1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 * Copyright (C) 2022 Christoph Hellwig.
12 #include "async-thread.h"
13 #include "check-integrity.h"
14 #include "dev-replace.h"
15 #include "rcu-string.h"
17 #include "file-item.h"
19 static struct bio_set btrfs_bioset;
20 static struct bio_set btrfs_clone_bioset;
21 static struct bio_set btrfs_repair_bioset;
22 static mempool_t btrfs_failed_bio_pool;
24 struct btrfs_failed_bio {
25 struct btrfs_bio *bbio;
27 atomic_t repair_count;
31 * Initialize a btrfs_bio structure. This skips the embedded bio itself as it
32 * is already initialized by the block layer.
34 void btrfs_bio_init(struct btrfs_bio *bbio, struct btrfs_fs_info *fs_info,
35 btrfs_bio_end_io_t end_io, void *private)
37 memset(bbio, 0, offsetof(struct btrfs_bio, bio));
38 bbio->fs_info = fs_info;
39 bbio->end_io = end_io;
40 bbio->private = private;
41 atomic_set(&bbio->pending_ios, 1);
45 * Allocate a btrfs_bio structure. The btrfs_bio is the main I/O container for
46 * btrfs, and is used for all I/O submitted through btrfs_submit_bio.
48 * Just like the underlying bio_alloc_bioset it will not fail as it is backed by
51 struct btrfs_bio *btrfs_bio_alloc(unsigned int nr_vecs, blk_opf_t opf,
52 struct btrfs_fs_info *fs_info,
53 btrfs_bio_end_io_t end_io, void *private)
55 struct btrfs_bio *bbio;
58 bio = bio_alloc_bioset(NULL, nr_vecs, opf, GFP_NOFS, &btrfs_bioset);
59 bbio = btrfs_bio(bio);
60 btrfs_bio_init(bbio, fs_info, end_io, private);
64 static blk_status_t btrfs_bio_extract_ordered_extent(struct btrfs_bio *bbio)
66 struct btrfs_ordered_extent *ordered;
69 ordered = btrfs_lookup_ordered_extent(bbio->inode, bbio->file_offset);
70 if (WARN_ON_ONCE(!ordered))
72 ret = btrfs_extract_ordered_extent(bbio, ordered);
73 btrfs_put_ordered_extent(ordered);
75 return errno_to_blk_status(ret);
78 static struct btrfs_bio *btrfs_split_bio(struct btrfs_fs_info *fs_info,
79 struct btrfs_bio *orig_bbio,
80 u64 map_length, bool use_append)
82 struct btrfs_bio *bbio;
88 bio = bio_split_rw(&orig_bbio->bio, &fs_info->limits, &nr_segs,
89 &btrfs_clone_bioset, map_length);
91 bio = bio_split(&orig_bbio->bio, map_length >> SECTOR_SHIFT,
92 GFP_NOFS, &btrfs_clone_bioset);
94 bbio = btrfs_bio(bio);
95 btrfs_bio_init(bbio, fs_info, NULL, orig_bbio);
96 bbio->inode = orig_bbio->inode;
97 bbio->file_offset = orig_bbio->file_offset;
98 if (!(orig_bbio->bio.bi_opf & REQ_BTRFS_ONE_ORDERED))
99 orig_bbio->file_offset += map_length;
101 atomic_inc(&orig_bbio->pending_ios);
105 static void btrfs_orig_write_end_io(struct bio *bio);
107 static void btrfs_bbio_propagate_error(struct btrfs_bio *bbio,
108 struct btrfs_bio *orig_bbio)
111 * For writes we tolerate nr_mirrors - 1 write failures, so we can't
112 * just blindly propagate a write failure here. Instead increment the
113 * error count in the original I/O context so that it is guaranteed to
114 * be larger than the error tolerance.
