]>
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 | 47 | static struct read_verify_pool * |
af9eb208 | 48 | dev_to_pool( |
f1bb1696 | 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 | |
af9eb208 | 64 | disk_to_dev( |
b364a9c0 DW |
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 | { | |
af9eb208 | 84 | dev_t dev = disk_to_dev(ctx, disk); |
c9b349bd DW |
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 * | |
af9eb208 | 119 | decode_special_owner( |
b364a9c0 DW |
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 | 234 | /* Iterate the extent mappings of a file to report errors. */ |
af9eb208 DW |
235 | static int |
236 | report_fd_loss( | |
b364a9c0 DW |
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); | |
af9eb208 | 255 | return ret; |
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); | |
af9eb208 | 263 | return ret; |
73ce9669 | 264 | } |
af9eb208 DW |
265 | |
266 | return 0; | |
b364a9c0 DW |
267 | } |
268 | ||
269 | /* Report read verify errors in unlinked (but still open) files. */ | |
270 | static int | |
af9eb208 | 271 | report_inode_loss( |
b364a9c0 DW |
272 | struct scrub_ctx *ctx, |
273 | struct xfs_handle *handle, | |
4cca629d | 274 | struct xfs_bulkstat *bstat, |
b364a9c0 DW |
275 | void *arg) |
276 | { | |
277 | char descr[DESCR_BUFSZ]; | |
b364a9c0 | 278 | int fd; |
af9eb208 | 279 | int error, err2; |
b364a9c0 | 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 | 290 | /* Try to open the inode. */ |
59f79e0a | 291 | fd = scrub_open_handle(handle); |
b364a9c0 DW |
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. */ | |
af9eb208 DW |
303 | error = report_fd_loss(ctx, descr, fd, arg); |
304 | ||
305 | err2 = close(fd); | |
306 | if (err2) | |
6c05cc5d | 307 | str_errno(ctx, descr); |
b364a9c0 | 308 | |
af9eb208 | 309 | return error; |
b364a9c0 DW |
310 | } |
311 | ||
312 | /* Scan a directory for matches in the read verify error list. */ | |
f544ec31 | 313 | static int |
af9eb208 | 314 | report_dir_loss( |
b364a9c0 DW |
315 | struct scrub_ctx *ctx, |
316 | const char *path, | |
317 | int dir_fd, | |
318 | void *arg) | |
319 | { | |
af9eb208 | 320 | return report_fd_loss(ctx, path, dir_fd, arg); |
b364a9c0 DW |
321 | } |
322 | ||
323 | /* | |
324 | * Scan the inode associated with a directory entry for matches with | |
325 | * the read verify error list. | |
326 | */ | |
f544ec31 | 327 | static int |
af9eb208 | 328 | report_dirent_loss( |
b364a9c0 DW |
329 | struct scrub_ctx *ctx, |
330 | const char *path, | |
331 | int dir_fd, | |
332 | struct dirent *dirent, | |
333 | struct stat *sb, | |
334 | void *arg) | |
335 | { | |
b364a9c0 | 336 | int fd; |
af9eb208 | 337 | int error, err2; |
b364a9c0 DW |
338 | |
339 | /* Ignore things we can't open. */ | |
340 | if (!S_ISREG(sb->st_mode) && !S_ISDIR(sb->st_mode)) | |
f544ec31 | 341 | return 0; |
b364a9c0 DW |
342 | |
343 | /* Ignore . and .. */ | |
