232 }
9233
9234 int bpf_object__pin_maps(struct bpf_object *obj, const char *path)
9235 {
9236 struct bpf_map *map;
9237 int err;
9238
9239 if (!obj)
9240 return libbpf_err(-ENOENT);
9241
9242 if (obj->state < OBJ_PREPARED) {
9243 pr_warn("object not yet loaded; load it first\n");
9244 return libbpf_err(-ENOENT);
9245 }
9246
9247 bpf_object__for_each_map(map, obj) {
9248 char *pin_path = NULL;
9249 char buf[PATH_MAX];
9250
9251 if (!map->autocreate)
9252 continue;
9253
9254 if (path) {
9255 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
9256 if (err)
9257 goto err_unpin_maps;
9258 sanitize_pin_path(buf);
9259 pin_path = buf;
9260 } else if (!map->pin_path) {
9261 continue;
9262 }
9263
9264 err = bpf_map__pin(map, pin_path);
9265 if (err)
9266 goto err_unpin_maps;
9267 }
9268
9269 return 0;
9270
9271 err_unpin_maps:
9272 while ((map = bpf_object__prev_map(obj, map))) {
9273 if (!map->pin_path)
9274 continue;
9275
9276 bpf_map__unpin(map, NULL);
9277 }
9278
9279 return libbpf_err(err);
9280 }
9281
9282 int bpf_object__unpin_maps(struct bpf_object *obj, const char *path)
9283 {
9284 struct bpf_map *map;
9285 int err;
9286
9287 if (!obj)
9288 return libbpf_err(-ENOENT);
9289
9290 bpf_object__for_each_map(map, obj) {
9291 char *pin_path = NULL;
9292 char buf[PATH_MAX];
9293
9294 if (path) {
9295 err = pathname_concat(buf, sizeof(buf), path, bpf_map__name(map));
9296 if (err)
9297 return libbpf_err(err);
9298 sanitize_pin_path(buf);
9299 pin_path = buf;
9300 } else if (!map->pin_path) {
9301 continue;
9302 }
9303
9304 err = bpf_map__unpin(map, pin_path);
9305 if (err)
9306 return libbpf_err(err);
9307 }
9308
9309 return 0;
9310 }
9311
9312 int bpf_object__pin_programs(struct bpf_object *obj, const char *path)
9313 {
9314 struct bpf_program *prog;
9315 char buf[PATH_MAX];
9316 int err;
9317
9318 if (!obj)
9319 return libbpf_err(-ENOENT);
9320
9321 if (obj->state < OBJ_LOADED) {
9322 pr_warn("object not yet loaded; load it first\n");
9323 return libbpf_err(-ENOENT);
9324 }
9325
9326 bpf_object__for_each_program(prog, obj) {
9327 err = pathname_concat(buf, sizeof(buf), path, prog->name);
9328 if (err)
9329 goto err_unpin_programs;
9330
9331 err = bpf_program__pin(prog, buf);
9332 if (err)
9333 goto err_unpin_programs;
9334 }
9335
9336 return 0;
9337
9338 err_unpin_programs:
9339 while ((prog = bpf_object__prev_program(obj, prog))) {
9340 if (pathname_concat(buf, sizeof(buf), path, prog->name))
9341 continue;
9342
9343 bpf_program__unpin(prog, buf);
9344 }
9345
9346 return libbpf_err(err);
9347 }
9348
9349 int bpf_object__unpin_programs(struct bpf_object *obj, const char *path)
9350 {
9351 struct bpf_program *prog;
9352 int err;
9353
9354 if (!obj)
9355 return libbpf_err(-ENOENT);
9356
9357 bpf_object__for_each_program(prog, obj) {
9358 char buf[PATH_MAX];
9359
9360 err = pathname_concat(buf, sizeof(buf), path, prog->name);
9361 if (err)
9362 return libbpf_err(err);
9363
9364 err = bpf_program__unpin(prog, buf);
9365 if (err)
9366 return libbpf_err(err);
9367 }
9368
9369 return 0;
9370 }
9371
9372 int bpf_object__pin(struct bpf_object *obj, const char *path)
9373 {
9374 int err;
9375
9376 err = bpf_object__pin_maps(obj, path);
9377 if (err)
9378 return libbpf_err(err);
9379
9380 err = bpf_object__pin_programs(obj, path);
9381 if (err) {
9382 bpf_object__unpin_maps(obj, path);
9383 return libbpf_err(err);
9384 }
9385
9386 return 0;
9387 }
9388
9389 int bpf_object__unpin(struct bpf_object *obj, const char *path)
9390 {
9391 int err;
9392
9393 err = bpf_object__unpin_programs(obj, path);
9394 if (err)
9395 return libbpf_err(err);
9396
9397 err = bpf_object__unpin_maps(obj, path);
9398 if (err)
9399 return libbpf_err(err);
9400
9401 return 0;
9402 }
9403
9404 static void bpf_map__destroy(struct bpf_map *map)
9405 {
9406 if (map->inner_map) {
9407 bpf_map__destroy(map->inner_map);
9408 zfree(&map->inner_map);
9409 }
9410
9411 zfree(&map->init_slots);
9412 map->init_slots_sz = 0;
9413
9414 if (map->mmaped && map->mmaped != map->obj->arena_data)
9415 munmap(map->mmaped, bpf_map_mmap_sz(map));
9416 map->mmaped = NULL;
9417
9418 if (map->st_ops) {
9419 zfree(&map->st_ops->data);
9420 zfree(&map->st_ops->progs);
9421 zfree(&map->st_ops->kern_func_off);
9422 zfree(&map->st_ops);
9423 }
9424
9425 zfree(&map->name);
9426 zfree(&map->real_name);
9427 zfree(&map->pin_path);
9428
9429 if (map->fd >= 0)
9430 zclose(map->fd);
9431 }
9432
9433 void bpf_object__close(struct bpf_object *obj)
9434 {
9435 size_t i;
9436
9437 if (IS_ERR_OR_NULL(obj))
9438 return;
9439
9440 /*
9441 * if user called bpf_object__prepare() without ever getting to
9442 * bpf_object__load(), we need to clean up stuff that is normally
9443 * cleaned up at the end of loading step
9444 */
9445 bpf_object_post_load_cleanup(obj);
9446
9447 usdt_manager_free(obj->usdt_man);
9448 obj->usdt_man = NULL;
9449
9450 bpf_gen__free(obj->gen_loader);
9451 bpf_object__elf_finish(obj);
9452 bpf_object_unload(obj);
9453 btf__free(obj->btf);
9454 btf__free(obj->btf_vmlinux);
9455 btf_ext__free(obj->btf_ext);
9456
9457 for (i = 0; i < obj->nr_maps; i++)
9458 bpf_map__destroy(&obj->maps[i]);
9459
9460 zfree(&obj->btf_custom_path);
9461 zfree(&obj->kconfig);
9462
9463 for (i = 0; i < obj->nr_extern; i++) {
9464 zfree(&obj->externs[i].name);
9465 zfree(&obj->externs[i].essent_name);
9466 }
9467
9468 zfree(&obj->externs);
9469 obj->nr_extern = 0;
9470
9471 zfree(&obj->maps);
9472 obj->nr_maps = 0;
9473
9474 if (obj->programs && obj->nr_programs) {
9475 for (i = 0; i < obj->nr_programs; i++)
9476 bpf_program__exit(&obj->programs[i]);
9477 }
9478 zfree(&obj->programs);
9479
9480 zfree(&obj->feat_cache);
9481 zfree(&obj->token_path);
9482 if (obj->token_fd > 0)
9483 close(obj->token_fd);
9484
9485 zfree(&obj->arena_data);
9486
9487 zfree(&obj->jumptables_data);
9488 obj->jumptables_data_sz = 0;
9489
9490 for (i = 0; i < obj->jumptable_map_cnt; i++)
9491 close(obj->jumptable_maps[i].fd);
9492 zfree(&obj->jumptable_maps);
9493
9494 free(obj);
9495 }
9496
9497 const char *bpf_object__name(const struct bpf_object *obj)
9498 {
9499 return obj ? obj->name : libbpf_err_ptr(-EINVAL);
9500 }
9501
9502 unsigned int bpf_object__kversion(const struct bpf_object *obj)
9503 {
9504 return obj ? obj->kern_version : 0;
9505 }
9506
9507 int bpf_object__token_fd(const struct bpf_object *obj)
9508 {
9509 return obj->token_fd ?: -1;
9510 }
9511
9512 struct btf *bpf_object__btf(const struct bpf_object *obj)
9513 {
9514 return obj ? obj->btf : NULL;
9515 }
9516
9517 int bpf_object__btf_fd(const struct bpf_object *obj)
9518 {
9519 return obj->btf ? btf__fd(obj->btf) : -1;
9520 }
9521
9522 int bpf_object__set_kversion(struct bpf_object *obj, __u32 kern_version)
9523 {
9524 if (obj->state >= OBJ_LOADED)
9525 return libbpf_err(-EINVAL);
9526
9527 obj->kern_version = kern_version;
9528
9529 return 0;
9530 }
9531
9532 int bpf_object__gen_loader(struct bpf_object *obj, struct gen_loader_opts *opts)
9533 {
9534 struct bpf_gen *gen;
9535
9536 if (!opts)
9537 return libbpf_err(-EFAULT);
9538 if (!OPTS_VALID(opts, gen_loader_opts))
9539 return libbpf_err(-EINVAL);
9540 gen = calloc(1, sizeof(*gen));
9541 if (!gen)
9542 return libbpf_err(-ENOMEM);
9543 gen->opts = opts;
9544 gen->swapped_endian = !is_native_endianness(obj);
9545 obj->gen_loader = gen;
9546 return 0;
9547 }
9548
9549 static struct bpf_program *
9550 __bpf_program__iter(const struct bpf_program *p, const struct bpf_object *obj,
9551 bool forward)
9552 {
9553 size_t nr_programs = obj->nr_programs;
9554 ssize_t idx;
9555
9556 if (!nr_programs)
9557 return NULL;
9558
9559 if (!p)
9560 /* Iter from the beginning */
9561 return forward ? &obj->programs[0] :
9562 &obj->programs[nr_programs - 1];
9563
9564 if (p->obj != obj) {
9565 pr_warn("error: program handler doesn't match object\n");
9566 return errno = EINVAL, NULL;
9567 }
9568
9569 idx = (p - obj->programs) + (forward ? 1 : -1);
9570 if (idx >= obj->nr_programs || idx < 0)
9571 return NULL;
9572 return &obj->programs[idx];
9573 }
9574
9575 struct bpf_program *
9576 bpf_object__next_program(const struct bpf_object *obj, struct bpf_program *prev)
9577 {
9578 struct bpf_program *prog = prev;
9579
9580 do {
9581 prog = __bpf_program__iter(prog, obj, true);
9582 } while (prog && prog_is_subprog(obj, prog));
9583
9584 return prog;
9585 }
9586
9587 struct bpf_program *
9588 bpf_object__prev_program(const struct bpf_object *obj, struct bpf_program *next)
9589 {
9590 struct bpf_program *prog = next;
9591
9592 do {
9593 prog = __bpf_program__iter(prog, obj, false);
9594 } while (prog && prog_is_subprog(obj, prog));
9595
9596 return prog;
9597 }
9598
9599 void bpf_program__set_ifindex(struct bpf_program *prog, __u32 ifindex)
9600 {
9601 prog->prog_ifindex = ifindex;
9602 }
9603
9604 const char *bpf_program__name(const struct bpf_program *prog)
9605 {
9606 return prog->name;
9607 }
9608
9609 const char *bpf_program__section_name(const struct bpf_program *prog)
9610 {
9611 return prog->sec_name;
9612 }
9613
9614 bool bpf_program__autoload(const struct bpf_program *prog)
9615 {
9616 return prog->autoload;
9617 }
9618
9619 int bpf_program__set_autoload(struct bpf_program *prog, bool autoload)
9620 {
9621 if (prog->obj->state >= OBJ_LOADED)
9622 return libbpf_err(-EINVAL);
9623
9624 prog->autoload = autoload;
9625 return 0;
9626 }
9627
9628 bool bpf_program__autoattach(const struct bpf_program *prog)
9629 {
9630 return prog->autoattach;
9631 }
9632
9633 void bpf_program__set_autoattach(struct bpf_program *prog, bool autoattach)
9634 {
9635 prog->autoattach = autoattach;
9636 }
9637
9638 const struct bpf_insn *bpf_program__insns(const struct bpf_program *prog)
9639 {
9640 return prog->insns;
9641 }
9642
9643 size_t bpf_program__insn_cnt(const struct bpf_program *prog)
9644 {
9645 return prog->insns_cnt;
9646 }
9647
9648 int bpf_program__set_insns(struct bpf_program *prog,
9649 struct bpf_insn *new_insns, size_t new_insn_cnt)
9650 {
9651 struct bpf_insn *insns;
9652
9653 if (prog->obj->state >= OBJ_LOADED)
9654 return libbpf_err(-EBUSY);
9655
9656 insns = libbpf_reallocarray(prog->insns, new_insn_cnt, sizeof(*insns));
9657 /* NULL is a valid return from reallocarray if the new count is zero */
9658 if (!insns && new_insn_cnt) {
9659 pr_warn("prog '%s': failed to realloc prog code\n", prog->name);
9660 return libbpf_err(-ENOMEM);
9661 }
9662 memcpy(insns, new_insns, new_insn_cnt * sizeof(*insns));
9663
9664 prog->insns = insns;
9665 prog->insns_cnt = new_insn_cnt;
9666 return 0;
9667 }
9668
9669 int bpf_program__fd(const struct bpf_program *prog)
9670 {
9671 if (!prog)
9672 return libbpf_err(-EINVAL);
9673
9674 if (prog->fd < 0)
9675 return libbpf_err(-ENOENT);
9676
9677 return prog->fd;
9678 }
9679
9680 __alias(bpf_program__type)
9681 enum bpf_prog_type bpf_program__get_type(const struct bpf_program *prog);
9682
9683 enum bpf_prog_type bpf_program__type(const struct bpf_program *prog)
9684 {
9685 return prog->type;
9686 }
9687
9688 static size_t custom_sec_def_cnt;
9689 static struct bpf_sec_def *custom_sec_defs;
9690 static struct bpf_sec_def custom_fallback_def;
9691 static bool has_custom_fallback_def;
9692 static int last_custom_sec_def_handler_id;
9693
9694 int bpf_program__set_type(struct bpf_program *prog, enum bpf_prog_type type)
9695 {
9696 if (prog->obj->state >= OBJ_LOADED)
9697 return libbpf_err(-EBUSY);
9698
9699 /* if type is not changed, do nothing */
9700 if (prog->type == type)
9701 return 0;
9702
9703 prog->type = type;
9704
9705 /* If a program type was changed, we need to reset associated SEC()
9706 * handler, as it will be invalid now. The only exception is a generic
9707 * fallback handler, which by definition is program type-agnostic and
9708 * is a catch-all custom handler, optionally set by the application,
9709 * so should be able to handle any type of BPF program.
9710 */
9711 if (prog->sec_def != &custom_fallback_def)
9712 prog->sec_def = NULL;
9713 return 0;
9714 }
9715
9716 __alias(bpf_program__expected_attach_type)
9717 enum bpf_attach_type bpf_program__get_expected_attach_type(const struct bpf_program *prog);
9718
9719 enum bpf_attach_type bpf_program__expected_attach_type(const struct bpf_program *prog)
9720 {
9721 return prog->expected_attach_type;
9722 }
9723
9724 int bpf_program__set_expected_attach_type(struct bpf_program *prog,
9725 enum bpf_attach_type type)
9726 {
9727 if (prog->obj->state >= OBJ_LOADED)
9728 return libbpf_err(-EBUSY);
9729
9730 prog->expected_attach_type = type;
9731 return 0;
9732 }
9733
9734 __u32 bpf_program__flags(const struct bpf_program *prog)
9735 {
9736 return prog->prog_flags;
9737 }
9738
9739 int bpf_program__set_flags(struct bpf_program *prog, __u32 flags)
9740 {
9741 if (prog->obj->state >= OBJ_LOADED)
9742 return libbpf_err(-EBUSY);
9743
9744 prog->prog_flags = flags;
9745 return 0;
9746 }
9747
9748 __u32 bpf_program__log_level(const struct bpf_program *prog)
9749 {
9750 return prog->log_level;
9751 }
9752
9753 int bpf_program__set_log_level(struct bpf_program *prog, __u32 log_level)
9754 {
9755 if (prog->obj->state >= OBJ_LOADED)
9756 return libbpf_err(-EBUSY);
9757
9758 prog->log_level = log_level;
9759 return 0;
9760 }
9761
9762 const char *bpf_program__log_buf(const struct bpf_program *prog, size_t *log_size)
9763 {
9764 *log_size = prog->log_size;
9765 return prog->log_buf;
9766 }
9767
9768 int bpf_program__set_log_buf(struct bpf_program *prog, char *log_buf, size_t log_size)
9769 {
9770 if (log_size && !log_buf)
9771 return libbpf_err(-EINVAL);
9772 if (prog->log_size > UINT_MAX)
9773 return libbpf_err(-EINVAL);
9774 if (prog->obj->state >= OBJ_LOADED)
9775 return libbpf_err(-EBUSY);
9776
9777 prog->log_buf = log_buf;
9778 prog->log_size = log_size;
9779 return 0;
9780 }
9781
9782 struct bpf_func_info *bpf_program__func_info(const struct bpf_program *prog)
9783 {
9784 if (prog->func_info_rec_size != sizeof(struct bpf_func_info))
9785 return libbpf_err_ptr(-EOPNOTSUPP);
9786 return prog->func_info;
9787 }
9788
9789 __u32 bpf_program__func_info_cnt(const struct bpf_program *prog)
9790 {
9791 return prog->func_info_cnt;
9792 }
9793
9794 struct bpf_line_info *bpf_program__line_info(const struct bpf_program *prog)
9795 {
9796 if (prog->line_info_rec_size != sizeof(struct bpf_line_info))
9797 return libbpf_err_ptr(-EOPNOTSUPP);
9798 return prog->line_info;
9799 }
9800
9801 __u32 bpf_program__line_info_cnt(const struct bpf_program *prog)
9802 {
9803 return prog->line_info_cnt;
9804 }
9805
9806 #define SEC_DEF(sec_pfx, ptype, atype, flags, ...) { \
9807 .sec = (char *)sec_pfx, \
9808 .prog_type = BPF_PROG_TYPE_##ptype, \
9809 .expected_attach_type = atype, \
9810 .cookie = (long)(flags), \
9811 .prog_prepare_load_fn = libbpf_prepare_prog_load, \
9812 __VA_ARGS__ \
9813 }
9814
9815 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9816 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9817 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9818 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9819 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9820 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9821 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9822 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9823 static int attach_kprobe_session(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9824 static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9825 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9826 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link);
9827
9828 static const struct bpf_sec_def section_defs[] = {
9829 SEC_DEF("socket", SOCKET_FILTER, 0, SEC_NONE),
9830 SEC_DEF("sk_reuseport/migrate", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT_OR_MIGRATE, SEC_ATTACHABLE),
9831 SEC_DEF("sk_reuseport", SK_REUSEPORT, BPF_SK_REUSEPORT_SELECT, SEC_ATTACHABLE),
9832 SEC_DEF("kprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
9833 SEC_DEF("uprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
9834 SEC_DEF("uprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
9835 SEC_DEF("kretprobe+", KPROBE, 0, SEC_NONE, attach_kprobe),
9836 SEC_DEF("uretprobe+", KPROBE, 0, SEC_NONE, attach_uprobe),
9837 SEC_DEF("uretprobe.s+", KPROBE, 0, SEC_SLEEPABLE, attach_uprobe),
9838 SEC_DEF("kprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
9839 SEC_DEF("kretprobe.multi+", KPROBE, BPF_TRACE_KPROBE_MULTI, SEC_NONE, attach_kprobe_multi),
9840 SEC_DEF("kprobe.session+", KPROBE, BPF_TRACE_KPROBE_SESSION, SEC_NONE, attach_kprobe_session),
9841 SEC_DEF("uprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi),
9842 SEC_DEF("uretprobe.multi+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_NONE, attach_uprobe_multi),
9843 SEC_DEF("uprobe.session+", KPROBE, BPF_TRACE_UPROBE_SESSION, SEC_NONE, attach_uprobe_multi),
9844 SEC_DEF("uprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi),
9845 SEC_DEF("uretprobe.multi.s+", KPROBE, BPF_TRACE_UPROBE_MULTI, SEC_SLEEPABLE, attach_uprobe_multi),
9846 SEC_DEF("uprobe.session.s+", KPROBE, BPF_TRACE_UPROBE_SESSION, SEC_SLEEPABLE, attach_uprobe_multi),
9847 SEC_DEF("ksyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
9848 SEC_DEF("kretsyscall+", KPROBE, 0, SEC_NONE, attach_ksyscall),
9849 SEC_DEF("usdt+", KPROBE, 0, SEC_USDT, attach_usdt),
9850 SEC_DEF("usdt.s+", KPROBE, 0, SEC_USDT | SEC_SLEEPABLE, attach_usdt),
9851 SEC_DEF("tc/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE), /* alias for tcx */
9852 SEC_DEF("tc/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE), /* alias for tcx */
9853 SEC_DEF("tcx/ingress", SCHED_CLS, BPF_TCX_INGRESS, SEC_NONE),
9854 SEC_DEF("tcx/egress", SCHED_CLS, BPF_TCX_EGRESS, SEC_NONE),
9855 SEC_DEF("tc", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */
9856 SEC_DEF("classifier", SCHED_CLS, 0, SEC_NONE), /* deprecated / legacy, use tcx */
9857 SEC_DEF("action", SCHED_ACT, 0, SEC_NONE), /* deprecated / legacy, use tcx */
9858 SEC_DEF("netkit/primary", SCHED_CLS, BPF_NETKIT_PRIMARY, SEC_NONE),
9859 SEC_DEF("netkit/peer", SCHED_CLS, BPF_NETKIT_PEER, SEC_NONE),
9860 SEC_DEF("tracepoint+", TRACEPOINT, 0, SEC_NONE, attach_tp),
9861 SEC_DEF("tp+", TRACEPOINT, 0, SEC_NONE, attach_tp),
9862 SEC_DEF("raw_tracepoint+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
9863 SEC_DEF("raw_tp+", RAW_TRACEPOINT, 0, SEC_NONE, attach_raw_tp),
9864 SEC_DEF("raw_tracepoint.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
9865 SEC_DEF("raw_tp.w+", RAW_TRACEPOINT_WRITABLE, 0, SEC_NONE, attach_raw_tp),
9866 SEC_DEF("tp_btf+", TRACING, BPF_TRACE_RAW_TP, SEC_ATTACH_BTF, attach_trace),
9867 SEC_DEF("fentry+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF, attach_trace),
9868 SEC_DEF("fmod_ret+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF, attach_trace),
9869 SEC_DEF("fexit+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF, attach_trace),
9870 SEC_DEF("fentry.s+", TRACING, BPF_TRACE_FENTRY, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
9871 SEC_DEF("fmod_ret.s+", TRACING, BPF_MODIFY_RETURN, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
9872 SEC_DEF("fexit.s+", TRACING, BPF_TRACE_FEXIT, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
9873 SEC_DEF("fsession+", TRACING, BPF_TRACE_FSESSION, SEC_ATTACH_BTF, attach_trace),
9874 SEC_DEF("fsession.s+", TRACING, BPF_TRACE_FSESSION, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_trace),
9875 SEC_DEF("freplace+", EXT, 0, SEC_ATTACH_BTF, attach_trace),
9876 SEC_DEF("lsm+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF, attach_lsm),
9877 SEC_DEF("lsm.s+", LSM, BPF_LSM_MAC, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_lsm),
9878 SEC_DEF("lsm_cgroup+", LSM, BPF_LSM_CGROUP, SEC_ATTACH_BTF),
9879 SEC_DEF("iter+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF, attach_iter),
9880 SEC_DEF("iter.s+", TRACING, BPF_TRACE_ITER, SEC_ATTACH_BTF | SEC_SLEEPABLE, attach_iter),
9881 SEC_DEF("syscall", SYSCALL, 0, SEC_SLEEPABLE),
9882 SEC_DEF("xdp.frags/devmap", XDP, BPF_XDP_DEVMAP, SEC_XDP_FRAGS),
9883 SEC_DEF("xdp/devmap", XDP, BPF_XDP_DEVMAP, SEC_ATTACHABLE),
9884 SEC_DEF("xdp.frags/cpumap", XDP, BPF_XDP_CPUMAP, SEC_XDP_FRAGS),
9885 SEC_DEF("xdp/cpumap", XDP, BPF_XDP_CPUMAP, SEC_ATTACHABLE),
9886 SEC_DEF("xdp.frags", XDP, BPF_XDP, SEC_XDP_FRAGS),
9887 SEC_DEF("xdp", XDP, BPF_XDP, SEC_ATTACHABLE_OPT),
9888 SEC_DEF("perf_event", PERF_EVENT, 0, SEC_NONE),
9889 SEC_DEF("lwt_in", LWT_IN, 0, SEC_NONE),
9890 SEC_DEF("lwt_out", LWT_OUT, 0, SEC_NONE),
9891 SEC_DEF("lwt_xmit", LWT_XMIT, 0, SEC_NONE),
9892 SEC_DEF("lwt_seg6local", LWT_SEG6LOCAL, 0, SEC_NONE),
9893 SEC_DEF("sockops", SOCK_OPS, BPF_CGROUP_SOCK_OPS, SEC_ATTACHABLE_OPT),
9894 SEC_DEF("sk_skb/stream_parser", SK_SKB, BPF_SK_SKB_STREAM_PARSER, SEC_ATTACHABLE_OPT),
9895 SEC_DEF("sk_skb/stream_verdict",SK_SKB, BPF_SK_SKB_STREAM_VERDICT, SEC_ATTACHABLE_OPT),
9896 SEC_DEF("sk_skb/verdict", SK_SKB, BPF_SK_SKB_VERDICT, SEC_ATTACHABLE_OPT),
9897 SEC_DEF("sk_skb", SK_SKB, 0, SEC_NONE),
9898 SEC_DEF("sk_msg", SK_MSG, BPF_SK_MSG_VERDICT, SEC_ATTACHABLE_OPT),
9899 SEC_DEF("lirc_mode2", LIRC_MODE2, BPF_LIRC_MODE2, SEC_ATTACHABLE_OPT),
9900 SEC_DEF("flow_dissector", FLOW_DISSECTOR, BPF_FLOW_DISSECTOR, SEC_ATTACHABLE_OPT),
9901 SEC_DEF("cgroup_skb/ingress", CGROUP_SKB, BPF_CGROUP_INET_INGRESS, SEC_ATTACHABLE_OPT),
9902 SEC_DEF("cgroup_skb/egress", CGROUP_SKB, BPF_CGROUP_INET_EGRESS, SEC_ATTACHABLE_OPT),