116 if (bbio->bio.bi_end_io == &btrfs_orig_write_end_io) {
117 struct btrfs_io_stripe *orig_stripe = orig_bbio->bio.bi_private;
118 struct btrfs_io_context *orig_bioc = orig_stripe->bioc;
120 atomic_add(orig_bioc->max_errors, &orig_bioc->error);
122 orig_bbio->bio.bi_status = bbio->bio.bi_status;
126 static void btrfs_orig_bbio_end_io(struct btrfs_bio *bbio)
128 if (bbio->bio.bi_pool == &btrfs_clone_bioset) {
129 struct btrfs_bio *orig_bbio = bbio->private;
131 if (bbio->bio.bi_status)
132 btrfs_bbio_propagate_error(bbio, orig_bbio);
137 if (atomic_dec_and_test(&bbio->pending_ios))
141 static int next_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
143 if (cur_mirror == fbio->num_copies)
144 return cur_mirror + 1 - fbio->num_copies;
145 return cur_mirror + 1;
148 static int prev_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
151 return fbio->num_copies;
152 return cur_mirror - 1;
155 static void btrfs_repair_done(struct btrfs_failed_bio *fbio)
157 if (atomic_dec_and_test(&fbio->repair_count)) {
158 btrfs_orig_bbio_end_io(fbio->bbio);
159 mempool_free(fbio, &btrfs_failed_bio_pool);
163 static void btrfs_end_repair_bio(struct btrfs_bio *repair_bbio,
164 struct btrfs_device *dev)
166 struct btrfs_failed_bio *fbio = repair_bbio->private;
167 struct btrfs_inode *inode = repair_bbio->inode;
168 struct btrfs_fs_info *fs_info = inode->root->fs_info;
169 struct bio_vec *bv = bio_first_bvec_all(&repair_bbio->bio);
170 int mirror = repair_bbio->mirror_num;
172 if (repair_bbio->bio.bi_status ||
173 !btrfs_data_csum_ok(repair_bbio, dev, 0, bv)) {
174 bio_reset(&repair_bbio->bio, NULL, REQ_OP_READ);
175 repair_bbio->bio.bi_iter = repair_bbio->saved_iter;
177 mirror = next_repair_mirror(fbio, mirror);
178 if (mirror == fbio->bbio->mirror_num) {
179 btrfs_debug(fs_info, "no mirror left");
180 fbio->bbio->bio.bi_status = BLK_STS_IOERR;
184 btrfs_submit_bio(repair_bbio, mirror);
189 mirror = prev_repair_mirror(fbio, mirror);
190 btrfs_repair_io_failure(fs_info, btrfs_ino(inode),
191 repair_bbio->file_offset, fs_info->sectorsize,
192 repair_bbio->saved_iter.bi_sector << SECTOR_SHIFT,
193 bv->bv_page, bv->bv_offset, mirror);
194 } while (mirror != fbio->bbio->mirror_num);
197 btrfs_repair_done(fbio);
198 bio_put(&repair_bbio->bio);
202 * Try to kick off a repair read to the next available mirror for a bad sector.
204 * This primarily tries to recover good data to serve the actual read request,
205 * but also tries to write the good data back to the bad mirror(s) when a
206 * read succeeded to restore the redundancy.
208 static struct btrfs_failed_bio *repair_one_sector(struct btrfs_bio *failed_bbio,
211 struct btrfs_failed_bio *fbio)
213 struct btrfs_inode *inode = failed_bbio->inode;
214 struct btrfs_fs_info *fs_info = inode->root->fs_info;
215 const u32 sectorsize = fs_info->sectorsize;
216 const u64 logical = (failed_bbio->saved_iter.bi_sector << SECTOR_SHIFT);
217 struct btrfs_bio *repair_bbio;
218 struct bio *repair_bio;
222 btrfs_debug(fs_info, "repair read error: read error at %llu",
223 failed_bbio->file_offset + bio_offset);
225 num_copies = btrfs_num_copies(fs_info, logical, sectorsize);
226 if (num_copies == 1) {
227 btrfs_debug(fs_info, "no copy to repair from");
228 failed_bbio->bio.bi_status = BLK_STS_IOERR;
233 fbio = mempool_alloc(&btrfs_failed_bio_pool, GFP_NOFS);
234 fbio->bbio = failed_bbio;
235 fbio->num_copies = num_copies;
236 atomic_set(&fbio->repair_count, 1);
239 atomic_inc(&fbio->repair_count);
241 repair_bio = bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS,
242 &btrfs_repair_bioset);
243 repair_bio->bi_iter.bi_sector = failed_bbio->saved_iter.bi_sector;
244 __bio_add_page(repair_bio, bv->bv_page, bv->bv_len, bv->bv_offset);
246 repair_bbio = btrfs_bio(repair_bio);
247 btrfs_bio_init(repair_bbio, fs_info, NULL, fbio);
248 repair_bbio->inode = failed_bbio->inode;
249 repair_bbio->file_offset = failed_bbio->file_offset + bio_offset;
251 mirror = next_repair_mirror(fbio, failed_bbio->mirror_num);
252 btrfs_debug(fs_info, "submitting repair read to mirror %d", mirror);
253 btrfs_submit_bio(repair_bbio, mirror);
257 static void btrfs_check_read_bio(struct btrfs_bio *bbio, struct btrfs_device *dev)
259 struct btrfs_inode *inode = bbio->inode;
260 struct btrfs_fs_info *fs_info = inode->root->fs_info;
261 u32 sectorsize = fs_info->sectorsize;
262 struct bvec_iter *iter = &bbio->saved_iter;
263 blk_status_t status = bbio->bio.bi_status;
264 struct btrfs_failed_bio *fbio = NULL;
267 /* Read-repair requires the inode field to be set by the submitter. */
271 * Hand off repair bios to the repair code as there is no upper level
272 * submitter for them.