344 | if (!strcmp(".", dirent->d_name) || !strcmp("..", dirent->d_name)) | |
f544ec31 | 345 | return 0; |
b364a9c0 DW |
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); | |
f544ec31 DW |
354 | if (fd < 0) { |
355 | if (errno == ENOENT) | |
356 | return 0; | |
357 | str_errno(ctx, path); | |
358 | return errno; | |
359 | } | |
b364a9c0 DW |
360 | |
361 | /* Go find the badness. */ | |
af9eb208 | 362 | error = report_fd_loss(ctx, path, fd, arg); |
b364a9c0 | 363 | |
af9eb208 DW |
364 | err2 = close(fd); |
365 | if (err2) | |
6c05cc5d | 366 | str_errno(ctx, path); |
af9eb208 DW |
367 | if (!error && err2) |
368 | error = err2; | |
369 | ||
370 | return error; | |
b364a9c0 DW |
371 | } |
372 | ||
c9b349bd | 373 | /* Use a fsmap to report metadata lost to a media error. */ |
7a2eef2b | 374 | static int |
c9b349bd | 375 | report_ioerr_fsmap( |
b364a9c0 | 376 | struct scrub_ctx *ctx, |
b364a9c0 DW |
377 | struct fsmap *map, |
378 | void *arg) | |
379 | { | |
380 | const char *type; | |
f1f5fd3a | 381 | char buf[DESCR_BUFSZ]; |
b364a9c0 DW |
382 | uint64_t err_physical = *(uint64_t *)arg; |
383 | uint64_t err_off; | |
384 | ||
909c6a54 DW |
385 | /* Don't care about unwritten extents. */ |
386 | if (map->fmr_flags & FMR_OF_PREALLOC) | |
7a2eef2b | 387 | return 0; |
909c6a54 | 388 | |
b364a9c0 DW |
389 | if (err_physical > map->fmr_physical) |
390 | err_off = err_physical - map->fmr_physical; | |
391 | else | |
392 | err_off = 0; | |
393 | ||
f1f5fd3a | 394 | /* Report special owners */ |
b364a9c0 | 395 | if (map->fmr_flags & FMR_OF_SPECIAL_OWNER) { |
f1f5fd3a DW |
396 | snprintf(buf, DESCR_BUFSZ, _("disk offset %"PRIu64), |
397 | (uint64_t)map->fmr_physical + err_off); | |
af9eb208 | 398 | type = decode_special_owner(map->fmr_owner); |
abc2e70d | 399 | str_corrupt(ctx, buf, _("media error in %s."), type); |
b364a9c0 DW |
400 | } |
401 | ||
02d0069e DW |
402 | /* Report extent maps */ |
403 | if (map->fmr_flags & FMR_OF_EXTENT_MAP) { | |
404 | bool attr = (map->fmr_flags & FMR_OF_ATTR_FORK); | |
405 | ||
406 | scrub_render_ino_descr(ctx, buf, DESCR_BUFSZ, | |
407 | map->fmr_owner, 0, " %s", | |
408 | attr ? _("extended attribute") : | |
409 | _("file data")); | |
abc2e70d | 410 | str_corrupt(ctx, buf, _("media error in extent map")); |
02d0069e DW |
411 | } |
412 | ||
b364a9c0 DW |
413 | /* |
414 | * XXX: If we had a getparent() call we could report IO errors | |
415 | * efficiently. Until then, we'll have to scan the dir tree | |
416 | * to find the bad file's pathname. | |
417 | */ | |
418 | ||
7a2eef2b | 419 | return 0; |
b364a9c0 DW |
420 | } |
421 | ||
422 | /* | |
c9b349bd DW |
423 | * For a range of bad blocks, visit each space mapping that overlaps the bad |
424 | * range so that we can report lost metadata. | |
b364a9c0 | 425 | */ |
c9b349bd DW |
426 | static int |
427 | report_ioerr( | |
b364a9c0 DW |
428 | uint64_t start, |
