Debian Codegraph @

src/shared/hibernate-util.c

raw
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2 /***
3 Copyright © 2018 Dell Inc.
4 ***/
5
6 #include <linux/fiemap.h>
7 #include <linux/fs.h>
8 #include <linux/magic.h>
9 #include <sys/ioctl.h>
10
11 #include "alloc-util.h"
12 #include "blockdev-util.h"
13 #include "btrfs-util.h"
14 #include "devnum-util.h"
15 #include "efivars.h"
16 #include "env-util.h"
17 #include "fd-util.h"
18 #include "fileio.h"
19 #include "hibernate-util.h"
20 #include "log.h"
21 #include "parse-util.h"
22 #include "path-util.h"
23 #include "proc-cmdline.h"
24 #include "stat-util.h"
25
26 #define HIBERNATION_SWAP_THRESHOLD 0.98
27
28 void hibernation_device_done(HibernationDevice *device) {
29 assert(device);
30
31 free(device->path);
32 }
33
34 int read_fiemap(int fd, struct fiemap **ret) {
35 _cleanup_free_ struct fiemap *fiemap = NULL, *result_fiemap = NULL;
36 struct stat statinfo;
37 uint32_t result_extents = 0;
38 uint64_t fiemap_start = 0, fiemap_length;
39 const size_t n_extra = DIV_ROUND_UP(sizeof(struct fiemap), sizeof(struct fiemap_extent));
40 int r;
41
42 assert(fd >= 0);
43 assert(ret);
44
45 if (fstat(fd, &statinfo) < 0)
46 return log_debug_errno(errno, "Cannot determine file size: %m");
47 r = stat_verify_regular(&statinfo);
48 if (r < 0)
49 return r;
50 fiemap_length = statinfo.st_size;
51
52 /* Zero this out in case we run on a file with no extents */
53 fiemap = calloc(n_extra, sizeof(struct fiemap_extent));
54 if (!fiemap)
55 return -ENOMEM;
56
57 result_fiemap = malloc_multiply(n_extra, sizeof(struct fiemap_extent));
58 if (!result_fiemap)
59 return -ENOMEM;
60
61 /* XFS filesystem has incorrect implementation of fiemap ioctl and
62 * returns extents for only one block-group at a time, so we need
63 * to handle it manually, starting the next fiemap call from the end
64 * of the last extent. */
65 while (fiemap_start < fiemap_length) {
66 *fiemap = (struct fiemap) {
67 .fm_start = fiemap_start,
68 .fm_length = fiemap_length,
69 .fm_flags = FIEMAP_FLAG_SYNC,
70 };
71
72 /* Find out how many extents there are */
73 if (ioctl(fd, FS_IOC_FIEMAP, fiemap) < 0)
74 return log_debug_errno(errno, "Failed to read extents: %m");
75
76 /* Nothing to process */
77 if (fiemap->fm_mapped_extents == 0)
78 break;
79
80 /* Resize fiemap to allow us to read in the extents, result fiemap has to hold all
81 * the extents for the whole file. Add space for the initial struct fiemap. */
82 if (!greedy_realloc0((void**) &fiemap, n_extra + fiemap->fm_mapped_extents, sizeof(struct fiemap_extent)))
83 return -ENOMEM;
84
85 fiemap->fm_extent_count = fiemap->fm_mapped_extents;
86 fiemap->fm_mapped_extents = 0;
87
88 if (ioctl(fd, FS_IOC_FIEMAP, fiemap) < 0)
89 return log_debug_errno(errno, "Failed to read extents: %m");
90
91 /* Resize result_fiemap to allow us to copy in the extents */
92 if (!greedy_realloc((void**) &result_fiemap,
93 n_extra + result_extents + fiemap->fm_mapped_extents, sizeof(struct fiemap_extent)))
94 return -ENOMEM;
95