9903 SEC_DEF("cgroup/skb", CGROUP_SKB, 0, SEC_NONE),
9904 SEC_DEF("cgroup/sock_create", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE),
9905 SEC_DEF("cgroup/sock_release", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_RELEASE, SEC_ATTACHABLE),
9906 SEC_DEF("cgroup/sock", CGROUP_SOCK, BPF_CGROUP_INET_SOCK_CREATE, SEC_ATTACHABLE_OPT),
9907 SEC_DEF("cgroup/post_bind4", CGROUP_SOCK, BPF_CGROUP_INET4_POST_BIND, SEC_ATTACHABLE),
9908 SEC_DEF("cgroup/post_bind6", CGROUP_SOCK, BPF_CGROUP_INET6_POST_BIND, SEC_ATTACHABLE),
9909 SEC_DEF("cgroup/bind4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_BIND, SEC_ATTACHABLE),
9910 SEC_DEF("cgroup/bind6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_BIND, SEC_ATTACHABLE),
9911 SEC_DEF("cgroup/connect4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_CONNECT, SEC_ATTACHABLE),
9912 SEC_DEF("cgroup/connect6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_CONNECT, SEC_ATTACHABLE),
9913 SEC_DEF("cgroup/connect_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_CONNECT, SEC_ATTACHABLE),
9914 SEC_DEF("cgroup/sendmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_SENDMSG, SEC_ATTACHABLE),
9915 SEC_DEF("cgroup/sendmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_SENDMSG, SEC_ATTACHABLE),
9916 SEC_DEF("cgroup/sendmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_SENDMSG, SEC_ATTACHABLE),
9917 SEC_DEF("cgroup/recvmsg4", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP4_RECVMSG, SEC_ATTACHABLE),
9918 SEC_DEF("cgroup/recvmsg6", CGROUP_SOCK_ADDR, BPF_CGROUP_UDP6_RECVMSG, SEC_ATTACHABLE),
9919 SEC_DEF("cgroup/recvmsg_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_RECVMSG, SEC_ATTACHABLE),
9920 SEC_DEF("cgroup/getpeername4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETPEERNAME, SEC_ATTACHABLE),
9921 SEC_DEF("cgroup/getpeername6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETPEERNAME, SEC_ATTACHABLE),
9922 SEC_DEF("cgroup/getpeername_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETPEERNAME, SEC_ATTACHABLE),
9923 SEC_DEF("cgroup/getsockname4", CGROUP_SOCK_ADDR, BPF_CGROUP_INET4_GETSOCKNAME, SEC_ATTACHABLE),
9924 SEC_DEF("cgroup/getsockname6", CGROUP_SOCK_ADDR, BPF_CGROUP_INET6_GETSOCKNAME, SEC_ATTACHABLE),
9925 SEC_DEF("cgroup/getsockname_unix", CGROUP_SOCK_ADDR, BPF_CGROUP_UNIX_GETSOCKNAME, SEC_ATTACHABLE),
9926 SEC_DEF("cgroup/sysctl", CGROUP_SYSCTL, BPF_CGROUP_SYSCTL, SEC_ATTACHABLE),
9927 SEC_DEF("cgroup/getsockopt", CGROUP_SOCKOPT, BPF_CGROUP_GETSOCKOPT, SEC_ATTACHABLE),
9928 SEC_DEF("cgroup/setsockopt", CGROUP_SOCKOPT, BPF_CGROUP_SETSOCKOPT, SEC_ATTACHABLE),
9929 SEC_DEF("cgroup/dev", CGROUP_DEVICE, BPF_CGROUP_DEVICE, SEC_ATTACHABLE_OPT),
9930 SEC_DEF("struct_ops+", STRUCT_OPS, 0, SEC_NONE),
9931 SEC_DEF("struct_ops.s+", STRUCT_OPS, 0, SEC_SLEEPABLE),
9932 SEC_DEF("sk_lookup", SK_LOOKUP, BPF_SK_LOOKUP, SEC_ATTACHABLE),
9933 SEC_DEF("netfilter", NETFILTER, BPF_NETFILTER, SEC_NONE),
9934 };
9935
9936 int libbpf_register_prog_handler(const char *sec,
9937 enum bpf_prog_type prog_type,
9938 enum bpf_attach_type exp_attach_type,
9939 const struct libbpf_prog_handler_opts *opts)
9940 {
9941 struct bpf_sec_def *sec_def;
9942
9943 if (!OPTS_VALID(opts, libbpf_prog_handler_opts))
9944 return libbpf_err(-EINVAL);
9945
9946 if (last_custom_sec_def_handler_id == INT_MAX) /* prevent overflow */
9947 return libbpf_err(-E2BIG);
9948
9949 if (sec) {
9950 sec_def = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt + 1,
9951 sizeof(*sec_def));
9952 if (!sec_def)
9953 return libbpf_err(-ENOMEM);
9954
9955 custom_sec_defs = sec_def;
9956 sec_def = &custom_sec_defs[custom_sec_def_cnt];
9957 } else {
9958 if (has_custom_fallback_def)
9959 return libbpf_err(-EBUSY);
9960
9961 sec_def = &custom_fallback_def;
9962 }
9963
9964 sec_def->sec = sec ? strdup(sec) : NULL;
9965 if (sec && !sec_def->sec)
9966 return libbpf_err(-ENOMEM);
9967
9968 sec_def->prog_type = prog_type;
9969 sec_def->expected_attach_type = exp_attach_type;
9970 sec_def->cookie = OPTS_GET(opts, cookie, 0);
9971
9972 sec_def->prog_setup_fn = OPTS_GET(opts, prog_setup_fn, NULL);
9973 sec_def->prog_prepare_load_fn = OPTS_GET(opts, prog_prepare_load_fn, NULL);
9974 sec_def->prog_attach_fn = OPTS_GET(opts, prog_attach_fn, NULL);
9975
9976 sec_def->handler_id = ++last_custom_sec_def_handler_id;
9977
9978 if (sec)
9979 custom_sec_def_cnt++;
9980 else
9981 has_custom_fallback_def = true;
9982
9983 return sec_def->handler_id;
9984 }
9985
9986 int libbpf_unregister_prog_handler(int handler_id)
9987 {
9988 struct bpf_sec_def *sec_defs;
9989 int i;
9990
9991 if (handler_id <= 0)
9992 return libbpf_err(-EINVAL);
9993
9994 if (has_custom_fallback_def && custom_fallback_def.handler_id == handler_id) {
9995 memset(&custom_fallback_def, 0, sizeof(custom_fallback_def));
9996 has_custom_fallback_def = false;
9997 return 0;
9998 }
9999
10000 for (i = 0; i < custom_sec_def_cnt; i++) {
10001 if (custom_sec_defs[i].handler_id == handler_id)
10002 break;
10003 }
10004
10005 if (i == custom_sec_def_cnt)
10006 return libbpf_err(-ENOENT);
10007
10008 free(custom_sec_defs[i].sec);
10009 for (i = i + 1; i < custom_sec_def_cnt; i++)
10010 custom_sec_defs[i - 1] = custom_sec_defs[i];
10011 custom_sec_def_cnt--;
10012
10013 /* try to shrink the array, but it's ok if we couldn't */
10014 sec_defs = libbpf_reallocarray(custom_sec_defs, custom_sec_def_cnt, sizeof(*sec_defs));
10015 /* if new count is zero, reallocarray can return a valid NULL result;
10016 * in this case the previous pointer will be freed, so we *have to*
10017 * reassign old pointer to the new value (even if it's NULL)
10018 */
10019 if (sec_defs || custom_sec_def_cnt == 0)
10020 custom_sec_defs = sec_defs;
10021
10022 return 0;
10023 }
10024
10025 static bool sec_def_matches(const struct bpf_sec_def *sec_def, const char *sec_name)
10026 {
10027 size_t len = strlen(sec_def->sec);
10028
10029 /* "type/" always has to have proper SEC("type/extras") form */
10030 if (sec_def->sec[len - 1] == '/') {
10031 if (str_has_pfx(sec_name, sec_def->sec))
10032 return true;
10033 return false;
10034 }
10035
10036 /* "type+" means it can be either exact SEC("type") or
10037 * well-formed SEC("type/extras") with proper '/' separator
10038 */
10039 if (sec_def->sec[len - 1] == '+') {
10040 len--;
10041 /* not even a prefix */
10042 if (strncmp(sec_name, sec_def->sec, len) != 0)
10043 return false;
10044 /* exact match or has '/' separator */
10045 if (sec_name[len] == '\0' || sec_name[len] == '/')
10046 return true;
10047 return false;
10048 }
10049
10050 return strcmp(sec_name, sec_def->sec) == 0;
10051 }
10052
10053 static const struct bpf_sec_def *find_sec_def(const char *sec_name)
10054 {
10055 const struct bpf_sec_def *sec_def;
10056 int i, n;
10057
10058 n = custom_sec_def_cnt;
10059 for (i = 0; i < n; i++) {
10060 sec_def = &custom_sec_defs[i];
10061 if (sec_def_matches(sec_def, sec_name))
10062 return sec_def;
10063 }
10064
10065 n = ARRAY_SIZE(section_defs);
10066 for (i = 0; i < n; i++) {
10067 sec_def = &section_defs[i];
10068 if (sec_def_matches(sec_def, sec_name))
10069 return sec_def;
10070 }
10071
10072 if (has_custom_fallback_def)
10073 return &custom_fallback_def;
10074
10075 return NULL;
10076 }
10077
10078 #define MAX_TYPE_NAME_SIZE 32
10079
10080 static char *libbpf_get_type_names(bool attach_type)
10081 {
10082 int i, len = ARRAY_SIZE(section_defs) * MAX_TYPE_NAME_SIZE;
10083 char *buf;
10084
10085 buf = malloc(len);
10086 if (!buf)
10087 return NULL;
10088
10089 buf[0] = '\0';
10090 /* Forge string buf with all available names */
10091 for (i = 0; i < ARRAY_SIZE(section_defs); i++) {
10092 const struct bpf_sec_def *sec_def = &section_defs[i];
10093
10094 if (attach_type) {
10095 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
10096 continue;
10097
10098 if (!(sec_def->cookie & SEC_ATTACHABLE))
10099 continue;
10100 }
10101
10102 if (strlen(buf) + strlen(section_defs[i].sec) + 2 > len) {
10103 free(buf);
10104 return NULL;
10105 }
10106 strcat(buf, " ");
10107 strcat(buf, section_defs[i].sec);
10108 }
10109
10110 return buf;
10111 }
10112
10113 int libbpf_prog_type_by_name(const char *name, enum bpf_prog_type *prog_type,
10114 enum bpf_attach_type *expected_attach_type)
10115 {
10116 const struct bpf_sec_def *sec_def;
10117 char *type_names;
10118
10119 if (!name)
10120 return libbpf_err(-EINVAL);
10121
10122 sec_def = find_sec_def(name);
10123 if (sec_def) {
10124 *prog_type = sec_def->prog_type;
10125 *expected_attach_type = sec_def->expected_attach_type;
10126 return 0;
10127 }
10128
10129 pr_debug("failed to guess program type from ELF section '%s'\n", name);
10130 type_names = libbpf_get_type_names(false);
10131 if (type_names != NULL) {
10132 pr_debug("supported section(type) names are:%s\n", type_names);
10133 free(type_names);
10134 }
10135
10136 return libbpf_err(-ESRCH);
10137 }
10138
10139 const char *libbpf_bpf_attach_type_str(enum bpf_attach_type t)
10140 {
10141 if (t < 0 || t >= ARRAY_SIZE(attach_type_name))
10142 return NULL;
10143
10144 return attach_type_name[t];
10145 }
10146
10147 const char *libbpf_bpf_link_type_str(enum bpf_link_type t)
10148 {
10149 if (t < 0 || t >= ARRAY_SIZE(link_type_name))
10150 return NULL;
10151
10152 return link_type_name[t];
10153 }
10154
10155 const char *libbpf_bpf_map_type_str(enum bpf_map_type t)
10156 {
10157 if (t < 0 || t >= ARRAY_SIZE(map_type_name))
10158 return NULL;
10159
10160 return map_type_name[t];
10161 }
10162
10163 const char *libbpf_bpf_prog_type_str(enum bpf_prog_type t)
10164 {
10165 if (t < 0 || t >= ARRAY_SIZE(prog_type_name))
10166 return NULL;
10167
10168 return prog_type_name[t];
10169 }
10170
10171 static struct bpf_map *find_struct_ops_map_by_offset(struct bpf_object *obj,
10172 int sec_idx,
10173 size_t offset)
10174 {
10175 struct bpf_map *map;
10176 size_t i;
10177
10178 for (i = 0; i < obj->nr_maps; i++) {
10179 map = &obj->maps[i];
10180 if (!bpf_map__is_struct_ops(map))
10181 continue;
10182 if (map->sec_idx == sec_idx &&
10183 map->sec_offset <= offset &&
10184 offset - map->sec_offset < map->def.value_size)
10185 return map;
10186 }
10187
10188 return NULL;
10189 }
10190
10191 /* Collect the reloc from ELF, populate the st_ops->progs[], and update
10192 * st_ops->data for shadow type.
10193 */
10194 static int bpf_object__collect_st_ops_relos(struct bpf_object *obj,
10195 Elf64_Shdr *shdr, Elf_Data *data)
10196 {
10197 const struct btf_type *type;
10198 const struct btf_member *member;
10199 struct bpf_struct_ops *st_ops;
10200 struct bpf_program *prog;
10201 unsigned int shdr_idx;
10202 const struct btf *btf;
10203 struct bpf_map *map;
10204 unsigned int moff, insn_idx;
10205 const char *name;
10206 __u32 member_idx;
10207 Elf64_Sym *sym;
10208 Elf64_Rel *rel;
10209 int i, nrels;
10210
10211 btf = obj->btf;
10212 nrels = shdr->sh_size / shdr->sh_entsize;
10213 for (i = 0; i < nrels; i++) {
10214 rel = elf_rel_by_idx(data, i);
10215 if (!rel) {
10216 pr_warn("struct_ops reloc: failed to get %d reloc\n", i);
10217 return -LIBBPF_ERRNO__FORMAT;
10218 }
10219
10220 sym = elf_sym_by_idx(obj, ELF64_R_SYM(rel->r_info));
10221 if (!sym) {
10222 pr_warn("struct_ops reloc: symbol %zx not found\n",
10223 (size_t)ELF64_R_SYM(rel->r_info));
10224 return -LIBBPF_ERRNO__FORMAT;
10225 }
10226
10227 name = elf_sym_str(obj, sym->st_name) ?: "<?>";
10228 map = find_struct_ops_map_by_offset(obj, shdr->sh_info, rel->r_offset);
10229 if (!map) {
10230 pr_warn("struct_ops reloc: cannot find map at rel->r_offset %zu\n",
10231 (size_t)rel->r_offset);
10232 return -EINVAL;
10233 }
10234
10235 moff = rel->r_offset - map->sec_offset;
10236 shdr_idx = sym->st_shndx;
10237 st_ops = map->st_ops;
10238 pr_debug("struct_ops reloc %s: for %lld value %lld shdr_idx %u rel->r_offset %zu map->sec_offset %zu name %d (\'%s\')\n",
10239 map->name,
10240 (long long)(rel->r_info >> 32),
10241 (long long)sym->st_value,
10242 shdr_idx, (size_t)rel->r_offset,
10243 map->sec_offset, sym->st_name, name);
10244
10245 if (shdr_idx >= SHN_LORESERVE) {
10246 pr_warn("struct_ops reloc %s: rel->r_offset %zu shdr_idx %u unsupported non-static function\n",
10247 map->name, (size_t)rel->r_offset, shdr_idx);
10248 return -LIBBPF_ERRNO__RELOC;
10249 }
10250 if (sym->st_value % BPF_INSN_SZ) {
10251 pr_warn("struct_ops reloc %s: invalid target program offset %llu\n",
10252 map->name, (unsigned long long)sym->st_value);
10253 return -LIBBPF_ERRNO__FORMAT;
10254 }
10255 insn_idx = sym->st_value / BPF_INSN_SZ;
10256
10257 type = btf__type_by_id(btf, st_ops->type_id);
10258 member = find_member_by_offset(type, moff * 8);
10259 if (!member) {
10260 pr_warn("struct_ops reloc %s: cannot find member at moff %u\n",
10261 map->name, moff);
10262 return -EINVAL;
10263 }
10264 member_idx = member - btf_members(type);
10265 name = btf__name_by_offset(btf, member->name_off);
10266
10267 if (!resolve_func_ptr(btf, member->type, NULL)) {
10268 pr_warn("struct_ops reloc %s: cannot relocate non func ptr %s\n",
10269 map->name, name);
10270 return -EINVAL;
10271 }
10272
10273 prog = find_prog_by_sec_insn(obj, shdr_idx, insn_idx);
10274 if (!prog) {
10275 pr_warn("struct_ops reloc %s: cannot find prog at shdr_idx %u to relocate func ptr %s\n",
10276 map->name, shdr_idx, name);
10277 return -EINVAL;
10278 }
10279
10280 /* prevent the use of BPF prog with invalid type */
10281 if (prog->type != BPF_PROG_TYPE_STRUCT_OPS) {
10282 pr_warn("struct_ops reloc %s: prog %s is not struct_ops BPF program\n",
10283 map->name, prog->name);
10284 return -EINVAL;
10285 }
10286
10287 st_ops->progs[member_idx] = prog;
10288
10289 /* st_ops->data will be exposed to users, being returned by
10290 * bpf_map__initial_value() as a pointer to the shadow
10291 * type. All function pointers in the original struct type
10292 * should be converted to a pointer to struct bpf_program
10293 * in the shadow type.
10294 */
10295 *((struct bpf_program **)(st_ops->data + moff)) = prog;
10296 }
10297
10298 return 0;
10299 }
10300
10301 #define BTF_TRACE_PREFIX "btf_trace_"
10302 #define BTF_LSM_PREFIX "bpf_lsm_"
10303 #define BTF_ITER_PREFIX "bpf_iter_"
10304 #define BTF_MAX_NAME_SIZE 128
10305
10306 void btf_get_kernel_prefix_kind(enum bpf_attach_type attach_type,
10307 const char **prefix, int *kind)
10308 {
10309 switch (attach_type) {
10310 case BPF_TRACE_RAW_TP:
10311 *prefix = BTF_TRACE_PREFIX;
10312 *kind = BTF_KIND_TYPEDEF;
10313 break;
10314 case BPF_LSM_MAC:
10315 case BPF_LSM_CGROUP:
10316 *prefix = BTF_LSM_PREFIX;
10317 *kind = BTF_KIND_FUNC;
10318 break;
10319 case BPF_TRACE_ITER:
10320 *prefix = BTF_ITER_PREFIX;
10321 *kind = BTF_KIND_FUNC;
10322 break;
10323 default:
10324 *prefix = "";
10325 *kind = BTF_KIND_FUNC;
10326 }
10327 }
10328
10329 static int find_btf_by_prefix_kind(const struct btf *btf, const char *prefix,
10330 const char *name, __u32 kind)
10331 {
10332 char btf_type_name[BTF_MAX_NAME_SIZE];
10333 int ret;
10334
10335 ret = snprintf(btf_type_name, sizeof(btf_type_name),
10336 "%s%s", prefix, name);
10337 /* snprintf returns the number of characters written excluding the
10338 * terminating null. So, if >= BTF_MAX_NAME_SIZE are written, it
10339 * indicates truncation.
10340 */
10341 if (ret < 0 || ret >= sizeof(btf_type_name))
10342 return -ENAMETOOLONG;
10343 return btf__find_by_name_kind(btf, btf_type_name, kind);
10344 }
10345
10346 static inline int find_attach_btf_id(struct btf *btf, const char *name,
10347 enum bpf_attach_type attach_type)
10348 {
10349 const char *prefix;
10350 int kind;
10351
10352 btf_get_kernel_prefix_kind(attach_type, &prefix, &kind);
10353 return find_btf_by_prefix_kind(btf, prefix, name, kind);
10354 }
10355
10356 int libbpf_find_vmlinux_btf_id(const char *name,
10357 enum bpf_attach_type attach_type)
10358 {
10359 struct btf *btf;
10360 int err;
10361
10362 btf = btf__load_vmlinux_btf();
10363 err = libbpf_get_error(btf);
10364 if (err) {
10365 pr_warn("vmlinux BTF is not found\n");
10366 return libbpf_err(err);
10367 }
10368
10369 err = find_attach_btf_id(btf, name, attach_type);
10370 if (err <= 0)
10371 pr_warn("%s is not found in vmlinux BTF\n", name);
10372
10373 btf__free(btf);
10374 return libbpf_err(err);
10375 }
10376
10377 static int libbpf_find_prog_btf_id(const char *name, __u32 attach_prog_fd, int token_fd)
10378 {
10379 struct bpf_prog_info info;
10380 __u32 info_len = sizeof(info);
10381 struct btf *btf;
10382 int err;
10383
10384 memset(&info, 0, info_len);
10385 err = bpf_prog_get_info_by_fd(attach_prog_fd, &info, &info_len);
10386 if (err) {
10387 pr_warn("failed bpf_prog_get_info_by_fd for FD %d: %s\n",
10388 attach_prog_fd, errstr(err));
10389 return err;
10390 }
10391
10392 err = -EINVAL;
10393 if (!info.btf_id) {
10394 pr_warn("The target program doesn't have BTF\n");
10395 goto out;
10396 }
10397 btf = btf_load_from_kernel(info.btf_id, NULL, token_fd);
10398 err = libbpf_get_error(btf);
10399 if (err) {
10400 pr_warn("Failed to get BTF %d of the program: %s\n", info.btf_id, errstr(err));
10401 goto out;
10402 }
10403 err = btf__find_by_name_kind(btf, name, BTF_KIND_FUNC);
10404 btf__free(btf);
10405 if (err <= 0) {
10406 pr_warn("%s is not found in prog's BTF\n", name);
10407 goto out;
10408 }
10409 out:
10410 return err;
10411 }
10412
10413 static int find_kernel_btf_id(struct bpf_object *obj, const char *attach_name,
10414 enum bpf_attach_type attach_type,
10415 int *btf_obj_fd, int *btf_type_id)
10416 {
10417 int ret, i, mod_len = 0;
10418 const char *fn_name, *mod_name = NULL;
10419
10420 fn_name = strchr(attach_name, ':');
10421 if (fn_name) {
10422 mod_name = attach_name;
10423 mod_len = fn_name - mod_name;
10424 fn_name++;
10425 }
10426
10427 if (!mod_name || strncmp(mod_name, "vmlinux", mod_len) == 0) {
10428 ret = find_attach_btf_id(obj->btf_vmlinux,
10429 mod_name ? fn_name : attach_name,
10430 attach_type);
10431 if (ret > 0) {
10432 *btf_obj_fd = 0; /* vmlinux BTF */
10433 *btf_type_id = ret;
10434 return 0;
10435 }
10436 if (ret != -ENOENT)
10437 return ret;
10438 }
10439
10440 ret = load_module_btfs(obj);
10441 if (ret)
10442 return ret;
10443
10444 for (i = 0; i < obj->btf_module_cnt; i++) {
10445 const struct module_btf *mod = &obj->btf_modules[i];
10446
10447 if (mod_name && strncmp(mod->name, mod_name, mod_len) != 0)
10448 continue;
10449
10450 ret = find_attach_btf_id(mod->btf,
10451 mod_name ? fn_name : attach_name,
10452 attach_type);
10453 if (ret > 0) {
10454 *btf_obj_fd = mod->fd;
10455 *btf_type_id = ret;
10456 return 0;
10457 }
10458 if (ret == -ENOENT)
10459 continue;
10460
10461 return ret;
10462 }
10463
10464 return -ESRCH;
10465 }
10466
10467 static int libbpf_find_attach_btf_id(struct bpf_program *prog, const char *attach_name,
10468 int *btf_obj_fd, int *btf_type_id)
10469 {
10470 enum bpf_attach_type attach_type = prog->expected_attach_type;
10471 __u32 attach_prog_fd = prog->attach_prog_fd;
10472 int err = 0;
10473
10474 /* BPF program's BTF ID */
10475 if (prog->type == BPF_PROG_TYPE_EXT || attach_prog_fd) {
10476 if (!attach_prog_fd) {
10477 pr_warn("prog '%s': attach program FD is not set\n", prog->name);
10478 return -EINVAL;
10479 }
10480 err = libbpf_find_prog_btf_id(attach_name, attach_prog_fd, prog->obj->token_fd);
10481 if (err < 0) {
10482 pr_warn("prog '%s': failed to find BPF program (FD %d) BTF ID for '%s': %s\n",
10483 prog->name, attach_prog_fd, attach_name, errstr(err));
10484 return err;
10485 }
10486 *btf_obj_fd = 0;
10487 *btf_type_id = err;
10488 return 0;
10489 }
10490
10491 /* kernel/module BTF ID */
10492 if (prog->obj->gen_loader) {
10493 bpf_gen__record_attach_target(prog->obj->gen_loader, attach_name, attach_type);
10494 *btf_obj_fd = 0;
10495 *btf_type_id = 1;
10496 } else {
10497 err = find_kernel_btf_id(prog->obj, attach_name,
10498 attach_type, btf_obj_fd,
10499 btf_type_id);
10500 }
10501 if (err) {
10502 pr_warn("prog '%s': failed to find kernel BTF type ID of '%s': %s\n",
10503 prog->name, attach_name, errstr(err));
10504 return err;
10505 }
10506 return 0;
10507 }
10508
10509 int libbpf_attach_type_by_name(const char *name,
10510 enum bpf_attach_type *attach_type)
10511 {
10512 char *type_names;
10513 const struct bpf_sec_def *sec_def;
10514
10515 if (!name)
10516 return libbpf_err(-EINVAL);
10517
10518 sec_def = find_sec_def(name);
10519 if (!sec_def) {
10520 pr_debug("failed to guess attach type based on ELF section name '%s'\n", name);
10521 type_names = libbpf_get_type_names(true);
10522 if (type_names != NULL) {
10523 pr_debug("attachable section(type) names are:%s\n", type_names);
10524 free(type_names);
10525 }
10526
10527 return libbpf_err(-EINVAL);
10528 }
10529
10530 if (sec_def->prog_prepare_load_fn != libbpf_prepare_prog_load)
10531 return libbpf_err(-EINVAL);
10532 if (!(sec_def->cookie & SEC_ATTACHABLE))
10533 return libbpf_err(-EINVAL);
10534
10535 *attach_type = sec_def->expected_attach_type;
10536 return 0;
10537 }
10538
10539 int bpf_map__fd(const struct bpf_map *map)
10540 {
10541 if (!map)
10542 return libbpf_err(-EINVAL);
10543 if (!map_is_created(map))
10544 return -1;
10545 return map->fd;
10546 }
10547
10548 static bool map_uses_real_name(const struct bpf_map *map)
10549 {
10550 /* Since libbpf started to support custom .data.* and .rodata.* maps,
10551 * their user-visible name differs from kernel-visible name. Users see
10552 * such map's corresponding ELF section name as a map name.