429 | uint64_t length, | |
b364a9c0 DW |
430 | void *arg) |
431 | { | |
432 | struct fsmap keys[2]; | |
c9b349bd | 433 | struct disk_ioerr_report *dioerr = arg; |
b364a9c0 | 434 | dev_t dev; |
b364a9c0 | 435 | |
af9eb208 | 436 | dev = disk_to_dev(dioerr->ctx, dioerr->disk); |
b364a9c0 | 437 | |
b364a9c0 DW |
438 | /* Go figure out which blocks are bad from the fsmap. */ |
439 | memset(keys, 0, sizeof(struct fsmap) * 2); | |
440 | keys->fmr_device = dev; | |
441 | keys->fmr_physical = start; | |
442 | (keys + 1)->fmr_device = dev; | |
443 | (keys + 1)->fmr_physical = start + length - 1; | |
444 | (keys + 1)->fmr_owner = ULLONG_MAX; | |
445 | (keys + 1)->fmr_offset = ULLONG_MAX; | |
446 | (keys + 1)->fmr_flags = UINT_MAX; | |
7a2eef2b | 447 | return scrub_iterate_fsmap(dioerr->ctx, keys, report_ioerr_fsmap, |
b364a9c0 | 448 | &start); |
c9b349bd DW |
449 | } |
450 | ||
451 | /* Report all the media errors found on a disk. */ | |
452 | static int | |
453 | report_disk_ioerrs( | |
454 | struct scrub_ctx *ctx, | |
455 | struct disk *disk, | |
456 | struct media_verify_state *vs) | |
457 | { | |
458 | struct disk_ioerr_report dioerr = { | |
459 | .ctx = ctx, | |
460 | .disk = disk, | |
461 | }; | |
462 | struct bitmap *tree; | |
463 | ||
464 | if (!disk) | |
465 | return 0; | |
466 | tree = bitmap_for_disk(ctx, disk, vs); | |
467 | if (!tree) | |
468 | return 0; | |
469 | return bitmap_iterate(tree, report_ioerr, &dioerr); | |
470 | } | |
471 | ||
472 | /* Given bad extent lists for the data & rtdev, find bad files. */ | |
af9eb208 | 473 | static int |
c9b349bd DW |
474 | report_all_media_errors( |
475 | struct scrub_ctx *ctx, | |
476 | struct media_verify_state *vs) | |
477 | { | |
c9b349bd DW |
478 | int ret; |
479 | ||
480 | ret = report_disk_ioerrs(ctx, ctx->datadev, vs); | |
481 | if (ret) { | |
482 | str_liberror(ctx, ret, _("walking datadev io errors")); | |
af9eb208 | 483 | return ret; |
c9b349bd DW |
484 | } |
485 | ||
486 | ret = report_disk_ioerrs(ctx, ctx->rtdev, vs); | |
487 | if (ret) { | |
488 | str_liberror(ctx, ret, _("walking rtdev io errors")); | |
af9eb208 | 489 | return ret; |
c9b349bd DW |
490 | } |
491 | ||
492 | /* Scan the directory tree to get file paths. */ | |
af9eb208 | 493 | ret = scan_fs_tree(ctx, report_dir_loss, report_dirent_loss, vs); |
f544ec31 | 494 | if (ret) |
af9eb208 | 495 | return ret; |
c9b349bd DW |
496 | |
497 | /* Scan for unlinked files. */ | |
af9eb208 | 498 | return scrub_scan_all_inodes(ctx, report_inode_loss, vs); |
b364a9c0 DW |
499 | } |
500 | ||
501 | /* Schedule a read-verify of a (data block) extent. */ | |
7a2eef2b DW |
502 | static int |
503 | check_rmap( | |
b364a9c0 | 504 | struct scrub_ctx *ctx, |
b364a9c0 DW |
505 | struct fsmap *map, |
506 | void *arg) | |
507 | { | |
557f98d7 | 508 | struct media_verify_state *vs = arg; |
f1bb1696 | 509 | struct read_verify_pool *rvp; |
