projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: enhance transaction abort infrastructure
[linux-2.6-block.git] / fs / btrfs / scrub.c
CommitLineData
a2de733c
AJ		 1/*
2 * Copyright (C) 2011 STRATO. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
a2de733c 19#include <linux/blkdev.h>
558540c1 20#include <linux/ratelimit.h>
a2de733c
AJ		 21#include "ctree.h"
22#include "volumes.h"
23#include "disk-io.h"
24#include "ordered-data.h"
0ef8e451 25#include "transaction.h"
558540c1 26#include "backref.h"
5da6fcbc 27#include "extent_io.h"
21adbd5c 28#include "check-integrity.h"
a2de733c
AJ		 29
30/*
31 * This is only the first step towards a full-features scrub. It reads all
32 * extent and super block and verifies the checksums. In case a bad checksum
33 * is found or the extent cannot be read, good data will be written back if
34 * any can be found.
35 *
36 * Future enhancements:
a2de733c
AJ		 37 * - In case an unrepairable extent is encountered, track which files are
38 * affected and report them
39 * - In case of a read error on files with nodatasum, map the file and read
40 * the extent to trigger a writeback of the good copy
41 * - track and record media errors, throw out bad devices
a2de733c 42 * - add a mode to also read unallocated space
a2de733c
AJ		 43 */
44
45struct scrub_bio;
46struct scrub_page;
47struct scrub_dev;
a2de733c
AJ		 48static void scrub_bio_end_io(struct bio *bio, int err);
49static void scrub_checksum(struct btrfs_work *work);
50static int scrub_checksum_data(struct scrub_dev *sdev,
51 struct scrub_page *spag, void *buffer);
52static int scrub_checksum_tree_block(struct scrub_dev *sdev,
53 struct scrub_page *spag, u64 logical,
54 void *buffer);
55static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer);
96e36920
ID		 56static int scrub_fixup_check(struct scrub_bio *sbio, int ix);
57static void scrub_fixup_end_io(struct bio *bio, int err);
58static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
59 struct page *page);
60static void scrub_fixup(struct scrub_bio *sbio, int ix);
a2de733c
AJ		 61
62#define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */
63#define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */
64
65struct scrub_page {
66 u64 flags; /* extent flags */
67 u64 generation;
e12fa9cd 68 int mirror_num;
a2de733c
AJ		 69 int have_csum;
70 u8 csum[BTRFS_CSUM_SIZE];
71};
72
73struct scrub_bio {
74 int index;
75 struct scrub_dev *sdev;
76 struct bio *bio;
77 int err;
78 u64 logical;
79 u64 physical;
80 struct scrub_page spag[SCRUB_PAGES_PER_BIO];
81 u64 count;
82 int next_free;
83 struct btrfs_work work;
84};
85
86struct scrub_dev {
87 struct scrub_bio *bios[SCRUB_BIOS_PER_DEV];
88 struct btrfs_device *dev;
89 int first_free;
90 int curr;
91 atomic_t in_flight;
0ef8e451 92 atomic_t fixup_cnt;
a2de733c
AJ		 93 spinlock_t list_lock;
94 wait_queue_head_t list_wait;
95 u16 csum_size;
96 struct list_head csum_list;
97 atomic_t cancel_req;
8628764e 98 int readonly;
a2de733c
AJ		 99 /*
100 * statistics
101 */
102 struct btrfs_scrub_progress stat;
103 spinlock_t stat_lock;
104};
105
0ef8e451
JS		 106struct scrub_fixup_nodatasum {
107 struct scrub_dev *sdev;
108 u64 logical;
109 struct btrfs_root *root;
110 struct btrfs_work work;
111 int mirror_num;
112};
113
558540c1
JS		 114struct scrub_warning {
115 struct btrfs_path *path;
116 u64 extent_item_size;
117 char *scratch_buf;
118 char *msg_buf;
119 const char *errstr;
120 sector_t sector;
121 u64 logical;
122 struct btrfs_device *dev;
123 int msg_bufsize;
124 int scratch_bufsize;
125};
126
a2de733c
AJ		 127static void scrub_free_csums(struct scrub_dev *sdev)
128{
129 while (!list_empty(&sdev->csum_list)) {
130 struct btrfs_ordered_sum *sum;
131 sum = list_first_entry(&sdev->csum_list,
132 struct btrfs_ordered_sum, list);
133 list_del(&sum->list);
134 kfree(sum);
135 }
136}
137
1bc87793
AJ		 138static void scrub_free_bio(struct bio *bio)
139{
140 int i;
141 struct page *last_page = NULL;
142
143 if (!bio)
144 return;
145
146 for (i = 0; i < bio->bi_vcnt; ++i) {
147 if (bio->bi_io_vec[i].bv_page == last_page)
148 continue;
149 last_page = bio->bi_io_vec[i].bv_page;
150 __free_page(last_page);
151 }
152 bio_put(bio);
153}
154
a2de733c
AJ		 155static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev)
156{
157 int i;
a2de733c
AJ		 158
159 if (!sdev)
160 return;
161
162 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
163 struct scrub_bio *sbio = sdev->bios[i];
a2de733c
AJ		 164
165 if (!sbio)
166 break;
167
1bc87793 168 scrub_free_bio(sbio->bio);
a2de733c
AJ		 169 kfree(sbio);
170 }
171
172 scrub_free_csums(sdev);
173 kfree(sdev);
174}
175
176static noinline_for_stack
177struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev)
178{
179 struct scrub_dev *sdev;
180 int i;
a2de733c
AJ		 181 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
182
183 sdev = kzalloc(sizeof(*sdev), GFP_NOFS);
184 if (!sdev)
185 goto nomem;
186 sdev->dev = dev;
187 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) {
a2de733c
AJ		 188 struct scrub_bio *sbio;
189
190 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
191 if (!sbio)
192 goto nomem;
193 sdev->bios[i] = sbio;
194
a2de733c
AJ		 195 sbio->index = i;
196 sbio->sdev = sdev;
a2de733c
AJ		 197 sbio->count = 0;
198 sbio->work.func = scrub_checksum;
a2de733c
AJ		 199
200 if (i != SCRUB_BIOS_PER_DEV-1)
201 sdev->bios[i]->next_free = i + 1;
0ef8e451 202 else
a2de733c
AJ		 203 sdev->bios[i]->next_free = -1;
204 }
205 sdev->first_free = 0;
206 sdev->curr = -1;
207 atomic_set(&sdev->in_flight, 0);
0ef8e451 208 atomic_set(&sdev->fixup_cnt, 0);
a2de733c 209 atomic_set(&sdev->cancel_req, 0);
6c41761f 210 sdev->csum_size = btrfs_super_csum_size(fs_info->super_copy);
a2de733c
AJ		 211 INIT_LIST_HEAD(&sdev->csum_list);
212
213 spin_lock_init(&sdev->list_lock);
214 spin_lock_init(&sdev->stat_lock);
215 init_waitqueue_head(&sdev->list_wait);
216 return sdev;
217
218nomem:
219 scrub_free_dev(sdev);
220 return ERR_PTR(-ENOMEM);
221}
222
558540c1
JS		 223static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, void *ctx)
224{
225 u64 isize;
226 u32 nlink;
227 int ret;
228 int i;
229 struct extent_buffer *eb;
230 struct btrfs_inode_item *inode_item;
231 struct scrub_warning *swarn = ctx;
232 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
233 struct inode_fs_paths *ipath = NULL;
234 struct btrfs_root *local_root;
235 struct btrfs_key root_key;
236
237 root_key.objectid = root;
238 root_key.type = BTRFS_ROOT_ITEM_KEY;
239 root_key.offset = (u64)-1;
240 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
241 if (IS_ERR(local_root)) {
242 ret = PTR_ERR(local_root);
243 goto err;
244 }
245
246 ret = inode_item_info(inum, 0, local_root, swarn->path);
247 if (ret) {
248 btrfs_release_path(swarn->path);
249 goto err;
250 }
251
252 eb = swarn->path->nodes[0];
253 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
254 struct btrfs_inode_item);
255 isize = btrfs_inode_size(eb, inode_item);
256 nlink = btrfs_inode_nlink(eb, inode_item);
257 btrfs_release_path(swarn->path);
258
259 ipath = init_ipath(4096, local_root, swarn->path);
26bdef54
DC		 260 if (IS_ERR(ipath)) {
261 ret = PTR_ERR(ipath);
262 ipath = NULL;
263 goto err;
264 }
558540c1
JS		 265 ret = paths_from_inode(inum, ipath);
266
