]>
Commit | Line | Data |
---|---|---|
959ef981 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
b364a9c0 DW |
2 | /* |
3 | * Copyright (C) 2018 Oracle. All Rights Reserved. | |
b364a9c0 | 4 | * Author: Darrick J. Wong <darrick.wong@oracle.com> |
b364a9c0 | 5 | */ |
a440f877 | 6 | #include "xfs.h" |
b364a9c0 | 7 | #include <stdint.h> |
b364a9c0 DW |
8 | #include <dirent.h> |
9 | #include <sys/statvfs.h> | |
b364a9c0 | 10 | #include "handle.h" |
42b4c8e8 | 11 | #include "libfrog/paths.h" |
56598728 | 12 | #include "libfrog/workqueue.h" |
b364a9c0 DW |
13 | #include "xfs_scrub.h" |
14 | #include "common.h" | |
a58400ed | 15 | #include "libfrog/bitmap.h" |
b364a9c0 DW |
16 | #include "disk.h" |
17 | #include "filemap.h" | |
ed60d210 | 18 | #include "fscounters.h" |
b364a9c0 DW |
19 | #include "inodes.h" |
20 | #include "read_verify.h" | |
21 | #include "spacemap.h" | |
22 | #include "vfs.h" | |
23 | ||
24 | /* | |
25 | * Phase 6: Verify data file integrity. | |
26 | * | |
27 | * Identify potential data block extents with GETFSMAP, then feed those | |
28 | * extents to the read-verify pool to get the verify commands batched, | |
29 | * issued, and (if there are problems) reported back to us. If there | |
30 | * are errors, we'll record the bad regions and (if available) use rmap | |
31 | * to tell us if metadata are now corrupt. Otherwise, we'll scan the | |
32 | * whole directory tree looking for files that overlap the bad regions | |
33 | * and report the paths of the now corrupt files. | |
34 | */ | |
35 | ||
f1bb1696 DW |
36 | /* Verify disk blocks with GETFSMAP */ |
37 | ||
557f98d7 | 38 | struct media_verify_state { |
f1bb1696 DW |
39 | struct read_verify_pool *rvp_data; |
40 | struct read_verify_pool *rvp_log; | |
41 | struct read_verify_pool *rvp_realtime; | |
42 | struct bitmap *d_bad; /* bytes */ | |
43 | struct bitmap *r_bad; /* bytes */ | |
44 | }; | |
45 | ||
b364a9c0 | 46 | /* Find the fd for a given device identifier. */ |
f1bb1696 DW |
47 | static struct read_verify_pool * |
48 | xfs_dev_to_pool( | |
49 | struct scrub_ctx *ctx, | |
557f98d7 | 50 | struct media_verify_state *vs, |
f1bb1696 | 51 | dev_t dev) |
b364a9c0 DW |
52 | { |
53 | if (dev == ctx->fsinfo.fs_datadev) | |
557f98d7 | 54 | return vs->rvp_data; |
b364a9c0 | 55 | else if (dev == ctx->fsinfo.fs_logdev) |
557f98d7 | 56 | return vs->rvp_log; |
b364a9c0 | 57 | else if (dev == ctx->fsinfo.fs_rtdev) |
557f98d7 | 58 | return vs->rvp_realtime; |
b364a9c0 DW |
59 | abort(); |
60 | } | |
61 | ||
62 | /* Find the device major/minor for a given file descriptor. */ | |
63 | static dev_t | |
64 | xfs_disk_to_dev( | |
65 | struct scrub_ctx *ctx, | |
66 | struct disk *disk) | |
67 | { | |
68 | if (disk == ctx->datadev) | |
69 | return ctx->fsinfo.fs_datadev; | |
70 | else if (disk == ctx->logdev) | |
71 | return ctx->fsinfo.fs_logdev; | |
72 | else if (disk == ctx->rtdev) | |
73 | return ctx->fsinfo.fs_rtdev; | |
74 | abort(); | |
75 | } | |
76 | ||
c9b349bd DW |
77 | /* Find the incore bad blocks bitmap for a given disk. */ |
78 | static struct bitmap * | |
79 | bitmap_for_disk( | |
80 | struct scrub_ctx *ctx, | |
81 | struct disk *disk, | |
82 | struct media_verify_state *vs) | |
83 | { | |
84 | dev_t dev = xfs_disk_to_dev(ctx, disk); | |
85 | ||
86 | if (dev == ctx->fsinfo.fs_datadev) | |