8cab77d3 | 510 | int ret; |
f1bb1696 | 511 | |
af9eb208 | 512 | rvp = dev_to_pool(ctx, vs, map->fmr_device); |
b364a9c0 DW |
513 | |
514 | dbg_printf("rmap dev %d:%d phys %"PRIu64" owner %"PRId64 | |
515 | " offset %"PRIu64" len %"PRIu64" flags 0x%x\n", | |
516 | major(map->fmr_device), minor(map->fmr_device), | |
517 | (uint64_t)map->fmr_physical, (int64_t)map->fmr_owner, | |
518 | (uint64_t)map->fmr_offset, (uint64_t)map->fmr_length, | |
519 | map->fmr_flags); | |
520 | ||
521 | /* "Unknown" extents should be verified; they could be data. */ | |
522 | if ((map->fmr_flags & FMR_OF_SPECIAL_OWNER) && | |
523 | map->fmr_owner == XFS_FMR_OWN_UNKNOWN) | |
524 | map->fmr_flags &= ~FMR_OF_SPECIAL_OWNER; | |
525 | ||
526 | /* | |
527 | * We only care about read-verifying data extents that have been | |
528 | * written to disk. This means we can skip "special" owners | |
529 | * (metadata), xattr blocks, unwritten extents, and extent maps. | |
530 | * These should all get checked elsewhere in the scrubber. | |
531 | */ | |
532 | if (map->fmr_flags & (FMR_OF_PREALLOC | FMR_OF_ATTR_FORK | | |
533 | FMR_OF_EXTENT_MAP | FMR_OF_SPECIAL_OWNER)) | |
7a2eef2b | 534 | return 0; |
b364a9c0 DW |
535 | |
536 | /* XXX: Filter out directory data blocks. */ | |
537 | ||
538 | /* Schedule the read verify command for (eventual) running. */ | |
8cab77d3 DW |
539 | ret = read_verify_schedule_io(rvp, map->fmr_physical, map->fmr_length, |
540 | vs); | |
541 | if (ret) { | |
7a2eef2b DW |
542 | str_liberror(ctx, ret, _("scheduling media verify command")); |
543 | return ret; | |
8cab77d3 | 544 | } |
b364a9c0 | 545 | |
7a2eef2b | 546 | return 0; |
b364a9c0 DW |
547 | } |
548 | ||
f1bb1696 | 549 | /* Wait for read/verify actions to finish, then return # bytes checked. */ |
8cab77d3 | 550 | static int |
f1bb1696 | 551 | clean_pool( |
8cab77d3 DW |
552 | struct read_verify_pool *rvp, |
553 | unsigned long long *bytes_checked) | |
f1bb1696 | 554 | { |
8cab77d3 DW |
555 | uint64_t pool_checked; |
556 | int ret; | |
f1bb1696 DW |
557 | |
558 | if (!rvp) | |
559 | return 0; | |
560 | ||
22d658ec DW |
561 | ret = read_verify_force_io(rvp); |
562 | if (ret) | |
563 | return ret; | |
564 | ||
8cab77d3 DW |
565 | ret = read_verify_pool_flush(rvp); |
566 | if (ret) | |
567 | goto out_destroy; | |
568 | ||
569 | ret = read_verify_bytes(rvp, &pool_checked); | |
570 | if (ret) | |
571 | goto out_destroy; | |
572 | ||
573 | *bytes_checked += pool_checked; | |
574 | out_destroy: | |
f1bb1696 DW |
575 | read_verify_pool_destroy(rvp); |
576 | return ret; | |
577 | } | |
578 | ||
c9b349bd DW |
579 | /* Remember a media error for later. */ |
580 | static void | |
581 | remember_ioerr( | |
582 | struct scrub_ctx *ctx, | |
583 | struct disk *disk, | |
584 | uint64_t start, | |
585 | uint64_t length, | |
586 | int error, | |
587 | void *arg) | |
588 | { | |
589 | struct media_verify_state *vs = arg; | |