267 if (ret < 0)
268 goto err;
269
270 /*
271 * we deliberately ignore the bit ipath might have been too small to
272 * hold all of the paths here
273 */
274 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
275 printk(KERN_WARNING "btrfs: %s at logical %llu on dev "
276 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
277 "length %llu, links %u (path: %s)\n", swarn->errstr,
278 swarn->logical, swarn->dev->name,
279 (unsigned long long)swarn->sector, root, inum, offset,
280 min(isize - offset, (u64)PAGE_SIZE), nlink,
745c4d8e 281 (char *)(unsigned long)ipath->fspath->val[i]);
558540c1
JS		 282
283 free_ipath(ipath);
284 return 0;
285
286err:
287 printk(KERN_WARNING "btrfs: %s at logical %llu on dev "
288 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
289 "resolving failed with ret=%d\n", swarn->errstr,
290 swarn->logical, swarn->dev->name,
291 (unsigned long long)swarn->sector, root, inum, offset, ret);
292
293 free_ipath(ipath);
294 return 0;
295}
296
297static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio,
298 int ix)
299{
300 struct btrfs_device *dev = sbio->sdev->dev;
301 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
302 struct btrfs_path *path;
303 struct btrfs_key found_key;
304 struct extent_buffer *eb;
305 struct btrfs_extent_item *ei;
306 struct scrub_warning swarn;
307 u32 item_size;
308 int ret;
309 u64 ref_root;
310 u8 ref_level;
311 unsigned long ptr = 0;
312 const int bufsize = 4096;
4692cf58 313 u64 extent_item_pos;
558540c1
JS		 314
315 path = btrfs_alloc_path();
316
317 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
318 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
319 swarn.sector = (sbio->physical + ix * PAGE_SIZE) >> 9;
320 swarn.logical = sbio->logical + ix * PAGE_SIZE;
321 swarn.errstr = errstr;
322 swarn.dev = dev;
323 swarn.msg_bufsize = bufsize;
324 swarn.scratch_bufsize = bufsize;
325
326 if (!path || !swarn.scratch_buf || !swarn.msg_buf)
327 goto out;
328
329 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key);
330 if (ret < 0)
331 goto out;
332
4692cf58 333 extent_item_pos = swarn.logical - found_key.objectid;
558540c1
JS		 334 swarn.extent_item_size = found_key.offset;
335
336 eb = path->nodes[0];
337 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
338 item_size = btrfs_item_size_nr(eb, path->slots[0]);
4692cf58 339 btrfs_release_path(path);
558540c1
JS		 340
341 if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
342 do {
343 ret = tree_backref_for_extent(&ptr, eb, ei, item_size,
344 &ref_root, &ref_level);
345 printk(KERN_WARNING "%s at logical %llu on dev %s, "
346 "sector %llu: metadata %s (level %d) in tree "
347 "%llu\n", errstr, swarn.logical, dev->name,
348 (unsigned long long)swarn.sector,
349 ref_level ? "node" : "leaf",
350 ret < 0 ? -1 : ref_level,
351 ret < 0 ? -1 : ref_root);
352 } while (ret != 1);
353 } else {
354 swarn.path = path;
355 iterate_extent_inodes(fs_info, path, found_key.objectid,
4692cf58 356 extent_item_pos,
558540c1
JS		 357 scrub_print_warning_inode, &swarn);
358 }
359
360out:
361 btrfs_free_path(path);
362 kfree(swarn.scratch_buf);
363 kfree(swarn.msg_buf);
364}
365
0ef8e451
JS		 366static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *ctx)
367{
5da6fcbc 368 struct page *page = NULL;
0ef8e451
JS		 369 unsigned long index;
370 struct scrub_fixup_nodatasum *fixup = ctx;
371 int ret;
5da6fcbc 372 int corrected = 0;
0ef8e451 373 struct btrfs_key key;
5da6fcbc 374 struct inode *inode = NULL;
0ef8e451
JS		 375 u64 end = offset + PAGE_SIZE - 1;
376 struct btrfs_root *local_root;
377
378 key.objectid = root;
379 key.type = BTRFS_ROOT_ITEM_KEY;
380 key.offset = (u64)-1;
381 local_root = btrfs_read_fs_root_no_name(fixup->root->fs_info, &key);
382 if (IS_ERR(local_root))
383 return PTR_ERR(local_root);
384
385 key.type = BTRFS_INODE_ITEM_KEY;
386 key.objectid = inum;
387 key.offset = 0;
388 inode = btrfs_iget(fixup->root->fs_info->sb, &key, local_root, NULL);
389 if (IS_ERR(inode))
390 return PTR_ERR(inode);
391
0ef8e451
JS		 392 index = offset >> PAGE_CACHE_SHIFT;
393
394 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
5da6fcbc
JS		 395 if (!page) {
396 ret = -ENOMEM;
397 goto out;
398 }
399
400 if (PageUptodate(page)) {
401 struct btrfs_mapping_tree *map_tree;
402 if (PageDirty(page)) {
403 /*
404 * we need to write the data to the defect sector. the
405 * data that was in that sector is not in memory,
406 * because the page was modified. we must not write the
407 * modified page to that sector.
408 *
409 * TODO: what could be done here: wait for the delalloc
410 * runner to write out that page (might involve
411 * COW) and see whether the sector is still
412 * referenced afterwards.
413 *
414 * For the meantime, we'll treat this error
415 * incorrectable, although there is a chance that a
416 * later scrub will find the bad sector again and that
417 * there's no dirty page in memory, then.
418 */
419 ret = -EIO;
420 goto out;
421 }
422 map_tree = &BTRFS_I(inode)->root->fs_info->mapping_tree;
423 ret = repair_io_failure(map_tree, offset, PAGE_SIZE,
424 fixup->logical, page,
425 fixup->mirror_num);
426 unlock_page(page);
427 corrected = !ret;
428 } else {
429 /*
430 * we need to get good data first. the general readpage path
431 * will call repair_io_failure for us, we just have to make
432 * sure we read the bad mirror.
433 */
434 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
435 EXTENT_DAMAGED, GFP_NOFS);
436 if (ret) {
437 /* set_extent_bits should give proper error */
438 WARN_ON(ret > 0);
439 if (ret > 0)
440 ret = -EFAULT;
441 goto out;
442 }
443
444 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
445 btrfs_get_extent,
446 fixup->mirror_num);
447 wait_on_page_locked(page);
448
449 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
450 end, EXTENT_DAMAGED, 0, NULL);
451 if (!corrected)
452 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
453 EXTENT_DAMAGED, GFP_NOFS);
454 }
455
456out:
457 if (page)
458 put_page(page);
459 if (inode)
460 iput(inode);
0ef8e451
JS		 461
462 if (ret < 0)
463 return ret;
464
465 if (ret == 0 && corrected) {
466 /*
467 * we only need to call readpage for one of the inodes belonging
468 * to this extent. so make iterate_extent_inodes stop
469 */
470 return 1;
471 }
472
473 return -EIO;
474}
475
476static void scrub_fixup_nodatasum(struct btrfs_work *work)
477{
478 int ret;
479 struct scrub_fixup_nodatasum *fixup;
480 struct scrub_dev *sdev;
481 struct btrfs_trans_handle *trans = NULL;
482 struct btrfs_fs_info *fs_info;
483 struct btrfs_path *path;
484 int uncorrectable = 0;
485
486 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
487 sdev = fixup->sdev;
488 fs_info = fixup->root->fs_info;
489
490 path = btrfs_alloc_path();
491 if (!path) {
492 spin_lock(&sdev->stat_lock);
493 ++sdev->stat.malloc_errors;
494 spin_unlock(&sdev->stat_lock);
495 uncorrectable = 1;
496 goto out;
497 }
498
499 trans = btrfs_join_transaction(fixup->root);
500 if (IS_ERR(trans)) {
501 uncorrectable = 1;
502 goto out;
503 }
504
505 /*
506 * the idea is to trigger a regular read through the standard path. we
507 * read a page from the (failed) logical address by specifying the
508 * corresponding copynum of the failed sector. thus, that readpage is
509 * expected to fail.