10553 * This check distinguishes .data/.rodata from .data.* and .rodata.*
10554 * maps to know which name has to be returned to the user.
10555 */
10556 if (map->libbpf_type == LIBBPF_MAP_DATA && strcmp(map->real_name, DATA_SEC) != 0)
10557 return true;
10558 if (map->libbpf_type == LIBBPF_MAP_RODATA && strcmp(map->real_name, RODATA_SEC) != 0)
10559 return true;
10560 return false;
10561 }
10562
10563 const char *bpf_map__name(const struct bpf_map *map)
10564 {
10565 if (!map)
10566 return NULL;
10567
10568 if (map_uses_real_name(map))
10569 return map->real_name;
10570
10571 return map->name;
10572 }
10573
10574 enum bpf_map_type bpf_map__type(const struct bpf_map *map)
10575 {
10576 return map->def.type;
10577 }
10578
10579 int bpf_map__set_type(struct bpf_map *map, enum bpf_map_type type)
10580 {
10581 if (map_is_created(map))
10582 return libbpf_err(-EBUSY);
10583 map->def.type = type;
10584 return 0;
10585 }
10586
10587 __u32 bpf_map__map_flags(const struct bpf_map *map)
10588 {
10589 return map->def.map_flags;
10590 }
10591
10592 int bpf_map__set_map_flags(struct bpf_map *map, __u32 flags)
10593 {
10594 if (map_is_created(map))
10595 return libbpf_err(-EBUSY);
10596 map->def.map_flags = flags;
10597 return 0;
10598 }
10599
10600 __u64 bpf_map__map_extra(const struct bpf_map *map)
10601 {
10602 return map->map_extra;
10603 }
10604
10605 int bpf_map__set_map_extra(struct bpf_map *map, __u64 map_extra)
10606 {
10607 if (map_is_created(map))
10608 return libbpf_err(-EBUSY);
10609 map->map_extra = map_extra;
10610 return 0;
10611 }
10612
10613 __u32 bpf_map__numa_node(const struct bpf_map *map)
10614 {
10615 return map->numa_node;
10616 }
10617
10618 int bpf_map__set_numa_node(struct bpf_map *map, __u32 numa_node)
10619 {
10620 if (map_is_created(map))
10621 return libbpf_err(-EBUSY);
10622 map->numa_node = numa_node;
10623 return 0;
10624 }
10625
10626 __u32 bpf_map__key_size(const struct bpf_map *map)
10627 {
10628 return map->def.key_size;
10629 }
10630
10631 int bpf_map__set_key_size(struct bpf_map *map, __u32 size)
10632 {
10633 if (map_is_created(map))
10634 return libbpf_err(-EBUSY);
10635 map->def.key_size = size;
10636 return 0;
10637 }
10638
10639 __u32 bpf_map__value_size(const struct bpf_map *map)
10640 {
10641 return map->def.value_size;
10642 }
10643
10644 static int map_btf_datasec_resize(struct bpf_map *map, __u32 size)
10645 {
10646 struct btf *btf;
10647 struct btf_type *datasec_type, *var_type;
10648 struct btf_var_secinfo *var;
10649 const struct btf_type *array_type;
10650 const struct btf_array *array;
10651 int vlen, element_sz, new_array_id;
10652 __u32 nr_elements;
10653
10654 /* check btf existence */
10655 btf = bpf_object__btf(map->obj);
10656 if (!btf)
10657 return -ENOENT;
10658
10659 /* verify map is datasec */
10660 datasec_type = btf_type_by_id(btf, bpf_map__btf_value_type_id(map));
10661 if (!btf_is_datasec(datasec_type)) {
10662 pr_warn("map '%s': cannot be resized, map value type is not a datasec\n",
10663 bpf_map__name(map));
10664 return -EINVAL;
10665 }
10666
10667 /* verify datasec has at least one var */
10668 vlen = btf_vlen(datasec_type);
10669 if (vlen == 0) {
10670 pr_warn("map '%s': cannot be resized, map value datasec is empty\n",
10671 bpf_map__name(map));
10672 return -EINVAL;
10673 }
10674
10675 /* verify last var in the datasec is an array */
10676 var = &btf_var_secinfos(datasec_type)[vlen - 1];
10677 var_type = btf_type_by_id(btf, var->type);
10678 array_type = skip_mods_and_typedefs(btf, var_type->type, NULL);
10679 if (!btf_is_array(array_type)) {
10680 pr_warn("map '%s': cannot be resized, last var must be an array\n",
10681 bpf_map__name(map));
10682 return -EINVAL;
10683 }
10684
10685 /* verify request size aligns with array */
10686 array = btf_array(array_type);
10687 element_sz = btf__resolve_size(btf, array->type);
10688 if (element_sz <= 0 || (size - var->offset) % element_sz != 0) {
10689 pr_warn("map '%s': cannot be resized, element size (%d) doesn't align with new total size (%u)\n",
10690 bpf_map__name(map), element_sz, size);
10691 return -EINVAL;
10692 }
10693
10694 /* create a new array based on the existing array, but with new length */
10695 nr_elements = (size - var->offset) / element_sz;
10696 new_array_id = btf__add_array(btf, array->index_type, array->type, nr_elements);
10697 if (new_array_id < 0)
10698 return new_array_id;
10699
10700 /* adding a new btf type invalidates existing pointers to btf objects,
10701 * so refresh pointers before proceeding
10702 */
10703 datasec_type = btf_type_by_id(btf, map->btf_value_type_id);
10704 var = &btf_var_secinfos(datasec_type)[vlen - 1];
10705 var_type = btf_type_by_id(btf, var->type);
10706
10707 /* finally update btf info */
10708 datasec_type->size = size;
10709 var->size = size - var->offset;
10710 var_type->type = new_array_id;
10711
10712 return 0;
10713 }
10714
10715 int bpf_map__set_value_size(struct bpf_map *map, __u32 size)
10716 {
10717 if (map_is_created(map))
10718 return libbpf_err(-EBUSY);
10719
10720 if (map->mmaped) {
10721 size_t mmap_old_sz, mmap_new_sz;
10722 int err;
10723
10724 if (map->def.type != BPF_MAP_TYPE_ARRAY)
10725 return libbpf_err(-EOPNOTSUPP);
10726
10727 mmap_old_sz = bpf_map_mmap_sz(map);
10728 mmap_new_sz = array_map_mmap_sz(size, map->def.max_entries);
10729 err = bpf_map_mmap_resize(map, mmap_old_sz, mmap_new_sz);
10730 if (err) {
10731 pr_warn("map '%s': failed to resize memory-mapped region: %s\n",
10732 bpf_map__name(map), errstr(err));
10733 return libbpf_err(err);
10734 }
10735 err = map_btf_datasec_resize(map, size);
10736 if (err && err != -ENOENT) {
10737 pr_warn("map '%s': failed to adjust resized BTF, clearing BTF key/value info: %s\n",
10738 bpf_map__name(map), errstr(err));
10739 map->btf_value_type_id = 0;
10740 map->btf_key_type_id = 0;
10741 }
10742 }
10743
10744 map->def.value_size = size;
10745 return 0;
10746 }
10747
10748 __u32 bpf_map__btf_key_type_id(const struct bpf_map *map)
10749 {
10750 return map ? map->btf_key_type_id : 0;
10751 }
10752
10753 __u32 bpf_map__btf_value_type_id(const struct bpf_map *map)
10754 {
10755 return map ? map->btf_value_type_id : 0;
10756 }
10757
10758 int bpf_map__set_initial_value(struct bpf_map *map,
10759 const void *data, size_t size)
10760 {
10761 size_t actual_sz;
10762
10763 if (map_is_created(map))
10764 return libbpf_err(-EBUSY);
10765
10766 if (!map->mmaped || map->libbpf_type == LIBBPF_MAP_KCONFIG)
10767 return libbpf_err(-EINVAL);
10768
10769 if (map->def.type == BPF_MAP_TYPE_ARENA)
10770 actual_sz = map->obj->arena_data_sz;
10771 else
10772 actual_sz = map->def.value_size;
10773 if (size != actual_sz)
10774 return libbpf_err(-EINVAL);
10775
10776 memcpy(map->mmaped, data, size);
10777 return 0;
10778 }
10779
10780 void *bpf_map__initial_value(const struct bpf_map *map, size_t *psize)
10781 {
10782 if (bpf_map__is_struct_ops(map)) {
10783 if (psize)
10784 *psize = map->def.value_size;
10785 return map->st_ops->data;
10786 }
10787
10788 if (!map->mmaped)
10789 return NULL;
10790
10791 if (map->def.type == BPF_MAP_TYPE_ARENA)
10792 *psize = map->obj->arena_data_sz;
10793 else
10794 *psize = map->def.value_size;
10795
10796 return map->mmaped;
10797 }
10798
10799 bool bpf_map__is_internal(const struct bpf_map *map)
10800 {
10801 return map->libbpf_type != LIBBPF_MAP_UNSPEC;
10802 }
10803
10804 __u32 bpf_map__ifindex(const struct bpf_map *map)
10805 {
10806 return map->map_ifindex;
10807 }
10808
10809 int bpf_map__set_ifindex(struct bpf_map *map, __u32 ifindex)
10810 {
10811 if (map_is_created(map))
10812 return libbpf_err(-EBUSY);
10813 map->map_ifindex = ifindex;
10814 return 0;
10815 }
10816
10817 int bpf_map__set_inner_map_fd(struct bpf_map *map, int fd)
10818 {
10819 if (!bpf_map_type__is_map_in_map(map->def.type)) {
10820 pr_warn("error: unsupported map type\n");
10821 return libbpf_err(-EINVAL);
10822 }
10823 if (map->inner_map_fd != -1) {
10824 pr_warn("error: inner_map_fd already specified\n");
10825 return libbpf_err(-EINVAL);
10826 }
10827 if (map->inner_map) {
10828 bpf_map__destroy(map->inner_map);
10829 zfree(&map->inner_map);
10830 }
10831 map->inner_map_fd = fd;
10832 return 0;
10833 }
10834
10835 int bpf_map__set_exclusive_program(struct bpf_map *map, struct bpf_program *prog)
10836 {
10837 if (map_is_created(map)) {
10838 pr_warn("exclusive programs must be set before map creation\n");
10839 return libbpf_err(-EINVAL);
10840 }
10841
10842 if (map->obj != prog->obj) {
10843 pr_warn("excl_prog and map must be from the same bpf object\n");
10844 return libbpf_err(-EINVAL);
10845 }
10846
10847 map->excl_prog = prog;
10848 return 0;
10849 }
10850
10851 struct bpf_program *bpf_map__exclusive_program(struct bpf_map *map)
10852 {
10853 return map->excl_prog;
10854 }
10855
10856 static struct bpf_map *
10857 __bpf_map__iter(const struct bpf_map *m, const struct bpf_object *obj, int i)
10858 {
10859 ssize_t idx;
10860 struct bpf_map *s, *e;
10861
10862 if (!obj || !obj->maps)
10863 return errno = EINVAL, NULL;
10864
10865 s = obj->maps;
10866 e = obj->maps + obj->nr_maps;
10867
10868 if ((m < s) || (m >= e)) {
10869 pr_warn("error in %s: map handler doesn't belong to object\n",
10870 __func__);
10871 return errno = EINVAL, NULL;
10872 }
10873
10874 idx = (m - obj->maps) + i;
10875 if (idx >= obj->nr_maps || idx < 0)
10876 return NULL;
10877 return &obj->maps[idx];
10878 }
10879
10880 struct bpf_map *
10881 bpf_object__next_map(const struct bpf_object *obj, const struct bpf_map *prev)
10882 {
10883 if (prev == NULL && obj != NULL)
10884 return obj->maps;
10885
10886 return __bpf_map__iter(prev, obj, 1);
10887 }
10888
10889 struct bpf_map *
10890 bpf_object__prev_map(const struct bpf_object *obj, const struct bpf_map *next)
10891 {
10892 if (next == NULL && obj != NULL) {
10893 if (!obj->nr_maps)
10894 return NULL;
10895 return obj->maps + obj->nr_maps - 1;
10896 }
10897
10898 return __bpf_map__iter(next, obj, -1);
10899 }
10900
10901 struct bpf_map *
10902 bpf_object__find_map_by_name(const struct bpf_object *obj, const char *name)
10903 {
10904 struct bpf_map *pos;
10905
10906 bpf_object__for_each_map(pos, obj) {
10907 /* if it's a special internal map name (which always starts
10908 * with dot) then check if that special name matches the
10909 * real map name (ELF section name)
10910 */
10911 if (name[0] == '.') {
10912 if (pos->real_name && strcmp(pos->real_name, name) == 0)
10913 return pos;
10914 continue;
10915 }
10916 /* otherwise map name has to be an exact match */
10917 if (map_uses_real_name(pos)) {
10918 if (strcmp(pos->real_name, name) == 0)
10919 return pos;
10920 continue;
10921 }
10922 if (strcmp(pos->name, name) == 0)
10923 return pos;
10924 }
10925 return errno = ENOENT, NULL;
10926 }
10927
10928 int
10929 bpf_object__find_map_fd_by_name(const struct bpf_object *obj, const char *name)
10930 {
10931 return bpf_map__fd(bpf_object__find_map_by_name(obj, name));
10932 }
10933
10934 static int validate_map_op(const struct bpf_map *map, size_t key_sz,
10935 size_t value_sz, bool check_value_sz, __u64 flags)
10936 {
10937 if (!map_is_created(map)) /* map is not yet created */
10938 return -ENOENT;
10939
10940 if (map->def.key_size != key_sz) {
10941 pr_warn("map '%s': unexpected key size %zu provided, expected %u\n",
10942 map->name, key_sz, map->def.key_size);
10943 return -EINVAL;
10944 }
10945
10946 if (map->fd < 0) {
10947 pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name);
10948 return -EINVAL;
10949 }
10950
10951 if (!check_value_sz)
10952 return 0;
10953
10954 switch (map->def.type) {
10955 case BPF_MAP_TYPE_PERCPU_ARRAY:
10956 case BPF_MAP_TYPE_PERCPU_HASH:
10957 case BPF_MAP_TYPE_LRU_PERCPU_HASH:
10958 case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: {
10959 int num_cpu = libbpf_num_possible_cpus();
10960 size_t elem_sz = roundup(map->def.value_size, 8);
10961
10962 if (flags & (BPF_F_CPU | BPF_F_ALL_CPUS)) {
10963 if ((flags & BPF_F_CPU) && (flags & BPF_F_ALL_CPUS)) {
10964 pr_warn("map '%s': BPF_F_CPU and BPF_F_ALL_CPUS are mutually exclusive\n",
10965 map->name);
10966 return -EINVAL;
10967 }
10968 if (map->def.value_size != value_sz) {
10969 pr_warn("map '%s': unexpected value size %zu provided for either BPF_F_CPU or BPF_F_ALL_CPUS, expected %u\n",
10970 map->name, value_sz, map->def.value_size);
10971 return -EINVAL;
10972 }
10973 break;
10974 }
10975
10976 if (value_sz != num_cpu * elem_sz) {
10977 pr_warn("map '%s': unexpected value size %zu provided for per-CPU map, expected %d * %zu = %zd\n",
10978 map->name, value_sz, num_cpu, elem_sz, num_cpu * elem_sz);
10979 return -EINVAL;
10980 }
10981 break;
10982 }
10983 default:
10984 if (map->def.value_size != value_sz) {
10985 pr_warn("map '%s': unexpected value size %zu provided, expected %u\n",
10986 map->name, value_sz, map->def.value_size);
10987 return -EINVAL;
10988 }
10989 break;
10990 }
10991 return 0;
10992 }
10993
10994 int bpf_map__lookup_elem(const struct bpf_map *map,
10995 const void *key, size_t key_sz,
10996 void *value, size_t value_sz, __u64 flags)
10997 {
10998 int err;
10999
11000 err = validate_map_op(map, key_sz, value_sz, true, flags);
11001 if (err)
11002 return libbpf_err(err);
11003
11004 return bpf_map_lookup_elem_flags(map->fd, key, value, flags);
11005 }
11006
11007 int bpf_map__update_elem(const struct bpf_map *map,
11008 const void *key, size_t key_sz,
11009 const void *value, size_t value_sz, __u64 flags)
11010 {
11011 int err;
11012
11013 err = validate_map_op(map, key_sz, value_sz, true, flags);
11014 if (err)
11015 return libbpf_err(err);
11016
11017 return bpf_map_update_elem(map->fd, key, value, flags);
11018 }
11019
11020 int bpf_map__delete_elem(const struct bpf_map *map,
11021 const void *key, size_t key_sz, __u64 flags)
11022 {
11023 int err;
11024
11025 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */, flags);
11026 if (err)
11027 return libbpf_err(err);
11028
11029 return bpf_map_delete_elem_flags(map->fd, key, flags);
11030 }
11031
11032 int bpf_map__lookup_and_delete_elem(const struct bpf_map *map,
11033 const void *key, size_t key_sz,
11034 void *value, size_t value_sz, __u64 flags)
11035 {
11036 int err;
11037
11038 err = validate_map_op(map, key_sz, value_sz, true, flags);
11039 if (err)
11040 return libbpf_err(err);
11041
11042 return bpf_map_lookup_and_delete_elem_flags(map->fd, key, value, flags);
11043 }
11044
11045 int bpf_map__get_next_key(const struct bpf_map *map,
11046 const void *cur_key, void *next_key, size_t key_sz)
11047 {
11048 int err;
11049
11050 err = validate_map_op(map, key_sz, 0, false /* check_value_sz */, 0);
11051 if (err)
11052 return libbpf_err(err);
11053
11054 return bpf_map_get_next_key(map->fd, cur_key, next_key);
11055 }
11056
11057 long libbpf_get_error(const void *ptr)
11058 {
11059 if (!IS_ERR_OR_NULL(ptr))
11060 return 0;
11061
11062 if (IS_ERR(ptr))
11063 errno = -PTR_ERR(ptr);
11064
11065 /* If ptr == NULL, then errno should be already set by the failing
11066 * API, because libbpf never returns NULL on success and it now always
11067 * sets errno on error. So no extra errno handling for ptr == NULL
11068 * case.
11069 */
11070 return -errno;
11071 }
11072
11073 /* Replace link's underlying BPF program with the new one */
11074 int bpf_link__update_program(struct bpf_link *link, struct bpf_program *prog)
11075 {
11076 int ret;
11077 int prog_fd = bpf_program__fd(prog);
11078
11079 if (prog_fd < 0) {
11080 pr_warn("prog '%s': can't use BPF program without FD (was it loaded?)\n",
11081 prog->name);
11082 return libbpf_err(-EINVAL);
11083 }
11084
11085 ret = bpf_link_update(bpf_link__fd(link), prog_fd, NULL);
11086 return libbpf_err_errno(ret);
11087 }
11088
11089 /* Release "ownership" of underlying BPF resource (typically, BPF program
11090 * attached to some BPF hook, e.g., tracepoint, kprobe, etc). Disconnected
11091 * link, when destructed through bpf_link__destroy() call won't attempt to
11092 * detach/unregisted that BPF resource. This is useful in situations where,
11093 * say, attached BPF program has to outlive userspace program that attached it
11094 * in the system. Depending on type of BPF program, though, there might be
11095 * additional steps (like pinning BPF program in BPF FS) necessary to ensure
11096 * exit of userspace program doesn't trigger automatic detachment and clean up
11097 * inside the kernel.