590 | struct bitmap *tree; | |
591 | int ret; | |
592 | ||
593 | tree = bitmap_for_disk(ctx, disk, vs); | |
594 | if (!tree) { | |
595 | str_liberror(ctx, ENOENT, _("finding bad block bitmap")); | |
596 | return; | |
597 | } | |
598 | ||
599 | ret = bitmap_set(tree, start, length); | |
600 | if (ret) | |
601 | str_liberror(ctx, ret, _("setting bad block bitmap")); | |
602 | } | |
603 | ||
b364a9c0 DW |
604 | /* |
605 | * Read verify all the file data blocks in a filesystem. Since XFS doesn't | |
606 | * do data checksums, we trust that the underlying storage will pass back | |
607 | * an IO error if it can't retrieve whatever we previously stored there. | |
608 | * If we hit an IO error, we'll record the bad blocks in a bitmap and then | |
609 | * scan the extent maps of the entire fs tree to figure (and the unlinked | |
610 | * inodes) out which files are now broken. | |
611 | */ | |
af9eb208 DW |
612 | int |
613 | phase6_func( | |
b364a9c0 DW |
614 | struct scrub_ctx *ctx) |
615 | { | |
557f98d7 | 616 | struct media_verify_state vs = { NULL }; |
af9eb208 | 617 | int ret, ret2, ret3; |
b364a9c0 | 618 | |
233fabee DW |
619 | ret = bitmap_alloc(&vs.d_bad); |
620 | if (ret) { | |
621 | str_liberror(ctx, ret, _("creating datadev badblock bitmap")); | |
af9eb208 | 622 | return ret; |
b364a9c0 DW |
623 | } |
624 | ||
233fabee DW |
625 | ret = bitmap_alloc(&vs.r_bad); |
626 | if (ret) { | |
627 | str_liberror(ctx, ret, _("creating realtime badblock bitmap")); | |
b364a9c0 DW |
628 | goto out_dbad; |
629 | } | |
630 | ||
8cab77d3 | 631 | ret = read_verify_pool_alloc(ctx, ctx->datadev, |
c9b349bd | 632 | ctx->mnt.fsgeom.blocksize, remember_ioerr, |
8cab77d3 DW |
633 | scrub_nproc(ctx), &vs.rvp_data); |
634 | if (ret) { | |
635 | str_liberror(ctx, ret, _("creating datadev media verifier")); | |
b364a9c0 DW |
636 | goto out_rbad; |
637 | } | |
f1bb1696 | 638 | if (ctx->logdev) { |
8cab77d3 | 639 | ret = read_verify_pool_alloc(ctx, ctx->logdev, |
c9b349bd | 640 | ctx->mnt.fsgeom.blocksize, remember_ioerr, |
8cab77d3 DW |
641 | scrub_nproc(ctx), &vs.rvp_log); |
642 | if (ret) { | |
643 | str_liberror(ctx, ret, | |
644 | _("creating logdev media verifier")); | |
f1bb1696 DW |
645 | goto out_datapool; |
646 | } | |
647 | } | |
648 | if (ctx->rtdev) { | |
8cab77d3 | 649 | ret = read_verify_pool_alloc(ctx, ctx->rtdev, |
c9b349bd | 650 | ctx->mnt.fsgeom.blocksize, remember_ioerr, |
8cab77d3 DW |
651 | scrub_nproc(ctx), &vs.rvp_realtime); |
652 | if (ret) { | |
653 | str_liberror(ctx, ret, | |
654 | _("creating rtdev media verifier")); | |
f1bb1696 DW |
655 | goto out_logpool; |
656 | } | |
657 | } | |
7a2eef2b | 658 | ret = scrub_scan_all_spacemaps(ctx, check_rmap, &vs); |
af9eb208 | 659 | if (ret) |
f1bb1696 | 660 | goto out_rtpool; |
8cab77d3 DW |
661 | |
662 | ret = clean_pool(vs.rvp_data, &ctx->bytes_checked); | |
af9eb208 | 663 | if (ret) |