510 * that is the point where on-the-fly error correction will kick in
511 * (once it's finished) and rewrite the failed sector if a good copy
512 * can be found.
513 */
514 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
515 path, scrub_fixup_readpage,
516 fixup);
517 if (ret < 0) {
518 uncorrectable = 1;
519 goto out;
520 }
521 WARN_ON(ret != 1);
522
523 spin_lock(&sdev->stat_lock);
524 ++sdev->stat.corrected_errors;
525 spin_unlock(&sdev->stat_lock);
526
527out:
528 if (trans && !IS_ERR(trans))
529 btrfs_end_transaction(trans, fixup->root);
530 if (uncorrectable) {
531 spin_lock(&sdev->stat_lock);
532 ++sdev->stat.uncorrectable_errors;
533 spin_unlock(&sdev->stat_lock);
534 printk_ratelimited(KERN_ERR "btrfs: unable to fixup "
535 "(nodatasum) error at logical %llu\n",
536 fixup->logical);
537 }
538
539 btrfs_free_path(path);
540 kfree(fixup);
541
542 /* see caller why we're pretending to be paused in the scrub counters */
543 mutex_lock(&fs_info->scrub_lock);
544 atomic_dec(&fs_info->scrubs_running);
545 atomic_dec(&fs_info->scrubs_paused);
546 mutex_unlock(&fs_info->scrub_lock);
547 atomic_dec(&sdev->fixup_cnt);
548 wake_up(&fs_info->scrub_pause_wait);
549 wake_up(&sdev->list_wait);
550}
551
a2de733c
AJ		 552/*
553 * scrub_recheck_error gets called when either verification of the page
554 * failed or the bio failed to read, e.g. with EIO. In the latter case,
555 * recheck_error gets called for every page in the bio, even though only
556 * one may be bad
557 */
13db62b7 558static int scrub_recheck_error(struct scrub_bio *sbio, int ix)
a2de733c 559{
13db62b7
JS		 560 struct scrub_dev *sdev = sbio->sdev;
561 u64 sector = (sbio->physical + ix * PAGE_SIZE) >> 9;
558540c1
JS		 562 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
563 DEFAULT_RATELIMIT_BURST);
13db62b7 564
96e36920 565 if (sbio->err) {
13db62b7 566 if (scrub_fixup_io(READ, sbio->sdev->dev->bdev, sector,
96e36920
ID		 567 sbio->bio->bi_io_vec[ix].bv_page) == 0) {
568 if (scrub_fixup_check(sbio, ix) == 0)
13db62b7 569 return 0;
96e36920 570 }
558540c1
JS		 571 if (__ratelimit(&_rs))
572 scrub_print_warning("i/o error", sbio, ix);
573 } else {
574 if (__ratelimit(&_rs))
575 scrub_print_warning("checksum error", sbio, ix);
a2de733c
AJ		 576 }
577
13db62b7
JS		 578 spin_lock(&sdev->stat_lock);
579 ++sdev->stat.read_errors;
580 spin_unlock(&sdev->stat_lock);
581
96e36920 582 scrub_fixup(sbio, ix);
13db62b7 583 return 1;
a2de733c
AJ		 584}
585
96e36920 586static int scrub_fixup_check(struct scrub_bio *sbio, int ix)
a2de733c
AJ		 587{
588 int ret = 1;
589 struct page *page;
590 void *buffer;
96e36920 591 u64 flags = sbio->spag[ix].flags;
a2de733c 592
96e36920 593 page = sbio->bio->bi_io_vec[ix].bv_page;
a2de733c
AJ		 594 buffer = kmap_atomic(page, KM_USER0);
595 if (flags & BTRFS_EXTENT_FLAG_DATA) {
96e36920
ID		 596 ret = scrub_checksum_data(sbio->sdev,
597 sbio->spag + ix, buffer);
a2de733c 598 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
96e36920
ID		 599 ret = scrub_checksum_tree_block(sbio->sdev,
600 sbio->spag + ix,
601 sbio->logical + ix * PAGE_SIZE,
a2de733c
AJ		 602 buffer);
603 } else {
604 WARN_ON(1);
605 }
606 kunmap_atomic(buffer, KM_USER0);
607
608 return ret;
609}
610
a2de733c
AJ		 611static void scrub_fixup_end_io(struct bio *bio, int err)
612{
613 complete((struct completion *)bio->bi_private);
614}
615
96e36920 616static void scrub_fixup(struct scrub_bio *sbio, int ix)
a2de733c 617{
96e36920 618 struct scrub_dev *sdev = sbio->sdev;
a2de733c
AJ		 619 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
620 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
a1d3c478 621 struct btrfs_bio *bbio = NULL;
0ef8e451 622 struct scrub_fixup_nodatasum *fixup;
96e36920 623 u64 logical = sbio->logical + ix * PAGE_SIZE;
a2de733c
AJ		 624 u64 length;
625 int i;
626 int ret;
627 DECLARE_COMPLETION_ONSTACK(complete);
628
96e36920
ID		 629 if ((sbio->spag[ix].flags & BTRFS_EXTENT_FLAG_DATA) &&
630 (sbio->spag[ix].have_csum == 0)) {
0ef8e451
JS		 631 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
632 if (!fixup)
633 goto uncorrectable;
634 fixup->sdev = sdev;
635 fixup->logical = logical;
636 fixup->root = fs_info->extent_root;
637 fixup->mirror_num = sbio->spag[ix].mirror_num;
a2de733c 638 /*
0ef8e451
JS		 639 * increment scrubs_running to prevent cancel requests from
640 * completing as long as a fixup worker is running. we must also
641 * increment scrubs_paused to prevent deadlocking on pause
642 * requests used for transactions commits (as the worker uses a
643 * transaction context). it is safe to regard the fixup worker
644 * as paused for all matters practical. effectively, we only
645 * avoid cancellation requests from completing.