87 | return vs->d_bad; | |
88 | else if (dev == ctx->fsinfo.fs_rtdev) | |
89 | return vs->r_bad; | |
90 | return NULL; | |
91 | } | |
92 | ||
93 | struct disk_ioerr_report { | |
94 | struct scrub_ctx *ctx; | |
95 | struct disk *disk; | |
96 | }; | |
97 | ||
b364a9c0 DW |
98 | struct owner_decode { |
99 | uint64_t owner; | |
100 | const char *descr; | |
101 | }; | |
102 | ||
103 | static const struct owner_decode special_owners[] = { | |
104 | {XFS_FMR_OWN_FREE, "free space"}, | |
105 | {XFS_FMR_OWN_UNKNOWN, "unknown owner"}, | |
106 | {XFS_FMR_OWN_FS, "static FS metadata"}, | |
107 | {XFS_FMR_OWN_LOG, "journalling log"}, | |
108 | {XFS_FMR_OWN_AG, "per-AG metadata"}, | |
109 | {XFS_FMR_OWN_INOBT, "inode btree blocks"}, | |
110 | {XFS_FMR_OWN_INODES, "inodes"}, | |
111 | {XFS_FMR_OWN_REFC, "refcount btree"}, | |
112 | {XFS_FMR_OWN_COW, "CoW staging"}, | |
113 | {XFS_FMR_OWN_DEFECTIVE, "bad blocks"}, | |
114 | {0, NULL}, | |
115 | }; | |
116 | ||
117 | /* Decode a special owner. */ | |
118 | static const char * | |
119 | xfs_decode_special_owner( | |
120 | uint64_t owner) | |
121 | { | |
122 | const struct owner_decode *od = special_owners; | |
123 | ||
124 | while (od->descr) { | |
125 | if (od->owner == owner) | |
126 | return od->descr; | |
127 | od++; | |
128 | } | |
129 | ||
130 | return NULL; | |
131 | } | |
132 | ||
133 | /* Routines to translate bad physical extents into file paths and offsets. */ | |
134 | ||
ed953d26 | 135 | struct badfile_report { |
73ce9669 DW |
136 | struct scrub_ctx *ctx; |
137 | const char *descr; | |
138 | struct media_verify_state *vs; | |
139 | struct file_bmap *bmap; | |
ed953d26 DW |
140 | }; |
141 | ||
142 | /* Report on bad extents found during a media scan. */ | |
143 | static int | |
144 | report_badfile( | |
145 | uint64_t start, | |
146 | uint64_t length, | |
147 | void *arg) | |
148 | { | |
149 | struct badfile_report *br = arg; | |
150 | unsigned long long bad_offset; | |
151 | unsigned long long bad_length; | |
152 | ||
153 | /* Clamp the bad region to the file mapping. */ | |
154 | if (start < br->bmap->bm_physical) { | |
155 | length -= br->bmap->bm_physical - start; | |
156 | start = br->bmap->bm_physical; | |
157 | } | |
158 | length = min(length, br->bmap->bm_length); | |
159 | ||
160 | /* Figure out how far into the bmap is the bad mapping and report it. */ | |
161 | bad_offset = start - br->bmap->bm_physical; | |
162 | bad_length = min(start + length, | |
163 | br->bmap->bm_physical + br->bmap->bm_length) - start; | |
164 | ||
49e05cb0 | 165 | str_unfixable_error(br->ctx, br->descr, |
ed953d26 DW |
166 | _("media error at data offset %llu length %llu."), |
167 | br->bmap->bm_offset + bad_offset, bad_length); | |
168 | return 0; | |
169 | } | |
170 | ||
b364a9c0 | 171 | /* Report if this extent overlaps a bad region. */ |
73ce9669 | 172 | static int |
663e02a0 | 173 | report_data_loss( |
b364a9c0 | 174 | struct scrub_ctx *ctx, |
b364a9c0 DW |
175 | int fd, |
176 | int whichfork, | |
177 | struct fsxattr *fsx, | |
73ce9669 | 178 | struct file_bmap *bmap, |
b364a9c0 DW |
179 | void *arg) |
180 | { | |
73ce9669 DW |
181 | struct badfile_report *br = arg; |
182 | struct media_verify_state *vs = br->vs; | |
b364a9c0 | 183 | struct bitmap *bmp; |
73ce9669 DW |
184 | |
185 | br->bmap = bmap; | |