96 memcpy(result_fiemap->fm_extents + result_extents,
97 fiemap->fm_extents,
98 sizeof(struct fiemap_extent) * fiemap->fm_mapped_extents);
99
100 result_extents += fiemap->fm_mapped_extents;
101
102 /* Highly unlikely that it is zero */
103 if (_likely_(fiemap->fm_mapped_extents > 0)) {
104 uint32_t i = fiemap->fm_mapped_extents - 1;
105
106 fiemap_start = fiemap->fm_extents[i].fe_logical +
107 fiemap->fm_extents[i].fe_length;
108
109 if (fiemap->fm_extents[i].fe_flags & FIEMAP_EXTENT_LAST)
110 break;
111 }
112 }
113
114 memcpy(result_fiemap, fiemap, sizeof(struct fiemap));
115 result_fiemap->fm_mapped_extents = result_extents;
116 *ret = TAKE_PTR(result_fiemap);
117 return 0;
118 }
119
120 static int read_resume_config(dev_t *ret_devno, uint64_t *ret_offset) {
121 _cleanup_free_ char *devno_str = NULL, *offset_str = NULL;
122 uint64_t offset;
123 dev_t devno;
124 int r;
125
126 assert(ret_devno);
127 assert(ret_offset);
128
129 r = proc_cmdline_get_key("noresume", /* flags= */ 0, /* ret_value= */ NULL);
130 if (r < 0)
131 return log_debug_errno(r, "Failed to check if 'noresume' kernel command line option is set: %m");
132 if (r > 0)
133 return log_debug_errno(SYNTHETIC_ERRNO(ENOTRECOVERABLE),
134 "'noresume' kernel command line option is set, refusing hibernation device lookup.");
135
136 r = read_one_line_file("/sys/power/resume", &devno_str);
137 if (r < 0)
138 return log_debug_errno(r, "Failed to read /sys/power/resume: %m");
139
140 r = parse_devnum(devno_str, &devno);
141 if (r < 0)
142 return log_debug_errno(r, "Failed to parse /sys/power/resume devno '%s': %m", devno_str);
143
144 r = read_one_line_file("/sys/power/resume_offset", &offset_str);
145 if (r < 0)
146 return log_debug_errno(r, "Failed to read /sys/power/resume_offset: %m");
147
148 r = safe_atou64(offset_str, &offset);
149 if (r < 0)
150 return log_debug_errno(r, "Failed to parse /sys/power/resume_offset '%s': %m", offset_str);
151
152 if (devno == 0 && offset > 0)
153 return log_debug_errno(SYNTHETIC_ERRNO(ENOMEDIUM),
154 "Found populated /sys/power/resume_offset (%" PRIu64 ") but /sys/power/resume is not set, refusing.",
155 offset);
156
157 *ret_devno = devno;
158 *ret_offset = offset;
159
160 return 0;
161 }
162
163 /* entry in /proc/swaps */
164 typedef struct SwapEntry {
165 char *path;
166 bool swapfile;
167
168 uint64_t size;
169 uint64_t used;
170 int priority;
171
172 /* Not present in original entry */
173 dev_t devno;
174 uint64_t offset;
175 } SwapEntry;
176
177 typedef struct SwapEntries {
178 SwapEntry *swaps;
179 size_t n_swaps;
180 } SwapEntries;
181
182 static void swap_entry_done(SwapEntry *entry) {
183 assert(entry);
184
185 free(entry->path);
186 }
187
188 static void swap_entries_done(SwapEntries *entries) {
189 assert(entries);
190
191 FOREACH_ARRAY(i, entries->swaps, entries->n_swaps)
192 swap_entry_done(i);
193
194 free(entries->swaps);
195 }
196
197 static int swap_entry_get_resume_config(SwapEntry *swap) {
198 _cleanup_close_ int fd = -EBADF;
199 uint64_t offset_raw;
200 struct stat st;
201 int r;
202
203 assert(swap);
