| 1 |
/* eBPF mini library */ |
| 2 |
#include <stdlib.h> |
| 3 |
#include <stdio.h> |
| 4 |
#include <linux/unistd.h> |
| 5 |
#include <unistd.h> |
| 6 |
#include <string.h> |
| 7 |
#include <linux/netlink.h> |
| 8 |
#include <linux/bpf.h> |
| 9 |
#include <errno.h> |
| 10 |
#include <net/ethernet.h> |
| 11 |
#include <net/if.h> |
| 12 |
#include <linux/if_packet.h> |
| 13 |
#include <linux/perf_event.h> |
| 14 |
#include <arpa/inet.h> |
| 15 |
#include "libbpf.h" |
| 16 |
|
| 17 |
/* Older headers might not have this defined yet. */ |
| 18 |
#ifndef __NR_bpf |
| 19 |
# if defined(__i386__) |
| 20 |
# define __NR_bpf 357 |
| 21 |
# elif defined(__x86_64__) |
| 22 |
# define __NR_bpf 321 |
| 23 |
# elif defined(__aarch64__) |
| 24 |
# define __NR_bpf 280 |
| 25 |
# else |
| 26 |
# error __NR_bpf not defined. |
| 27 |
# endif |
| 28 |
#endif |
| 29 |
|
| 30 |
static __u64 ptr_to_u64(void *ptr) |
| 31 |
{ |
| 32 |
return (__u64) (unsigned long) ptr; |
| 33 |
} |
| 34 |
|
| 35 |
int bpf_create_map(enum bpf_map_type map_type, unsigned key_size, |
| 36 |
unsigned value_size, unsigned max_entries, |
| 37 |
unsigned flags __attribute__((unused))) |
| 38 |
{ |
| 39 |
union bpf_attr attr; |
| 40 |
memset(&attr, 0, sizeof(union bpf_attr)); |
| 41 |
attr.map_type = map_type; |
| 42 |
attr.key_size = key_size; |
| 43 |
attr.value_size = value_size; |
| 44 |
attr.max_entries = max_entries; |
| 45 |
|
| 46 |
return syscall(__NR_bpf, BPF_MAP_CREATE, &attr, sizeof(attr)); |
| 47 |
} |
| 48 |
|
| 49 |
int bpf_update_elem(int fd, void *key, void *value, unsigned long long flags) |
| 50 |
{ |
| 51 |
union bpf_attr attr; |
| 52 |
memset(&attr, 0, sizeof(union bpf_attr)); |
| 53 |
attr.map_fd = fd; |
| 54 |
attr.key = ptr_to_u64(key); |
| 55 |
attr.value = ptr_to_u64(value); |
| 56 |
attr.flags = flags; |
| 57 |
|
| 58 |
return syscall(__NR_bpf, BPF_MAP_UPDATE_ELEM, &attr, sizeof(attr)); |
| 59 |
} |
| 60 |
|
| 61 |
int bpf_lookup_elem(int fd, void *key, void *value) |
| 62 |
{ |
| 63 |
union bpf_attr attr; |
| 64 |
memset(&attr, 0, sizeof(union bpf_attr)); |
| 65 |
attr.map_fd = fd; |
| 66 |
attr.key = ptr_to_u64(key); |
| 67 |
attr.value = ptr_to_u64(value); |
| 68 |
|
| 69 |
return syscall(__NR_bpf, BPF_MAP_LOOKUP_ELEM, &attr, sizeof(attr)); |
| 70 |
} |
| 71 |
|
| 72 |
int bpf_delete_elem(int fd, void *key) |
| 73 |
{ |
| 74 |
union bpf_attr attr; |
| 75 |
memset(&attr, 0, sizeof(union bpf_attr)); |
| 76 |
attr.map_fd = fd; |
| 77 |
attr.key = ptr_to_u64(key); |
| 78 |
|
| 79 |
return syscall(__NR_bpf, BPF_MAP_DELETE_ELEM, &attr, sizeof(attr)); |
| 80 |
} |
| 81 |
|
| 82 |
int bpf_get_next_key(int fd, void *key, void *next_key) |
| 83 |
{ |
| 84 |
union bpf_attr attr; |
| 85 |
memset(&attr, 0, sizeof(union bpf_attr)); |
| 86 |
attr.map_fd = fd; |
| 87 |
attr.key = ptr_to_u64(key); |
| 88 |
attr.next_key = ptr_to_u64(next_key); |
| 89 |