b364a9c0 DW |
186 | |
187 | /* Only report errors for real extents. */ | |
188 | if (bmap->bm_flags & (BMV_OF_PREALLOC | BMV_OF_DELALLOC)) | |
73ce9669 | 189 | return 0; |
b364a9c0 DW |
190 | |
191 | if (fsx->fsx_xflags & FS_XFLAG_REALTIME) | |
ed5f9cc7 | 192 | bmp = vs->r_bad; |
b364a9c0 | 193 | else |
ed5f9cc7 | 194 | bmp = vs->d_bad; |
b364a9c0 | 195 | |
73ce9669 DW |
196 | return bitmap_iterate_range(bmp, bmap->bm_physical, bmap->bm_length, |
197 | report_badfile, br); | |
b364a9c0 DW |
198 | } |
199 | ||
663e02a0 | 200 | /* Report if the extended attribute data overlaps a bad region. */ |
73ce9669 | 201 | static int |
663e02a0 DW |
202 | report_attr_loss( |
203 | struct scrub_ctx *ctx, | |
663e02a0 DW |
204 | int fd, |
205 | int whichfork, | |
206 | struct fsxattr *fsx, | |
73ce9669 | 207 | struct file_bmap *bmap, |
663e02a0 DW |
208 | void *arg) |
209 | { | |
73ce9669 DW |
210 | struct badfile_report *br = arg; |
211 | struct media_verify_state *vs = br->vs; | |
663e02a0 DW |
212 | struct bitmap *bmp = vs->d_bad; |
213 | ||
214 | /* Complain about attr fork extents that don't look right. */ | |
215 | if (bmap->bm_flags & (BMV_OF_PREALLOC | BMV_OF_DELALLOC)) { | |
73ce9669 | 216 | str_info(ctx, br->descr, |
663e02a0 | 217 | _("found unexpected unwritten/delalloc attr fork extent.")); |
73ce9669 | 218 | return 0; |
663e02a0 DW |
219 | } |
220 | ||
221 | if (fsx->fsx_xflags & FS_XFLAG_REALTIME) { | |
73ce9669 | 222 | str_info(ctx, br->descr, |
663e02a0 | 223 | _("found unexpected realtime attr fork extent.")); |
73ce9669 | 224 | return 0; |
663e02a0 DW |
225 | } |
226 | ||
227 | if (bitmap_test(bmp, bmap->bm_physical, bmap->bm_length)) | |
73ce9669 | 228 | str_corrupt(ctx, br->descr, |
663e02a0 DW |
229 | _("media error in extended attribute data.")); |
230 | ||
73ce9669 | 231 | return 0; |
663e02a0 DW |
232 | } |
233 | ||
b364a9c0 DW |
234 | /* Iterate the extent mappings of a file to report errors. */ |
235 | static bool | |
236 | xfs_report_verify_fd( | |
237 | struct scrub_ctx *ctx, | |
238 | const char *descr, | |
239 | int fd, | |
240 | void *arg) | |
241 | { | |
73ce9669 DW |
242 | struct badfile_report br = { |
243 | .ctx = ctx, | |
244 | .vs = arg, | |
245 | .descr = descr, | |
246 | }; | |
247 | struct file_bmap key = {0}; | |
248 | int ret; | |
b364a9c0 DW |
249 | |
250 | /* data fork */ | |
73ce9669 DW |
251 | ret = scrub_iterate_filemaps(ctx, fd, XFS_DATA_FORK, &key, |
252 | report_data_loss, &br); | |
253 | if (ret) { | |
254 | str_liberror(ctx, ret, descr); | |
b364a9c0 | 255 | return false; |
73ce9669 | 256 | } |
b364a9c0 DW |
257 | |
258 | /* attr fork */ | |
73ce9669 DW |
259 | ret = scrub_iterate_filemaps(ctx, fd, XFS_ATTR_FORK, &key, |
260 | report_attr_loss, &br); | |
261 | if (ret) { | |
262 | str_liberror(ctx, ret, descr); | |
263 | return false; | |
264 | } | |
265 | return true; | |
b364a9c0 DW |
266 | } |
267 | ||
268 | /* Report read verify errors in unlinked (but still open) files. */ | |
269 | static int | |
270 | xfs_report_verify_inode( | |
271 | struct scrub_ctx *ctx, | |
272 | struct xfs_handle *handle, | |
4cca629d | 273 | struct xfs_bulkstat *bstat, |
b364a9c0 DW |
274 | void *arg) |
275 | { | |
276 | char descr[DESCR_BUFSZ]; | |
b364a9c0 DW |
277 | bool moveon; |
278 | int fd; | |
279 | int error; | |