11098 */
11099 void bpf_link__disconnect(struct bpf_link *link)
11100 {
11101 link->disconnected = true;
11102 }
11103
11104 int bpf_link__destroy(struct bpf_link *link)
11105 {
11106 int err = 0;
11107
11108 if (IS_ERR_OR_NULL(link))
11109 return 0;
11110
11111 if (!link->disconnected && link->detach)
11112 err = link->detach(link);
11113 if (link->pin_path)
11114 free(link->pin_path);
11115 if (link->dealloc)
11116 link->dealloc(link);
11117 else
11118 free(link);
11119
11120 return libbpf_err(err);
11121 }
11122
11123 int bpf_link__fd(const struct bpf_link *link)
11124 {
11125 return link->fd;
11126 }
11127
11128 const char *bpf_link__pin_path(const struct bpf_link *link)
11129 {
11130 return link->pin_path;
11131 }
11132
11133 static int bpf_link__detach_fd(struct bpf_link *link)
11134 {
11135 return libbpf_err_errno(close(link->fd));
11136 }
11137
11138 struct bpf_link *bpf_link__open(const char *path)
11139 {
11140 struct bpf_link *link;
11141 int fd;
11142
11143 fd = bpf_obj_get(path);
11144 if (fd < 0) {
11145 fd = -errno;
11146 pr_warn("failed to open link at %s: %d\n", path, fd);
11147 return libbpf_err_ptr(fd);
11148 }
11149
11150 link = calloc(1, sizeof(*link));
11151 if (!link) {
11152 close(fd);
11153 return libbpf_err_ptr(-ENOMEM);
11154 }
11155 link->detach = &bpf_link__detach_fd;
11156 link->fd = fd;
11157
11158 link->pin_path = strdup(path);
11159 if (!link->pin_path) {
11160 bpf_link__destroy(link);
11161 return libbpf_err_ptr(-ENOMEM);
11162 }
11163
11164 return link;
11165 }
11166
11167 int bpf_link__detach(struct bpf_link *link)
11168 {
11169 return bpf_link_detach(link->fd) ? -errno : 0;
11170 }
11171
11172 int bpf_link__pin(struct bpf_link *link, const char *path)
11173 {
11174 int err;
11175
11176 if (link->pin_path)
11177 return libbpf_err(-EBUSY);
11178 err = make_parent_dir(path);
11179 if (err)
11180 return libbpf_err(err);
11181 err = check_path(path);
11182 if (err)
11183 return libbpf_err(err);
11184
11185 link->pin_path = strdup(path);
11186 if (!link->pin_path)
11187 return libbpf_err(-ENOMEM);
11188
11189 if (bpf_obj_pin(link->fd, link->pin_path)) {
11190 err = -errno;
11191 zfree(&link->pin_path);
11192 return libbpf_err(err);
11193 }
11194
11195 pr_debug("link fd=%d: pinned at %s\n", link->fd, link->pin_path);
11196 return 0;
11197 }
11198
11199 int bpf_link__unpin(struct bpf_link *link)
11200 {
11201 int err;
11202
11203 if (!link->pin_path)
11204 return libbpf_err(-EINVAL);
11205
11206 err = unlink(link->pin_path);
11207 if (err != 0)
11208 return -errno;
11209
11210 pr_debug("link fd=%d: unpinned from %s\n", link->fd, link->pin_path);
11211 zfree(&link->pin_path);
11212 return 0;
11213 }
11214
11215 struct bpf_link_perf {
11216 struct bpf_link link;
11217 int perf_event_fd;
11218 /* legacy kprobe support: keep track of probe identifier and type */
11219 char *legacy_probe_name;
11220 bool legacy_is_kprobe;
11221 bool legacy_is_retprobe;
11222 };
11223
11224 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe);
11225 static int remove_uprobe_event_legacy(const char *probe_name, bool retprobe);
11226
11227 static int bpf_link_perf_detach(struct bpf_link *link)
11228 {
11229 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
11230 int err = 0;
11231
11232 if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0)
11233 err = -errno;
11234
11235 if (perf_link->perf_event_fd != link->fd)
11236 close(perf_link->perf_event_fd);
11237 close(link->fd);
11238
11239 /* legacy uprobe/kprobe needs to be removed after perf event fd closure */
11240 if (perf_link->legacy_probe_name) {
11241 if (perf_link->legacy_is_kprobe) {
11242 err = remove_kprobe_event_legacy(perf_link->legacy_probe_name,
11243 perf_link->legacy_is_retprobe);
11244 } else {
11245 err = remove_uprobe_event_legacy(perf_link->legacy_probe_name,
11246 perf_link->legacy_is_retprobe);
11247 }
11248 }
11249
11250 return err;
11251 }
11252
11253 static void bpf_link_perf_dealloc(struct bpf_link *link)
11254 {
11255 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
11256
11257 free(perf_link->legacy_probe_name);
11258 free(perf_link);
11259 }
11260
11261 struct bpf_link *bpf_program__attach_perf_event_opts(const struct bpf_program *prog, int pfd,
11262 const struct bpf_perf_event_opts *opts)
11263 {
11264 struct bpf_link_perf *link;
11265 int prog_fd, link_fd = -1, err;
11266 bool force_ioctl_attach;
11267
11268 if (!OPTS_VALID(opts, bpf_perf_event_opts))
11269 return libbpf_err_ptr(-EINVAL);
11270
11271 if (pfd < 0) {
11272 pr_warn("prog '%s': invalid perf event FD %d\n",
11273 prog->name, pfd);
11274 return libbpf_err_ptr(-EINVAL);
11275 }
11276 prog_fd = bpf_program__fd(prog);
11277 if (prog_fd < 0) {
11278 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
11279 prog->name);
11280 return libbpf_err_ptr(-EINVAL);
11281 }
11282
11283 link = calloc(1, sizeof(*link));
11284 if (!link)
11285 return libbpf_err_ptr(-ENOMEM);
11286 link->link.detach = &bpf_link_perf_detach;
11287 link->link.dealloc = &bpf_link_perf_dealloc;
11288 link->perf_event_fd = pfd;
11289
11290 force_ioctl_attach = OPTS_GET(opts, force_ioctl_attach, false);
11291 if (kernel_supports(prog->obj, FEAT_PERF_LINK) && !force_ioctl_attach) {
11292 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_opts,
11293 .perf_event.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0));
11294
11295 link_fd = bpf_link_create(prog_fd, pfd, BPF_PERF_EVENT, &link_opts);
11296 if (link_fd < 0) {
11297 err = -errno;
11298 pr_warn("prog '%s': failed to create BPF link for perf_event FD %d: %s\n",
11299 prog->name, pfd, errstr(err));
11300 goto err_out;
11301 }
11302 link->link.fd = link_fd;
11303 } else {
11304 if (OPTS_GET(opts, bpf_cookie, 0)) {
11305 pr_warn("prog '%s': user context value is not supported\n", prog->name);
11306 err = -EOPNOTSUPP;
11307 goto err_out;
11308 }
11309
11310 if (ioctl(pfd, PERF_EVENT_IOC_SET_BPF, prog_fd) < 0) {
11311 err = -errno;
11312 pr_warn("prog '%s': failed to attach to perf_event FD %d: %s\n",
11313 prog->name, pfd, errstr(err));
11314 if (err == -EPROTO)
11315 pr_warn("prog '%s': try add PERF_SAMPLE_CALLCHAIN to or remove exclude_callchain_[kernel|user] from pfd %d\n",
11316 prog->name, pfd);
11317 goto err_out;
11318 }
11319 link->link.fd = pfd;
11320 }
11321
11322 if (!OPTS_GET(opts, dont_enable, false)) {
11323 if (ioctl(pfd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
11324 err = -errno;
11325 pr_warn("prog '%s': failed to enable perf_event FD %d: %s\n",
11326 prog->name, pfd, errstr(err));
11327 goto err_out;
11328 }
11329 }
11330
11331 return &link->link;
11332 err_out:
11333 if (link_fd >= 0)
11334 close(link_fd);
11335 free(link);
11336 return libbpf_err_ptr(err);
11337 }
11338
11339 struct bpf_link *bpf_program__attach_perf_event(const struct bpf_program *prog, int pfd)
11340 {
11341 return bpf_program__attach_perf_event_opts(prog, pfd, NULL);
11342 }
11343
11344 /*
11345 * this function is expected to parse integer in the range of [0, 2^31-1] from
11346 * given file using scanf format string fmt. If actual parsed value is
11347 * negative, the result might be indistinguishable from error
11348 */
11349 static int parse_uint_from_file(const char *file, const char *fmt)
11350 {
11351 int err, ret;
11352 FILE *f;
11353
11354 f = fopen(file, "re");
11355 if (!f) {
11356 err = -errno;
11357 pr_debug("failed to open '%s': %s\n", file, errstr(err));
11358 return err;
11359 }
11360 err = fscanf(f, fmt, &ret);
11361 if (err != 1) {
11362 err = err == EOF ? -EIO : -errno;
11363 pr_debug("failed to parse '%s': %s\n", file, errstr(err));
11364 fclose(f);
11365 return err;
11366 }
11367 fclose(f);
11368 return ret;
11369 }
11370
11371 static int determine_kprobe_perf_type(void)
11372 {
11373 const char *file = "/sys/bus/event_source/devices/kprobe/type";
11374
11375 return parse_uint_from_file(file, "%d\n");
11376 }
11377
11378 static int determine_uprobe_perf_type(void)
11379 {
11380 const char *file = "/sys/bus/event_source/devices/uprobe/type";
11381
11382 return parse_uint_from_file(file, "%d\n");
11383 }
11384
11385 static int determine_kprobe_retprobe_bit(void)
11386 {
11387 const char *file = "/sys/bus/event_source/devices/kprobe/format/retprobe";
11388
11389 return parse_uint_from_file(file, "config:%d\n");
11390 }
11391
11392 static int determine_uprobe_retprobe_bit(void)
11393 {
11394 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe";
11395
11396 return parse_uint_from_file(file, "config:%d\n");
11397 }
11398
11399 #define PERF_UPROBE_REF_CTR_OFFSET_BITS 32
11400 #define PERF_UPROBE_REF_CTR_OFFSET_SHIFT 32
11401
11402 static int perf_event_open_probe(bool uprobe, bool retprobe, const char *name,
11403 uint64_t offset, int pid, size_t ref_ctr_off)
11404 {
11405 const size_t attr_sz = sizeof(struct perf_event_attr);
11406 struct perf_event_attr attr;
11407 int type, pfd;
11408
11409 if ((__u64)ref_ctr_off >= (1ULL << PERF_UPROBE_REF_CTR_OFFSET_BITS))
11410 return -EINVAL;
11411
11412 memset(&attr, 0, attr_sz);
11413
11414 type = uprobe ? determine_uprobe_perf_type()
11415 : determine_kprobe_perf_type();
11416 if (type < 0) {
11417 pr_warn("failed to determine %s perf type: %s\n",
11418 uprobe ? "uprobe" : "kprobe",
11419 errstr(type));
11420 return type;
11421 }
11422 if (retprobe) {
11423 int bit = uprobe ? determine_uprobe_retprobe_bit()
11424 : determine_kprobe_retprobe_bit();
11425
11426 if (bit < 0) {
11427 pr_warn("failed to determine %s retprobe bit: %s\n",
11428 uprobe ? "uprobe" : "kprobe",
11429 errstr(bit));
11430 return bit;
11431 }
11432 attr.config |= 1 << bit;
11433 }
11434 attr.size = attr_sz;
11435 attr.type = type;
11436 attr.config |= (__u64)ref_ctr_off << PERF_UPROBE_REF_CTR_OFFSET_SHIFT;
11437 attr.config1 = ptr_to_u64(name); /* kprobe_func or uprobe_path */
11438 attr.config2 = offset; /* kprobe_addr or probe_offset */
11439
11440 /* pid filter is meaningful only for uprobes */
11441 pfd = syscall(__NR_perf_event_open, &attr,
11442 pid < 0 ? -1 : pid /* pid */,
11443 pid == -1 ? 0 : -1 /* cpu */,
11444 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
11445 return pfd >= 0 ? pfd : -errno;
11446 }
11447
11448 static int append_to_file(const char *file, const char *fmt, ...)
11449 {
11450 int fd, n, err = 0;
11451 va_list ap;
11452 char buf[1024];
11453
11454 va_start(ap, fmt);
11455 n = vsnprintf(buf, sizeof(buf), fmt, ap);
11456 va_end(ap);
11457
11458 if (n < 0 || n >= sizeof(buf))
11459 return -EINVAL;
11460
11461 fd = open(file, O_WRONLY | O_APPEND | O_CLOEXEC, 0);
11462 if (fd < 0)
11463 return -errno;
11464
11465 if (write(fd, buf, n) < 0)
11466 err = -errno;
11467
11468 close(fd);
11469 return err;
11470 }
11471
11472 #define DEBUGFS "/sys/kernel/debug/tracing"
11473 #define TRACEFS "/sys/kernel/tracing"
11474
11475 static bool use_debugfs(void)
11476 {
11477 static int has_debugfs = -1;
11478
11479 if (has_debugfs < 0)
11480 has_debugfs = faccessat(AT_FDCWD, DEBUGFS, F_OK, AT_EACCESS) == 0;
11481
11482 return has_debugfs == 1;
11483 }
11484
11485 static const char *tracefs_path(void)
11486 {
11487 return use_debugfs() ? DEBUGFS : TRACEFS;
11488 }
11489
11490 static const char *tracefs_kprobe_events(void)
11491 {
11492 return use_debugfs() ? DEBUGFS"/kprobe_events" : TRACEFS"/kprobe_events";
11493 }
11494
11495 static const char *tracefs_uprobe_events(void)
11496 {
11497 return use_debugfs() ? DEBUGFS"/uprobe_events" : TRACEFS"/uprobe_events";
11498 }
11499
11500 static const char *tracefs_available_filter_functions(void)
11501 {
11502 return use_debugfs() ? DEBUGFS"/available_filter_functions"
11503 : TRACEFS"/available_filter_functions";
11504 }
11505
11506 static const char *tracefs_available_filter_functions_addrs(void)
11507 {
11508 return use_debugfs() ? DEBUGFS"/available_filter_functions_addrs"
11509 : TRACEFS"/available_filter_functions_addrs";
11510 }
11511
11512 static void gen_probe_legacy_event_name(char *buf, size_t buf_sz,
11513 const char *name, size_t offset)
11514 {
11515 static int index = 0;
11516 int i;
11517
11518 snprintf(buf, buf_sz, "libbpf_%u_%d_%s_0x%zx", getpid(),
11519 __sync_fetch_and_add(&index, 1), name, offset);
11520
11521 /* sanitize name in the probe name */
11522 for (i = 0; buf[i]; i++) {
11523 if (!isalnum(buf[i]))
11524 buf[i] = '_';
11525 }
11526 }
11527
11528 static int add_kprobe_event_legacy(const char *probe_name, bool retprobe,
11529 const char *kfunc_name, size_t offset)
11530 {
11531 return append_to_file(tracefs_kprobe_events(), "%c:%s/%s %s+0x%zx",
11532 retprobe ? 'r' : 'p',
11533 retprobe ? "kretprobes" : "kprobes",
11534 probe_name, kfunc_name, offset);
11535 }
11536
11537 static int remove_kprobe_event_legacy(const char *probe_name, bool retprobe)
11538 {
11539 return append_to_file(tracefs_kprobe_events(), "-:%s/%s",
11540 retprobe ? "kretprobes" : "kprobes", probe_name);
11541 }
11542
11543 static int determine_kprobe_perf_type_legacy(const char *probe_name, bool retprobe)
11544 {
11545 char file[256];
11546
11547 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
11548 tracefs_path(), retprobe ? "kretprobes" : "kprobes", probe_name);
11549
11550 return parse_uint_from_file(file, "%d\n");
11551 }
11552
11553 static int perf_event_kprobe_open_legacy(const char *probe_name, bool retprobe,
11554 const char *kfunc_name, size_t offset, int pid)
11555 {
11556 const size_t attr_sz = sizeof(struct perf_event_attr);
11557 struct perf_event_attr attr;
11558 int type, pfd, err;
11559
11560 err = add_kprobe_event_legacy(probe_name, retprobe, kfunc_name, offset);
11561 if (err < 0) {
11562 pr_warn("failed to add legacy kprobe event for '%s+0x%zx': %s\n",
11563 kfunc_name, offset,
11564 errstr(err));
11565 return err;
11566 }
11567 type = determine_kprobe_perf_type_legacy(probe_name, retprobe);
11568 if (type < 0) {
11569 err = type;
11570 pr_warn("failed to determine legacy kprobe event id for '%s+0x%zx': %s\n",
11571 kfunc_name, offset,
11572 errstr(err));
11573 goto err_clean_legacy;
11574 }
11575
11576 memset(&attr, 0, attr_sz);
11577 attr.size = attr_sz;
11578 attr.config = type;
11579 attr.type = PERF_TYPE_TRACEPOINT;
11580
11581 pfd = syscall(__NR_perf_event_open, &attr,
11582 pid < 0 ? -1 : pid, /* pid */
11583 pid == -1 ? 0 : -1, /* cpu */
11584 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
11585 if (pfd < 0) {
11586 err = -errno;
11587 pr_warn("legacy kprobe perf_event_open() failed: %s\n",
11588 errstr(err));
11589 goto err_clean_legacy;
11590 }
11591 return pfd;
11592
11593 err_clean_legacy:
11594 /* Clear the newly added legacy kprobe_event */
11595 remove_kprobe_event_legacy(probe_name, retprobe);
11596 return err;
11597 }
11598
11599 static const char *arch_specific_syscall_pfx(void)
11600 {
11601 #if defined(__x86_64__)
11602 return "x64";
11603 #elif defined(__i386__)
11604 return "ia32";
11605 #elif defined(__s390x__)
11606 return "s390x";
11607 #elif defined(__arm__)
11608 return "arm";
11609 #elif defined(__aarch64__)
11610 return "arm64";
11611 #elif defined(__mips__)
11612 return "mips";
11613 #elif defined(__riscv)
11614 return "riscv";
11615 #elif defined(__powerpc__)
11616 return "powerpc";
11617 #elif defined(__powerpc64__)
11618 return "powerpc64";
11619 #else
11620 return NULL;
11621 #endif
11622 }
11623
11624 int probe_kern_syscall_wrapper(int token_fd)
11625 {
11626 char syscall_name[64];
11627 const char *ksys_pfx;
11628
11629 ksys_pfx = arch_specific_syscall_pfx();
11630 if (!ksys_pfx)
11631 return 0;
11632
11633 snprintf(syscall_name, sizeof(syscall_name), "__%s_sys_bpf", ksys_pfx);
11634
11635 if (determine_kprobe_perf_type() >= 0) {
11636 int pfd;
11637
11638 pfd = perf_event_open_probe(false, false, syscall_name, 0, getpid(), 0);
11639 if (pfd >= 0)
11640 close(pfd);
11641
11642 return pfd >= 0 ? 1 : 0;
11643 } else { /* legacy mode */
11644 char probe_name[MAX_EVENT_NAME_LEN];
11645
11646 gen_probe_legacy_event_name(probe_name, sizeof(probe_name), syscall_name, 0);
11647 if (add_kprobe_event_legacy(probe_name, false, syscall_name, 0) < 0)
11648 return 0;
11649
11650 (void)remove_kprobe_event_legacy(probe_name, false);
11651 return 1;
11652 }
11653 }
11654
11655 struct bpf_link *
11656 bpf_program__attach_kprobe_opts(const struct bpf_program *prog,
11657 const char *func_name,
11658 const struct bpf_kprobe_opts *opts)
11659 {
11660 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
11661 enum probe_attach_mode attach_mode;
11662 char *legacy_probe = NULL;
11663 struct bpf_link *link;
11664 size_t offset;
11665 bool retprobe, legacy;
11666 int pfd, err;
11667
11668 if (!OPTS_VALID(opts, bpf_kprobe_opts))
11669 return libbpf_err_ptr(-EINVAL);
11670
11671 attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
11672 retprobe = OPTS_GET(opts, retprobe, false);
11673 offset = OPTS_GET(opts, offset, 0);
11674 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11675
11676 legacy = determine_kprobe_perf_type() < 0;
11677 switch (attach_mode) {
11678 case PROBE_ATTACH_MODE_LEGACY:
11679 legacy = true;
11680 pe_opts.force_ioctl_attach = true;
11681 break;
11682 case PROBE_ATTACH_MODE_PERF:
11683 if (legacy)
11684 return libbpf_err_ptr(-ENOTSUP);
11685 pe_opts.force_ioctl_attach = true;
11686 break;
11687 case PROBE_ATTACH_MODE_LINK:
11688 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
11689 return libbpf_err_ptr(-ENOTSUP);
11690 break;
11691 case PROBE_ATTACH_MODE_DEFAULT:
11692 break;
11693 default:
11694 return libbpf_err_ptr(-EINVAL);
11695 }
11696
11697 if (!legacy) {
11698 pfd = perf_event_open_probe(false /* uprobe */, retprobe,
11699 func_name, offset,
11700 -1 /* pid */, 0 /* ref_ctr_off */);
11701 } else {
11702 char probe_name[MAX_EVENT_NAME_LEN];
11703
11704 gen_probe_legacy_event_name(probe_name, sizeof(probe_name),
11705 func_name, offset);
11706
11707 legacy_probe = strdup(probe_name);
11708 if (!legacy_probe)
11709 return libbpf_err_ptr(-ENOMEM);
11710
11711 pfd = perf_event_kprobe_open_legacy(legacy_probe, retprobe, func_name,
11712 offset, -1 /* pid */);
11713 }
11714 if (pfd < 0) {
11715 err = -errno;
11716 pr_warn("prog '%s': failed to create %s '%s+0x%zx' perf event: %s\n",
11717 prog->name, retprobe ? "kretprobe" : "kprobe",
11718 func_name, offset,
11719 errstr(err));
11720 goto err_out;
11721 }
11722 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
11723 err = libbpf_get_error(link);
11724 if (err) {
11725 close(pfd);
11726 pr_warn("prog '%s': failed to attach to %s '%s+0x%zx': %s\n",
11727 prog->name, retprobe ? "kretprobe" : "kprobe",
11728 func_name, offset,
11729 errstr(err));
11730 goto err_clean_legacy;
11731 }
11732 if (legacy) {
11733 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
11734
11735 perf_link->legacy_probe_name = legacy_probe;
11736 perf_link->legacy_is_kprobe = true;
11737 perf_link->legacy_is_retprobe = retprobe;
11738 }
11739
11740 return link;
11741
11742 err_clean_legacy:
11743 if (legacy)
11744 remove_kprobe_event_legacy(legacy_probe, retprobe);
11745 err_out:
11746 free(legacy_probe);
11747 return libbpf_err_ptr(err);
11748 }
11749
11750 struct bpf_link *bpf_program__attach_kprobe(const struct bpf_program *prog,
11751 bool retprobe,
11752 const char *func_name)
11753 {
11754 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts,
11755 .retprobe = retprobe,
11756 );
11757
11758 return bpf_program__attach_kprobe_opts(prog, func_name, &opts);
11759 }
11760
11761 struct bpf_link *bpf_program__attach_ksyscall(const struct bpf_program *prog,
11762 const char *syscall_name,
11763 const struct bpf_ksyscall_opts *opts)
11764 {
11765 LIBBPF_OPTS(bpf_kprobe_opts, kprobe_opts);
11766 char func_name[128];
11767
11768 if (!OPTS_VALID(opts, bpf_ksyscall_opts))
11769 return libbpf_err_ptr(-EINVAL);
11770
11771 if (kernel_supports(prog->obj, FEAT_SYSCALL_WRAPPER)) {
11772 /* arch_specific_syscall_pfx() should never return NULL here
11773 * because it is guarded by kernel_supports(). However, since
11774 * compiler does not know that we have an explicit conditional
11775 * as well.
11776 */
11777 snprintf(func_name, sizeof(func_name), "__%s_sys_%s",
11778 arch_specific_syscall_pfx() ? : "", syscall_name);
11779 } else {
11780 snprintf(func_name, sizeof(func_name), "__se_sys_%s", syscall_name);
11781 }
11782
11783 kprobe_opts.retprobe = OPTS_GET(opts, retprobe, false);
11784 kprobe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
11785
11786 return bpf_program__attach_kprobe_opts(prog, func_name, &kprobe_opts);
11787 }
11788
11789 /* Adapted from perf/util/string.c */
11790 bool glob_match(const char *str, const char *pat)
11791 {
11792 while (*str && *pat && *pat != '*') {
11793 if (*pat == '?') { /* Matches any single character */
11794 str++;
11795 pat++;
11796 continue;
11797 }
11798 if (*str != *pat)
11799 return false;
11800 str++;
11801 pat++;
11802 }
11803 /* Check wild card */
11804 if (*pat == '*') {
11805 while (*pat == '*')
11806 pat++;
11807 if (!*pat) /* Tail wild card matches all */
11808 return true;
11809 while (*str)
11810 if (glob_match(str++, pat))
11811 return true;
11812 }
11813 return !*str && !*pat;
11814 }
11815
11816 struct kprobe_multi_resolve {
11817 const char *pattern;
11818 unsigned long *addrs;
11819 size_t cap;
11820 size_t cnt;
11821 };
11822
11823 struct avail_kallsyms_data {
11824 char **syms;
11825 size_t cnt;
11826 struct kprobe_multi_resolve *res;
11827 };
11828
11829 static int avail_func_cmp(const void *a, const void *b)
11830 {
11831 return strcmp(*(const char **)a, *(const char **)b);
11832 }
11833
11834 static int avail_kallsyms_cb(unsigned long long sym_addr, char sym_type,
11835 const char *sym_name, void *ctx)
11836 {
11837 struct avail_kallsyms_data *data = ctx;
11838 struct kprobe_multi_resolve *res = data->res;
11839 int err;
11840
11841 if (!glob_match(sym_name, res->pattern))
11842 return 0;
11843
11844 if (!bsearch(&sym_name, data->syms, data->cnt, sizeof(*data->syms), avail_func_cmp)) {
11845 /* Some versions of kernel strip out .llvm.<hash> suffix from
11846 * function names reported in available_filter_functions, but
11847 * don't do so for kallsyms. While this is clearly a kernel
11848 * bug (fixed by [0]) we try to accommodate that in libbpf to
11849 * make multi-kprobe usability a bit better: if no match is
11850 * found, we will strip .llvm. suffix and try one more time.
11851 *
11852 * [0] fb6a421fb615 ("kallsyms: Match symbols exactly with CONFIG_LTO_CLANG")
11853 */
11854 char sym_trim[256], *psym_trim = sym_trim;
11855 const char *sym_sfx;
11856
11857 if (!(sym_sfx = strstr(sym_name, ".llvm.")))