204 assert(swap->path);
205
206 fd = open(swap->path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
207 if (fd < 0)
208 return -errno;
209
210 if (fstat(fd, &st) < 0)
211 return -errno;
212
213 if (!swap->swapfile) {
214 r = stat_verify_block(&st);
215 if (r < 0)
216 return r;
217
218 swap->devno = st.st_rdev;
219 swap->offset = 0;
220 return 0;
221 }
222
223 r = stat_verify_regular(&st);
224 if (r < 0)
225 return r;
226
227 r = get_block_device_fd(fd, &swap->devno);
228 if (r < 0)
229 return r;
230 if (r == 0)
231 return -EMEDIUMTYPE;
232
233 r = fd_is_fs_type(fd, BTRFS_SUPER_MAGIC);
234 if (r < 0)
235 return log_debug_errno(r, "Failed to check if swap file '%s' is on Btrfs: %m", swap->path);
236 if (r > 0) {
237 r = btrfs_get_file_physical_offset_fd(fd, &offset_raw);
238 if (r < 0)
239 return r;
240 } else {
241 _cleanup_free_ struct fiemap *fiemap = NULL;
242
243 r = read_fiemap(fd, &fiemap);
244 if (r < 0)
245 return log_debug_errno(r, "Failed to read extent map for swap file '%s': %m", swap->path);
246
247 offset_raw = fiemap->fm_extents[0].fe_physical;
248 }
249
250 swap->offset = offset_raw / page_size();
251 return 0;
252 }
253
254 static int read_swap_entries(SwapEntries *ret) {
255 _cleanup_(swap_entries_done) SwapEntries entries = {};
256 _cleanup_fclose_ FILE *f = NULL;
257
258 assert(ret);
259
260 f = fopen("/proc/swaps", "re");
261 if (!f)
262 return log_debug_errno(errno, "Failed to open %s: %m", "/proc/swaps");
263
264 /* Remove header */
265 (void) fscanf(f, "%*s %*s %*s %*s %*s\n");
266
267 for (unsigned i = 1;; i++) {
268 _cleanup_(swap_entry_done) SwapEntry swap = {};
269 _cleanup_free_ char *type = NULL;
270 int k;
271
272 k = fscanf(f,
273 "%ms " /* device/file path */
274 "%ms " /* type of swap */
275 "%" PRIu64 /* swap size */
276 "%" PRIu64 /* used */
277 "%i" /* priority */
278 "\n",
279 &swap.path, &type, &swap.size, &swap.used, &swap.priority);
280 if (k == EOF)
281 break;
282 if (k != 5)
283 return log_debug_errno(SYNTHETIC_ERRNO(EIO), "Failed to parse /proc/swaps line %u.", i);
284
285 if (streq(type, "file")) {
286 if (endswith(swap.path, "\\040(deleted)")) {
287 log_debug("Swap file '%s' has been deleted, ignoring.", swap.path);
288 continue;
289 }
290
291 swap.swapfile = true;
292
293 } else if (streq(type, "partition")) {
294 const char *node;
295
296 node = path_startswith(swap.path, "/dev/");
297 if (node && startswith(node, "zram")) {
298 log_debug("Swap partition '%s' is a zram device, ignoring.", swap.path);
299 continue;
300 }
301
302 swap.swapfile = false;
303
304 } else {
305 log_debug("Swap type %s is not supported for hibernation, ignoring device: %s",
306 type, swap.path);
307 continue;
308 }
309
310 if (!GREEDY_REALLOC(entries.swaps, entries.n_swaps + 1))
311 return log_oom_debug();
312
313 entries.swaps[entries.n_swaps++] = TAKE_STRUCT(swap);
314 }
315
316 *ret = TAKE_STRUCT(entries);
317 return 0;
318 }
319
320 /* Attempt to find a suitable device for hibernation by parsing /proc/swaps, /sys/power/resume, and
321 * /sys/power/resume_offset.