280 | ||
b364a9c0 DW |
281 | /* Ignore linked files and things we can't open. */ |
282 | if (bstat->bs_nlink != 0) | |
283 | return 0; | |
284 | if (!S_ISREG(bstat->bs_mode) && !S_ISDIR(bstat->bs_mode)) | |
285 | return 0; | |
286 | ||
15589f0a DW |
287 | scrub_render_ino_descr(ctx, descr, DESCR_BUFSZ, |
288 | bstat->bs_ino, bstat->bs_gen, _("(unlinked)")); | |
289 | ||
b364a9c0 DW |
290 | /* Try to open the inode. */ |
291 | fd = xfs_open_handle(handle); | |
292 | if (fd < 0) { | |
293 | error = errno; | |
294 | if (error == ESTALE) | |
295 | return error; | |
296 | ||
bb5dbd06 DW |
297 | str_info(ctx, descr, |
298 | _("Disappeared during read error reporting.")); | |
b364a9c0 DW |
299 | return error; |
300 | } | |
301 | ||
302 | /* Go find the badness. */ | |
303 | moveon = xfs_report_verify_fd(ctx, descr, fd, arg); | |
6c05cc5d DW |
304 | error = close(fd); |
305 | if (error) | |
306 | str_errno(ctx, descr); | |
b364a9c0 DW |
307 | |
308 | return moveon ? 0 : XFS_ITERATE_INODES_ABORT; | |
309 | } | |
310 | ||
311 | /* Scan a directory for matches in the read verify error list. */ | |
312 | static bool | |
313 | xfs_report_verify_dir( | |
314 | struct scrub_ctx *ctx, | |
315 | const char *path, | |
316 | int dir_fd, | |
317 | void *arg) | |
318 | { | |
319 | return xfs_report_verify_fd(ctx, path, dir_fd, arg); | |
320 | } | |
321 | ||
322 | /* | |
323 | * Scan the inode associated with a directory entry for matches with | |
324 | * the read verify error list. | |
325 | */ | |
326 | static bool | |
327 | xfs_report_verify_dirent( | |
328 | struct scrub_ctx *ctx, | |
329 | const char *path, | |
330 | int dir_fd, | |
331 | struct dirent *dirent, | |
332 | struct stat *sb, | |
333 | void *arg) | |
334 | { | |
335 | bool moveon; | |
336 | int fd; | |
6c05cc5d | 337 | int error; |
b364a9c0 DW |
338 | |
339 | /* Ignore things we can't open. */ | |
340 | if (!S_ISREG(sb->st_mode) && !S_ISDIR(sb->st_mode)) | |
341 | return true; | |
342 | ||
343 | /* Ignore . and .. */ | |
344 | if (!strcmp(".", dirent->d_name) || !strcmp("..", dirent->d_name)) | |
345 | return true; | |
346 | ||
347 | /* | |
348 | * If we were given a dirent, open the associated file under | |
349 | * dir_fd for badblocks scanning. If dirent is NULL, then it's | |
350 | * the directory itself we want to scan. | |
351 | */ | |
352 | fd = openat(dir_fd, dirent->d_name, | |
353 | O_RDONLY | O_NOATIME | O_NOFOLLOW | O_NOCTTY); | |
354 | if (fd < 0) | |
355 | return true; | |
356 | ||
357 | /* Go find the badness. */ | |
358 | moveon = xfs_report_verify_fd(ctx, path, fd, arg); | |
359 | if (moveon) | |
360 | goto out; | |
361 | ||
362 | out: | |
6c05cc5d DW |
363 | error = close(fd); |
364 | if (error) | |
365 | str_errno(ctx, path); | |
b364a9c0 DW |
366 | return moveon; |
367 | } | |
368 | ||
c9b349bd | 369 | /* Use a fsmap to report metadata lost to a media error. */ |
b364a9c0 | 370 | static bool |
c9b349bd | 371 | report_ioerr_fsmap( |
b364a9c0 DW |
372 | struct scrub_ctx *ctx, |
373 | const char *descr, | |
374 | struct fsmap *map, | |
375 | void *arg) | |
376 | { | |
377 | const char *type; | |
f1f5fd3a | 378 | char buf[DESCR_BUFSZ]; |
b364a9c0 DW |
379 | uint64_t err_physical = *(uint64_t *)arg; |
380 | uint64_t err_off; | |
381 | ||
909c6a54 DW |
382 | /* Don't care about unwritten extents. */ |