11858 return 0;
11859
11860 /* psym_trim vs sym_trim dance is done to avoid pointer vs array
11861 * coercion differences and get proper `const char **` pointer
11862 * which avail_func_cmp() expects
11863 */
11864 snprintf(sym_trim, sizeof(sym_trim), "%.*s", (int)(sym_sfx - sym_name), sym_name);
11865 if (!bsearch(&psym_trim, data->syms, data->cnt, sizeof(*data->syms), avail_func_cmp))
11866 return 0;
11867 }
11868
11869 err = libbpf_ensure_mem((void **)&res->addrs, &res->cap, sizeof(*res->addrs), res->cnt + 1);
11870 if (err)
11871 return err;
11872
11873 res->addrs[res->cnt++] = (unsigned long)sym_addr;
11874 return 0;
11875 }
11876
11877 static int libbpf_available_kallsyms_parse(struct kprobe_multi_resolve *res)
11878 {
11879 const char *available_functions_file = tracefs_available_filter_functions();
11880 struct avail_kallsyms_data data;
11881 char sym_name[500];
11882 FILE *f;
11883 int err = 0, ret, i;
11884 char **syms = NULL;
11885 size_t cap = 0, cnt = 0;
11886
11887 f = fopen(available_functions_file, "re");
11888 if (!f) {
11889 err = -errno;
11890 pr_warn("failed to open %s: %s\n", available_functions_file, errstr(err));
11891 return err;
11892 }
11893
11894 while (true) {
11895 char *name;
11896
11897 ret = fscanf(f, "%499s%*[^\n]\n", sym_name);
11898 if (ret == EOF && feof(f))
11899 break;
11900
11901 if (ret != 1) {
11902 pr_warn("failed to parse available_filter_functions entry: %d\n", ret);
11903 err = -EINVAL;
11904 goto cleanup;
11905 }
11906
11907 if (!glob_match(sym_name, res->pattern))
11908 continue;
11909
11910 err = libbpf_ensure_mem((void **)&syms, &cap, sizeof(*syms), cnt + 1);
11911 if (err)
11912 goto cleanup;
11913
11914 name = strdup(sym_name);
11915 if (!name) {
11916 err = -errno;
11917 goto cleanup;
11918 }
11919
11920 syms[cnt++] = name;
11921 }
11922
11923 /* no entries found, bail out */
11924 if (cnt == 0) {
11925 err = -ENOENT;
11926 goto cleanup;
11927 }
11928
11929 /* sort available functions */
11930 qsort(syms, cnt, sizeof(*syms), avail_func_cmp);
11931
11932 data.syms = syms;
11933 data.res = res;
11934 data.cnt = cnt;
11935 libbpf_kallsyms_parse(avail_kallsyms_cb, &data);
11936
11937 if (res->cnt == 0)
11938 err = -ENOENT;
11939
11940 cleanup:
11941 for (i = 0; i < cnt; i++)
11942 free((char *)syms[i]);
11943 free(syms);
11944
11945 fclose(f);
11946 return err;
11947 }
11948
11949 static bool has_available_filter_functions_addrs(void)
11950 {
11951 return access(tracefs_available_filter_functions_addrs(), R_OK) != -1;
11952 }
11953
11954 static int libbpf_available_kprobes_parse(struct kprobe_multi_resolve *res)
11955 {
11956 const char *available_path = tracefs_available_filter_functions_addrs();
11957 char sym_name[500];
11958 FILE *f;
11959 int ret, err = 0;
11960 unsigned long long sym_addr;
11961
11962 f = fopen(available_path, "re");
11963 if (!f) {
11964 err = -errno;
11965 pr_warn("failed to open %s: %s\n", available_path, errstr(err));
11966 return err;
11967 }
11968
11969 while (true) {
11970 ret = fscanf(f, "%llx %499s%*[^\n]\n", &sym_addr, sym_name);
11971 if (ret == EOF && feof(f))
11972 break;
11973
11974 if (ret != 2) {
11975 pr_warn("failed to parse available_filter_functions_addrs entry: %d\n",
11976 ret);
11977 err = -EINVAL;
11978 goto cleanup;
11979 }
11980
11981 if (!glob_match(sym_name, res->pattern))
11982 continue;
11983
11984 err = libbpf_ensure_mem((void **)&res->addrs, &res->cap,
11985 sizeof(*res->addrs), res->cnt + 1);
11986 if (err)
11987 goto cleanup;
11988
11989 res->addrs[res->cnt++] = (unsigned long)sym_addr;
11990 }
11991
11992 if (res->cnt == 0)
11993 err = -ENOENT;
11994
11995 cleanup:
11996 fclose(f);
11997 return err;
11998 }
11999
12000 struct bpf_link *
12001 bpf_program__attach_kprobe_multi_opts(const struct bpf_program *prog,
12002 const char *pattern,
12003 const struct bpf_kprobe_multi_opts *opts)
12004 {
12005 LIBBPF_OPTS(bpf_link_create_opts, lopts);
12006 struct kprobe_multi_resolve res = {
12007 .pattern = pattern,
12008 };
12009 enum bpf_attach_type attach_type;
12010 struct bpf_link *link = NULL;
12011 const unsigned long *addrs;
12012 int err, link_fd, prog_fd;
12013 bool retprobe, session, unique_match;
12014 const __u64 *cookies;
12015 const char **syms;
12016 size_t cnt;
12017
12018 if (!OPTS_VALID(opts, bpf_kprobe_multi_opts))
12019 return libbpf_err_ptr(-EINVAL);
12020
12021 prog_fd = bpf_program__fd(prog);
12022 if (prog_fd < 0) {
12023 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
12024 prog->name);
12025 return libbpf_err_ptr(-EINVAL);
12026 }
12027
12028 syms = OPTS_GET(opts, syms, false);
12029 addrs = OPTS_GET(opts, addrs, false);
12030 cnt = OPTS_GET(opts, cnt, false);
12031 cookies = OPTS_GET(opts, cookies, false);
12032 unique_match = OPTS_GET(opts, unique_match, false);
12033
12034 if (!pattern && !addrs && !syms)
12035 return libbpf_err_ptr(-EINVAL);
12036 if (pattern && (addrs || syms || cookies || cnt))
12037 return libbpf_err_ptr(-EINVAL);
12038 if (!pattern && !cnt)
12039 return libbpf_err_ptr(-EINVAL);
12040 if (!pattern && unique_match)
12041 return libbpf_err_ptr(-EINVAL);
12042 if (addrs && syms)
12043 return libbpf_err_ptr(-EINVAL);
12044
12045 if (pattern) {
12046 if (has_available_filter_functions_addrs())
12047 err = libbpf_available_kprobes_parse(&res);
12048 else
12049 err = libbpf_available_kallsyms_parse(&res);
12050 if (err)
12051 goto error;
12052
12053 if (unique_match && res.cnt != 1) {
12054 pr_warn("prog '%s': failed to find a unique match for '%s' (%zu matches)\n",
12055 prog->name, pattern, res.cnt);
12056 err = -EINVAL;
12057 goto error;
12058 }
12059
12060 addrs = res.addrs;
12061 cnt = res.cnt;
12062 }
12063
12064 retprobe = OPTS_GET(opts, retprobe, false);
12065 session = OPTS_GET(opts, session, false);
12066
12067 if (retprobe && session)
12068 return libbpf_err_ptr(-EINVAL);
12069
12070 attach_type = session ? BPF_TRACE_KPROBE_SESSION : BPF_TRACE_KPROBE_MULTI;
12071
12072 lopts.kprobe_multi.syms = syms;
12073 lopts.kprobe_multi.addrs = addrs;
12074 lopts.kprobe_multi.cookies = cookies;
12075 lopts.kprobe_multi.cnt = cnt;
12076 lopts.kprobe_multi.flags = retprobe ? BPF_F_KPROBE_MULTI_RETURN : 0;
12077
12078 link = calloc(1, sizeof(*link));
12079 if (!link) {
12080 err = -ENOMEM;
12081 goto error;
12082 }
12083 link->detach = &bpf_link__detach_fd;
12084
12085 link_fd = bpf_link_create(prog_fd, 0, attach_type, &lopts);
12086 if (link_fd < 0) {
12087 err = -errno;
12088 pr_warn("prog '%s': failed to attach: %s\n",
12089 prog->name, errstr(err));
12090 goto error;
12091 }
12092 link->fd = link_fd;
12093 free(res.addrs);
12094 return link;
12095
12096 error:
12097 free(link);
12098 free(res.addrs);
12099 return libbpf_err_ptr(err);
12100 }
12101
12102 static int attach_kprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12103 {
12104 DECLARE_LIBBPF_OPTS(bpf_kprobe_opts, opts);
12105 unsigned long offset = 0;
12106 const char *func_name;
12107 char *func;
12108 int n;
12109
12110 *link = NULL;
12111
12112 /* no auto-attach for SEC("kprobe") and SEC("kretprobe") */
12113 if (strcmp(prog->sec_name, "kprobe") == 0 || strcmp(prog->sec_name, "kretprobe") == 0)
12114 return 0;
12115
12116 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe/");
12117 if (opts.retprobe)
12118 func_name = prog->sec_name + sizeof("kretprobe/") - 1;
12119 else
12120 func_name = prog->sec_name + sizeof("kprobe/") - 1;
12121
12122 n = sscanf(func_name, "%m[a-zA-Z0-9_.]+%li", &func, &offset);
12123 if (n < 1) {
12124 pr_warn("kprobe name is invalid: %s\n", func_name);
12125 return -EINVAL;
12126 }
12127 if (opts.retprobe && offset != 0) {
12128 free(func);
12129 pr_warn("kretprobes do not support offset specification\n");
12130 return -EINVAL;
12131 }
12132
12133 opts.offset = offset;
12134 *link = bpf_program__attach_kprobe_opts(prog, func, &opts);
12135 free(func);
12136 return libbpf_get_error(*link);
12137 }
12138
12139 static int attach_ksyscall(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12140 {
12141 LIBBPF_OPTS(bpf_ksyscall_opts, opts);
12142 const char *syscall_name;
12143
12144 *link = NULL;
12145
12146 /* no auto-attach for SEC("ksyscall") and SEC("kretsyscall") */
12147 if (strcmp(prog->sec_name, "ksyscall") == 0 || strcmp(prog->sec_name, "kretsyscall") == 0)
12148 return 0;
12149
12150 opts.retprobe = str_has_pfx(prog->sec_name, "kretsyscall/");
12151 if (opts.retprobe)
12152 syscall_name = prog->sec_name + sizeof("kretsyscall/") - 1;
12153 else
12154 syscall_name = prog->sec_name + sizeof("ksyscall/") - 1;
12155
12156 *link = bpf_program__attach_ksyscall(prog, syscall_name, &opts);
12157 return *link ? 0 : -errno;
12158 }
12159
12160 static int attach_kprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12161 {
12162 LIBBPF_OPTS(bpf_kprobe_multi_opts, opts);
12163 const char *spec;
12164 char *pattern;
12165 int n;
12166
12167 *link = NULL;
12168
12169 /* no auto-attach for SEC("kprobe.multi") and SEC("kretprobe.multi") */
12170 if (strcmp(prog->sec_name, "kprobe.multi") == 0 ||
12171 strcmp(prog->sec_name, "kretprobe.multi") == 0)
12172 return 0;
12173
12174 opts.retprobe = str_has_pfx(prog->sec_name, "kretprobe.multi/");
12175 if (opts.retprobe)
12176 spec = prog->sec_name + sizeof("kretprobe.multi/") - 1;
12177 else
12178 spec = prog->sec_name + sizeof("kprobe.multi/") - 1;
12179
12180 n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
12181 if (n < 1) {
12182 pr_warn("kprobe multi pattern is invalid: %s\n", spec);
12183 return -EINVAL;
12184 }
12185
12186 *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
12187 free(pattern);
12188 return libbpf_get_error(*link);
12189 }
12190
12191 static int attach_kprobe_session(const struct bpf_program *prog, long cookie,
12192 struct bpf_link **link)
12193 {
12194 LIBBPF_OPTS(bpf_kprobe_multi_opts, opts, .session = true);
12195 const char *spec;
12196 char *pattern;
12197 int n;
12198
12199 *link = NULL;
12200
12201 /* no auto-attach for SEC("kprobe.session") */
12202 if (strcmp(prog->sec_name, "kprobe.session") == 0)
12203 return 0;
12204
12205 spec = prog->sec_name + sizeof("kprobe.session/") - 1;
12206 n = sscanf(spec, "%m[a-zA-Z0-9_.*?]", &pattern);
12207 if (n < 1) {
12208 pr_warn("kprobe session pattern is invalid: %s\n", spec);
12209 return -EINVAL;
12210 }
12211
12212 *link = bpf_program__attach_kprobe_multi_opts(prog, pattern, &opts);
12213 free(pattern);
12214 return *link ? 0 : -errno;
12215 }
12216
12217 static int attach_uprobe_multi(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12218 {
12219 char *probe_type = NULL, *binary_path = NULL, *func_name = NULL;
12220 LIBBPF_OPTS(bpf_uprobe_multi_opts, opts);
12221 int n, ret = -EINVAL;
12222
12223 *link = NULL;
12224
12225 n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]",
12226 &probe_type, &binary_path, &func_name);
12227 switch (n) {
12228 case 1:
12229 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
12230 ret = 0;
12231 break;
12232 case 3:
12233 opts.session = str_has_pfx(probe_type, "uprobe.session");
12234 opts.retprobe = str_has_pfx(probe_type, "uretprobe.multi");
12235
12236 *link = bpf_program__attach_uprobe_multi(prog, -1, binary_path, func_name, &opts);
12237 ret = libbpf_get_error(*link);
12238 break;
12239 default:
12240 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
12241 prog->sec_name);
12242 break;
12243 }
12244 free(probe_type);
12245 free(binary_path);
12246 free(func_name);
12247 return ret;
12248 }
12249
12250 static inline int add_uprobe_event_legacy(const char *probe_name, bool retprobe,
12251 const char *binary_path, size_t offset)
12252 {
12253 return append_to_file(tracefs_uprobe_events(), "%c:%s/%s %s:0x%zx",
12254 retprobe ? 'r' : 'p',
12255 retprobe ? "uretprobes" : "uprobes",
12256 probe_name, binary_path, offset);
12257 }
12258
12259 static inline int remove_uprobe_event_legacy(const char *probe_name, bool retprobe)
12260 {
12261 return append_to_file(tracefs_uprobe_events(), "-:%s/%s",
12262 retprobe ? "uretprobes" : "uprobes", probe_name);
12263 }
12264
12265 static int determine_uprobe_perf_type_legacy(const char *probe_name, bool retprobe)
12266 {
12267 char file[512];
12268
12269 snprintf(file, sizeof(file), "%s/events/%s/%s/id",
12270 tracefs_path(), retprobe ? "uretprobes" : "uprobes", probe_name);
12271
12272 return parse_uint_from_file(file, "%d\n");
12273 }
12274
12275 static int perf_event_uprobe_open_legacy(const char *probe_name, bool retprobe,
12276 const char *binary_path, size_t offset, int pid)
12277 {
12278 const size_t attr_sz = sizeof(struct perf_event_attr);
12279 struct perf_event_attr attr;
12280 int type, pfd, err;
12281
12282 err = add_uprobe_event_legacy(probe_name, retprobe, binary_path, offset);
12283 if (err < 0) {
12284 pr_warn("failed to add legacy uprobe event for %s:0x%zx: %s\n",
12285 binary_path, (size_t)offset, errstr(err));
12286 return err;
12287 }
12288 type = determine_uprobe_perf_type_legacy(probe_name, retprobe);
12289 if (type < 0) {
12290 err = type;
12291 pr_warn("failed to determine legacy uprobe event id for %s:0x%zx: %s\n",
12292 binary_path, offset, errstr(err));
12293 goto err_clean_legacy;
12294 }
12295
12296 memset(&attr, 0, attr_sz);
12297 attr.size = attr_sz;
12298 attr.config = type;
12299 attr.type = PERF_TYPE_TRACEPOINT;
12300
12301 pfd = syscall(__NR_perf_event_open, &attr,
12302 pid < 0 ? -1 : pid, /* pid */
12303 pid == -1 ? 0 : -1, /* cpu */
12304 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
12305 if (pfd < 0) {
12306 err = -errno;
12307 pr_warn("legacy uprobe perf_event_open() failed: %s\n", errstr(err));
12308 goto err_clean_legacy;
12309 }
12310 return pfd;
12311
12312 err_clean_legacy:
12313 /* Clear the newly added legacy uprobe_event */
12314 remove_uprobe_event_legacy(probe_name, retprobe);
12315 return err;
12316 }
12317
12318 /* Find offset of function name in archive specified by path. Currently
12319 * supported are .zip files that do not compress their contents, as used on
12320 * Android in the form of APKs, for example. "file_name" is the name of the ELF
12321 * file inside the archive. "func_name" matches symbol name or name@@LIB for
12322 * library functions.
12323 *
12324 * An overview of the APK format specifically provided here:
12325 * https://en.wikipedia.org/w/index.php?title=Apk_(file_format)&oldid=1139099120#Package_contents
12326 */
12327 static long elf_find_func_offset_from_archive(const char *archive_path, const char *file_name,
12328 const char *func_name)
12329 {
12330 struct zip_archive *archive;
12331 struct zip_entry entry;
12332 long ret;
12333 Elf *elf;
12334
12335 archive = zip_archive_open(archive_path);
12336 if (IS_ERR(archive)) {
12337 ret = PTR_ERR(archive);
12338 pr_warn("zip: failed to open %s: %ld\n", archive_path, ret);
12339 return ret;
12340 }
12341
12342 ret = zip_archive_find_entry(archive, file_name, &entry);
12343 if (ret) {
12344 pr_warn("zip: could not find archive member %s in %s: %ld\n", file_name,
12345 archive_path, ret);
12346 goto out;
12347 }
12348 pr_debug("zip: found entry for %s in %s at 0x%lx\n", file_name, archive_path,
12349 (unsigned long)entry.data_offset);
12350
12351 if (entry.compression) {
12352 pr_warn("zip: entry %s of %s is compressed and cannot be handled\n", file_name,
12353 archive_path);
12354 ret = -LIBBPF_ERRNO__FORMAT;
12355 goto out;
12356 }
12357
12358 elf = elf_memory((void *)entry.data, entry.data_length);
12359 if (!elf) {
12360 pr_warn("elf: could not read elf file %s from %s: %s\n", file_name, archive_path,
12361 elf_errmsg(-1));
12362 ret = -LIBBPF_ERRNO__LIBELF;
12363 goto out;
12364 }
12365
12366 ret = elf_find_func_offset(elf, file_name, func_name);
12367 if (ret > 0) {
12368 pr_debug("elf: symbol address match for %s of %s in %s: 0x%x + 0x%lx = 0x%lx\n",
12369 func_name, file_name, archive_path, entry.data_offset, ret,
12370 ret + entry.data_offset);
12371 ret += entry.data_offset;
12372 }
12373 elf_end(elf);
12374
12375 out:
12376 zip_archive_close(archive);
12377 return ret;
12378 }
12379
12380 static const char *arch_specific_lib_paths(void)
12381 {
12382 /*
12383 * Based on https://packages.debian.org/sid/libc6.
12384 *
12385 * Assume that the traced program is built for the same architecture
12386 * as libbpf, which should cover the vast majority of cases.
12387 */
12388 #if defined(__x86_64__)
12389 return "/lib/x86_64-linux-gnu";
12390 #elif defined(__i386__)
12391 return "/lib/i386-linux-gnu";
12392 #elif defined(__s390x__)
12393 return "/lib/s390x-linux-gnu";
12394 #elif defined(__arm__) && defined(__SOFTFP__)
12395 return "/lib/arm-linux-gnueabi";
12396 #elif defined(__arm__) && !defined(__SOFTFP__)
12397 return "/lib/arm-linux-gnueabihf";
12398 #elif defined(__aarch64__)
12399 return "/lib/aarch64-linux-gnu";
12400 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 64
12401 return "/lib/mips64el-linux-gnuabi64";
12402 #elif defined(__mips__) && defined(__MIPSEL__) && _MIPS_SZLONG == 32
12403 return "/lib/mipsel-linux-gnu";
12404 #elif defined(__powerpc64__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
12405 return "/lib/powerpc64le-linux-gnu";
12406 #elif defined(__sparc__) && defined(__arch64__)
12407 return "/lib/sparc64-linux-gnu";
12408 #elif defined(__riscv) && __riscv_xlen == 64
12409 return "/lib/riscv64-linux-gnu";
12410 #else
12411 return NULL;
12412 #endif
12413 }
12414
12415 /* Get full path to program/shared library. */
12416 static int resolve_full_path(const char *file, char *result, size_t result_sz)
12417 {
12418 const char *search_paths[3] = {};
12419 int i, perm;
12420
12421 if (str_has_sfx(file, ".so") || strstr(file, ".so.")) {
12422 search_paths[0] = getenv("LD_LIBRARY_PATH");
12423 search_paths[1] = "/usr/lib64:/usr/lib";
12424 search_paths[2] = arch_specific_lib_paths();
12425 perm = R_OK;
12426 } else {
12427 search_paths[0] = getenv("PATH");
12428 search_paths[1] = "/usr/bin:/usr/sbin";
12429 perm = R_OK | X_OK;
12430 }
12431
12432 for (i = 0; i < ARRAY_SIZE(search_paths); i++) {
12433 const char *s;
12434
12435 if (!search_paths[i])
12436 continue;
12437 for (s = search_paths[i]; s != NULL; s = strchr(s, ':')) {
12438 const char *next_path;
12439 int seg_len;
12440
12441 if (s[0] == ':')
12442 s++;
12443 next_path = strchr(s, ':');
12444 seg_len = next_path ? next_path - s : strlen(s);
12445 if (!seg_len)
12446 continue;
12447 snprintf(result, result_sz, "%.*s/%s", seg_len, s, file);
12448 /* ensure it has required permissions */
12449 if (faccessat(AT_FDCWD, result, perm, AT_EACCESS) < 0)
12450 continue;
12451 pr_debug("resolved '%s' to '%s'\n", file, result);
12452 return 0;
12453 }
12454 }
12455 return -ENOENT;
12456 }
12457
12458 struct bpf_link *
12459 bpf_program__attach_uprobe_multi(const struct bpf_program *prog,
12460 pid_t pid,
12461 const char *path,
12462 const char *func_pattern,
12463 const struct bpf_uprobe_multi_opts *opts)
12464 {
12465 const unsigned long *ref_ctr_offsets = NULL, *offsets = NULL;
12466 LIBBPF_OPTS(bpf_link_create_opts, lopts);
12467 unsigned long *resolved_offsets = NULL;
12468 enum bpf_attach_type attach_type;
12469 int err = 0, link_fd, prog_fd;
12470 struct bpf_link *link = NULL;
12471 char full_path[PATH_MAX];
12472 bool retprobe, session;
12473 const __u64 *cookies;
12474 const char **syms;
12475 size_t cnt;
12476
12477 if (!OPTS_VALID(opts, bpf_uprobe_multi_opts))
12478 return libbpf_err_ptr(-EINVAL);
12479
12480 prog_fd = bpf_program__fd(prog);
12481 if (prog_fd < 0) {
12482 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
12483 prog->name);
12484 return libbpf_err_ptr(-EINVAL);
12485 }
12486
12487 syms = OPTS_GET(opts, syms, NULL);
12488 offsets = OPTS_GET(opts, offsets, NULL);
12489 ref_ctr_offsets = OPTS_GET(opts, ref_ctr_offsets, NULL);
12490 cookies = OPTS_GET(opts, cookies, NULL);
12491 cnt = OPTS_GET(opts, cnt, 0);
12492 retprobe = OPTS_GET(opts, retprobe, false);
12493 session = OPTS_GET(opts, session, false);
12494
12495 /*
12496 * User can specify 2 mutually exclusive set of inputs:
12497 *
12498 * 1) use only path/func_pattern/pid arguments
12499 *
12500 * 2) use path/pid with allowed combinations of:
12501 * syms/offsets/ref_ctr_offsets/cookies/cnt
12502 *
12503 * - syms and offsets are mutually exclusive
12504 * - ref_ctr_offsets and cookies are optional
12505 *
12506 * Any other usage results in error.