|
| 90 |
return syscall(__NR_bpf, BPF_MAP_GET_NEXT_KEY, &attr, sizeof(attr)); |
| 91 |
} |
| 92 |
|
| 93 |
#define ROUND_UP(x, n) (((x) + (n) - 1u) & ~((n) - 1u)) |
| 94 |
|
| 95 |
char bpf_log_buf[LOG_BUF_SIZE]; |
| 96 |
extern int verbose; // set from stapbpf command line |
| 97 |
|
| 98 |
int bpf_prog_load(enum bpf_prog_type prog_type, |
| 99 |
const struct bpf_insn *insns, int prog_len, |
| 100 |
const char *license, int kern_version) |
| 101 |
{ |
| 102 |
union bpf_attr attr; |
| 103 |
memset (&attr, 0, sizeof(attr)); // kernel asserts 0 pad values |
| 104 |
attr.prog_type = prog_type; |
| 105 |
attr.insns = ptr_to_u64((void *) insns); |
| 106 |
attr.insn_cnt = prog_len / sizeof(struct bpf_insn); |
| 107 |
attr.license = ptr_to_u64((void *) license); |
| 108 |
|
| 109 |
/* If the syscall fails, retry with higher verbosity to get |
| 110 |
the eBPF verifier output */ |
| 111 |
int retry = 0; |
| 112 |
do_retry: |
| 113 |
if (verbose || retry) |
| 114 |
{ |
| 115 |
attr.log_buf = ptr_to_u64(bpf_log_buf); |
| 116 |
attr.log_size = LOG_BUF_SIZE; |
| 117 |
attr.log_level = retry ? verbose + 1 : verbose; |
| 118 |
/* they hang together, or they hang separately with -EINVAL */ |
| 119 |
} |
| 120 |
|
| 121 |
/* assign one field outside of struct init to make sure any |
| 122 |
* padding is zero initialized |
| 123 |
*/ |
| 124 |
attr.kern_version = kern_version; |
| 125 |
|
| 126 |
bpf_log_buf[0] = 0; |
| 127 |
|
| 128 |
if (verbose > 1) |
| 129 |
fprintf(stderr, "Loading probe type %d, size %d\n", prog_type, prog_len); |
| 130 |
|
| 131 |
int rc = syscall(__NR_bpf, BPF_PROG_LOAD, &attr, sizeof(attr)); |
| 132 |
if (rc < 0 && verbose == 0 && !retry) |
| 133 |
{ |
| 134 |
retry = 1; goto do_retry; |
| 135 |
} |
| 136 |
return rc; |
| 137 |
} |
| 138 |
|
| 139 |
int bpf_obj_pin(int fd, const char *pathname) |
| 140 |
{ |
| 141 |
union bpf_attr attr; |
| 142 |
memset(&attr, 0, sizeof(union bpf_attr)); |
| 143 |
attr.pathname = ptr_to_u64((void *)pathname); |
| 144 |
attr.bpf_fd = fd; |
| 145 |
|
| 146 |
return syscall(__NR_bpf, BPF_OBJ_PIN, &attr, sizeof(attr)); |
| 147 |
} |
| 148 |
|
| 149 |
// XXX experimental, imitates tools/include/linux/ring_buffer.h and bcc |
| 150 |
static inline __u64 |
| 151 |
ring_buffer_read_head(volatile struct perf_event_mmap_page *base) |
| 152 |
{ |
| 153 |
__u64 head = base->data_head; |
| 154 |
asm volatile ("" ::: "memory"); // memory fence |
| 155 |
return head; |
| 156 |
} |
| 157 |
|
| 158 |
// XXX experimental, imitates tools/include/linux/ring_buffer.h and bcc |
| 159 |
static inline void |
| 160 |
ring_buffer_write_tail(volatile struct perf_event_mmap_page *base, |
| 161 |
__u64 tail) |
| 162 |
{ |
| 163 |
asm volatile("" ::: "memory"); // memory fence |
| 164 |
base->data_tail = tail; |
| 165 |
} |
| 166 |
|
| 167 |
|
| 168 |
enum bpf_perf_event_ret |
| 169 |
