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 "dev-replace.h"
15 #include "file-item.h"
16 #include "raid-stripe-tree.h"
18 static struct bio_set btrfs_bioset;
19 static struct bio_set btrfs_clone_bioset;
20 static struct bio_set btrfs_repair_bioset;
21 static mempool_t btrfs_failed_bio_pool;
23 struct btrfs_failed_bio {
24 struct btrfs_bio *bbio;
26 atomic_t repair_count;
29 /* Is this a data path I/O that needs storage layer checksum and repair? */
30 static inline bool is_data_bbio(struct btrfs_bio *bbio)
32 return bbio->inode && is_data_inode(bbio->inode);
35 static bool bbio_has_ordered_extent(struct btrfs_bio *bbio)
37 return is_data_bbio(bbio) && btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE;
41 * Initialize a btrfs_bio structure. This skips the embedded bio itself as it
42 * is already initialized by the block layer.
44 void btrfs_bio_init(struct btrfs_bio *bbio, struct btrfs_fs_info *fs_info,
45 btrfs_bio_end_io_t end_io, void *private)
47 memset(bbio, 0, offsetof(struct btrfs_bio, bio));
48 bbio->fs_info = fs_info;
49 bbio->end_io = end_io;
50 bbio->private = private;
51 atomic_set(&bbio->pending_ios, 1);
52 WRITE_ONCE(bbio->status, BLK_STS_OK);
56 * Allocate a btrfs_bio structure. The btrfs_bio is the main I/O container for
57 * btrfs, and is used for all I/O submitted through btrfs_submit_bbio().
59 * Just like the underlying bio_alloc_bioset it will not fail as it is backed by
62 struct btrfs_bio *btrfs_bio_alloc(unsigned int nr_vecs, blk_opf_t opf,
63 struct btrfs_fs_info *fs_info,
64 btrfs_bio_end_io_t end_io, void *private)
66 struct btrfs_bio *bbio;
69 bio = bio_alloc_bioset(NULL, nr_vecs, opf, GFP_NOFS, &btrfs_bioset);
70 bbio = btrfs_bio(bio);
71 btrfs_bio_init(bbio, fs_info, end_io, private);
75 static struct btrfs_bio *btrfs_split_bio(struct btrfs_fs_info *fs_info,
76 struct btrfs_bio *orig_bbio,
79 struct btrfs_bio *bbio;
82 bio = bio_split(&orig_bbio->bio, map_length >> SECTOR_SHIFT, GFP_NOFS,
87 bbio = btrfs_bio(bio);
88 btrfs_bio_init(bbio, fs_info, NULL, orig_bbio);
89 bbio->inode = orig_bbio->inode;
90 bbio->file_offset = orig_bbio->file_offset;
91 orig_bbio->file_offset += map_length;
92 if (bbio_has_ordered_extent(bbio)) {
93 refcount_inc(&orig_bbio->ordered->refs);
94 bbio->ordered = orig_bbio->ordered;
96 atomic_inc(&orig_bbio->pending_ios);
100 /* Free a bio that was never submitted to the underlying device. */
101 static void btrfs_cleanup_bio(struct btrfs_bio *bbio)
103 if (bbio_has_ordered_extent(bbio))
104 btrfs_put_ordered_extent(bbio->ordered);
108 static void __btrfs_bio_end_io(struct btrfs_bio *bbio)
110 if (bbio_has_ordered_extent(bbio)) {
111 struct btrfs_ordered_extent *ordered = bbio->ordered;
114 btrfs_put_ordered_extent(ordered);
120 void btrfs_bio_end_io(struct btrfs_bio *bbio, blk_status_t status)
122 bbio->bio.bi_status = status;
123 if (bbio->bio.bi_pool == &btrfs_clone_bioset) {
124 struct btrfs_bio *orig_bbio = bbio->private;
126 btrfs_cleanup_bio(bbio);
131 * At this point, bbio always points to the original btrfs_bio. Save
132 * the first error in it.