383 | if (map->fmr_flags & FMR_OF_PREALLOC) | |
384 | return true; | |
385 | ||
b364a9c0 DW |
386 | if (err_physical > map->fmr_physical) |
387 | err_off = err_physical - map->fmr_physical; | |
388 | else | |
389 | err_off = 0; | |
390 | ||
f1f5fd3a | 391 | /* Report special owners */ |
b364a9c0 | 392 | if (map->fmr_flags & FMR_OF_SPECIAL_OWNER) { |
f1f5fd3a DW |
393 | snprintf(buf, DESCR_BUFSZ, _("disk offset %"PRIu64), |
394 | (uint64_t)map->fmr_physical + err_off); | |
b364a9c0 | 395 | type = xfs_decode_special_owner(map->fmr_owner); |
abc2e70d | 396 | str_corrupt(ctx, buf, _("media error in %s."), type); |
b364a9c0 DW |
397 | } |
398 | ||
02d0069e DW |
399 | /* Report extent maps */ |
400 | if (map->fmr_flags & FMR_OF_EXTENT_MAP) { | |
401 | bool attr = (map->fmr_flags & FMR_OF_ATTR_FORK); | |
402 | ||
403 | scrub_render_ino_descr(ctx, buf, DESCR_BUFSZ, | |
404 | map->fmr_owner, 0, " %s", | |
405 | attr ? _("extended attribute") : | |
406 | _("file data")); | |
abc2e70d | 407 | str_corrupt(ctx, buf, _("media error in extent map")); |
02d0069e DW |
408 | } |
409 | ||
b364a9c0 DW |
410 | /* |
411 | * XXX: If we had a getparent() call we could report IO errors | |
412 | * efficiently. Until then, we'll have to scan the dir tree | |
413 | * to find the bad file's pathname. | |
414 | */ | |
415 | ||
416 | return true; | |
417 | } | |
418 | ||
419 | /* | |
c9b349bd DW |
420 | * For a range of bad blocks, visit each space mapping that overlaps the bad |
421 | * range so that we can report lost metadata. | |
b364a9c0 | 422 | */ |
c9b349bd DW |
423 | static int |
424 | report_ioerr( | |
b364a9c0 DW |
425 | uint64_t start, |
426 | uint64_t length, | |
b364a9c0 DW |
427 | void *arg) |
428 | { | |
429 | struct fsmap keys[2]; | |
430 | char descr[DESCR_BUFSZ]; | |
c9b349bd | 431 | struct disk_ioerr_report *dioerr = arg; |
b364a9c0 | 432 | dev_t dev; |
b364a9c0 | 433 | |
c9b349bd | 434 | dev = xfs_disk_to_dev(dioerr->ctx, dioerr->disk); |
b364a9c0 | 435 | |
c9b349bd DW |
436 | snprintf(descr, DESCR_BUFSZ, |
437 | _("dev %d:%d ioerr @ %"PRIu64":%"PRIu64" "), | |
b364a9c0 DW |
438 | major(dev), minor(dev), start, length); |
439 | ||
440 | /* Go figure out which blocks are bad from the fsmap. */ | |
441 | memset(keys, 0, sizeof(struct fsmap) * 2); | |
442 | keys->fmr_device = dev; | |
443 | keys->fmr_physical = start; | |
444 | (keys + 1)->fmr_device = dev; | |
445 | (keys + 1)->fmr_physical = start + length - 1; | |
446 | (keys + 1)->fmr_owner = ULLONG_MAX; | |
447 | (keys + 1)->fmr_offset = ULLONG_MAX; | |
448 | (keys + 1)->fmr_flags = UINT_MAX; | |
c9b349bd | 449 | xfs_iterate_fsmap(dioerr->ctx, descr, keys, report_ioerr_fsmap, |
b364a9c0 | 450 | &start); |
c9b349bd DW |
451 | return 0; |
452 | } | |
453 | ||
454 | /* Report all the media errors found on a disk. */ | |
455 | static int | |
456 | report_disk_ioerrs( | |
457 | struct scrub_ctx *ctx, | |
458 | struct disk *disk, | |
459 | struct media_verify_state *vs) | |
460 | { | |
461 | struct disk_ioerr_report dioerr = { | |
462 | .ctx = ctx, | |
463 | .disk = disk, | |
464 | }; | |
465 | struct bitmap *tree; | |
466 | ||
467 | if (!disk) | |
468 | return 0; | |
469 | tree = bitmap_for_disk(ctx, disk, vs); | |
470 | if (!tree) | |
471 | return 0; | |