274 if (bbio->bio.bi_pool == &btrfs_repair_bioset) {
275 btrfs_end_repair_bio(bbio, dev);
279 /* Clear the I/O error. A failed repair will reset it. */
280 bbio->bio.bi_status = BLK_STS_OK;
282 while (iter->bi_size) {
283 struct bio_vec bv = bio_iter_iovec(&bbio->bio, *iter);
285 bv.bv_len = min(bv.bv_len, sectorsize);
286 if (status || !btrfs_data_csum_ok(bbio, dev, offset, &bv))
287 fbio = repair_one_sector(bbio, offset, &bv, fbio);
289 bio_advance_iter_single(&bbio->bio, iter, sectorsize);
290 offset += sectorsize;
293 if (bbio->csum != bbio->csum_inline)
297 btrfs_repair_done(fbio);
299 btrfs_orig_bbio_end_io(bbio);
302 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev)
304 if (!dev || !dev->bdev)
306 if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET)
309 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
310 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
311 else if (!(bio->bi_opf & REQ_RAHEAD))
312 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
313 if (bio->bi_opf & REQ_PREFLUSH)
314 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS);
317 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info,
320 if (bio->bi_opf & REQ_META)
321 return fs_info->endio_meta_workers;
322 return fs_info->endio_workers;
325 static void btrfs_end_bio_work(struct work_struct *work)
327 struct btrfs_bio *bbio = container_of(work, struct btrfs_bio, end_io_work);
329 /* Metadata reads are checked and repaired by the submitter. */
330 if (bbio->inode && !(bbio->bio.bi_opf & REQ_META))
331 btrfs_check_read_bio(bbio, bbio->bio.bi_private);
333 btrfs_orig_bbio_end_io(bbio);
336 static void btrfs_simple_end_io(struct bio *bio)
338 struct btrfs_bio *bbio = btrfs_bio(bio);
339 struct btrfs_device *dev = bio->bi_private;
340 struct btrfs_fs_info *fs_info = bbio->fs_info;
342 btrfs_bio_counter_dec(fs_info);
345 btrfs_log_dev_io_error(bio, dev);
347 if (bio_op(bio) == REQ_OP_READ) {
348 INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work);
349 queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work);
351 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
352 btrfs_record_physical_zoned(bbio);
353 btrfs_orig_bbio_end_io(bbio);
357 static void btrfs_raid56_end_io(struct bio *bio)
359 struct btrfs_io_context *bioc = bio->bi_private;
360 struct btrfs_bio *bbio = btrfs_bio(bio);
362 btrfs_bio_counter_dec(bioc->fs_info);
363 bbio->mirror_num = bioc->mirror_num;
364 if (bio_op(bio) == REQ_OP_READ && bbio->inode &&
365 !(bbio->bio.bi_opf & REQ_META))
366 btrfs_check_read_bio(bbio, NULL);
368 btrfs_orig_bbio_end_io(bbio);
370 btrfs_put_bioc(bioc);
373 static void btrfs_orig_write_end_io(struct bio *bio)
375 struct btrfs_io_stripe *stripe = bio->bi_private;
376 struct btrfs_io_context *bioc = stripe->bioc;
377 struct btrfs_bio *bbio = btrfs_bio(bio);
379 btrfs_bio_counter_dec(bioc->fs_info);
381 if (bio->bi_status) {
382 atomic_inc(&bioc->error);
383 btrfs_log_dev_io_error(bio, stripe->dev);
387 * Only send an error to the higher layers if it is beyond the tolerance
390 if (atomic_read(&bioc->error) > bioc->max_errors)
391 bio->bi_status = BLK_STS_IOERR;
393 bio->bi_status = BLK_STS_OK;
395 btrfs_orig_bbio_end_io(bbio);
396 btrfs_put_bioc(bioc);
399 static void btrfs_clone_write_end_io(struct bio *bio)
401 struct btrfs_io_stripe *stripe = bio->bi_private;
403 if (bio->bi_status) {
404 atomic_inc(&stripe->bioc->error);
405 btrfs_log_dev_io_error(bio, stripe->dev);
408 /* Pass on control to the original bio this one was cloned from */
409 bio_endio(stripe->bioc->orig_bio);
413 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio)
415 if (!dev || !dev->bdev ||
416 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