8cab77d3 | 664 | str_liberror(ctx, ret, _("flushing datadev verify pool")); |
8cab77d3 | 665 | |
af9eb208 DW |
666 | ret2 = clean_pool(vs.rvp_log, &ctx->bytes_checked); |
667 | if (ret2) | |
668 | str_liberror(ctx, ret2, _("flushing logdev verify pool")); | |
8cab77d3 | 669 | |
af9eb208 DW |
670 | ret3 = clean_pool(vs.rvp_realtime, &ctx->bytes_checked); |
671 | if (ret3) | |
672 | str_liberror(ctx, ret3, _("flushing rtdev verify pool")); | |
673 | ||
674 | /* | |
675 | * If the verify flush didn't work or we found no bad blocks, we're | |
676 | * done! No errors detected. | |
677 | */ | |
678 | if (ret || ret2 || ret3) | |
679 | goto out_rbad; | |
680 | if (bitmap_empty(vs.d_bad) && bitmap_empty(vs.r_bad)) | |
681 | goto out_rbad; | |
b364a9c0 DW |
682 | |
683 | /* Scan the whole dir tree to see what matches the bad extents. */ | |
af9eb208 | 684 | ret = report_all_media_errors(ctx, &vs); |
b364a9c0 | 685 | |
557f98d7 DW |
686 | bitmap_free(&vs.r_bad); |
687 | bitmap_free(&vs.d_bad); | |
af9eb208 | 688 | return ret; |
b364a9c0 | 689 | |
f1bb1696 | 690 | out_rtpool: |
7668d01d | 691 | if (vs.rvp_realtime) { |
4cd869e5 | 692 | read_verify_pool_abort(vs.rvp_realtime); |
557f98d7 | 693 | read_verify_pool_destroy(vs.rvp_realtime); |
7668d01d | 694 | } |
f1bb1696 | 695 | out_logpool: |
7668d01d | 696 | if (vs.rvp_log) { |
4cd869e5 | 697 | read_verify_pool_abort(vs.rvp_log); |
557f98d7 | 698 | read_verify_pool_destroy(vs.rvp_log); |
7668d01d | 699 | } |
f1bb1696 | 700 | out_datapool: |
4cd869e5 | 701 | read_verify_pool_abort(vs.rvp_data); |
557f98d7 | 702 | read_verify_pool_destroy(vs.rvp_data); |
b364a9c0 | 703 | out_rbad: |
557f98d7 | 704 | bitmap_free(&vs.r_bad); |
b364a9c0 | 705 | out_dbad: |
557f98d7 | 706 | bitmap_free(&vs.d_bad); |
af9eb208 | 707 | return ret; |
b364a9c0 | 708 | } |
ed60d210 | 709 | |
af9eb208 DW |
710 | /* Estimate how much work we're going to do. */ |
711 | int | |
712 | phase6_estimate( | |
ed60d210 DW |
713 | struct scrub_ctx *ctx, |
714 | uint64_t *items, | |
715 | unsigned int *nr_threads, | |
716 | int *rshift) | |
717 | { | |
718 | unsigned long long d_blocks; | |
719 | unsigned long long d_bfree; | |
720 | unsigned long long r_blocks; | |
721 | unsigned long long r_bfree; | |
722 | unsigned long long f_files; | |
723 | unsigned long long f_free; | |
934d8d3a | 724 | int ret; |
ed60d210 | 725 | |
934d8d3a | 726 | ret = scrub_scan_estimate_blocks(ctx, &d_blocks, &d_bfree, |
ed60d210 | 727 | &r_blocks, &r_bfree, &f_files, &f_free); |
934d8d3a DW |
728 | if (ret) { |
729 | str_liberror(ctx, ret, _("estimating verify work")); | |
af9eb208 | 730 | return ret; |
934d8d3a | 731 | } |
ed60d210 | 732 | |
a749451c DW |
733 | *items = cvt_off_fsb_to_b(&ctx->mnt, |
734 | (d_blocks - d_bfree) + (r_blocks - r_bfree)); | |
ed60d210 DW |
735 | *nr_threads = disk_heads(ctx->datadev); |
736 | *rshift = 20; | |
af9eb208 DW |
737 | return 0; |
738 | } |