472 | return bitmap_iterate(tree, report_ioerr, &dioerr); | |
473 | } | |
474 | ||
475 | /* Given bad extent lists for the data & rtdev, find bad files. */ | |
476 | static bool | |
477 | report_all_media_errors( | |
478 | struct scrub_ctx *ctx, | |
479 | struct media_verify_state *vs) | |
480 | { | |
481 | bool moveon; | |
482 | int ret; | |
483 | ||
484 | ret = report_disk_ioerrs(ctx, ctx->datadev, vs); | |
485 | if (ret) { | |
486 | str_liberror(ctx, ret, _("walking datadev io errors")); | |
487 | return false; | |
488 | } | |
489 | ||
490 | ret = report_disk_ioerrs(ctx, ctx->rtdev, vs); | |
491 | if (ret) { | |
492 | str_liberror(ctx, ret, _("walking rtdev io errors")); | |
493 | return false; | |
494 | } | |
495 | ||
496 | /* Scan the directory tree to get file paths. */ | |
497 | moveon = scan_fs_tree(ctx, xfs_report_verify_dir, | |
498 | xfs_report_verify_dirent, vs); | |
499 | if (!moveon) | |
500 | return false; | |
501 | ||
502 | /* Scan for unlinked files. */ | |
503 | return xfs_scan_all_inodes(ctx, xfs_report_verify_inode, vs); | |
b364a9c0 DW |
504 | } |
505 | ||
506 | /* Schedule a read-verify of a (data block) extent. */ | |
507 | static bool | |
508 | xfs_check_rmap( | |
509 | struct scrub_ctx *ctx, | |
510 | const char *descr, | |
511 | struct fsmap *map, | |
512 | void *arg) | |
513 | { | |
557f98d7 | 514 | struct media_verify_state *vs = arg; |
f1bb1696 | 515 | struct read_verify_pool *rvp; |
8cab77d3 | 516 | int ret; |
f1bb1696 | 517 | |
557f98d7 | 518 | rvp = xfs_dev_to_pool(ctx, vs, map->fmr_device); |
b364a9c0 DW |
519 | |
520 | dbg_printf("rmap dev %d:%d phys %"PRIu64" owner %"PRId64 | |
521 | " offset %"PRIu64" len %"PRIu64" flags 0x%x\n", | |
522 | major(map->fmr_device), minor(map->fmr_device), | |
523 | (uint64_t)map->fmr_physical, (int64_t)map->fmr_owner, | |
524 | (uint64_t)map->fmr_offset, (uint64_t)map->fmr_length, | |
525 | map->fmr_flags); | |
526 | ||
527 | /* "Unknown" extents should be verified; they could be data. */ | |
528 | if ((map->fmr_flags & FMR_OF_SPECIAL_OWNER) && | |
529 | map->fmr_owner == XFS_FMR_OWN_UNKNOWN) | |
530 | map->fmr_flags &= ~FMR_OF_SPECIAL_OWNER; | |
531 | ||
532 | /* | |
533 | * We only care about read-verifying data extents that have been | |
534 | * written to disk. This means we can skip "special" owners | |
535 | * (metadata), xattr blocks, unwritten extents, and extent maps. | |
536 | * These should all get checked elsewhere in the scrubber. | |
537 | */ | |
538 | if (map->fmr_flags & (FMR_OF_PREALLOC | FMR_OF_ATTR_FORK | | |
539 | FMR_OF_EXTENT_MAP | FMR_OF_SPECIAL_OWNER)) | |
22d658ec | 540 | return true; |
b364a9c0 DW |
541 | |
542 | /* XXX: Filter out directory data blocks. */ | |
543 | ||
544 | /* Schedule the read verify command for (eventual) running. */ | |
8cab77d3 DW |
545 | ret = read_verify_schedule_io(rvp, map->fmr_physical, map->fmr_length, |
546 | vs); | |
547 | if (ret) { | |
548 | str_liberror(ctx, ret, descr); | |
549 | return false; | |
550 | } | |
b364a9c0 | 551 | |
b364a9c0 DW |
552 | return true; |
553 | } | |
554 | ||
f1bb1696 | 555 | /* Wait for read/verify actions to finish, then return # bytes checked. */ |
8cab77d3 | 556 | static int |
f1bb1696 | 557 | clean_pool( |
8cab77d3 DW |
558 | struct read_verify_pool *rvp, |
559 | unsigned long long *bytes_checked) | |
f1bb1696 | 560 | { |