417 (btrfs_op(bio) == BTRFS_MAP_WRITE &&
418 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
423 bio_set_dev(bio, dev->bdev);
426 * For zone append writing, bi_sector must point the beginning of the
429 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
430 u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
431 u64 zone_start = round_down(physical, dev->fs_info->zone_size);
433 ASSERT(btrfs_dev_is_sequential(dev, physical));
434 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
436 btrfs_debug_in_rcu(dev->fs_info,
437 "%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
438 __func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
439 (unsigned long)dev->bdev->bd_dev, btrfs_dev_name(dev),
440 dev->devid, bio->bi_iter.bi_size);
442 btrfsic_check_bio(bio);
444 if (bio->bi_opf & REQ_BTRFS_CGROUP_PUNT)
445 blkcg_punt_bio_submit(bio);
450 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr)
452 struct bio *orig_bio = bioc->orig_bio, *bio;
454 ASSERT(bio_op(orig_bio) != REQ_OP_READ);
456 /* Reuse the bio embedded into the btrfs_bio for the last mirror */
457 if (dev_nr == bioc->num_stripes - 1) {
459 bio->bi_end_io = btrfs_orig_write_end_io;
461 bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set);
462 bio_inc_remaining(orig_bio);
463 bio->bi_end_io = btrfs_clone_write_end_io;
466 bio->bi_private = &bioc->stripes[dev_nr];
467 bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT;
468 bioc->stripes[dev_nr].bioc = bioc;
469 btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio);
472 static void __btrfs_submit_bio(struct bio *bio, struct btrfs_io_context *bioc,
473 struct btrfs_io_stripe *smap, int mirror_num)
475 /* Do not leak our private flag into the block layer. */
476 bio->bi_opf &= ~REQ_BTRFS_ONE_ORDERED;
479 /* Single mirror read/write fast path. */
480 btrfs_bio(bio)->mirror_num = mirror_num;
481 bio->bi_iter.bi_sector = smap->physical >> SECTOR_SHIFT;
482 bio->bi_private = smap->dev;
483 bio->bi_end_io = btrfs_simple_end_io;
484 btrfs_submit_dev_bio(smap->dev, bio);
485 } else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
486 /* Parity RAID write or read recovery. */
487 bio->bi_private = bioc;
488 bio->bi_end_io = btrfs_raid56_end_io;
489 if (bio_op(bio) == REQ_OP_READ)
490 raid56_parity_recover(bio, bioc, mirror_num);
492 raid56_parity_write(bio, bioc);
494 /* Write to multiple mirrors. */
495 int total_devs = bioc->num_stripes;
497 bioc->orig_bio = bio;
498 for (int dev_nr = 0; dev_nr < total_devs; dev_nr++)
499 btrfs_submit_mirrored_bio(bioc, dev_nr);
503 static blk_status_t btrfs_bio_csum(struct btrfs_bio *bbio)
505 if (bbio->bio.bi_opf & REQ_META)
506 return btree_csum_one_bio(bbio);
507 return btrfs_csum_one_bio(bbio);
511 * Async submit bios are used to offload expensive checksumming onto the worker
514 struct async_submit_bio {
515 struct btrfs_bio *bbio;
516 struct btrfs_io_context *bioc;
517 struct btrfs_io_stripe smap;
519 struct btrfs_work work;
523 * In order to insert checksums into the metadata in large chunks, we wait
524 * until bio submission time. All the pages in the bio are checksummed and
525 * sums are attached onto the ordered extent record.
527 * At IO completion time the csums attached on the ordered extent record are
528 * inserted into the btree.
530 static void run_one_async_start(struct btrfs_work *work)
532 struct async_submit_bio *async =
533 container_of(work, struct async_submit_bio, work);
536 ret = btrfs_bio_csum(async->bbio);
538 async->bbio->bio.bi_status = ret;
542 * In order to insert checksums into the metadata in large chunks, we wait
543 * until bio submission time. All the pages in the bio are checksummed and
544 * sums are attached onto the ordered extent record.