a2de733c 646 */
0ef8e451
JS		 647 mutex_lock(&fs_info->scrub_lock);
648 atomic_inc(&fs_info->scrubs_running);
649 atomic_inc(&fs_info->scrubs_paused);
650 mutex_unlock(&fs_info->scrub_lock);
651 atomic_inc(&sdev->fixup_cnt);
652 fixup->work.func = scrub_fixup_nodatasum;
653 btrfs_queue_worker(&fs_info->scrub_workers, &fixup->work);
654 return;
a2de733c
AJ		 655 }
656
657 length = PAGE_SIZE;
96e36920 658 ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length,
a1d3c478
JS		 659 &bbio, 0);
660 if (ret || !bbio || length < PAGE_SIZE) {
a2de733c
AJ		 661 printk(KERN_ERR
662 "scrub_fixup: btrfs_map_block failed us for %llu\n",
96e36920 663 (unsigned long long)logical);
a2de733c 664 WARN_ON(1);
56d2a48f 665 kfree(bbio);
a2de733c
AJ		 666 return;
667 }
668
a1d3c478 669 if (bbio->num_stripes == 1)
a2de733c
AJ		 670 /* there aren't any replicas */
671 goto uncorrectable;
a2de733c
AJ		 672
673 /*
674 * first find a good copy
675 */
a1d3c478 676 for (i = 0; i < bbio->num_stripes; ++i) {
193ea74b 677 if (i + 1 == sbio->spag[ix].mirror_num)
a2de733c
AJ		 678 continue;
679
a1d3c478
JS		 680 if (scrub_fixup_io(READ, bbio->stripes[i].dev->bdev,
681 bbio->stripes[i].physical >> 9,
96e36920 682 sbio->bio->bi_io_vec[ix].bv_page)) {
a2de733c
AJ		 683 /* I/O-error, this is not a good copy */
684 continue;
96e36920 685 }
a2de733c 686
96e36920 687 if (scrub_fixup_check(sbio, ix) == 0)
a2de733c
AJ		 688 break;
689 }
a1d3c478 690 if (i == bbio->num_stripes)
a2de733c
AJ		 691 goto uncorrectable;
692
8628764e
AJ		 693 if (!sdev->readonly) {
694 /*
695 * bi_io_vec[ix].bv_page now contains good data, write it back
696 */
697 if (scrub_fixup_io(WRITE, sdev->dev->bdev,
698 (sbio->physical + ix * PAGE_SIZE) >> 9,
699 sbio->bio->bi_io_vec[ix].bv_page)) {
700 /* I/O-error, writeback failed, give up */
701 goto uncorrectable;
702 }
96e36920 703 }
a2de733c 704
a1d3c478 705 kfree(bbio);
a2de733c
AJ		 706 spin_lock(&sdev->stat_lock);
707 ++sdev->stat.corrected_errors;
708 spin_unlock(&sdev->stat_lock);
709
558540c1
JS		 710 printk_ratelimited(KERN_ERR "btrfs: fixed up error at logical %llu\n",
711 (unsigned long long)logical);
a2de733c
AJ		 712 return;
713
714uncorrectable:
a1d3c478 715 kfree(bbio);
a2de733c
AJ		 716 spin_lock(&sdev->stat_lock);
717 ++sdev->stat.uncorrectable_errors;
718 spin_unlock(&sdev->stat_lock);
719
558540c1
JS		 720 printk_ratelimited(KERN_ERR "btrfs: unable to fixup (regular) error at "
721 "logical %llu\n", (unsigned long long)logical);
96e36920
ID		 722}
723
724static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector,
725 struct page *page)
726{
727 struct bio *bio = NULL;
728 int ret;
729 DECLARE_COMPLETION_ONSTACK(complete);
730
96e36920
ID		 731 bio = bio_alloc(GFP_NOFS, 1);
732 bio->bi_bdev = bdev;
733 bio->bi_sector = sector;
734 bio_add_page(bio, page, PAGE_SIZE, 0);
735 bio->bi_end_io = scrub_fixup_end_io;
736 bio->bi_private = &complete;
21adbd5c 737 btrfsic_submit_bio(rw, bio);
96e36920 738
e7786c3a 739 /* this will also unplug the queue */
96e36920
ID		 740 wait_for_completion(&complete);
741
742 ret = !test_bit(BIO_UPTODATE, &bio->bi_flags);
743 bio_put(bio);
744 return ret;
a2de733c
AJ		 745}
746
747static void scrub_bio_end_io(struct bio *bio, int err)
748{
749 struct scrub_bio *sbio = bio->bi_private;
750 struct scrub_dev *sdev = sbio->sdev;
751 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
752
753 sbio->err = err;
1bc87793 754 sbio->bio = bio;
a2de733c
AJ		 755
756 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work);
757}
758
759static void scrub_checksum(struct btrfs_work *work)
760{
761 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
762 struct scrub_dev *sdev = sbio->sdev;
763 struct page *page;
764 void *buffer;
765 int i;
766 u64 flags;
767 u64 logical;
768 int ret;
769
770 if (sbio->err) {
13db62b7 771 ret = 0;
a2de733c 772 for (i = 0; i < sbio->count; ++i)
13db62b7
JS		 773 ret |= scrub_recheck_error(sbio, i);
774 if (!ret) {
775 spin_lock(&sdev->stat_lock);
776 ++sdev->stat.unverified_errors;
777 spin_unlock(&sdev->stat_lock);
778 }
96e36920
ID		 779
780 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1);
781 sbio->bio->bi_flags |= 1 << BIO_UPTODATE;
782 sbio->bio->bi_phys_segments = 0;
783 sbio->bio->bi_idx = 0;
784
785 for (i = 0; i < sbio->count; i++) {
786 struct bio_vec *bi;
787 bi = &sbio->bio->bi_io_vec[i];
788 bi->bv_offset = 0;
789 bi->bv_len = PAGE_SIZE;
790 }
a2de733c
AJ		 791 goto out;
792 }
793 for (i = 0; i < sbio->count; ++i) {
794 page = sbio->bio->bi_io_vec[i].bv_page;
795 buffer = kmap_atomic(page, KM_USER0);
796 flags = sbio->spag[i].flags;
797 logical = sbio->logical + i * PAGE_SIZE;
798 ret = 0;
799 if (flags & BTRFS_EXTENT_FLAG_DATA) {
800 ret = scrub_checksum_data(sdev, sbio->spag + i, buffer);
801 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
802 ret = scrub_checksum_tree_block(sdev, sbio->spag + i,
803 logical, buffer);
804 } else if (flags & BTRFS_EXTENT_FLAG_SUPER) {
805 BUG_ON(i);
806 (void)scrub_checksum_super(sbio, buffer);
807 } else {
808 WARN_ON(1);
809 }
810 kunmap_atomic(buffer, KM_USER0);
13db62b7
JS		 811 if (ret) {
812 ret = scrub_recheck_error(sbio, i);
813 if (!ret) {
814 spin_lock(&sdev->stat_lock);
815 ++sdev->stat.unverified_errors;
816 spin_unlock(&sdev->stat_lock);
817 }
818 }
a2de733c
AJ		 819 }
820
821out:
1bc87793
AJ		 822 scrub_free_bio(sbio->bio);
823 sbio->bio = NULL;
a2de733c
AJ		 824 spin_lock(&sdev->list_lock);
825 sbio->next_free = sdev->first_free;
826 sdev->first_free = sbio->index;
827 spin_unlock(&sdev->list_lock);
a2de733c
AJ		 828 atomic_dec(&sdev->in_flight);
829 wake_up(&sdev->list_wait);
830}
831
832static int scrub_checksum_data(struct scrub_dev *sdev,
833 struct scrub_page *spag, void *buffer)
834{
835 u8 csum[BTRFS_CSUM_SIZE];
836 u32 crc = ~(u32)0;
837 int fail = 0;
838 struct btrfs_root *root = sdev->dev->dev_root;
839
840 if (!spag->have_csum)
841 return 0;
842
843 crc = btrfs_csum_data(root, buffer, crc, PAGE_SIZE);
844 btrfs_csum_final(crc, csum);
845 if (memcmp(csum, spag->csum, sdev->csum_size))
846 fail = 1;
847
848 spin_lock(&sdev->stat_lock);
849 ++sdev->stat.data_extents_scrubbed;
850 sdev->stat.data_bytes_scrubbed += PAGE_SIZE;
851 if (fail)
852 ++sdev->stat.csum_errors;
853 spin_unlock(&sdev->stat_lock);
854
855 return fail;
856}
857
858static int scrub_checksum_tree_block(struct scrub_dev *sdev,
859 struct scrub_page *spag, u64 logical,
860 void *buffer)
861{
862 struct btrfs_header *h;
863 struct btrfs_root *root = sdev->dev->dev_root;
864 struct btrfs_fs_info *fs_info = root->fs_info;
865 u8 csum[BTRFS_CSUM_SIZE];
866 u32 crc = ~(u32)0;
867 int fail = 0;
868 int crc_fail = 0;
869
870 /*
871 * we don't use the getter functions here, as we
872 * a) don't have an extent buffer and
873 * b) the page is already kmapped
874 */
875 h = (struct btrfs_header *)buffer;
876
877 if (logical != le64_to_cpu(h->bytenr))
878 ++fail;
879
880 if (spag->generation != le64_to_cpu(h->generation))
881 ++fail;
882
883 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
884 ++fail;
885
886 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
887 BTRFS_UUID_SIZE))
888 ++fail;
889
890 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
891 PAGE_SIZE - BTRFS_CSUM_SIZE);
892 btrfs_csum_final(crc, csum);
893 if (memcmp(csum, h->csum, sdev->csum_size))
894 ++crc_fail;
895
896 spin_lock(&sdev->stat_lock);
897 ++sdev->stat.tree_extents_scrubbed;
898 sdev->stat.tree_bytes_scrubbed += PAGE_SIZE;
899 if (crc_fail)
900 ++sdev->stat.csum_errors;
901 if (fail)
902 ++sdev->stat.verify_errors;
903 spin_unlock(&sdev->stat_lock);
904
905 return fail || crc_fail;
906}
907
908static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer)
909{
910 struct btrfs_super_block *s;
911 u64 logical;
912 struct scrub_dev *sdev = sbio->sdev;
913 struct btrfs_root *root = sdev->dev->dev_root;
914 struct btrfs_fs_info *fs_info = root->fs_info;
915 u8 csum[BTRFS_CSUM_SIZE];
916 u32 crc = ~(u32)0;
917 int fail = 0;
918
919 s = (struct btrfs_super_block *)buffer;
920 logical = sbio->logical;
921
922 if (logical != le64_to_cpu(s->bytenr))
923 ++fail;
924
925 if (sbio->spag[0].generation != le64_to_cpu(s->generation))
926 ++fail;
927
928 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
929 ++fail;
930
931 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc,
932 PAGE_SIZE - BTRFS_CSUM_SIZE);
933 btrfs_csum_final(crc, csum);
934 if (memcmp(csum, s->csum, sbio->sdev->csum_size))
935 ++fail;
936
937 if (fail) {
938 /*
939 * if we find an error in a super block, we just report it.