8cab77d3 DW |
561 | uint64_t pool_checked; |
562 | int ret; | |
f1bb1696 DW |
563 | |
564 | if (!rvp) | |
565 | return 0; | |
566 | ||
22d658ec DW |
567 | ret = read_verify_force_io(rvp); |
568 | if (ret) | |
569 | return ret; | |
570 | ||
8cab77d3 DW |
571 | ret = read_verify_pool_flush(rvp); |
572 | if (ret) | |
573 | goto out_destroy; | |
574 | ||
575 | ret = read_verify_bytes(rvp, &pool_checked); | |
576 | if (ret) | |
577 | goto out_destroy; | |
578 | ||
579 | *bytes_checked += pool_checked; | |
580 | out_destroy: | |
f1bb1696 DW |
581 | read_verify_pool_destroy(rvp); |
582 | return ret; | |
583 | } | |
584 | ||
c9b349bd DW |
585 | /* Remember a media error for later. */ |
586 | static void | |
587 | remember_ioerr( | |
588 | struct scrub_ctx *ctx, | |
589 | struct disk *disk, | |
590 | uint64_t start, | |
591 | uint64_t length, | |
592 | int error, | |
593 | void *arg) | |
594 | { | |
595 | struct media_verify_state *vs = arg; | |
596 | struct bitmap *tree; | |
597 | int ret; | |
598 | ||
599 | tree = bitmap_for_disk(ctx, disk, vs); | |
600 | if (!tree) { | |
601 | str_liberror(ctx, ENOENT, _("finding bad block bitmap")); | |
602 | return; | |
603 | } | |
604 | ||
605 | ret = bitmap_set(tree, start, length); | |
606 | if (ret) | |
607 | str_liberror(ctx, ret, _("setting bad block bitmap")); | |
608 | } | |
609 | ||
b364a9c0 DW |
610 | /* |
611 | * Read verify all the file data blocks in a filesystem. Since XFS doesn't | |
612 | * do data checksums, we trust that the underlying storage will pass back | |
613 | * an IO error if it can't retrieve whatever we previously stored there. | |
614 | * If we hit an IO error, we'll record the bad blocks in a bitmap and then | |
615 | * scan the extent maps of the entire fs tree to figure (and the unlinked | |
616 | * inodes) out which files are now broken. | |
617 | */ | |
618 | bool | |
619 | xfs_scan_blocks( | |
620 | struct scrub_ctx *ctx) | |
621 | { | |
557f98d7 | 622 | struct media_verify_state vs = { NULL }; |
93ab49dd | 623 | bool moveon = false; |
233fabee | 624 | int ret; |
b364a9c0 | 625 | |
233fabee DW |
626 | ret = bitmap_alloc(&vs.d_bad); |
627 | if (ret) { | |
628 | str_liberror(ctx, ret, _("creating datadev badblock bitmap")); | |
41c08606 | 629 | goto out; |
b364a9c0 DW |
630 | } |
631 | ||
233fabee DW |
632 | ret = bitmap_alloc(&vs.r_bad); |
633 | if (ret) { | |
634 | str_liberror(ctx, ret, _("creating realtime badblock bitmap")); | |
b364a9c0 DW |
635 | goto out_dbad; |
636 | } | |
637 | ||
8cab77d3 | 638 | ret = read_verify_pool_alloc(ctx, ctx->datadev, |
c9b349bd | 639 | ctx->mnt.fsgeom.blocksize, remember_ioerr, |
8cab77d3 DW |
640 | scrub_nproc(ctx), &vs.rvp_data); |
641 | if (ret) { | |
642 | str_liberror(ctx, ret, _("creating datadev media verifier")); | |
b364a9c0 DW |
643 | goto out_rbad; |
644 | } | |
f1bb1696 | 645 | if (ctx->logdev) { |
8cab77d3 | 646 | ret = read_verify_pool_alloc(ctx, ctx->logdev, |
c9b349bd | 647 | ctx->mnt.fsgeom.blocksize, remember_ioerr, |
8cab77d3 DW |
648 | scrub_nproc(ctx), &vs.rvp_log); |
649 | if (ret) { | |
650 | str_liberror(ctx, ret, | |
651 | _("creating logdev media verifier")); | |
f1bb1696 DW |
652 | goto out_datapool; |
653 | } | |
654 | } | |
655 | if (ctx->rtdev) { | |
8cab77d3 | 656 | ret = read_verify_pool_alloc(ctx, ctx->rtdev, |