12507 */
12508
12509 if (!path)
12510 return libbpf_err_ptr(-EINVAL);
12511 if (!func_pattern && cnt == 0)
12512 return libbpf_err_ptr(-EINVAL);
12513
12514 if (func_pattern) {
12515 if (syms || offsets || ref_ctr_offsets || cookies || cnt)
12516 return libbpf_err_ptr(-EINVAL);
12517 } else {
12518 if (!!syms == !!offsets)
12519 return libbpf_err_ptr(-EINVAL);
12520 }
12521
12522 if (retprobe && session)
12523 return libbpf_err_ptr(-EINVAL);
12524
12525 if (func_pattern) {
12526 if (!strchr(path, '/')) {
12527 err = resolve_full_path(path, full_path, sizeof(full_path));
12528 if (err) {
12529 pr_warn("prog '%s': failed to resolve full path for '%s': %s\n",
12530 prog->name, path, errstr(err));
12531 return libbpf_err_ptr(err);
12532 }
12533 path = full_path;
12534 }
12535
12536 err = elf_resolve_pattern_offsets(path, func_pattern,
12537 &resolved_offsets, &cnt);
12538 if (err < 0)
12539 return libbpf_err_ptr(err);
12540 offsets = resolved_offsets;
12541 } else if (syms) {
12542 err = elf_resolve_syms_offsets(path, cnt, syms, &resolved_offsets, STT_FUNC);
12543 if (err < 0)
12544 return libbpf_err_ptr(err);
12545 offsets = resolved_offsets;
12546 }
12547
12548 attach_type = session ? BPF_TRACE_UPROBE_SESSION : BPF_TRACE_UPROBE_MULTI;
12549
12550 lopts.uprobe_multi.path = path;
12551 lopts.uprobe_multi.offsets = offsets;
12552 lopts.uprobe_multi.ref_ctr_offsets = ref_ctr_offsets;
12553 lopts.uprobe_multi.cookies = cookies;
12554 lopts.uprobe_multi.cnt = cnt;
12555 lopts.uprobe_multi.flags = retprobe ? BPF_F_UPROBE_MULTI_RETURN : 0;
12556
12557 if (pid == 0)
12558 pid = getpid();
12559 if (pid > 0)
12560 lopts.uprobe_multi.pid = pid;
12561
12562 link = calloc(1, sizeof(*link));
12563 if (!link) {
12564 err = -ENOMEM;
12565 goto error;
12566 }
12567 link->detach = &bpf_link__detach_fd;
12568
12569 link_fd = bpf_link_create(prog_fd, 0, attach_type, &lopts);
12570 if (link_fd < 0) {
12571 err = -errno;
12572 pr_warn("prog '%s': failed to attach multi-uprobe: %s\n",
12573 prog->name, errstr(err));
12574 goto error;
12575 }
12576 link->fd = link_fd;
12577 free(resolved_offsets);
12578 return link;
12579
12580 error:
12581 free(resolved_offsets);
12582 free(link);
12583 return libbpf_err_ptr(err);
12584 }
12585
12586 LIBBPF_API struct bpf_link *
12587 bpf_program__attach_uprobe_opts(const struct bpf_program *prog, pid_t pid,
12588 const char *binary_path, size_t func_offset,
12589 const struct bpf_uprobe_opts *opts)
12590 {
12591 const char *archive_path = NULL, *archive_sep = NULL;
12592 char *legacy_probe = NULL;
12593 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
12594 enum probe_attach_mode attach_mode;
12595 char full_path[PATH_MAX];
12596 struct bpf_link *link;
12597 size_t ref_ctr_off;
12598 int pfd, err;
12599 bool retprobe, legacy;
12600 const char *func_name;
12601
12602 if (!OPTS_VALID(opts, bpf_uprobe_opts))
12603 return libbpf_err_ptr(-EINVAL);
12604
12605 attach_mode = OPTS_GET(opts, attach_mode, PROBE_ATTACH_MODE_DEFAULT);
12606 retprobe = OPTS_GET(opts, retprobe, false);
12607 ref_ctr_off = OPTS_GET(opts, ref_ctr_offset, 0);
12608 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
12609
12610 if (!binary_path)
12611 return libbpf_err_ptr(-EINVAL);
12612
12613 /* Check if "binary_path" refers to an archive. */
12614 archive_sep = strstr(binary_path, "!/");
12615 if (archive_sep) {
12616 full_path[0] = '\0';
12617 libbpf_strlcpy(full_path, binary_path,
12618 min(sizeof(full_path), (size_t)(archive_sep - binary_path + 1)));
12619 archive_path = full_path;
12620 binary_path = archive_sep + 2;
12621 } else if (!strchr(binary_path, '/')) {
12622 err = resolve_full_path(binary_path, full_path, sizeof(full_path));
12623 if (err) {
12624 pr_warn("prog '%s': failed to resolve full path for '%s': %s\n",
12625 prog->name, binary_path, errstr(err));
12626 return libbpf_err_ptr(err);
12627 }
12628 binary_path = full_path;
12629 }
12630 func_name = OPTS_GET(opts, func_name, NULL);
12631 if (func_name) {
12632 long sym_off;
12633
12634 if (archive_path) {
12635 sym_off = elf_find_func_offset_from_archive(archive_path, binary_path,
12636 func_name);
12637 binary_path = archive_path;
12638 } else {
12639 sym_off = elf_find_func_offset_from_file(binary_path, func_name);
12640 }
12641 if (sym_off < 0)
12642 return libbpf_err_ptr(sym_off);
12643 func_offset += sym_off;
12644 }
12645
12646 legacy = determine_uprobe_perf_type() < 0;
12647 switch (attach_mode) {
12648 case PROBE_ATTACH_MODE_LEGACY:
12649 legacy = true;
12650 pe_opts.force_ioctl_attach = true;
12651 break;
12652 case PROBE_ATTACH_MODE_PERF:
12653 if (legacy)
12654 return libbpf_err_ptr(-ENOTSUP);
12655 pe_opts.force_ioctl_attach = true;
12656 break;
12657 case PROBE_ATTACH_MODE_LINK:
12658 if (legacy || !kernel_supports(prog->obj, FEAT_PERF_LINK))
12659 return libbpf_err_ptr(-ENOTSUP);
12660 break;
12661 case PROBE_ATTACH_MODE_DEFAULT:
12662 break;
12663 default:
12664 return libbpf_err_ptr(-EINVAL);
12665 }
12666
12667 if (!legacy) {
12668 pfd = perf_event_open_probe(true /* uprobe */, retprobe, binary_path,
12669 func_offset, pid, ref_ctr_off);
12670 } else {
12671 char probe_name[MAX_EVENT_NAME_LEN];
12672
12673 if (ref_ctr_off)
12674 return libbpf_err_ptr(-EINVAL);
12675
12676 gen_probe_legacy_event_name(probe_name, sizeof(probe_name),
12677 strrchr(binary_path, '/') ? : binary_path,
12678 func_offset);
12679
12680 legacy_probe = strdup(probe_name);
12681 if (!legacy_probe)
12682 return libbpf_err_ptr(-ENOMEM);
12683
12684 pfd = perf_event_uprobe_open_legacy(legacy_probe, retprobe,
12685 binary_path, func_offset, pid);
12686 }
12687 if (pfd < 0) {
12688 err = -errno;
12689 pr_warn("prog '%s': failed to create %s '%s:0x%zx' perf event: %s\n",
12690 prog->name, retprobe ? "uretprobe" : "uprobe",
12691 binary_path, func_offset,
12692 errstr(err));
12693 goto err_out;
12694 }
12695
12696 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
12697 err = libbpf_get_error(link);
12698 if (err) {
12699 close(pfd);
12700 pr_warn("prog '%s': failed to attach to %s '%s:0x%zx': %s\n",
12701 prog->name, retprobe ? "uretprobe" : "uprobe",
12702 binary_path, func_offset,
12703 errstr(err));
12704 goto err_clean_legacy;
12705 }
12706 if (legacy) {
12707 struct bpf_link_perf *perf_link = container_of(link, struct bpf_link_perf, link);
12708
12709 perf_link->legacy_probe_name = legacy_probe;
12710 perf_link->legacy_is_kprobe = false;
12711 perf_link->legacy_is_retprobe = retprobe;
12712 }
12713 return link;
12714
12715 err_clean_legacy:
12716 if (legacy)
12717 remove_uprobe_event_legacy(legacy_probe, retprobe);
12718 err_out:
12719 free(legacy_probe);
12720 return libbpf_err_ptr(err);
12721 }
12722
12723 /* Format of u[ret]probe section definition supporting auto-attach:
12724 * u[ret]probe/binary:function[+offset]
12725 *
12726 * binary can be an absolute/relative path or a filename; the latter is resolved to a
12727 * full binary path via bpf_program__attach_uprobe_opts.
12728 *
12729 * Specifying uprobe+ ensures we carry out strict matching; either "uprobe" must be
12730 * specified (and auto-attach is not possible) or the above format is specified for
12731 * auto-attach.
12732 */
12733 static int attach_uprobe(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12734 {
12735 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts);
12736 char *probe_type = NULL, *binary_path = NULL, *func_name = NULL, *func_off;
12737 int n, c, ret = -EINVAL;
12738 long offset = 0;
12739
12740 *link = NULL;
12741
12742 n = sscanf(prog->sec_name, "%m[^/]/%m[^:]:%m[^\n]",
12743 &probe_type, &binary_path, &func_name);
12744 switch (n) {
12745 case 1:
12746 /* handle SEC("u[ret]probe") - format is valid, but auto-attach is impossible. */
12747 ret = 0;
12748 break;
12749 case 2:
12750 pr_warn("prog '%s': section '%s' missing ':function[+offset]' specification\n",
12751 prog->name, prog->sec_name);
12752 break;
12753 case 3:
12754 /* check if user specifies `+offset`, if yes, this should be
12755 * the last part of the string, make sure sscanf read to EOL
12756 */
12757 func_off = strrchr(func_name, '+');
12758 if (func_off) {
12759 n = sscanf(func_off, "+%li%n", &offset, &c);
12760 if (n == 1 && *(func_off + c) == '\0')
12761 func_off[0] = '\0';
12762 else
12763 offset = 0;
12764 }
12765 opts.retprobe = strcmp(probe_type, "uretprobe") == 0 ||
12766 strcmp(probe_type, "uretprobe.s") == 0;
12767 if (opts.retprobe && offset != 0) {
12768 pr_warn("prog '%s': uretprobes do not support offset specification\n",
12769 prog->name);
12770 break;
12771 }
12772 opts.func_name = func_name;
12773 *link = bpf_program__attach_uprobe_opts(prog, -1, binary_path, offset, &opts);
12774 ret = libbpf_get_error(*link);
12775 break;
12776 default:
12777 pr_warn("prog '%s': invalid format of section definition '%s'\n", prog->name,
12778 prog->sec_name);
12779 break;
12780 }
12781 free(probe_type);
12782 free(binary_path);
12783 free(func_name);
12784
12785 return ret;
12786 }
12787
12788 struct bpf_link *bpf_program__attach_uprobe(const struct bpf_program *prog,
12789 bool retprobe, pid_t pid,
12790 const char *binary_path,
12791 size_t func_offset)
12792 {
12793 DECLARE_LIBBPF_OPTS(bpf_uprobe_opts, opts, .retprobe = retprobe);
12794
12795 return bpf_program__attach_uprobe_opts(prog, pid, binary_path, func_offset, &opts);
12796 }
12797
12798 struct bpf_link *bpf_program__attach_usdt(const struct bpf_program *prog,
12799 pid_t pid, const char *binary_path,
12800 const char *usdt_provider, const char *usdt_name,
12801 const struct bpf_usdt_opts *opts)
12802 {
12803 char resolved_path[512];
12804 struct bpf_object *obj = prog->obj;
12805 struct bpf_link *link;
12806 __u64 usdt_cookie;
12807 int err;
12808
12809 if (!OPTS_VALID(opts, bpf_uprobe_opts))
12810 return libbpf_err_ptr(-EINVAL);
12811
12812 if (bpf_program__fd(prog) < 0) {
12813 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
12814 prog->name);
12815 return libbpf_err_ptr(-EINVAL);
12816 }
12817
12818 if (!binary_path)
12819 return libbpf_err_ptr(-EINVAL);
12820
12821 if (!strchr(binary_path, '/')) {
12822 err = resolve_full_path(binary_path, resolved_path, sizeof(resolved_path));
12823 if (err) {
12824 pr_warn("prog '%s': failed to resolve full path for '%s': %s\n",
12825 prog->name, binary_path, errstr(err));
12826 return libbpf_err_ptr(err);
12827 }
12828 binary_path = resolved_path;
12829 }
12830
12831 /* USDT manager is instantiated lazily on first USDT attach. It will
12832 * be destroyed together with BPF object in bpf_object__close().
12833 */
12834 if (IS_ERR(obj->usdt_man))
12835 return libbpf_ptr(obj->usdt_man);
12836 if (!obj->usdt_man) {
12837 obj->usdt_man = usdt_manager_new(obj);
12838 if (IS_ERR(obj->usdt_man))
12839 return libbpf_ptr(obj->usdt_man);
12840 }
12841
12842 usdt_cookie = OPTS_GET(opts, usdt_cookie, 0);
12843 link = usdt_manager_attach_usdt(obj->usdt_man, prog, pid, binary_path,
12844 usdt_provider, usdt_name, usdt_cookie);
12845 err = libbpf_get_error(link);
12846 if (err)
12847 return libbpf_err_ptr(err);
12848 return link;
12849 }
12850
12851 static int attach_usdt(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12852 {
12853 char *path = NULL, *provider = NULL, *name = NULL;
12854 const char *sec_name;
12855 int n, err;
12856
12857 sec_name = bpf_program__section_name(prog);
12858 if (strcmp(sec_name, "usdt") == 0) {
12859 /* no auto-attach for just SEC("usdt") */
12860 *link = NULL;
12861 return 0;
12862 }
12863
12864 n = sscanf(sec_name, "usdt/%m[^:]:%m[^:]:%m[^:]", &path, &provider, &name);
12865 if (n != 3) {
12866 pr_warn("invalid section '%s', expected SEC(\"usdt/<path>:<provider>:<name>\")\n",
12867 sec_name);
12868 err = -EINVAL;
12869 } else {
12870 *link = bpf_program__attach_usdt(prog, -1 /* any process */, path,
12871 provider, name, NULL);
12872 err = libbpf_get_error(*link);
12873 }
12874 free(path);
12875 free(provider);
12876 free(name);
12877 return err;
12878 }
12879
12880 static int determine_tracepoint_id(const char *tp_category,
12881 const char *tp_name)
12882 {
12883 char file[PATH_MAX];
12884 int ret;
12885
12886 ret = snprintf(file, sizeof(file), "%s/events/%s/%s/id",
12887 tracefs_path(), tp_category, tp_name);
12888 if (ret < 0)
12889 return -errno;
12890 if (ret >= sizeof(file)) {
12891 pr_debug("tracepoint %s/%s path is too long\n",
12892 tp_category, tp_name);
12893 return -E2BIG;
12894 }
12895 return parse_uint_from_file(file, "%d\n");
12896 }
12897
12898 static int perf_event_open_tracepoint(const char *tp_category,
12899 const char *tp_name)
12900 {
12901 const size_t attr_sz = sizeof(struct perf_event_attr);
12902 struct perf_event_attr attr;
12903 int tp_id, pfd, err;
12904
12905 tp_id = determine_tracepoint_id(tp_category, tp_name);
12906 if (tp_id < 0) {
12907 pr_warn("failed to determine tracepoint '%s/%s' perf event ID: %s\n",
12908 tp_category, tp_name,
12909 errstr(tp_id));
12910 return tp_id;
12911 }
12912
12913 memset(&attr, 0, attr_sz);
12914 attr.type = PERF_TYPE_TRACEPOINT;
12915 attr.size = attr_sz;
12916 attr.config = tp_id;
12917
12918 pfd = syscall(__NR_perf_event_open, &attr, -1 /* pid */, 0 /* cpu */,
12919 -1 /* group_fd */, PERF_FLAG_FD_CLOEXEC);
12920 if (pfd < 0) {
12921 err = -errno;
12922 pr_warn("tracepoint '%s/%s' perf_event_open() failed: %s\n",
12923 tp_category, tp_name,
12924 errstr(err));
12925 return err;
12926 }
12927 return pfd;
12928 }
12929
12930 struct bpf_link *bpf_program__attach_tracepoint_opts(const struct bpf_program *prog,
12931 const char *tp_category,
12932 const char *tp_name,
12933 const struct bpf_tracepoint_opts *opts)
12934 {
12935 DECLARE_LIBBPF_OPTS(bpf_perf_event_opts, pe_opts);
12936 struct bpf_link *link;
12937 int pfd, err;
12938
12939 if (!OPTS_VALID(opts, bpf_tracepoint_opts))
12940 return libbpf_err_ptr(-EINVAL);
12941
12942 pe_opts.bpf_cookie = OPTS_GET(opts, bpf_cookie, 0);
12943
12944 pfd = perf_event_open_tracepoint(tp_category, tp_name);
12945 if (pfd < 0) {
12946 pr_warn("prog '%s': failed to create tracepoint '%s/%s' perf event: %s\n",
12947 prog->name, tp_category, tp_name,
12948 errstr(pfd));
12949 return libbpf_err_ptr(pfd);
12950 }
12951 link = bpf_program__attach_perf_event_opts(prog, pfd, &pe_opts);
12952 err = libbpf_get_error(link);
12953 if (err) {
12954 close(pfd);
12955 pr_warn("prog '%s': failed to attach to tracepoint '%s/%s': %s\n",
12956 prog->name, tp_category, tp_name,
12957 errstr(err));
12958 return libbpf_err_ptr(err);
12959 }
12960 return link;
12961 }
12962
12963 struct bpf_link *bpf_program__attach_tracepoint(const struct bpf_program *prog,
12964 const char *tp_category,
12965 const char *tp_name)
12966 {
12967 return bpf_program__attach_tracepoint_opts(prog, tp_category, tp_name, NULL);
12968 }
12969
12970 static int attach_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
12971 {
12972 char *sec_name, *tp_cat, *tp_name;
12973
12974 *link = NULL;
12975
12976 /* no auto-attach for SEC("tp") or SEC("tracepoint") */
12977 if (strcmp(prog->sec_name, "tp") == 0 || strcmp(prog->sec_name, "tracepoint") == 0)
12978 return 0;
12979
12980 sec_name = strdup(prog->sec_name);
12981 if (!sec_name)
12982 return -ENOMEM;
12983
12984 /* extract "tp/<category>/<name>" or "tracepoint/<category>/<name>" */
12985 if (str_has_pfx(prog->sec_name, "tp/"))
12986 tp_cat = sec_name + sizeof("tp/") - 1;
12987 else
12988 tp_cat = sec_name + sizeof("tracepoint/") - 1;
12989 tp_name = strchr(tp_cat, '/');
12990 if (!tp_name) {
12991 free(sec_name);
12992 return -EINVAL;
12993 }
12994 *tp_name = '\0';
12995 tp_name++;
12996
12997 *link = bpf_program__attach_tracepoint(prog, tp_cat, tp_name);
12998 free(sec_name);
12999 return libbpf_get_error(*link);
13000 }
13001
13002 struct bpf_link *
13003 bpf_program__attach_raw_tracepoint_opts(const struct bpf_program *prog,
13004 const char *tp_name,
13005 struct bpf_raw_tracepoint_opts *opts)
13006 {
13007 LIBBPF_OPTS(bpf_raw_tp_opts, raw_opts);
13008 struct bpf_link *link;
13009 int prog_fd, pfd;
13010
13011 if (!OPTS_VALID(opts, bpf_raw_tracepoint_opts))
13012 return libbpf_err_ptr(-EINVAL);
13013
13014 prog_fd = bpf_program__fd(prog);
13015 if (prog_fd < 0) {
13016 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
13017 return libbpf_err_ptr(-EINVAL);
13018 }
13019
13020 link = calloc(1, sizeof(*link));
13021 if (!link)
13022 return libbpf_err_ptr(-ENOMEM);
13023 link->detach = &bpf_link__detach_fd;
13024
13025 raw_opts.tp_name = tp_name;
13026 raw_opts.cookie = OPTS_GET(opts, cookie, 0);
13027 pfd = bpf_raw_tracepoint_open_opts(prog_fd, &raw_opts);
13028 if (pfd < 0) {
13029 pfd = -errno;
13030 free(link);
13031 pr_warn("prog '%s': failed to attach to raw tracepoint '%s': %s\n",
13032 prog->name, tp_name, errstr(pfd));
13033 return libbpf_err_ptr(pfd);
13034 }
13035 link->fd = pfd;
13036 return link;
13037 }
13038
13039 struct bpf_link *bpf_program__attach_raw_tracepoint(const struct bpf_program *prog,
13040 const char *tp_name)
13041 {
13042 return bpf_program__attach_raw_tracepoint_opts(prog, tp_name, NULL);
13043 }
13044
13045 static int attach_raw_tp(const struct bpf_program *prog, long cookie, struct bpf_link **link)
13046 {
13047 static const char *const prefixes[] = {
13048 "raw_tp",
13049 "raw_tracepoint",
13050 "raw_tp.w",
13051 "raw_tracepoint.w",
13052 };
13053 size_t i;
13054 const char *tp_name = NULL;
13055
13056 *link = NULL;
13057
13058 for (i = 0; i < ARRAY_SIZE(prefixes); i++) {
13059 size_t pfx_len;
13060
13061 if (!str_has_pfx(prog->sec_name, prefixes[i]))
13062 continue;
13063
13064 pfx_len = strlen(prefixes[i]);
13065 /* no auto-attach case of, e.g., SEC("raw_tp") */
13066 if (prog->sec_name[pfx_len] == '\0')
13067 return 0;
13068
13069 if (prog->sec_name[pfx_len] != '/')
13070 continue;
13071
13072 tp_name = prog->sec_name + pfx_len + 1;
13073 break;
13074 }
13075
13076 if (!tp_name) {
13077 pr_warn("prog '%s': invalid section name '%s'\n",
13078 prog->name, prog->sec_name);
13079 return -EINVAL;
13080 }
13081
13082 *link = bpf_program__attach_raw_tracepoint(prog, tp_name);
13083 return libbpf_get_error(*link);
13084 }
13085
13086 /* Common logic for all BPF program types that attach to a btf_id */
13087 static struct bpf_link *bpf_program__attach_btf_id(const struct bpf_program *prog,
13088 const struct bpf_trace_opts *opts)
13089 {
13090 LIBBPF_OPTS(bpf_link_create_opts, link_opts);
13091 struct bpf_link *link;
13092 int prog_fd, pfd;
13093
13094 if (!OPTS_VALID(opts, bpf_trace_opts))
13095 return libbpf_err_ptr(-EINVAL);
13096
13097 prog_fd = bpf_program__fd(prog);
13098 if (prog_fd < 0) {
13099 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
13100 return libbpf_err_ptr(-EINVAL);
13101 }
13102
13103 link = calloc(1, sizeof(*link));
13104 if (!link)
13105 return libbpf_err_ptr(-ENOMEM);
13106 link->detach = &bpf_link__detach_fd;
13107
13108 /* libbpf is smart enough to redirect to BPF_RAW_TRACEPOINT_OPEN on old kernels */
13109 link_opts.tracing.cookie = OPTS_GET(opts, cookie, 0);
13110 pfd = bpf_link_create(prog_fd, 0, bpf_program__expected_attach_type(prog), &link_opts);
13111 if (pfd < 0) {
13112 pfd = -errno;
13113 free(link);
13114 pr_warn("prog '%s': failed to attach: %s\n",
13115 prog->name, errstr(pfd));
13116 return libbpf_err_ptr(pfd);
13117 }
13118 link->fd = pfd;
13119 return link;
13120 }
13121
13122 struct bpf_link *bpf_program__attach_trace(const struct bpf_program *prog)
13123 {
13124 return bpf_program__attach_btf_id(prog, NULL);
13125 }
13126
13127 struct bpf_link *bpf_program__attach_trace_opts(const struct bpf_program *prog,
13128 const struct bpf_trace_opts *opts)
13129 {
13130 return bpf_program__attach_btf_id(prog, opts);
13131 }
13132
13133 struct bpf_link *bpf_program__attach_lsm(const struct bpf_program *prog)
13134 {
13135 return bpf_program__attach_btf_id(prog, NULL);
13136 }
13137
13138 static int attach_trace(const struct bpf_program *prog, long cookie, struct bpf_link **link)
13139 {
13140 *link = bpf_program__attach_trace(prog);
13141 return libbpf_get_error(*link);
13142 }
13143
13144 static int attach_lsm(const struct bpf_program *prog, long cookie, struct bpf_link **link)
13145 {
13146 *link = bpf_program__attach_lsm(prog);
13147 return libbpf_get_error(*link);
13148 }
13149
13150 static struct bpf_link *
13151 bpf_program_attach_fd(const struct bpf_program *prog,
13152 int target_fd, const char *target_name,
13153 const struct bpf_link_create_opts *opts)
13154 {
13155 enum bpf_attach_type attach_type;
13156 struct bpf_link *link;
13157 int prog_fd, link_fd;
13158
13159 prog_fd = bpf_program__fd(prog);
13160 if (prog_fd < 0) {
13161 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
13162 return libbpf_err_ptr(-EINVAL);
13163 }
13164
13165 link = calloc(1, sizeof(*link));
13166 if (!link)
13167 return libbpf_err_ptr(-ENOMEM);
13168 link->detach = &bpf_link__detach_fd;
13169
13170 attach_type = bpf_program__expected_attach_type(prog);
13171 link_fd = bpf_link_create(prog_fd, target_fd, attach_type, opts);
13172 if (link_fd < 0) {
13173 link_fd = -errno;
13174 free(link);
13175 pr_warn("prog '%s': failed to attach to %s: %s\n",
13176 prog->name, target_name,
13177 errstr(link_fd));
13178 return libbpf_err_ptr(link_fd);
13179 }
13180 link->fd = link_fd;
13181 return link;
13182 }
13183
13184 struct bpf_link *
13185 bpf_program__attach_cgroup(const struct bpf_program *prog, int cgroup_fd)
13186 {
13187 return bpf_program_attach_fd(prog, cgroup_fd, "cgroup", NULL);
13188 }
13189
13190 struct bpf_link *
13191 bpf_program__attach_netns(const struct bpf_program *prog, int netns_fd)
13192 {
13193 return bpf_program_attach_fd(prog, netns_fd, "netns", NULL);
13194 }
13195
13196 struct bpf_link *
13197 bpf_program__attach_sockmap(const struct bpf_program *prog, int map_fd)
13198 {
13199 return bpf_program_attach_fd(prog, map_fd, "sockmap", NULL);
13200 }
13201
13202 struct bpf_link *bpf_program__attach_xdp(const struct bpf_program *prog, int ifindex)
13203 {
13204 /* target_fd/target_ifindex use the same field in LINK_CREATE */
13205 return bpf_program_attach_fd(prog, ifindex, "xdp", NULL);
13206 }
13207
13208 struct bpf_link *
13209 bpf_program__attach_cgroup_opts(const struct bpf_program *prog, int cgroup_fd,
13210 const struct bpf_cgroup_opts *opts)
13211 {
13212 LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
13213 __u32 relative_id;
13214 int relative_fd;
13215
13216 if (!OPTS_VALID(opts, bpf_cgroup_opts))
13217 return libbpf_err_ptr(-EINVAL);
13218
13219 relative_id = OPTS_GET(opts, relative_id, 0);
13220 relative_fd = OPTS_GET(opts, relative_fd, 0);
13221
13222 if (relative_fd && relative_id) {
13223 pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n",
13224 prog->name);
13225 return libbpf_err_ptr(-EINVAL);
13226 }
13227
13228 link_create_opts.cgroup.expected_revision = OPTS_GET(opts, expected_revision, 0);
13229 link_create_opts.cgroup.relative_fd = relative_fd;
13230 link_create_opts.cgroup.relative_id = relative_id;
13231 link_create_opts.flags = OPTS_GET(opts, flags, 0);
13232
13233 return bpf_program_attach_fd(prog, cgroup_fd, "cgroup", &link_create_opts);
13234 }
13235
13236 struct bpf_link *
13237 bpf_program__attach_tcx(const struct bpf_program *prog, int ifindex,
13238 const struct bpf_tcx_opts *opts)
13239 {
13240 LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
13241 __u32 relative_id;
13242 int relative_fd;
13243
13244 if (!OPTS_VALID(opts, bpf_tcx_opts))
13245 return libbpf_err_ptr(-EINVAL);
13246
13247 relative_id = OPTS_GET(opts, relative_id, 0);
13248 relative_fd = OPTS_GET(opts, relative_fd, 0);
13249
13250 /* validate we don't have unexpected combinations of non-zero fields */
13251 if (!ifindex) {