bpf_perf_event_read_simple(void *mmap_mem, size_t mmap_size, size_t page_size, |
| 170 |
void **copy_mem, size_t *copy_size, |
| 171 |
bpf_perf_event_print_t fn, void *private_data) |
| 172 |
{ |
| 173 |
struct perf_event_mmap_page *header = mmap_mem; |
| 174 |
__u64 data_head = ring_buffer_read_head(header); |
| 175 |
__u64 data_tail = header->data_tail; |
| 176 |
void *base = ((__u8 *)header) + page_size; |
| 177 |
int ret = LIBBPF_PERF_EVENT_CONT; |
| 178 |
struct perf_event_header *ehdr; |
| 179 |
size_t ehdr_size; |
| 180 |
|
| 181 |
while (data_head != data_tail) { |
| 182 |
ehdr = base + (data_tail & (mmap_size - 1)); |
| 183 |
ehdr_size = ehdr->size; |
| 184 |
|
| 185 |
if (((void *)ehdr) + ehdr_size > base + mmap_size) { |
| 186 |
void *copy_start = ehdr; |
| 187 |
size_t len_first = base + mmap_size - copy_start; |
| 188 |
size_t len_secnd = ehdr_size - len_first; |
| 189 |
|
| 190 |
if (*copy_size < ehdr_size) { |
| 191 |
free(*copy_mem); |
| 192 |
*copy_mem = malloc(ehdr_size); |
| 193 |
if (!*copy_mem) { |
| 194 |
*copy_size = 0; |
| 195 |
ret = LIBBPF_PERF_EVENT_ERROR; |
| 196 |
break; |
| 197 |
} |
| 198 |
*copy_size = ehdr_size; |
| 199 |
} |
| 200 |
|
| 201 |
memcpy(*copy_mem, copy_start, len_first); |
| 202 |
memcpy(*copy_mem + len_first, base, len_secnd); |
| 203 |
ehdr = *copy_mem; |
| 204 |
} |
| 205 |
|
| 206 |
ret = fn(ehdr, private_data); |
| 207 |
data_tail += ehdr_size; |
| 208 |
if (ret != LIBBPF_PERF_EVENT_CONT) |
| 209 |
break; |
| 210 |
} |
| 211 |
|
| 212 |
ring_buffer_write_tail(header, data_tail); |
| 213 |
return ret; |
| 214 |
} |
| 215 |
|
| 216 |
int bpf_obj_get(const char *pathname) |
| 217 |
{ |
| 218 |
union bpf_attr attr; |
| 219 |
memset(&attr, 0, sizeof(union bpf_attr)); |
| 220 |
attr.pathname = ptr_to_u64((void *)pathname); |
| 221 |
|
| 222 |
return syscall(__NR_bpf, BPF_OBJ_GET, &attr, sizeof(attr)); |
| 223 |
} |
| 224 |
|
| 225 |
int open_raw_sock(const char *name) |
| 226 |
{ |
| 227 |
struct sockaddr_ll sll; |
| 228 |
int sock; |
| 229 |
|
| 230 |
sock = socket(PF_PACKET, SOCK_RAW | SOCK_NONBLOCK | SOCK_CLOEXEC, htons(ETH_P_ALL)); |
| 231 |
if (sock < 0) { |
| 232 |
printf("cannot create raw socket\n"); |
| 233 |
return -1; |
| 234 |
} |
| 235 |
|
| 236 |
memset(&sll, 0, sizeof(sll)); |
| 237 |
sll.sll_family = AF_PACKET; |
| 238 |
sll.sll_ifindex = if_nametoindex(name); |
| 239 |
sll.sll_protocol = htons(ETH_P_ALL); |
| 240 |
if (bind(sock, (struct sockaddr *)&sll, sizeof(sll)) < 0) { |
| 241 |
printf("bind to %s: %s\n", name, strerror(errno)); |
| 242 |
close(sock); |
| 243 |
return -1; |
| 244 |
} |
| 245 |
|
| 246 |
return sock; |
| 247 |
} |
| 248 |
|
| 249 |
int perf_event_open(struct perf_event_attr *attr, int pid, int cpu, |
| 250 |
int group_fd, unsigned long flags) |
| 251 |
{ |
| 252 |
return syscall(__NR_perf_event_open, attr, pid, cpu, |
| 253 |
group_fd, flags); |
| 254 |
} |