c9b349bd | 657 | ctx->mnt.fsgeom.blocksize, remember_ioerr, |
8cab77d3 DW |
658 | scrub_nproc(ctx), &vs.rvp_realtime); |
659 | if (ret) { | |
660 | str_liberror(ctx, ret, | |
661 | _("creating rtdev media verifier")); | |
f1bb1696 DW |
662 | goto out_logpool; |
663 | } | |
664 | } | |
557f98d7 | 665 | moveon = xfs_scan_all_spacemaps(ctx, xfs_check_rmap, &vs); |
b364a9c0 | 666 | if (!moveon) |
f1bb1696 | 667 | goto out_rtpool; |
8cab77d3 DW |
668 | |
669 | ret = clean_pool(vs.rvp_data, &ctx->bytes_checked); | |
670 | if (ret) { | |
671 | str_liberror(ctx, ret, _("flushing datadev verify pool")); | |
672 | moveon = false; | |
673 | } | |
674 | ||
675 | ret = clean_pool(vs.rvp_log, &ctx->bytes_checked); | |
676 | if (ret) { | |
677 | str_liberror(ctx, ret, _("flushing logdev verify pool")); | |
678 | moveon = false; | |
679 | } | |
680 | ||
681 | ret = clean_pool(vs.rvp_realtime, &ctx->bytes_checked); | |
682 | if (ret) { | |
683 | str_liberror(ctx, ret, _("flushing rtdev verify pool")); | |
684 | moveon = false; | |
685 | } | |
b364a9c0 DW |
686 | |
687 | /* Scan the whole dir tree to see what matches the bad extents. */ | |
8cab77d3 | 688 | if (moveon && (!bitmap_empty(vs.d_bad) || !bitmap_empty(vs.r_bad))) |
c9b349bd | 689 | moveon = report_all_media_errors(ctx, &vs); |
b364a9c0 | 690 | |
557f98d7 DW |
691 | bitmap_free(&vs.r_bad); |
692 | bitmap_free(&vs.d_bad); | |
b364a9c0 DW |
693 | return moveon; |
694 | ||
f1bb1696 | 695 | out_rtpool: |
7668d01d | 696 | if (vs.rvp_realtime) { |
4cd869e5 | 697 | read_verify_pool_abort(vs.rvp_realtime); |
557f98d7 | 698 | read_verify_pool_destroy(vs.rvp_realtime); |
7668d01d | 699 | } |
f1bb1696 | 700 | out_logpool: |
7668d01d | 701 | if (vs.rvp_log) { |
4cd869e5 | 702 | read_verify_pool_abort(vs.rvp_log); |
557f98d7 | 703 | read_verify_pool_destroy(vs.rvp_log); |
7668d01d | 704 | } |
f1bb1696 | 705 | out_datapool: |
4cd869e5 | 706 | read_verify_pool_abort(vs.rvp_data); |
557f98d7 | 707 | read_verify_pool_destroy(vs.rvp_data); |
b364a9c0 | 708 | out_rbad: |
557f98d7 | 709 | bitmap_free(&vs.r_bad); |
b364a9c0 | 710 | out_dbad: |
557f98d7 | 711 | bitmap_free(&vs.d_bad); |
41c08606 | 712 | out: |
b364a9c0 DW |
713 | return moveon; |
714 | } | |
ed60d210 DW |
715 | |
716 | /* Estimate how much work we're going to do. */ | |
717 | bool | |
718 | xfs_estimate_verify_work( | |
719 | struct scrub_ctx *ctx, | |
720 | uint64_t *items, | |
721 | unsigned int *nr_threads, | |
722 | int *rshift) | |
723 | { | |
724 | unsigned long long d_blocks; | |
725 | unsigned long long d_bfree; | |
726 | unsigned long long r_blocks; | |
727 | unsigned long long r_bfree; | |
728 | unsigned long long f_files; | |
729 | unsigned long long f_free; | |
934d8d3a | 730 | int ret; |
ed60d210 | 731 | |
934d8d3a | 732 | ret = scrub_scan_estimate_blocks(ctx, &d_blocks, &d_bfree, |
ed60d210 | 733 | &r_blocks, &r_bfree, &f_files, &f_free); |
934d8d3a DW |
734 | if (ret) { |
735 | str_liberror(ctx, ret, _("estimating verify work")); | |
736 | return false; | |
737 | } | |
ed60d210 | 738 | |
a749451c DW |
739 | *items = cvt_off_fsb_to_b(&ctx->mnt, |
740 | (d_blocks - d_bfree) + (r_blocks - r_bfree)); | |
ed60d210 DW |
741 | *nr_threads = disk_heads(ctx->datadev); |
742 | *rshift = 20; | |
934d8d3a | 743 | return true; |
ed60d210 | 744 | } |