940 * They will get written with the next transaction commit
941 * anyway
942 */
943 spin_lock(&sdev->stat_lock);
944 ++sdev->stat.super_errors;
945 spin_unlock(&sdev->stat_lock);
946 }
947
948 return fail;
949}
950
143bede5 951static void scrub_submit(struct scrub_dev *sdev)
a2de733c
AJ		 952{
953 struct scrub_bio *sbio;
954
955 if (sdev->curr == -1)
143bede5 956 return;
a2de733c
AJ		 957
958 sbio = sdev->bios[sdev->curr];
a2de733c
AJ		 959 sbio->err = 0;
960 sdev->curr = -1;
961 atomic_inc(&sdev->in_flight);
962
21adbd5c 963 btrfsic_submit_bio(READ, sbio->bio);
a2de733c
AJ		 964}
965
966static int scrub_page(struct scrub_dev *sdev, u64 logical, u64 len,
e12fa9cd 967 u64 physical, u64 flags, u64 gen, int mirror_num,
a2de733c
AJ		 968 u8 *csum, int force)
969{
970 struct scrub_bio *sbio;
69f4cb52
AJ		 971 struct page *page;
972 int ret;
a2de733c
AJ		 973
974again:
975 /*
976 * grab a fresh bio or wait for one to become available
977 */
978 while (sdev->curr == -1) {
979 spin_lock(&sdev->list_lock);
980 sdev->curr = sdev->first_free;
981 if (sdev->curr != -1) {
982 sdev->first_free = sdev->bios[sdev->curr]->next_free;
983 sdev->bios[sdev->curr]->next_free = -1;
984 sdev->bios[sdev->curr]->count = 0;
985 spin_unlock(&sdev->list_lock);
986 } else {
987 spin_unlock(&sdev->list_lock);
988 wait_event(sdev->list_wait, sdev->first_free != -1);
989 }
990 }
991 sbio = sdev->bios[sdev->curr];
992 if (sbio->count == 0) {
69f4cb52
AJ		 993 struct bio *bio;
994
a2de733c
AJ		 995 sbio->physical = physical;
996 sbio->logical = logical;
69f4cb52
AJ		 997 bio = bio_alloc(GFP_NOFS, SCRUB_PAGES_PER_BIO);
998 if (!bio)
999 return -ENOMEM;
1000
1001 bio->bi_private = sbio;
1002 bio->bi_end_io = scrub_bio_end_io;
1003 bio->bi_bdev = sdev->dev->bdev;
1004 bio->bi_sector = sbio->physical >> 9;
1005 sbio->err = 0;
1006 sbio->bio = bio;
00d01bc1
AJ		 1007 } else if (sbio->physical + sbio->count * PAGE_SIZE != physical ||
1008 sbio->logical + sbio->count * PAGE_SIZE != logical) {
143bede5 1009 scrub_submit(sdev);
a2de733c
AJ		 1010 goto again;
1011 }
1012 sbio->spag[sbio->count].flags = flags;
1013 sbio->spag[sbio->count].generation = gen;
1014 sbio->spag[sbio->count].have_csum = 0;
1015 sbio->spag[sbio->count].mirror_num = mirror_num;
69f4cb52
AJ		 1016
1017 page = alloc_page(GFP_NOFS);
1018 if (!page)
1019 return -ENOMEM;
1020
1021 ret = bio_add_page(sbio->bio, page, PAGE_SIZE, 0);
1022 if (!ret) {
1023 __free_page(page);
143bede5 1024 scrub_submit(sdev);
69f4cb52
AJ		 1025 goto again;
1026 }
1027
a2de733c
AJ		 1028 if (csum) {
1029 sbio->spag[sbio->count].have_csum = 1;
1030 memcpy(sbio->spag[sbio->count].csum, csum, sdev->csum_size);
1031 }
1032 ++sbio->count;
143bede5
JM		 1033 if (sbio->count == SCRUB_PAGES_PER_BIO || force)
1034 scrub_submit(sdev);
a2de733c
AJ		 1035
1036 return 0;
1037}
1038
1039static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len,
1040 u8 *csum)
1041{
1042 struct btrfs_ordered_sum *sum = NULL;
1043 int ret = 0;
1044 unsigned long i;
1045 unsigned long num_sectors;
1046 u32 sectorsize = sdev->dev->dev_root->sectorsize;
1047
1048 while (!list_empty(&sdev->csum_list)) {
1049 sum = list_first_entry(&sdev->csum_list,
1050 struct btrfs_ordered_sum, list);
1051 if (sum->bytenr > logical)
1052 return 0;
1053 if (sum->bytenr + sum->len > logical)
1054 break;
1055
1056 ++sdev->stat.csum_discards;
1057 list_del(&sum->list);
1058 kfree(sum);
1059 sum = NULL;
1060 }
1061 if (!sum)
1062 return 0;
1063
1064 num_sectors = sum->len / sectorsize;
1065 for (i = 0; i < num_sectors; ++i) {
1066 if (sum->sums[i].bytenr == logical) {
1067 memcpy(csum, &sum->sums[i].sum, sdev->csum_size);
1068 ret = 1;
1069 break;
1070 }
1071 }
1072 if (ret && i == num_sectors - 1) {
1073 list_del(&sum->list);
1074 kfree(sum);
1075 }
1076 return ret;
1077}
1078
1079/* scrub extent tries to collect up to 64 kB for each bio */
1080static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len,
e12fa9cd 1081 u64 physical, u64 flags, u64 gen, int mirror_num)
a2de733c
AJ		 1082{
1083 int ret;
1084 u8 csum[BTRFS_CSUM_SIZE];
1085
1086 while (len) {
1087 u64 l = min_t(u64, len, PAGE_SIZE);
1088 int have_csum = 0;
1089
1090 if (flags & BTRFS_EXTENT_FLAG_DATA) {
1091 /* push csums to sbio */
1092 have_csum = scrub_find_csum(sdev, logical, l, csum);
1093 if (have_csum == 0)
1094 ++sdev->stat.no_csum;
1095 }
1096 ret = scrub_page(sdev, logical, l, physical, flags, gen,
1097 mirror_num, have_csum ? csum : NULL, 0);
1098 if (ret)
1099 return ret;
1100 len -= l;
1101 logical += l;
1102 physical += l;
1103 }
1104 return 0;
1105}
1106
1107static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev,
1108 struct map_lookup *map, int num, u64 base, u64 length)
1109{
1110 struct btrfs_path *path;
1111 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info;
1112 struct btrfs_root *root = fs_info->extent_root;
1113 struct btrfs_root *csum_root = fs_info->csum_root;
1114 struct btrfs_extent_item *extent;
e7786c3a 1115 struct blk_plug plug;
a2de733c
AJ		 1116 u64 flags;
1117 int ret;
1118 int slot;
1119 int i;
1120 u64 nstripes;
a2de733c
AJ		 1121 struct extent_buffer *l;
1122 struct btrfs_key key;
1123 u64 physical;
1124 u64 logical;
1125 u64 generation;
e12fa9cd 1126 int mirror_num;
7a26285e
AJ		 1127 struct reada_control *reada1;
1128 struct reada_control *reada2;
1129 struct btrfs_key key_start;
1130 struct btrfs_key key_end;
a2de733c
AJ		 1131
1132 u64 increment = map->stripe_len;
1133 u64 offset;
1134
1135 nstripes = length;
1136 offset = 0;
1137 do_div(nstripes, map->stripe_len);
1138 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1139 offset = map->stripe_len * num;
1140 increment = map->stripe_len * map->num_stripes;
193ea74b 1141 mirror_num = 1;
a2de733c
AJ		 1142 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1143 int factor = map->num_stripes / map->sub_stripes;
1144 offset = map->stripe_len * (num / map->sub_stripes);
1145 increment = map->stripe_len * factor;
193ea74b 1146 mirror_num = num % map->sub_stripes + 1;
a2de733c
AJ		 1147 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1148 increment = map->stripe_len;