322 *
323 * Beware:
324 * Never use a device or file that hasn't been somehow specified by a user who would also be entrusted
325 * with full system memory access (for example via /sys/power/resume) or that isn't an already active
326 * swap area! Otherwise various security attacks might become possible, for example an attacker could
327 * silently attach such a device and circumvent full disk encryption when it would be automatically used
328 * for hibernation. Also, having a swap area on top of encryption is not per se enough to protect from all
329 * such attacks.
330 *
331 * Returns:
332 * 1 - Values are set in /sys/power/resume and /sys/power/resume_offset.
333 *
334 * 0 - No values are set in /sys/power/resume and /sys/power/resume_offset.
335 * ret will represent the highest priority swap with most remaining space discovered in /proc/swaps.
336 *
337 * Negative value in the case of error */
338 int find_suitable_hibernation_device_full(HibernationDevice *ret_device, uint64_t *ret_size, uint64_t *ret_used) {
339 _cleanup_(swap_entries_done) SwapEntries entries = {};
340 SwapEntry *entry = NULL;
341 uint64_t resume_config_offset;
342 dev_t resume_config_devno;
343 int r;
344
345 assert(!ret_size == !ret_used);
346
347 r = read_resume_config(&resume_config_devno, &resume_config_offset);
348 if (r < 0)
349 return r;
350
351 r = read_swap_entries(&entries);
352 if (r < 0)
353 return r;
354
355 FOREACH_ARRAY(swap, entries.swaps, entries.n_swaps) {
356 r = swap_entry_get_resume_config(swap);
357 if (r == -EMEDIUMTYPE) {
358 assert(swap->swapfile);
359
360 log_debug_errno(r, "Unable to acquire backing block device for swap file '%s' (maybe on a RAID btrfs?), ignoring.",
361 swap->path);
362 continue;
363 }
364 if (r < 0)
365 return log_debug_errno(r, "Failed to get devno and offset for swap '%s': %m", swap->path);
366 assert(swap->devno > 0);
367
368 if (resume_config_devno > 0) {
369 if (swap->devno == resume_config_devno &&
370 (!swap->swapfile || swap->offset == resume_config_offset)) {
371 /* /sys/power/resume (resume=) is set, and the calculated swap file offset
372 * matches with /sys/power/resume_offset. */
373 entry = swap;
374 break;
375 }
376
377 /* If resume= is set, don't try to use other swap spaces. */
378 continue;
379 }
380
381 if (!entry ||
382 swap->priority > entry->priority ||
383 swap->size - swap->used > entry->size - entry->used)
384 entry = swap;
385 }
386
387 if (!entry) {
388 if (resume_config_devno > 0)
389 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Cannot find swap entry corresponding to /sys/power/resume.");
390
391 return log_debug_errno(SYNTHETIC_ERRNO(ENOSPC), "No swap space available for hibernation.");
392 }
393
394 if (ret_device) {
395 char *path;
396
397 if (entry->swapfile) {
398 r = device_path_make_canonical(S_IFBLK, entry->devno, &path);
399 if (r < 0)
400 return log_debug_errno(r,
401 "Failed to format canonical device path for devno '" DEVNUM_FORMAT_STR "': %m",
402 DEVNUM_FORMAT_VAL(entry->devno));
403 } else
404 path = TAKE_PTR(entry->path);
405
406 *ret_device = (HibernationDevice) {
407 .devno = entry->devno,
408 .offset = entry->offset,
409 .path = path,
410 };
411 }
412
413 if (ret_size) {
414 *ret_size = entry->size;
415 *ret_used = entry->used;
416 }
417
418 return resume_config_devno > 0;
419 }
420
421 static int get_proc_meminfo_active(unsigned long long *ret) {