13252 pr_warn("prog '%s': target netdevice ifindex cannot be zero\n",
13253 prog->name);
13254 return libbpf_err_ptr(-EINVAL);
13255 }
13256 if (relative_fd && relative_id) {
13257 pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n",
13258 prog->name);
13259 return libbpf_err_ptr(-EINVAL);
13260 }
13261
13262 link_create_opts.tcx.expected_revision = OPTS_GET(opts, expected_revision, 0);
13263 link_create_opts.tcx.relative_fd = relative_fd;
13264 link_create_opts.tcx.relative_id = relative_id;
13265 link_create_opts.flags = OPTS_GET(opts, flags, 0);
13266
13267 /* target_fd/target_ifindex use the same field in LINK_CREATE */
13268 return bpf_program_attach_fd(prog, ifindex, "tcx", &link_create_opts);
13269 }
13270
13271 struct bpf_link *
13272 bpf_program__attach_netkit(const struct bpf_program *prog, int ifindex,
13273 const struct bpf_netkit_opts *opts)
13274 {
13275 LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
13276 __u32 relative_id;
13277 int relative_fd;
13278
13279 if (!OPTS_VALID(opts, bpf_netkit_opts))
13280 return libbpf_err_ptr(-EINVAL);
13281
13282 relative_id = OPTS_GET(opts, relative_id, 0);
13283 relative_fd = OPTS_GET(opts, relative_fd, 0);
13284
13285 /* validate we don't have unexpected combinations of non-zero fields */
13286 if (!ifindex) {
13287 pr_warn("prog '%s': target netdevice ifindex cannot be zero\n",
13288 prog->name);
13289 return libbpf_err_ptr(-EINVAL);
13290 }
13291 if (relative_fd && relative_id) {
13292 pr_warn("prog '%s': relative_fd and relative_id cannot be set at the same time\n",
13293 prog->name);
13294 return libbpf_err_ptr(-EINVAL);
13295 }
13296
13297 link_create_opts.netkit.expected_revision = OPTS_GET(opts, expected_revision, 0);
13298 link_create_opts.netkit.relative_fd = relative_fd;
13299 link_create_opts.netkit.relative_id = relative_id;
13300 link_create_opts.flags = OPTS_GET(opts, flags, 0);
13301
13302 return bpf_program_attach_fd(prog, ifindex, "netkit", &link_create_opts);
13303 }
13304
13305 struct bpf_link *bpf_program__attach_freplace(const struct bpf_program *prog,
13306 int target_fd,
13307 const char *attach_func_name)
13308 {
13309 int btf_id;
13310
13311 if (!!target_fd != !!attach_func_name) {
13312 pr_warn("prog '%s': supply none or both of target_fd and attach_func_name\n",
13313 prog->name);
13314 return libbpf_err_ptr(-EINVAL);
13315 }
13316
13317 if (prog->type != BPF_PROG_TYPE_EXT) {
13318 pr_warn("prog '%s': only BPF_PROG_TYPE_EXT can attach as freplace\n",
13319 prog->name);
13320 return libbpf_err_ptr(-EINVAL);
13321 }
13322
13323 if (target_fd) {
13324 LIBBPF_OPTS(bpf_link_create_opts, target_opts);
13325
13326 btf_id = libbpf_find_prog_btf_id(attach_func_name, target_fd, prog->obj->token_fd);
13327 if (btf_id < 0)
13328 return libbpf_err_ptr(btf_id);
13329
13330 target_opts.target_btf_id = btf_id;
13331
13332 return bpf_program_attach_fd(prog, target_fd, "freplace",
13333 &target_opts);
13334 } else {
13335 /* no target, so use raw_tracepoint_open for compatibility
13336 * with old kernels
13337 */
13338 return bpf_program__attach_trace(prog);
13339 }
13340 }
13341
13342 struct bpf_link *
13343 bpf_program__attach_iter(const struct bpf_program *prog,
13344 const struct bpf_iter_attach_opts *opts)
13345 {
13346 DECLARE_LIBBPF_OPTS(bpf_link_create_opts, link_create_opts);
13347 struct bpf_link *link;
13348 int prog_fd, link_fd;
13349 __u32 target_fd = 0;
13350
13351 if (!OPTS_VALID(opts, bpf_iter_attach_opts))
13352 return libbpf_err_ptr(-EINVAL);
13353
13354 link_create_opts.iter_info = OPTS_GET(opts, link_info, (void *)0);
13355 link_create_opts.iter_info_len = OPTS_GET(opts, link_info_len, 0);
13356
13357 prog_fd = bpf_program__fd(prog);
13358 if (prog_fd < 0) {
13359 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
13360 return libbpf_err_ptr(-EINVAL);
13361 }
13362
13363 link = calloc(1, sizeof(*link));
13364 if (!link)
13365 return libbpf_err_ptr(-ENOMEM);
13366 link->detach = &bpf_link__detach_fd;
13367
13368 link_fd = bpf_link_create(prog_fd, target_fd, BPF_TRACE_ITER,
13369 &link_create_opts);
13370 if (link_fd < 0) {
13371 link_fd = -errno;
13372 free(link);
13373 pr_warn("prog '%s': failed to attach to iterator: %s\n",
13374 prog->name, errstr(link_fd));
13375 return libbpf_err_ptr(link_fd);
13376 }
13377 link->fd = link_fd;
13378 return link;
13379 }
13380
13381 static int attach_iter(const struct bpf_program *prog, long cookie, struct bpf_link **link)
13382 {
13383 *link = bpf_program__attach_iter(prog, NULL);
13384 return libbpf_get_error(*link);
13385 }
13386
13387 struct bpf_link *bpf_program__attach_netfilter(const struct bpf_program *prog,
13388 const struct bpf_netfilter_opts *opts)
13389 {
13390 LIBBPF_OPTS(bpf_link_create_opts, lopts);
13391 struct bpf_link *link;
13392 int prog_fd, link_fd;
13393
13394 if (!OPTS_VALID(opts, bpf_netfilter_opts))
13395 return libbpf_err_ptr(-EINVAL);
13396
13397 prog_fd = bpf_program__fd(prog);
13398 if (prog_fd < 0) {
13399 pr_warn("prog '%s': can't attach before loaded\n", prog->name);
13400 return libbpf_err_ptr(-EINVAL);
13401 }
13402
13403 link = calloc(1, sizeof(*link));
13404 if (!link)
13405 return libbpf_err_ptr(-ENOMEM);
13406
13407 link->detach = &bpf_link__detach_fd;
13408
13409 lopts.netfilter.pf = OPTS_GET(opts, pf, 0);
13410 lopts.netfilter.hooknum = OPTS_GET(opts, hooknum, 0);
13411 lopts.netfilter.priority = OPTS_GET(opts, priority, 0);
13412 lopts.netfilter.flags = OPTS_GET(opts, flags, 0);
13413
13414 link_fd = bpf_link_create(prog_fd, 0, BPF_NETFILTER, &lopts);
13415 if (link_fd < 0) {
13416 link_fd = -errno;
13417 free(link);
13418 pr_warn("prog '%s': failed to attach to netfilter: %s\n",
13419 prog->name, errstr(link_fd));
13420 return libbpf_err_ptr(link_fd);
13421 }
13422 link->fd = link_fd;
13423
13424 return link;
13425 }
13426
13427 struct bpf_link *bpf_program__attach(const struct bpf_program *prog)
13428 {
13429 struct bpf_link *link = NULL;
13430 int err;
13431
13432 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
13433 return libbpf_err_ptr(-EOPNOTSUPP);
13434
13435 if (bpf_program__fd(prog) < 0) {
13436 pr_warn("prog '%s': can't attach BPF program without FD (was it loaded?)\n",
13437 prog->name);
13438 return libbpf_err_ptr(-EINVAL);
13439 }
13440
13441 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, &link);
13442 if (err)
13443 return libbpf_err_ptr(err);
13444
13445 /* When calling bpf_program__attach() explicitly, auto-attach support
13446 * is expected to work, so NULL returned link is considered an error.
13447 * This is different for skeleton's attach, see comment in
13448 * bpf_object__attach_skeleton().
13449 */
13450 if (!link)
13451 return libbpf_err_ptr(-EOPNOTSUPP);
13452
13453 return link;
13454 }
13455
13456 struct bpf_link_struct_ops {
13457 struct bpf_link link;
13458 int map_fd;
13459 };
13460
13461 static int bpf_link__detach_struct_ops(struct bpf_link *link)
13462 {
13463 struct bpf_link_struct_ops *st_link;
13464 __u32 zero = 0;
13465
13466 st_link = container_of(link, struct bpf_link_struct_ops, link);
13467
13468 if (st_link->map_fd < 0)
13469 /* w/o a real link */
13470 return bpf_map_delete_elem(link->fd, &zero);
13471
13472 return close(link->fd);
13473 }
13474
13475 struct bpf_link *bpf_map__attach_struct_ops(const struct bpf_map *map)
13476 {
13477 struct bpf_link_struct_ops *link;
13478 __u32 zero = 0;
13479 int err, fd;
13480
13481 if (!bpf_map__is_struct_ops(map)) {
13482 pr_warn("map '%s': can't attach non-struct_ops map\n", map->name);
13483 return libbpf_err_ptr(-EINVAL);
13484 }
13485
13486 if (map->fd < 0) {
13487 pr_warn("map '%s': can't attach BPF map without FD (was it created?)\n", map->name);
13488 return libbpf_err_ptr(-EINVAL);
13489 }
13490
13491 link = calloc(1, sizeof(*link));
13492 if (!link)
13493 return libbpf_err_ptr(-EINVAL);
13494
13495 /* kern_vdata should be prepared during the loading phase. */
13496 err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
13497 /* It can be EBUSY if the map has been used to create or
13498 * update a link before. We don't allow updating the value of
13499 * a struct_ops once it is set. That ensures that the value
13500 * never changed. So, it is safe to skip EBUSY.
13501 */
13502 if (err && (!(map->def.map_flags & BPF_F_LINK) || err != -EBUSY)) {
13503 free(link);
13504 return libbpf_err_ptr(err);
13505 }
13506
13507 link->link.detach = bpf_link__detach_struct_ops;
13508
13509 if (!(map->def.map_flags & BPF_F_LINK)) {
13510 /* w/o a real link */
13511 link->link.fd = map->fd;
13512 link->map_fd = -1;
13513 return &link->link;
13514 }
13515
13516 fd = bpf_link_create(map->fd, 0, BPF_STRUCT_OPS, NULL);
13517 if (fd < 0) {
13518 free(link);
13519 return libbpf_err_ptr(fd);
13520 }
13521
13522 link->link.fd = fd;
13523 link->map_fd = map->fd;
13524
13525 return &link->link;
13526 }
13527
13528 /*
13529 * Swap the back struct_ops of a link with a new struct_ops map.
13530 */
13531 int bpf_link__update_map(struct bpf_link *link, const struct bpf_map *map)
13532 {
13533 struct bpf_link_struct_ops *st_ops_link;
13534 __u32 zero = 0;
13535 int err;
13536
13537 if (!bpf_map__is_struct_ops(map))
13538 return libbpf_err(-EINVAL);
13539
13540 if (map->fd < 0) {
13541 pr_warn("map '%s': can't use BPF map without FD (was it created?)\n", map->name);
13542 return libbpf_err(-EINVAL);
13543 }
13544
13545 st_ops_link = container_of(link, struct bpf_link_struct_ops, link);
13546 /* Ensure the type of a link is correct */
13547 if (st_ops_link->map_fd < 0)
13548 return libbpf_err(-EINVAL);
13549
13550 err = bpf_map_update_elem(map->fd, &zero, map->st_ops->kern_vdata, 0);
13551 /* It can be EBUSY if the map has been used to create or
13552 * update a link before. We don't allow updating the value of
13553 * a struct_ops once it is set. That ensures that the value
13554 * never changed. So, it is safe to skip EBUSY.
13555 */
13556 if (err && err != -EBUSY)
13557 return err;
13558
13559 err = bpf_link_update(link->fd, map->fd, NULL);
13560 if (err < 0)
13561 return err;
13562
13563 st_ops_link->map_fd = map->fd;
13564
13565 return 0;
13566 }
13567
13568 typedef enum bpf_perf_event_ret (*bpf_perf_event_print_t)(struct perf_event_header *hdr,
13569 void *private_data);
13570
13571 static enum bpf_perf_event_ret
13572 perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size,
13573 void **copy_mem, size_t *copy_size,
13574 bpf_perf_event_print_t fn, void *private_data)
13575 {
13576 struct perf_event_mmap_page *header = mmap_mem;
13577 __u64 data_head = ring_buffer_read_head(header);
13578 __u64 data_tail = header->data_tail;
13579 void *base = ((__u8 *)header) + page_size;
13580 int ret = LIBBPF_PERF_EVENT_CONT;
13581 struct perf_event_header *ehdr;
13582 size_t ehdr_size;
13583
13584 while (data_head != data_tail) {
13585 ehdr = base + (data_tail & (mmap_size - 1));
13586 ehdr_size = ehdr->size;
13587
13588 if (((void *)ehdr) + ehdr_size > base + mmap_size) {
13589 void *copy_start = ehdr;
13590 size_t len_first = base + mmap_size - copy_start;
13591 size_t len_secnd = ehdr_size - len_first;
13592
13593 if (*copy_size < ehdr_size) {
13594 free(*copy_mem);
13595 *copy_mem = malloc(ehdr_size);
13596 if (!*copy_mem) {
13597 *copy_size = 0;
13598 ret = LIBBPF_PERF_EVENT_ERROR;
13599 break;
13600 }
13601 *copy_size = ehdr_size;
13602 }
13603
13604 memcpy(*copy_mem, copy_start, len_first);
13605 memcpy(*copy_mem + len_first, base, len_secnd);
13606 ehdr = *copy_mem;
13607 }
13608
13609 ret = fn(ehdr, private_data);
13610 data_tail += ehdr_size;
13611 if (ret != LIBBPF_PERF_EVENT_CONT)
13612 break;
13613 }
13614
13615 ring_buffer_write_tail(header, data_tail);
13616 return libbpf_err(ret);
13617 }
13618
13619 struct perf_buffer;
13620
13621 struct perf_buffer_params {
13622 struct perf_event_attr *attr;
13623 /* if event_cb is specified, it takes precendence */
13624 perf_buffer_event_fn event_cb;
13625 /* sample_cb and lost_cb are higher-level common-case callbacks */
13626 perf_buffer_sample_fn sample_cb;
13627 perf_buffer_lost_fn lost_cb;
13628 void *ctx;
13629 int cpu_cnt;
13630 int *cpus;
13631 int *map_keys;
13632 };
13633
13634 struct perf_cpu_buf {
13635 struct perf_buffer *pb;
13636 void *base; /* mmap()'ed memory */
13637 void *buf; /* for reconstructing segmented data */
13638 size_t buf_size;
13639 int fd;
13640 int cpu;
13641 int map_key;
13642 };
13643
13644 struct perf_buffer {
13645 perf_buffer_event_fn event_cb;
13646 perf_buffer_sample_fn sample_cb;
13647 perf_buffer_lost_fn lost_cb;
13648 void *ctx; /* passed into callbacks */
13649
13650 size_t page_size;
13651 size_t mmap_size;
13652 struct perf_cpu_buf **cpu_bufs;
13653 struct epoll_event *events;
13654 int cpu_cnt; /* number of allocated CPU buffers */
13655 int epoll_fd; /* perf event FD */
13656 int map_fd; /* BPF_MAP_TYPE_PERF_EVENT_ARRAY BPF map FD */
13657 };
13658
13659 static void perf_buffer__free_cpu_buf(struct perf_buffer *pb,
13660 struct perf_cpu_buf *cpu_buf)
13661 {
13662 if (!cpu_buf)
13663 return;
13664 if (cpu_buf->base &&
13665 munmap(cpu_buf->base, pb->mmap_size + pb->page_size))
13666 pr_warn("failed to munmap cpu_buf #%d\n", cpu_buf->cpu);
13667 if (cpu_buf->fd >= 0) {
13668 ioctl(cpu_buf->fd, PERF_EVENT_IOC_DISABLE, 0);
13669 close(cpu_buf->fd);
13670 }
13671 free(cpu_buf->buf);
13672 free(cpu_buf);
13673 }
13674
13675 void perf_buffer__free(struct perf_buffer *pb)
13676 {
13677 int i;
13678
13679 if (IS_ERR_OR_NULL(pb))
13680 return;
13681 if (pb->cpu_bufs) {
13682 for (i = 0; i < pb->cpu_cnt; i++) {
13683 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
13684
13685 if (!cpu_buf)
13686 continue;
13687
13688 bpf_map_delete_elem(pb->map_fd, &cpu_buf->map_key);
13689 perf_buffer__free_cpu_buf(pb, cpu_buf);
13690 }
13691 free(pb->cpu_bufs);
13692 }
13693 if (pb->epoll_fd >= 0)
13694 close(pb->epoll_fd);
13695 free(pb->events);
13696 free(pb);
13697 }
13698
13699 static struct perf_cpu_buf *
13700 perf_buffer__open_cpu_buf(struct perf_buffer *pb, struct perf_event_attr *attr,
13701 int cpu, int map_key)
13702 {
13703 struct perf_cpu_buf *cpu_buf;
13704 int err;
13705
13706 cpu_buf = calloc(1, sizeof(*cpu_buf));
13707 if (!cpu_buf)
13708 return ERR_PTR(-ENOMEM);
13709
13710 cpu_buf->pb = pb;
13711 cpu_buf->cpu = cpu;
13712 cpu_buf->map_key = map_key;
13713
13714 cpu_buf->fd = syscall(__NR_perf_event_open, attr, -1 /* pid */, cpu,
13715 -1, PERF_FLAG_FD_CLOEXEC);
13716 if (cpu_buf->fd < 0) {
13717 err = -errno;
13718 pr_warn("failed to open perf buffer event on cpu #%d: %s\n",
13719 cpu, errstr(err));
13720 goto error;
13721 }
13722
13723 cpu_buf->base = mmap(NULL, pb->mmap_size + pb->page_size,
13724 PROT_READ | PROT_WRITE, MAP_SHARED,
13725 cpu_buf->fd, 0);
13726 if (cpu_buf->base == MAP_FAILED) {
13727 cpu_buf->base = NULL;
13728 err = -errno;
13729 pr_warn("failed to mmap perf buffer on cpu #%d: %s\n",
13730 cpu, errstr(err));
13731 goto error;
13732 }
13733
13734 if (ioctl(cpu_buf->fd, PERF_EVENT_IOC_ENABLE, 0) < 0) {
13735 err = -errno;
13736 pr_warn("failed to enable perf buffer event on cpu #%d: %s\n",
13737 cpu, errstr(err));
13738 goto error;
13739 }
13740
13741 return cpu_buf;
13742
13743 error:
13744 perf_buffer__free_cpu_buf(pb, cpu_buf);
13745 return (struct perf_cpu_buf *)ERR_PTR(err);
13746 }
13747
13748 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
13749 struct perf_buffer_params *p);
13750
13751 struct perf_buffer *perf_buffer__new(int map_fd, size_t page_cnt,
13752 perf_buffer_sample_fn sample_cb,
13753 perf_buffer_lost_fn lost_cb,
13754 void *ctx,
13755 const struct perf_buffer_opts *opts)
13756 {
13757 const size_t attr_sz = sizeof(struct perf_event_attr);
13758 struct perf_buffer_params p = {};
13759 struct perf_event_attr attr;
13760 __u32 sample_period;
13761
13762 if (!OPTS_VALID(opts, perf_buffer_opts))
13763 return libbpf_err_ptr(-EINVAL);
13764
13765 sample_period = OPTS_GET(opts, sample_period, 1);
13766 if (!sample_period)
13767 sample_period = 1;
13768
13769 memset(&attr, 0, attr_sz);
13770 attr.size = attr_sz;
13771 attr.config = PERF_COUNT_SW_BPF_OUTPUT;
13772 attr.type = PERF_TYPE_SOFTWARE;
13773 attr.sample_type = PERF_SAMPLE_RAW;
13774 attr.wakeup_events = sample_period;
13775
13776 p.attr = &attr;
13777 p.sample_cb = sample_cb;
13778 p.lost_cb = lost_cb;
13779 p.ctx = ctx;
13780
13781 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
13782 }
13783
13784 struct perf_buffer *perf_buffer__new_raw(int map_fd, size_t page_cnt,
13785 struct perf_event_attr *attr,
13786 perf_buffer_event_fn event_cb, void *ctx,
13787 const struct perf_buffer_raw_opts *opts)
13788 {
13789 struct perf_buffer_params p = {};
13790
13791 if (!attr)
13792 return libbpf_err_ptr(-EINVAL);
13793
13794 if (!OPTS_VALID(opts, perf_buffer_raw_opts))
13795 return libbpf_err_ptr(-EINVAL);
13796
13797 p.attr = attr;
13798 p.event_cb = event_cb;
13799 p.ctx = ctx;
13800 p.cpu_cnt = OPTS_GET(opts, cpu_cnt, 0);
13801 p.cpus = OPTS_GET(opts, cpus, NULL);
13802 p.map_keys = OPTS_GET(opts, map_keys, NULL);
13803
13804 return libbpf_ptr(__perf_buffer__new(map_fd, page_cnt, &p));
13805 }
13806
13807 static struct perf_buffer *__perf_buffer__new(int map_fd, size_t page_cnt,
13808 struct perf_buffer_params *p)
13809 {
13810 const char *online_cpus_file = "/sys/devices/system/cpu/online";
13811 struct bpf_map_info map;
13812 struct perf_buffer *pb;
13813 bool *online = NULL;
13814 __u32 map_info_len;
13815 int err, i, j, n;
13816
13817 if (page_cnt == 0 || (page_cnt & (page_cnt - 1))) {
13818 pr_warn("page count should be power of two, but is %zu\n",
13819 page_cnt);
13820 return ERR_PTR(-EINVAL);
13821 }
13822
13823 /* best-effort sanity checks */
13824 memset(&map, 0, sizeof(map));
13825 map_info_len = sizeof(map);
13826 err = bpf_map_get_info_by_fd(map_fd, &map, &map_info_len);
13827 if (err) {
13828 err = -errno;
13829 /* if BPF_OBJ_GET_INFO_BY_FD is supported, will return
13830 * -EBADFD, -EFAULT, or -E2BIG on real error
13831 */
13832 if (err != -EINVAL) {
13833 pr_warn("failed to get map info for map FD %d: %s\n",
13834 map_fd, errstr(err));
13835 return ERR_PTR(err);
13836 }
13837 pr_debug("failed to get map info for FD %d; API not supported? Ignoring...\n",
13838 map_fd);
13839 } else {
13840 if (map.type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) {
13841 pr_warn("map '%s' should be BPF_MAP_TYPE_PERF_EVENT_ARRAY\n",
13842 map.name);
13843 return ERR_PTR(-EINVAL);
13844 }
13845 }
13846
13847 pb = calloc(1, sizeof(*pb));
13848 if (!pb)
13849 return ERR_PTR(-ENOMEM);
13850
13851 pb->event_cb = p->event_cb;
13852 pb->sample_cb = p->sample_cb;
13853 pb->lost_cb = p->lost_cb;
13854 pb->ctx = p->ctx;
13855
13856 pb->page_size = getpagesize();
13857 pb->mmap_size = pb->page_size * page_cnt;
13858 pb->map_fd = map_fd;
13859
13860 pb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
13861 if (pb->epoll_fd < 0) {
13862 err = -errno;
13863 pr_warn("failed to create epoll instance: %s\n",
13864 errstr(err));
13865 goto error;
13866 }
13867
13868 if (p->cpu_cnt > 0) {
13869 pb->cpu_cnt = p->cpu_cnt;
13870 } else {
13871 pb->cpu_cnt = libbpf_num_possible_cpus();
13872 if (pb->cpu_cnt < 0) {
13873 err = pb->cpu_cnt;
13874 goto error;
13875 }
13876 if (map.max_entries && map.max_entries < pb->cpu_cnt)
13877 pb->cpu_cnt = map.max_entries;
13878 }
13879
13880 pb->events = calloc(pb->cpu_cnt, sizeof(*pb->events));
13881 if (!pb->events) {
13882 err = -ENOMEM;
13883 pr_warn("failed to allocate events: out of memory\n");
13884 goto error;
13885 }
13886 pb->cpu_bufs = calloc(pb->cpu_cnt, sizeof(*pb->cpu_bufs));
13887 if (!pb->cpu_bufs) {
13888 err = -ENOMEM;
13889 pr_warn("failed to allocate buffers: out of memory\n");
13890 goto error;
13891 }
13892
13893 err = parse_cpu_mask_file(online_cpus_file, &online, &n);
13894 if (err) {
13895 pr_warn("failed to get online CPU mask: %s\n", errstr(err));
13896 goto error;
13897 }
13898
13899 for (i = 0, j = 0; i < pb->cpu_cnt; i++) {
13900 struct perf_cpu_buf *cpu_buf;
13901 int cpu, map_key;
13902
13903 cpu = p->cpu_cnt > 0 ? p->cpus[i] : i;
13904 map_key = p->cpu_cnt > 0 ? p->map_keys[i] : i;
13905
13906 /* in case user didn't explicitly requested particular CPUs to
13907 * be attached to, skip offline/not present CPUs
13908 */
13909 if (p->cpu_cnt <= 0 && (cpu >= n || !online[cpu]))
13910 continue;
13911
13912 cpu_buf = perf_buffer__open_cpu_buf(pb, p->attr, cpu, map_key);
13913 if (IS_ERR(cpu_buf)) {
13914 err = PTR_ERR(cpu_buf);
13915 goto error;
13916 }
13917
13918 pb->cpu_bufs[j] = cpu_buf;
13919
13920 err = bpf_map_update_elem(pb->map_fd, &map_key,
13921 &cpu_buf->fd, 0);
13922 if (err) {
13923 err = -errno;
13924 pr_warn("failed to set cpu #%d, key %d -> perf FD %d: %s\n",
13925 cpu, map_key, cpu_buf->fd,
13926 errstr(err));
13927 goto error;
13928 }
13929
13930 pb->events[j].events = EPOLLIN;
13931 pb->events[j].data.ptr = cpu_buf;
13932 if (epoll_ctl(pb->epoll_fd, EPOLL_CTL_ADD, cpu_buf->fd,
13933 &pb->events[j]) < 0) {
13934 err = -errno;
13935 pr_warn("failed to epoll_ctl cpu #%d perf FD %d: %s\n",
13936 cpu, cpu_buf->fd,
13937 errstr(err));
13938 goto error;
13939 }
13940 j++;
13941 }
13942 pb->cpu_cnt = j;
13943 free(online);
13944
13945 return pb;
13946
13947 error:
13948 free(online);
13949 if (pb)
13950 perf_buffer__free(pb);
13951 return ERR_PTR(err);
13952 }
13953
13954 struct perf_sample_raw {
13955 struct perf_event_header header;
13956 uint32_t size;
13957 char data[];
13958 };
13959
13960 struct perf_sample_lost {
13961 struct perf_event_header header;
13962 uint64_t id;
13963 uint64_t lost;
13964 uint64_t sample_id;
13965 };
13966
13967 static enum bpf_perf_event_ret
13968 perf_buffer__process_record(struct perf_event_header *e, void *ctx)
13969 {
13970 struct perf_cpu_buf *cpu_buf = ctx;
13971 struct perf_buffer *pb = cpu_buf->pb;
13972 void *data = e;
13973
13974 /* user wants full control over parsing perf event */
13975 if (pb->event_cb)
13976 return pb->event_cb(pb->ctx, cpu_buf->cpu, e);
13977
13978 switch (e->type) {
13979 case PERF_RECORD_SAMPLE: {
13980 struct perf_sample_raw *s = data;
13981
13982 if (pb->sample_cb)
13983 pb->sample_cb(pb->ctx, cpu_buf->cpu, s->data, s->size);
13984 break;
13985 }
13986 case PERF_RECORD_LOST: {
13987 struct perf_sample_lost *s = data;
13988
13989 if (pb->lost_cb)
13990 pb->lost_cb(pb->ctx, cpu_buf->cpu, s->lost);
13991 break;
13992 }
13993 default:
13994 pr_warn("unknown perf sample type %d\n", e->type);
13995 return LIBBPF_PERF_EVENT_ERROR;
13996 }
13997 return LIBBPF_PERF_EVENT_CONT;
13998 }
13999
14000 static int perf_buffer__process_records(struct perf_buffer *pb,
14001 struct perf_cpu_buf *cpu_buf)
14002 {
14003 enum bpf_perf_event_ret ret;
14004
14005 ret = perf_event_read_simple(cpu_buf->base, pb->mmap_size,
14006 pb->page_size, &cpu_buf->buf,
14007 &cpu_buf->buf_size,
14008 perf_buffer__process_record, cpu_buf);
14009 if (ret != LIBBPF_PERF_EVENT_CONT)
14010 return ret;
14011 return 0;
14012 }
14013
14014 int perf_buffer__epoll_fd(const struct perf_buffer *pb)
14015 {
14016 return pb->epoll_fd;
14017 }
14018
14019 int perf_buffer__poll(struct perf_buffer *pb, int timeout_ms)
14020 {
14021 int i, cnt, err;
14022
14023 cnt = epoll_wait(pb->epoll_fd, pb->events, pb->cpu_cnt, timeout_ms);
14024 if (cnt < 0)
14025 return -errno;
14026
14027 for (i = 0; i < cnt; i++) {
14028 struct perf_cpu_buf *cpu_buf = pb->events[i].data.ptr;
14029
14030 err = perf_buffer__process_records(pb, cpu_buf);
14031 if (err) {
14032 pr_warn("error while processing records: %s\n", errstr(err));
14033 return libbpf_err(err);
14034 }
14035 }
14036 return cnt;
14037 }
14038
14039 /* Return number of PERF_EVENT_ARRAY map slots set up by this perf_buffer
14040 * manager.