193ea74b 1149 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 1150 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1151 increment = map->stripe_len;
193ea74b 1152 mirror_num = num % map->num_stripes + 1;
a2de733c
AJ		 1153 } else {
1154 increment = map->stripe_len;
193ea74b 1155 mirror_num = 1;
a2de733c
AJ		 1156 }
1157
1158 path = btrfs_alloc_path();
1159 if (!path)
1160 return -ENOMEM;
1161
a2de733c
AJ		 1162 path->search_commit_root = 1;
1163 path->skip_locking = 1;
1164
1165 /*
7a26285e
AJ		 1166 * trigger the readahead for extent tree csum tree and wait for
1167 * completion. During readahead, the scrub is officially paused
1168 * to not hold off transaction commits
a2de733c
AJ		 1169 */
1170 logical = base + offset;
a2de733c 1171
7a26285e
AJ		 1172 wait_event(sdev->list_wait,
1173 atomic_read(&sdev->in_flight) == 0);
1174 atomic_inc(&fs_info->scrubs_paused);
1175 wake_up(&fs_info->scrub_pause_wait);
1176
1177 /* FIXME it might be better to start readahead at commit root */
1178 key_start.objectid = logical;
1179 key_start.type = BTRFS_EXTENT_ITEM_KEY;
1180 key_start.offset = (u64)0;
1181 key_end.objectid = base + offset + nstripes * increment;
1182 key_end.type = BTRFS_EXTENT_ITEM_KEY;
1183 key_end.offset = (u64)0;
1184 reada1 = btrfs_reada_add(root, &key_start, &key_end);
1185
1186 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1187 key_start.type = BTRFS_EXTENT_CSUM_KEY;
1188 key_start.offset = logical;
1189 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1190 key_end.type = BTRFS_EXTENT_CSUM_KEY;
1191 key_end.offset = base + offset + nstripes * increment;
1192 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
1193
1194 if (!IS_ERR(reada1))
1195 btrfs_reada_wait(reada1);
1196 if (!IS_ERR(reada2))
1197 btrfs_reada_wait(reada2);
1198
1199 mutex_lock(&fs_info->scrub_lock);
1200 while (atomic_read(&fs_info->scrub_pause_req)) {
1201 mutex_unlock(&fs_info->scrub_lock);
1202 wait_event(fs_info->scrub_pause_wait,
1203 atomic_read(&fs_info->scrub_pause_req) == 0);
1204 mutex_lock(&fs_info->scrub_lock);
a2de733c 1205 }
7a26285e
AJ		 1206 atomic_dec(&fs_info->scrubs_paused);
1207 mutex_unlock(&fs_info->scrub_lock);
1208 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 1209
1210 /*
1211 * collect all data csums for the stripe to avoid seeking during
1212 * the scrub. This might currently (crc32) end up to be about 1MB
1213 */
e7786c3a 1214 blk_start_plug(&plug);
a2de733c 1215
a2de733c
AJ		 1216 /*
1217 * now find all extents for each stripe and scrub them
1218 */
7a26285e
AJ		 1219 logical = base + offset;
1220 physical = map->stripes[num].physical;
a2de733c 1221 ret = 0;
7a26285e 1222 for (i = 0; i < nstripes; ++i) {
a2de733c
AJ		 1223 /*
1224 * canceled?
1225 */
1226 if (atomic_read(&fs_info->scrub_cancel_req) ||
1227 atomic_read(&sdev->cancel_req)) {
1228 ret = -ECANCELED;
1229 goto out;
1230 }
1231 /*
1232 * check to see if we have to pause
1233 */
1234 if (atomic_read(&fs_info->scrub_pause_req)) {
1235 /* push queued extents */
1236 scrub_submit(sdev);
1237 wait_event(sdev->list_wait,
1238 atomic_read(&sdev->in_flight) == 0);
1239 atomic_inc(&fs_info->scrubs_paused);
1240 wake_up(&fs_info->scrub_pause_wait);
1241 mutex_lock(&fs_info->scrub_lock);
1242 while (atomic_read(&fs_info->scrub_pause_req)) {
1243 mutex_unlock(&fs_info->scrub_lock);
1244 wait_event(fs_info->scrub_pause_wait,
1245 atomic_read(&fs_info->scrub_pause_req) == 0);
1246 mutex_lock(&fs_info->scrub_lock);
1247 }
1248 atomic_dec(&fs_info->scrubs_paused);
1249 mutex_unlock(&fs_info->scrub_lock);
1250 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 1251 }
1252
7a26285e
AJ		 1253 ret = btrfs_lookup_csums_range(csum_root, logical,
1254 logical + map->stripe_len - 1,
1255 &sdev->csum_list, 1);
1256 if (ret)
1257 goto out;
1258
a2de733c
AJ		 1259 key.objectid = logical;
1260 key.type = BTRFS_EXTENT_ITEM_KEY;
1261 key.offset = (u64)0;
1262
1263 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1264 if (ret < 0)
1265 goto out;
8c51032f 1266 if (ret > 0) {
a2de733c
AJ		 1267 ret = btrfs_previous_item(root, path, 0,
1268 BTRFS_EXTENT_ITEM_KEY);
1269 if (ret < 0)
1270 goto out;
8c51032f
AJ		 1271 if (ret > 0) {
1272 /* there's no smaller item, so stick with the
1273 * larger one */
1274 btrfs_release_path(path);
1275 ret = btrfs_search_slot(NULL, root, &key,
1276 path, 0, 0);
1277 if (ret < 0)
1278 goto out;
1279 }
a2de733c
AJ		 1280 }
1281
1282 while (1) {
1283 l = path->nodes[0];
1284 slot = path->slots[0];
1285 if (slot >= btrfs_header_nritems(l)) {
1286 ret = btrfs_next_leaf(root, path);
1287 if (ret == 0)
1288 continue;
1289 if (ret < 0)
1290 goto out;
1291
1292 break;
1293 }
1294 btrfs_item_key_to_cpu(l, &key, slot);
1295
1296 if (key.objectid + key.offset <= logical)
1297 goto next;
1298
1299 if (key.objectid >= logical + map->stripe_len)
1300 break;
1301
1302 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
1303 goto next;
1304
1305 extent = btrfs_item_ptr(l, slot,
1306 struct btrfs_extent_item);
1307 flags = btrfs_extent_flags(l, extent);
1308 generation = btrfs_extent_generation(l, extent);
1309
1310 if (key.objectid < logical &&
1311 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1312 printk(KERN_ERR
1313 "btrfs scrub: tree block %llu spanning "
1314 "stripes, ignored. logical=%llu\n",
1315 (unsigned long long)key.objectid,
1316 (unsigned long long)logical);
1317 goto next;
1318 }
1319
1320 /*
1321 * trim extent to this stripe
1322 */
1323 if (key.objectid < logical) {
1324 key.offset -= logical - key.objectid;
1325 key.objectid = logical;
1326 }
1327 if (key.objectid + key.offset >
1328 logical + map->stripe_len) {
1329 key.offset = logical + map->stripe_len -
1330 key.objectid;
1331 }
1332
1333 ret = scrub_extent(sdev, key.objectid, key.offset,
1334 key.objectid - logical + physical,
1335 flags, generation, mirror_num);
1336 if (ret)
1337 goto out;
1338
1339next:
1340 path->slots[0]++;
1341 }
71267333 1342 btrfs_release_path(path);
a2de733c
AJ		 1343 logical += increment;
1344 physical += map->stripe_len;
1345 spin_lock(&sdev->stat_lock);
1346 sdev->stat.last_physical = physical;
1347 spin_unlock(&sdev->stat_lock);
1348 }
1349 /* push queued extents */
1350 scrub_submit(sdev);
1351
1352out:
e7786c3a 1353 blk_finish_plug(&plug);
a2de733c
AJ		 1354 btrfs_free_path(path);
1355 return ret < 0 ? ret : 0;
1356}
1357
1358static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev,
859acaf1
AJ		 1359 u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length,
1360 u64 dev_offset)
a2de733c
AJ		 1361{
1362 struct btrfs_mapping_tree *map_tree =
1363 &sdev->dev->dev_root->fs_info->mapping_tree;
1364 struct map_lookup *map;
1365 struct extent_map *em;
1366 int i;
1367 int ret = -EINVAL;
1368
1369 read_lock(&map_tree->map_tree.lock);
1370 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
1371 read_unlock(&map_tree->map_tree.lock);
1372
1373 if (!em)
1374 return -EINVAL;
1375
1376 map = (struct map_lookup *)em->bdev;
1377 if (em->start != chunk_offset)
1378 goto out;
1379
1380 if (em->len < length)
1381 goto out;
1382
1383 for (i = 0; i < map->num_stripes; ++i) {
859acaf1
AJ		 1384 if (map->stripes[i].dev == sdev->dev &&
1385 map->stripes[i].physical == dev_offset) {
a2de733c
AJ		 1386 ret = scrub_stripe(sdev, map, i, chunk_offset, length);
1387 if (ret)
1388 goto out;
1389 }
1390 }
1391out:
1392 free_extent_map(em);
1393
1394 return ret;
1395}
1396
1397static noinline_for_stack
1398int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end)
1399{
1400 struct btrfs_dev_extent *dev_extent = NULL;
1401 struct btrfs_path *path;
1402 struct btrfs_root *root = sdev->dev->dev_root;
1403 struct btrfs_fs_info *fs_info = root->fs_info;
1404 u64 length;
1405 u64 chunk_tree;
1406 u64 chunk_objectid;
1407 u64 chunk_offset;
1408 int ret;
1409 int slot;
1410 struct extent_buffer *l;
1411 struct btrfs_key key;
1412 struct btrfs_key found_key;
1413 struct btrfs_block_group_cache *cache;
1414
1415 path = btrfs_alloc_path();
1416 if (!path)
1417 return -ENOMEM;
1418
1419 path->reada = 2;
1420 path->search_commit_root = 1;
1421 path->skip_locking = 1;
1422
1423 key.objectid = sdev->dev->devid;
1424 key.offset = 0ull;
1425 key.type = BTRFS_DEV_EXTENT_KEY;
1426
1427
1428 while (1) {
1429 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1430 if (ret < 0)
8c51032f
AJ		 1431 break;
1432 if (ret > 0) {
1433 if (path->slots[0] >=
1434 btrfs_header_nritems(path->nodes[0])) {
1435 ret = btrfs_next_leaf(root, path);
1436 if (ret)
1437 break;
1438 }
1439 }
a2de733c
AJ		 1440
1441 l = path->nodes[0];
1442 slot = path->slots[0];
1443
1444 btrfs_item_key_to_cpu(l, &found_key, slot);
1445
1446 if (found_key.objectid != sdev->dev->devid)
1447 break;
1448
8c51032f 1449 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
a2de733c
AJ		 1450 break;
1451
1452 if (found_key.offset >= end)
1453 break;
1454
1455 if (found_key.offset < key.offset)
1456 break;
1457
1458 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1459 length = btrfs_dev_extent_length(l, dev_extent);
1460
1461 if (found_key.offset + length <= start) {
1462 key.offset = found_key.offset + length;
71267333 1463 btrfs_release_path(path);
a2de733c
AJ		 1464 continue;
1465 }
1466
1467 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1468 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1469 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1470
1471 /*
1472 * get a reference on the corresponding block group to prevent
1473 * the chunk from going away while we scrub it
1474 */
1475 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
1476 if (!cache) {
1477 ret = -ENOENT;
8c51032f 1478 break;
a2de733c
AJ		 1479 }
1480 ret = scrub_chunk(sdev, chunk_tree, chunk_objectid,
859acaf1 1481 chunk_offset, length, found_key.offset);
a2de733c
AJ		 1482 btrfs_put_block_group(cache);
1483 if (ret)
1484 break;
1485
1486 key.offset = found_key.offset + length;
71267333 1487 btrfs_release_path(path);
a2de733c
AJ		 1488 }
1489
a2de733c 1490 btrfs_free_path(path);
8c51032f
AJ		 1491
1492 /*
1493 * ret can still be 1 from search_slot or next_leaf,
1494 * that's not an error
1495 */
1496 return ret < 0 ? ret : 0;
a2de733c
AJ		 1497}
1498
1499static noinline_for_stack int scrub_supers(struct scrub_dev *sdev)
1500{
1501 int i;
1502 u64 bytenr;
1503 u64 gen;
1504 int ret;
1505 struct btrfs_device *device = sdev->dev;
1506 struct btrfs_root *root = device->dev_root;
1507
1508 gen = root->fs_info->last_trans_committed;
1509
1510 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1511 bytenr = btrfs_sb_offset(i);
1512 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
1513 break;
1514
1515 ret = scrub_page(sdev, bytenr, PAGE_SIZE, bytenr,
1516 BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1);
1517 if (ret)
1518 return ret;
1519 }
1520 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
1521
1522 return 0;
1523}
1524
1525/*
1526 * get a reference count on fs_info->scrub_workers. start worker if necessary
1527 */
1528static noinline_for_stack int scrub_workers_get(struct btrfs_root *root)
1529{
1530 struct btrfs_fs_info *fs_info = root->fs_info;
0dc3b84a 1531 int ret = 0;
a2de733c
AJ		 1532
1533 mutex_lock(&fs_info->scrub_lock);
632dd772
AJ		 1534 if (fs_info->scrub_workers_refcnt == 0) {
1535 btrfs_init_workers(&fs_info->scrub_workers, "scrub",
1536 fs_info->thread_pool_size, &fs_info->generic_worker);
1537 fs_info->scrub_workers.idle_thresh = 4;
0dc3b84a
JB		 1538 ret = btrfs_start_workers(&fs_info->scrub_workers);
1539 if (ret)
1540 goto out;
632dd772 1541 }
a2de733c 1542 ++fs_info->scrub_workers_refcnt;
0dc3b84a 1543out:
a2de733c
AJ		 1544 mutex_unlock(&fs_info->scrub_lock);
1545
0dc3b84a 1546 return ret;
a2de733c
AJ		 1547}
1548
1549static noinline_for_stack void scrub_workers_put(struct btrfs_root *root)
1550{
1551 struct btrfs_fs_info *fs_info = root->fs_info;
1552
1553 mutex_lock(&fs_info->scrub_lock);
1554 if (--fs_info->scrub_workers_refcnt == 0)
1555 btrfs_stop_workers(&fs_info->scrub_workers);
1556 WARN_ON(fs_info->scrub_workers_refcnt < 0);
1557 mutex_unlock(&fs_info->scrub_lock);
1558}
1559
1560
1561int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end,
8628764e 1562 struct btrfs_scrub_progress *progress, int readonly)
a2de733c
AJ		 1563{
1564 struct scrub_dev *sdev;
1565 struct btrfs_fs_info *fs_info = root->fs_info;
1566 int ret;
1567 struct btrfs_device *dev;
1568
7841cb28 1569 if (btrfs_fs_closing(root->fs_info))