546 * At IO completion time the csums attached on the ordered extent record are
547 * inserted into the tree.
549 static void run_one_async_done(struct btrfs_work *work)
551 struct async_submit_bio *async =
552 container_of(work, struct async_submit_bio, work);
553 struct bio *bio = &async->bbio->bio;
555 /* If an error occurred we just want to clean up the bio and move on. */
556 if (bio->bi_status) {
557 btrfs_orig_bbio_end_io(async->bbio);
562 * All of the bios that pass through here are from async helpers.
563 * Use REQ_BTRFS_CGROUP_PUNT to issue them from the owning cgroup's
564 * context. This changes nothing when cgroups aren't in use.
566 bio->bi_opf |= REQ_BTRFS_CGROUP_PUNT;
567 __btrfs_submit_bio(bio, async->bioc, &async->smap, async->mirror_num);
570 static void run_one_async_free(struct btrfs_work *work)
572 kfree(container_of(work, struct async_submit_bio, work));
575 static bool should_async_write(struct btrfs_bio *bbio)
578 * If the I/O is not issued by fsync and friends, (->sync_writers != 0),
579 * then try to defer the submission to a workqueue to parallelize the
580 * checksum calculation.
582 if (atomic_read(&bbio->inode->sync_writers))
586 * Submit metadata writes synchronously if the checksum implementation
587 * is fast, or we are on a zoned device that wants I/O to be submitted
590 if (bbio->bio.bi_opf & REQ_META) {
591 struct btrfs_fs_info *fs_info = bbio->fs_info;
593 if (btrfs_is_zoned(fs_info))
595 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
603 * Submit bio to an async queue.
605 * Return true if the work has been succesfuly submitted, else false.
607 static bool btrfs_wq_submit_bio(struct btrfs_bio *bbio,
608 struct btrfs_io_context *bioc,
609 struct btrfs_io_stripe *smap, int mirror_num)
611 struct btrfs_fs_info *fs_info = bbio->fs_info;
612 struct async_submit_bio *async;
614 async = kmalloc(sizeof(*async), GFP_NOFS);
621 async->mirror_num = mirror_num;
623 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
625 if (op_is_sync(bbio->bio.bi_opf))
626 btrfs_queue_work(fs_info->hipri_workers, &async->work);
628 btrfs_queue_work(fs_info->workers, &async->work);
632 static bool btrfs_submit_chunk(struct btrfs_bio *bbio, int mirror_num)
634 struct btrfs_inode *inode = bbio->inode;
635 struct btrfs_fs_info *fs_info = bbio->fs_info;
636 struct btrfs_bio *orig_bbio = bbio;
637 struct bio *bio = &bbio->bio;
638 u64 logical = bio->bi_iter.bi_sector << 9;
639 u64 length = bio->bi_iter.bi_size;
640 u64 map_length = length;
641 bool use_append = btrfs_use_zone_append(bbio);
642 struct btrfs_io_context *bioc = NULL;
643 struct btrfs_io_stripe smap;
647 btrfs_bio_counter_inc_blocked(fs_info);
648 error = __btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
649 &bioc, &smap, &mirror_num, 1);
651 ret = errno_to_blk_status(error);
655 map_length = min(map_length, length);
657 map_length = min(map_length, fs_info->max_zone_append_size);
659 if (map_length < length) {
660 bbio = btrfs_split_bio(fs_info, bbio, map_length, use_append);
665 * Save the iter for the end_io handler and preload the checksums for
668 if (bio_op(bio) == REQ_OP_READ && inode && !(bio->bi_opf & REQ_META)) {
669 bbio->saved_iter = bio->bi_iter;
670 ret = btrfs_lookup_bio_sums(bbio);
675 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
677 bio->bi_opf &= ~REQ_OP_WRITE;
678 bio->bi_opf |= REQ_OP_ZONE_APPEND;
679 ret = btrfs_bio_extract_ordered_extent(bbio);
685 * Csum items for reloc roots have already been cloned at this
686 * point, so they are handled as part of the no-checksum case.