14041 */
14042 size_t perf_buffer__buffer_cnt(const struct perf_buffer *pb)
14043 {
14044 return pb->cpu_cnt;
14045 }
14046
14047 /*
14048 * Return perf_event FD of a ring buffer in *buf_idx* slot of
14049 * PERF_EVENT_ARRAY BPF map. This FD can be polled for new data using
14050 * select()/poll()/epoll() Linux syscalls.
14051 */
14052 int perf_buffer__buffer_fd(const struct perf_buffer *pb, size_t buf_idx)
14053 {
14054 struct perf_cpu_buf *cpu_buf;
14055
14056 if (buf_idx >= pb->cpu_cnt)
14057 return libbpf_err(-EINVAL);
14058
14059 cpu_buf = pb->cpu_bufs[buf_idx];
14060 if (!cpu_buf)
14061 return libbpf_err(-ENOENT);
14062
14063 return cpu_buf->fd;
14064 }
14065
14066 int perf_buffer__buffer(struct perf_buffer *pb, int buf_idx, void **buf, size_t *buf_size)
14067 {
14068 struct perf_cpu_buf *cpu_buf;
14069
14070 if (buf_idx >= pb->cpu_cnt)
14071 return libbpf_err(-EINVAL);
14072
14073 cpu_buf = pb->cpu_bufs[buf_idx];
14074 if (!cpu_buf)
14075 return libbpf_err(-ENOENT);
14076
14077 *buf = cpu_buf->base;
14078 *buf_size = pb->mmap_size;
14079 return 0;
14080 }
14081
14082 /*
14083 * Consume data from perf ring buffer corresponding to slot *buf_idx* in
14084 * PERF_EVENT_ARRAY BPF map without waiting/polling. If there is no data to
14085 * consume, do nothing and return success.
14086 * Returns:
14087 * - 0 on success;
14088 * - <0 on failure.
14089 */
14090 int perf_buffer__consume_buffer(struct perf_buffer *pb, size_t buf_idx)
14091 {
14092 struct perf_cpu_buf *cpu_buf;
14093
14094 if (buf_idx >= pb->cpu_cnt)
14095 return libbpf_err(-EINVAL);
14096
14097 cpu_buf = pb->cpu_bufs[buf_idx];
14098 if (!cpu_buf)
14099 return libbpf_err(-ENOENT);
14100
14101 return perf_buffer__process_records(pb, cpu_buf);
14102 }
14103
14104 int perf_buffer__consume(struct perf_buffer *pb)
14105 {
14106 int i, err;
14107
14108 for (i = 0; i < pb->cpu_cnt; i++) {
14109 struct perf_cpu_buf *cpu_buf = pb->cpu_bufs[i];
14110
14111 if (!cpu_buf)
14112 continue;
14113
14114 err = perf_buffer__process_records(pb, cpu_buf);
14115 if (err) {
14116 pr_warn("perf_buffer: failed to process records in buffer #%d: %s\n",
14117 i, errstr(err));
14118 return libbpf_err(err);
14119 }
14120 }
14121 return 0;
14122 }
14123
14124 int bpf_program__set_attach_target(struct bpf_program *prog,
14125 int attach_prog_fd,
14126 const char *attach_func_name)
14127 {
14128 int btf_obj_fd = 0, btf_id = 0, err;
14129
14130 if (!prog || attach_prog_fd < 0)
14131 return libbpf_err(-EINVAL);
14132
14133 if (prog->obj->state >= OBJ_LOADED)
14134 return libbpf_err(-EINVAL);
14135
14136 if (attach_prog_fd && !attach_func_name) {
14137 /* Store attach_prog_fd. The BTF ID will be resolved later during
14138 * the normal object/program load phase.
14139 */
14140 prog->attach_prog_fd = attach_prog_fd;
14141 return 0;
14142 }
14143
14144 if (attach_prog_fd) {
14145 btf_id = libbpf_find_prog_btf_id(attach_func_name,
14146 attach_prog_fd, prog->obj->token_fd);
14147 if (btf_id < 0)
14148 return libbpf_err(btf_id);
14149 } else {
14150 if (!attach_func_name)
14151 return libbpf_err(-EINVAL);
14152
14153 /* load btf_vmlinux, if not yet */
14154 err = bpf_object__load_vmlinux_btf(prog->obj, true);
14155 if (err)
14156 return libbpf_err(err);
14157 err = find_kernel_btf_id(prog->obj, attach_func_name,
14158 prog->expected_attach_type,
14159 &btf_obj_fd, &btf_id);
14160 if (err)
14161 return libbpf_err(err);
14162 }
14163
14164 prog->attach_btf_id = btf_id;
14165 prog->attach_btf_obj_fd = btf_obj_fd;
14166 prog->attach_prog_fd = attach_prog_fd;
14167 return 0;
14168 }
14169
14170 int bpf_program__assoc_struct_ops(struct bpf_program *prog, struct bpf_map *map,
14171 struct bpf_prog_assoc_struct_ops_opts *opts)
14172 {
14173 int prog_fd, map_fd;
14174
14175 prog_fd = bpf_program__fd(prog);
14176 if (prog_fd < 0) {
14177 pr_warn("prog '%s': can't associate BPF program without FD (was it loaded?)\n",
14178 prog->name);
14179 return libbpf_err(-EINVAL);
14180 }
14181
14182 if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) {
14183 pr_warn("prog '%s': can't associate struct_ops program\n", prog->name);
14184 return libbpf_err(-EINVAL);
14185 }
14186
14187 map_fd = bpf_map__fd(map);
14188 if (map_fd < 0) {
14189 pr_warn("map '%s': can't associate BPF map without FD (was it created?)\n", map->name);
14190 return libbpf_err(-EINVAL);
14191 }
14192
14193 if (!bpf_map__is_struct_ops(map)) {
14194 pr_warn("map '%s': can't associate non-struct_ops map\n", map->name);
14195 return libbpf_err(-EINVAL);
14196 }
14197
14198 return bpf_prog_assoc_struct_ops(prog_fd, map_fd, opts);
14199 }
14200
14201 int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz)
14202 {
14203 int err = 0, n, len, start, end = -1;
14204 bool *tmp;
14205
14206 *mask = NULL;
14207 *mask_sz = 0;
14208
14209 /* Each sub string separated by ',' has format \d+-\d+ or \d+ */
14210 while (*s) {
14211 if (*s == ',' || *s == '\n') {
14212 s++;
14213 continue;
14214 }
14215 n = sscanf(s, "%d%n-%d%n", &start, &len, &end, &len);
14216 if (n <= 0 || n > 2) {
14217 pr_warn("Failed to get CPU range %s: %d\n", s, n);
14218 err = -EINVAL;
14219 goto cleanup;
14220 } else if (n == 1) {
14221 end = start;
14222 }
14223 if (start < 0 || start > end) {
14224 pr_warn("Invalid CPU range [%d,%d] in %s\n",
14225 start, end, s);
14226 err = -EINVAL;
14227 goto cleanup;
14228 }
14229 tmp = realloc(*mask, end + 1);
14230 if (!tmp) {
14231 err = -ENOMEM;
14232 goto cleanup;
14233 }
14234 *mask = tmp;
14235 memset(tmp + *mask_sz, 0, start - *mask_sz);
14236 memset(tmp + start, 1, end - start + 1);
14237 *mask_sz = end + 1;
14238 s += len;
14239 }
14240 if (!*mask_sz) {
14241 pr_warn("Empty CPU range\n");
14242 return -EINVAL;
14243 }
14244 return 0;
14245 cleanup:
14246 free(*mask);
14247 *mask = NULL;
14248 return err;
14249 }
14250
14251 int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz)
14252 {
14253 int fd, err = 0, len;
14254 char buf[128];
14255
14256 fd = open(fcpu, O_RDONLY | O_CLOEXEC);
14257 if (fd < 0) {
14258 err = -errno;
14259 pr_warn("Failed to open cpu mask file %s: %s\n", fcpu, errstr(err));
14260 return err;
14261 }
14262 len = read(fd, buf, sizeof(buf));
14263 close(fd);
14264 if (len <= 0) {
14265 err = len ? -errno : -EINVAL;
14266 pr_warn("Failed to read cpu mask from %s: %s\n", fcpu, errstr(err));
14267 return err;
14268 }
14269 if (len >= sizeof(buf)) {
14270 pr_warn("CPU mask is too big in file %s\n", fcpu);
14271 return -E2BIG;
14272 }
14273 buf[len] = '\0';
14274
14275 return parse_cpu_mask_str(buf, mask, mask_sz);
14276 }
14277
14278 int libbpf_num_possible_cpus(void)
14279 {
14280 static const char *fcpu = "/sys/devices/system/cpu/possible";
14281 static int cpus;
14282 int err, n, i, tmp_cpus;
14283 bool *mask;
14284
14285 tmp_cpus = READ_ONCE(cpus);
14286 if (tmp_cpus > 0)
14287 return tmp_cpus;
14288
14289 err = parse_cpu_mask_file(fcpu, &mask, &n);
14290 if (err)
14291 return libbpf_err(err);
14292
14293 tmp_cpus = 0;
14294 for (i = 0; i < n; i++) {
14295 if (mask[i])
14296 tmp_cpus++;
14297 }
14298 free(mask);
14299
14300 WRITE_ONCE(cpus, tmp_cpus);
14301 return tmp_cpus;
14302 }
14303
14304 static int populate_skeleton_maps(const struct bpf_object *obj,
14305 struct bpf_map_skeleton *maps,
14306 size_t map_cnt, size_t map_skel_sz)
14307 {
14308 int i;
14309
14310 for (i = 0; i < map_cnt; i++) {
14311 struct bpf_map_skeleton *map_skel = (void *)maps + i * map_skel_sz;
14312 struct bpf_map **map = map_skel->map;
14313 const char *name = map_skel->name;
14314 void **mmaped = map_skel->mmaped;
14315
14316 *map = bpf_object__find_map_by_name(obj, name);
14317 if (!*map) {
14318 pr_warn("failed to find skeleton map '%s'\n", name);
14319 return -ESRCH;
14320 }
14321
14322 /* externs shouldn't be pre-setup from user code */
14323 if (mmaped && (*map)->libbpf_type != LIBBPF_MAP_KCONFIG)
14324 *mmaped = (*map)->mmaped;
14325 }
14326 return 0;
14327 }
14328
14329 static int populate_skeleton_progs(const struct bpf_object *obj,
14330 struct bpf_prog_skeleton *progs,
14331 size_t prog_cnt, size_t prog_skel_sz)
14332 {
14333 int i;
14334
14335 for (i = 0; i < prog_cnt; i++) {
14336 struct bpf_prog_skeleton *prog_skel = (void *)progs + i * prog_skel_sz;
14337 struct bpf_program **prog = prog_skel->prog;
14338 const char *name = prog_skel->name;
14339
14340 *prog = bpf_object__find_program_by_name(obj, name);
14341 if (!*prog) {
14342 pr_warn("failed to find skeleton program '%s'\n", name);
14343 return -ESRCH;
14344 }
14345 }
14346 return 0;
14347 }
14348
14349 int bpf_object__open_skeleton(struct bpf_object_skeleton *s,
14350 const struct bpf_object_open_opts *opts)
14351 {
14352 struct bpf_object *obj;
14353 int err;
14354
14355 obj = bpf_object_open(NULL, s->data, s->data_sz, s->name, opts);
14356 if (IS_ERR(obj)) {
14357 err = PTR_ERR(obj);
14358 pr_warn("failed to initialize skeleton BPF object '%s': %s\n",
14359 s->name, errstr(err));
14360 return libbpf_err(err);
14361 }
14362
14363 *s->obj = obj;
14364 err = populate_skeleton_maps(obj, s->maps, s->map_cnt, s->map_skel_sz);
14365 if (err) {
14366 pr_warn("failed to populate skeleton maps for '%s': %s\n", s->name, errstr(err));
14367 return libbpf_err(err);
14368 }
14369
14370 err = populate_skeleton_progs(obj, s->progs, s->prog_cnt, s->prog_skel_sz);
14371 if (err) {
14372 pr_warn("failed to populate skeleton progs for '%s': %s\n", s->name, errstr(err));
14373 return libbpf_err(err);
14374 }
14375
14376 return 0;
14377 }
14378
14379 int bpf_object__open_subskeleton(struct bpf_object_subskeleton *s)
14380 {
14381 int err, len, var_idx, i;
14382 const char *var_name;
14383 const struct bpf_map *map;
14384 struct btf *btf;
14385 __u32 map_type_id;
14386 const struct btf_type *map_type, *var_type;
14387 const struct bpf_var_skeleton *var_skel;
14388 struct btf_var_secinfo *var;
14389
14390 if (!s->obj)
14391 return libbpf_err(-EINVAL);
14392
14393 btf = bpf_object__btf(s->obj);
14394 if (!btf) {
14395 pr_warn("subskeletons require BTF at runtime (object %s)\n",
14396 bpf_object__name(s->obj));
14397 return libbpf_err(-errno);
14398 }
14399
14400 err = populate_skeleton_maps(s->obj, s->maps, s->map_cnt, s->map_skel_sz);
14401 if (err) {
14402 pr_warn("failed to populate subskeleton maps: %s\n", errstr(err));
14403 return libbpf_err(err);
14404 }
14405
14406 err = populate_skeleton_progs(s->obj, s->progs, s->prog_cnt, s->prog_skel_sz);
14407 if (err) {
14408 pr_warn("failed to populate subskeleton maps: %s\n", errstr(err));
14409 return libbpf_err(err);
14410 }
14411
14412 for (var_idx = 0; var_idx < s->var_cnt; var_idx++) {
14413 var_skel = (void *)s->vars + var_idx * s->var_skel_sz;
14414 map = *var_skel->map;
14415 map_type_id = bpf_map__btf_value_type_id(map);
14416 map_type = btf__type_by_id(btf, map_type_id);
14417
14418 if (!btf_is_datasec(map_type)) {
14419 pr_warn("type for map '%1$s' is not a datasec: %2$s\n",
14420 bpf_map__name(map),
14421 __btf_kind_str(btf_kind(map_type)));
14422 return libbpf_err(-EINVAL);
14423 }
14424
14425 len = btf_vlen(map_type);
14426 var = btf_var_secinfos(map_type);
14427 for (i = 0; i < len; i++, var++) {
14428 var_type = btf__type_by_id(btf, var->type);
14429 var_name = btf__name_by_offset(btf, var_type->name_off);
14430 if (strcmp(var_name, var_skel->name) == 0) {
14431 *var_skel->addr = map->mmaped + var->offset;
14432 break;
14433 }
14434 }
14435 }
14436 return 0;
14437 }
14438
14439 void bpf_object__destroy_subskeleton(struct bpf_object_subskeleton *s)
14440 {
14441 if (!s)
14442 return;
14443 free(s->maps);
14444 free(s->progs);
14445 free(s->vars);
14446 free(s);
14447 }
14448
14449 int bpf_object__load_skeleton(struct bpf_object_skeleton *s)
14450 {
14451 int i, err;
14452
14453 err = bpf_object__load(*s->obj);
14454 if (err) {
14455 pr_warn("failed to load BPF skeleton '%s': %s\n", s->name, errstr(err));
14456 return libbpf_err(err);
14457 }
14458
14459 for (i = 0; i < s->map_cnt; i++) {
14460 struct bpf_map_skeleton *map_skel = (void *)s->maps + i * s->map_skel_sz;
14461 struct bpf_map *map = *map_skel->map;
14462
14463 if (!map_skel->mmaped)
14464 continue;
14465
14466 if (map->def.type == BPF_MAP_TYPE_ARENA)
14467 *map_skel->mmaped = map->mmaped + map->obj->arena_data_off;
14468 else
14469 *map_skel->mmaped = map->mmaped;
14470 }
14471
14472 return 0;
14473 }
14474
14475 int bpf_object__attach_skeleton(struct bpf_object_skeleton *s)
14476 {
14477 int i, err;
14478
14479 for (i = 0; i < s->prog_cnt; i++) {
14480 struct bpf_prog_skeleton *prog_skel = (void *)s->progs + i * s->prog_skel_sz;
14481 struct bpf_program *prog = *prog_skel->prog;
14482 struct bpf_link **link = prog_skel->link;
14483
14484 if (!prog->autoload || !prog->autoattach)
14485 continue;
14486
14487 /* auto-attaching not supported for this program */
14488 if (!prog->sec_def || !prog->sec_def->prog_attach_fn)
14489 continue;
14490
14491 /* if user already set the link manually, don't attempt auto-attach */
14492 if (*link)
14493 continue;
14494
14495 err = prog->sec_def->prog_attach_fn(prog, prog->sec_def->cookie, link);
14496 if (err) {
14497 pr_warn("prog '%s': failed to auto-attach: %s\n",
14498 bpf_program__name(prog), errstr(err));
14499 return libbpf_err(err);
14500 }
14501
14502 /* It's possible that for some SEC() definitions auto-attach
14503 * is supported in some cases (e.g., if definition completely
14504 * specifies target information), but is not in other cases.
14505 * SEC("uprobe") is one such case. If user specified target
14506 * binary and function name, such BPF program can be
14507 * auto-attached. But if not, it shouldn't trigger skeleton's
14508 * attach to fail. It should just be skipped.
14509 * attach_fn signals such case with returning 0 (no error) and
14510 * setting link to NULL.
14511 */
14512 }
14513
14514
14515 for (i = 0; i < s->map_cnt; i++) {
14516 struct bpf_map_skeleton *map_skel = (void *)s->maps + i * s->map_skel_sz;
14517 struct bpf_map *map = *map_skel->map;
14518 struct bpf_link **link;
14519
14520 if (!map->autocreate || !map->autoattach)
14521 continue;
14522
14523 /* only struct_ops maps can be attached */
14524 if (!bpf_map__is_struct_ops(map))
14525 continue;
14526
14527 /* skeleton is created with earlier version of bpftool, notify user */
14528 if (s->map_skel_sz < offsetofend(struct bpf_map_skeleton, link)) {
14529 pr_warn("map '%s': BPF skeleton version is old, skipping map auto-attachment...\n",
14530 bpf_map__name(map));
14531 continue;
14532 }
14533
14534 link = map_skel->link;
14535 if (!link) {
14536 pr_warn("map '%s': BPF map skeleton link is uninitialized\n",
14537 bpf_map__name(map));
14538 continue;
14539 }
14540
14541 if (*link)
14542 continue;
14543
14544 *link = bpf_map__attach_struct_ops(map);
14545 if (!*link) {
14546 err = -errno;
14547 pr_warn("map '%s': failed to auto-attach: %s\n",
14548 bpf_map__name(map), errstr(err));
14549 return libbpf_err(err);
14550 }
14551 }
14552
14553 return 0;
14554 }
14555
14556 void bpf_object__detach_skeleton(struct bpf_object_skeleton *s)
14557 {
14558 int i;
14559
14560 for (i = 0; i < s->prog_cnt; i++) {
14561 struct bpf_prog_skeleton *prog_skel = (void *)s->progs + i * s->prog_skel_sz;
14562 struct bpf_link **link = prog_skel->link;
14563
14564 bpf_link__destroy(*link);
14565 *link = NULL;
14566 }
14567
14568 if (s->map_skel_sz < sizeof(struct bpf_map_skeleton))
14569 return;
14570
14571 for (i = 0; i < s->map_cnt; i++) {
14572 struct bpf_map_skeleton *map_skel = (void *)s->maps + i * s->map_skel_sz;
14573 struct bpf_link **link = map_skel->link;
14574
14575 if (link) {
14576 bpf_link__destroy(*link);
14577 *link = NULL;
14578 }
14579 }
14580 }
14581
14582 void bpf_object__destroy_skeleton(struct bpf_object_skeleton *s)
14583 {
14584 if (!s)
14585 return;
14586
14587 bpf_object__detach_skeleton(s);
14588 if (s->obj)
14589 bpf_object__close(*s->obj);
14590 free(s->maps);
14591 free(s->progs);
14592 free(s);
14593 }