422 _cleanup_free_ char *active_str = NULL;
423 unsigned long long active;
424 int r;
425
426 assert(ret);
427
428 r = get_proc_field("/proc/meminfo", "Active(anon)", &active_str);
429 if (r < 0)
430 return log_debug_errno(r, "Failed to retrieve Active(anon) from /proc/meminfo: %m");
431
432 r = safe_atollu(active_str, &active);
433 if (r < 0)
434 return log_debug_errno(r, "Failed to parse Active(anon) '%s' from /proc/meminfo: %m", active_str);
435
436 *ret = active;
437 return 0;
438 }
439
440 int hibernation_is_safe(void) {
441 unsigned long long active;
442 uint64_t size, used;
443 bool resume_set, bypass_space_check;
444 int r;
445
446 bypass_space_check = getenv_bool("SYSTEMD_BYPASS_HIBERNATION_MEMORY_CHECK") > 0;
447
448 r = find_suitable_hibernation_device_full(NULL, &size, &used);
449 if (IN_SET(r, -ENOSPC, -ESTALE) && bypass_space_check)
450 /* If we don't have any available swap space at all, or the specified resume device is missing,
451 * and $SYSTEMD_BYPASS_HIBERNATION_MEMORY_CHECK is set, skip all remaining checks since
452 * we can't do that properly anyway. It is quite possible that the user is using a setup
453 * similar to #30083. When we actually perform hibernation in sleep.c we'll check everything again. */
454 return 0;
455 if (r < 0)
456 return r;
457 resume_set = r > 0;
458
459 if (!resume_set && !is_efi_boot())
460 return log_debug_errno(SYNTHETIC_ERRNO(ENOTRECOVERABLE),
461 "Not running on EFI and resume= is not set. Hibernation is not safe.");
462
463 if (bypass_space_check)
464 return 0;
465
466 r = get_proc_meminfo_active(&active);
467 if (r < 0)
468 return r;
469
470 r = active <= (size - used) * HIBERNATION_SWAP_THRESHOLD;
471 log_debug("Detected %s swap for hibernation: Active(anon)=%llu kB, size=%" PRIu64 " kB, used=%" PRIu64 " kB, threshold=%.2g%%",
472 r ? "enough" : "not enough", active, size, used, 100 * HIBERNATION_SWAP_THRESHOLD);
473 if (!r)
474 return -ENOSPC;
475
476 return resume_set;
477 }
478
479 int write_resume_config(dev_t devno, uint64_t offset, const char *device) {
480 char offset_str[DECIMAL_STR_MAX(uint64_t)];
481 const char *devno_str;
482 int r;
483
484 assert(devno > 0);
485 assert(device);
486
487 devno_str = FORMAT_DEVNUM(devno);
488 xsprintf(offset_str, "%" PRIu64, offset);
489
490 r = write_string_file("/sys/power/resume_offset", offset_str, WRITE_STRING_FILE_DISABLE_BUFFER);
491 if (r < 0)
492 return log_error_errno(r, "Failed to write swap file offset %s to /sys/power/resume_offset for device '%s': %m",
493 offset_str, device);
494 log_debug("Wrote resume_offset=%s for device '%s' to /sys/power/resume_offset.",
495 offset_str, device);
496
497 r = write_string_file("/sys/power/resume", devno_str, WRITE_STRING_FILE_DISABLE_BUFFER);
498 if (r < 0)
499 return log_error_errno(r, "Failed to write device '%s' (%s) to /sys/power/resume: %m",
500 device, devno_str);
501 log_debug("Wrote resume=%s for device '%s' to /sys/power/resume.", devno_str, device);
502
503 return 0;
504 }
505
506 int clear_efi_hibernate_location_and_warn(void) {
507 int r;
508
509 if (!is_efi_boot())
510 return 0;
511
512 r = efi_set_variable(EFI_SYSTEMD_VARIABLE_STR("HibernateLocation"), NULL, 0);
513 if (r == -ENOENT)
514 return 0;
515 if (r < 0)
516 return log_warning_errno(r, "Failed to clear HibernateLocation EFI variable: %m");
517
518 return 1;
519 }