688 if (inode && !(inode->flags & BTRFS_INODE_NODATASUM) &&
689 !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state) &&
690 !btrfs_is_data_reloc_root(inode->root)) {
691 if (should_async_write(bbio) &&
692 btrfs_wq_submit_bio(bbio, bioc, &smap, mirror_num))
695 ret = btrfs_bio_csum(bbio);
701 __btrfs_submit_bio(bio, bioc, &smap, mirror_num);
703 return map_length == length;
706 if (map_length < length)
709 btrfs_bio_counter_dec(fs_info);
710 btrfs_bio_end_io(orig_bbio, ret);
711 /* Do not submit another chunk */
715 void btrfs_submit_bio(struct btrfs_bio *bbio, int mirror_num)
717 /* If bbio->inode is not populated, its file_offset must be 0. */
718 ASSERT(bbio->inode || bbio->file_offset == 0);
720 while (!btrfs_submit_chunk(bbio, mirror_num))
725 * Submit a repair write.
727 * This bypasses btrfs_submit_bio deliberately, as that writes all copies in a
728 * RAID setup. Here we only want to write the one bad copy, so we do the
729 * mapping ourselves and submit the bio directly.
731 * The I/O is issued synchronously to block the repair read completion from
734 int btrfs_repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
735 u64 length, u64 logical, struct page *page,
736 unsigned int pg_offset, int mirror_num)
738 struct btrfs_io_stripe smap = { 0 };
743 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
746 if (btrfs_repair_one_zone(fs_info, logical))
750 * Avoid races with device replace and make sure our bioc has devices
751 * associated to its stripes that don't go away while we are doing the
752 * read repair operation.
754 btrfs_bio_counter_inc_blocked(fs_info);
755 ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
757 goto out_counter_dec;
759 if (!smap.dev->bdev ||
760 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &smap.dev->dev_state)) {
762 goto out_counter_dec;
765 bio_init(&bio, smap.dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
766 bio.bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT;
767 __bio_add_page(&bio, page, length, pg_offset);
769 btrfsic_check_bio(&bio);
770 ret = submit_bio_wait(&bio);
772 /* try to remap that extent elsewhere? */
773 btrfs_dev_stat_inc_and_print(smap.dev, BTRFS_DEV_STAT_WRITE_ERRS);
777 btrfs_info_rl_in_rcu(fs_info,
778 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
779 ino, start, btrfs_dev_name(smap.dev),
780 smap.physical >> SECTOR_SHIFT);
786 btrfs_bio_counter_dec(fs_info);
791 * Submit a btrfs_bio based repair write.
793 * If @dev_replace is true, the write would be submitted to dev-replace target.
795 void btrfs_submit_repair_write(struct btrfs_bio *bbio, int mirror_num, bool dev_replace)
797 struct btrfs_fs_info *fs_info = bbio->fs_info;
798 u64 logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
799 u64 length = bbio->bio.bi_iter.bi_size;
800 struct btrfs_io_stripe smap = { 0 };
804 ASSERT(mirror_num > 0);
805 ASSERT(btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE);
806 ASSERT(!bbio->inode);
808 btrfs_bio_counter_inc_blocked(fs_info);
809 ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
814 ASSERT(smap.dev == fs_info->dev_replace.srcdev);
815 smap.dev = fs_info->dev_replace.tgtdev;
817 __btrfs_submit_bio(&bbio->bio, NULL, &smap, mirror_num);
821 btrfs_bio_counter_dec(fs_info);
822 btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
825 int __init btrfs_bioset_init(void)
827 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
828 offsetof(struct btrfs_bio, bio),
831 if (bioset_init(&btrfs_clone_bioset, BIO_POOL_SIZE,
832 offsetof(struct btrfs_bio, bio), 0))
833 goto out_free_bioset;
834 if (bioset_init(&btrfs_repair_bioset, BIO_POOL_SIZE,
835 offsetof(struct btrfs_bio, bio),
837 goto out_free_clone_bioset;
838 if (mempool_init_kmalloc_pool(&btrfs_failed_bio_pool, BIO_POOL_SIZE,
839 sizeof(struct btrfs_failed_bio)))
840 goto out_free_repair_bioset;
843 out_free_repair_bioset:
844 bioset_exit(&btrfs_repair_bioset);
845 out_free_clone_bioset:
846 bioset_exit(&btrfs_clone_bioset);
848 bioset_exit(&btrfs_bioset);
852 void __cold btrfs_bioset_exit(void)
854 mempool_exit(&btrfs_failed_bio_pool);
855 bioset_exit(&btrfs_repair_bioset);
856 bioset_exit(&btrfs_clone_bioset);
857 bioset_exit(&btrfs_bioset);