a2de733c
AJ		 1570 return -EINVAL;
1571
1572 /*
1573 * check some assumptions
1574 */
1575 if (root->sectorsize != PAGE_SIZE ||
1576 root->sectorsize != root->leafsize ||
1577 root->sectorsize != root->nodesize) {
1578 printk(KERN_ERR "btrfs_scrub: size assumptions fail\n");
1579 return -EINVAL;
1580 }
1581
1582 ret = scrub_workers_get(root);
1583 if (ret)
1584 return ret;
1585
1586 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1587 dev = btrfs_find_device(root, devid, NULL, NULL);
1588 if (!dev || dev->missing) {
1589 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1590 scrub_workers_put(root);
1591 return -ENODEV;
1592 }
1593 mutex_lock(&fs_info->scrub_lock);
1594
1595 if (!dev->in_fs_metadata) {
1596 mutex_unlock(&fs_info->scrub_lock);
1597 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1598 scrub_workers_put(root);
1599 return -ENODEV;
1600 }
1601
1602 if (dev->scrub_device) {
1603 mutex_unlock(&fs_info->scrub_lock);
1604 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1605 scrub_workers_put(root);
1606 return -EINPROGRESS;
1607 }
1608 sdev = scrub_setup_dev(dev);
1609 if (IS_ERR(sdev)) {
1610 mutex_unlock(&fs_info->scrub_lock);
1611 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1612 scrub_workers_put(root);
1613 return PTR_ERR(sdev);
1614 }
8628764e 1615 sdev->readonly = readonly;
a2de733c
AJ		 1616 dev->scrub_device = sdev;
1617
1618 atomic_inc(&fs_info->scrubs_running);
1619 mutex_unlock(&fs_info->scrub_lock);
1620 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1621
1622 down_read(&fs_info->scrub_super_lock);
1623 ret = scrub_supers(sdev);
1624 up_read(&fs_info->scrub_super_lock);
1625
1626 if (!ret)
1627 ret = scrub_enumerate_chunks(sdev, start, end);
1628
1629 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0);
a2de733c
AJ		 1630 atomic_dec(&fs_info->scrubs_running);
1631 wake_up(&fs_info->scrub_pause_wait);
1632
0ef8e451
JS		 1633 wait_event(sdev->list_wait, atomic_read(&sdev->fixup_cnt) == 0);
1634
a2de733c
AJ		 1635 if (progress)
1636 memcpy(progress, &sdev->stat, sizeof(*progress));
1637
1638 mutex_lock(&fs_info->scrub_lock);
1639 dev->scrub_device = NULL;
1640 mutex_unlock(&fs_info->scrub_lock);
1641
1642 scrub_free_dev(sdev);
1643 scrub_workers_put(root);
1644
1645 return ret;
1646}
1647
143bede5 1648void btrfs_scrub_pause(struct btrfs_root *root)
a2de733c
AJ		 1649{
1650 struct btrfs_fs_info *fs_info = root->fs_info;
1651
1652 mutex_lock(&fs_info->scrub_lock);
1653 atomic_inc(&fs_info->scrub_pause_req);
1654 while (atomic_read(&fs_info->scrubs_paused) !=
1655 atomic_read(&fs_info->scrubs_running)) {
1656 mutex_unlock(&fs_info->scrub_lock);
1657 wait_event(fs_info->scrub_pause_wait,
1658 atomic_read(&fs_info->scrubs_paused) ==
1659 atomic_read(&fs_info->scrubs_running));
1660 mutex_lock(&fs_info->scrub_lock);
1661 }
1662 mutex_unlock(&fs_info->scrub_lock);
a2de733c
AJ		 1663}
1664
143bede5 1665void btrfs_scrub_continue(struct btrfs_root *root)
a2de733c
AJ		 1666{
1667 struct btrfs_fs_info *fs_info = root->fs_info;
1668
1669 atomic_dec(&fs_info->scrub_pause_req);
1670 wake_up(&fs_info->scrub_pause_wait);
a2de733c
AJ		 1671}
1672
143bede5 1673void btrfs_scrub_pause_super(struct btrfs_root *root)
a2de733c
AJ		 1674{
1675 down_write(&root->fs_info->scrub_super_lock);
a2de733c
AJ		 1676}
1677
143bede5 1678void btrfs_scrub_continue_super(struct btrfs_root *root)
a2de733c
AJ		 1679{
1680 up_write(&root->fs_info->scrub_super_lock);
a2de733c
AJ		 1681}
1682
49b25e05 1683int __btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
a2de733c 1684{
a2de733c
AJ		 1685
1686 mutex_lock(&fs_info->scrub_lock);
1687 if (!atomic_read(&fs_info->scrubs_running)) {
1688 mutex_unlock(&fs_info->scrub_lock);
1689 return -ENOTCONN;
1690 }
1691
1692 atomic_inc(&fs_info->scrub_cancel_req);
1693 while (atomic_read(&fs_info->scrubs_running)) {
1694 mutex_unlock(&fs_info->scrub_lock);
1695 wait_event(fs_info->scrub_pause_wait,
1696 atomic_read(&fs_info->scrubs_running) == 0);
1697 mutex_lock(&fs_info->scrub_lock);
1698 }
1699 atomic_dec(&fs_info->scrub_cancel_req);
1700 mutex_unlock(&fs_info->scrub_lock);
1701
1702 return 0;
1703}
1704
49b25e05
JM		 1705int btrfs_scrub_cancel(struct btrfs_root *root)
1706{
1707 return __btrfs_scrub_cancel(root->fs_info);
1708}
1709
a2de733c
AJ		 1710int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev)
1711{
1712 struct btrfs_fs_info *fs_info = root->fs_info;
1713 struct scrub_dev *sdev;
1714
1715 mutex_lock(&fs_info->scrub_lock);
1716 sdev = dev->scrub_device;
1717 if (!sdev) {
1718 mutex_unlock(&fs_info->scrub_lock);
1719 return -ENOTCONN;
1720 }
1721 atomic_inc(&sdev->cancel_req);
1722 while (dev->scrub_device) {
1723 mutex_unlock(&fs_info->scrub_lock);
1724 wait_event(fs_info->scrub_pause_wait,
1725 dev->scrub_device == NULL);
1726 mutex_lock(&fs_info->scrub_lock);
1727 }
1728 mutex_unlock(&fs_info->scrub_lock);
1729
1730 return 0;
1731}
1732int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid)
1733{
1734 struct btrfs_fs_info *fs_info = root->fs_info;
1735 struct btrfs_device *dev;
1736 int ret;
1737
1738 /*
1739 * we have to hold the device_list_mutex here so the device
1740 * does not go away in cancel_dev. FIXME: find a better solution
1741 */
1742 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1743 dev = btrfs_find_device(root, devid, NULL, NULL);
1744 if (!dev) {
1745 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1746 return -ENODEV;
1747 }
1748 ret = btrfs_scrub_cancel_dev(root, dev);
1749 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1750
1751 return ret;
1752}
1753
1754int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
1755 struct btrfs_scrub_progress *progress)
1756{
1757 struct btrfs_device *dev;
1758 struct scrub_dev *sdev = NULL;
1759
1760 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1761 dev = btrfs_find_device(root, devid, NULL, NULL);
1762 if (dev)
1763 sdev = dev->scrub_device;
1764 if (sdev)
1765 memcpy(progress, &sdev->stat, sizeof(*progress));
1766 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1767
1768 return dev ? (sdev ? 0 : -ENOTCONN) : -ENODEV;
1769}
Linux 6.x block layer and io_uring tree(s)