134 if (status != BLK_STS_OK)
135 cmpxchg(&bbio->status, BLK_STS_OK, status);
137 if (atomic_dec_and_test(&bbio->pending_ios)) {
138 /* Load split bio's error which might be set above. */
139 if (status == BLK_STS_OK)
140 bbio->bio.bi_status = READ_ONCE(bbio->status);
141 __btrfs_bio_end_io(bbio);
145 static int next_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
147 if (cur_mirror == fbio->num_copies)
148 return cur_mirror + 1 - fbio->num_copies;
149 return cur_mirror + 1;
152 static int prev_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
155 return fbio->num_copies;
156 return cur_mirror - 1;
159 static void btrfs_repair_done(struct btrfs_failed_bio *fbio)
161 if (atomic_dec_and_test(&fbio->repair_count)) {
162 btrfs_bio_end_io(fbio->bbio, fbio->bbio->bio.bi_status);
163 mempool_free(fbio, &btrfs_failed_bio_pool);
167 static void btrfs_end_repair_bio(struct btrfs_bio *repair_bbio,
168 struct btrfs_device *dev)
170 struct btrfs_failed_bio *fbio = repair_bbio->private;
171 struct btrfs_inode *inode = repair_bbio->inode;
172 struct btrfs_fs_info *fs_info = inode->root->fs_info;
173 struct bio_vec *bv = bio_first_bvec_all(&repair_bbio->bio);
174 int mirror = repair_bbio->mirror_num;
177 * We can only trigger this for data bio, which doesn't support larger
180 ASSERT(folio_order(page_folio(bv->bv_page)) == 0);
182 if (repair_bbio->bio.bi_status ||
183 !btrfs_data_csum_ok(repair_bbio, dev, 0, bv)) {
184 bio_reset(&repair_bbio->bio, NULL, REQ_OP_READ);
185 repair_bbio->bio.bi_iter = repair_bbio->saved_iter;
187 mirror = next_repair_mirror(fbio, mirror);
188 if (mirror == fbio->bbio->mirror_num) {
189 btrfs_debug(fs_info, "no mirror left");
190 fbio->bbio->bio.bi_status = BLK_STS_IOERR;
194 btrfs_submit_bbio(repair_bbio, mirror);
199 mirror = prev_repair_mirror(fbio, mirror);
200 btrfs_repair_io_failure(fs_info, btrfs_ino(inode),
201 repair_bbio->file_offset, fs_info->sectorsize,
202 repair_bbio->saved_iter.bi_sector << SECTOR_SHIFT,
203 page_folio(bv->bv_page), bv->bv_offset, mirror);
204 } while (mirror != fbio->bbio->mirror_num);
207 btrfs_repair_done(fbio);
208 bio_put(&repair_bbio->bio);
212 * Try to kick off a repair read to the next available mirror for a bad sector.
214 * This primarily tries to recover good data to serve the actual read request,
215 * but also tries to write the good data back to the bad mirror(s) when a
216 * read succeeded to restore the redundancy.
218 static struct btrfs_failed_bio *repair_one_sector(struct btrfs_bio *failed_bbio,
221 struct btrfs_failed_bio *fbio)
223 struct btrfs_inode *inode = failed_bbio->inode;
224 struct btrfs_fs_info *fs_info = inode->root->fs_info;
225 const u32 sectorsize = fs_info->sectorsize;
226 const u64 logical = (failed_bbio->saved_iter.bi_sector << SECTOR_SHIFT);
227 struct btrfs_bio *repair_bbio;
228 struct bio *repair_bio;
232 btrfs_debug(fs_info, "repair read error: read error at %llu",
233 failed_bbio->file_offset + bio_offset);
235 num_copies = btrfs_num_copies(fs_info, logical, sectorsize);
236 if (num_copies == 1) {
237 btrfs_debug(fs_info, "no copy to repair from");
238 failed_bbio->bio.bi_status = BLK_STS_IOERR;
243 fbio = mempool_alloc(&btrfs_failed_bio_pool, GFP_NOFS);
244 fbio->bbio = failed_bbio;
245 fbio->num_copies = num_copies;
246 atomic_set(&fbio->repair_count, 1);
249 atomic_inc(&fbio->repair_count);
251 repair_bio = bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS,
252 &btrfs_repair_bioset);
253 repair_bio->bi_iter.bi_sector = failed_bbio->saved_iter.bi_sector;
254 __bio_add_page(repair_bio, bv->bv_page, bv->bv_len, bv->bv_offset);
256 repair_bbio = btrfs_bio(repair_bio);
257 btrfs_bio_init(repair_bbio, fs_info, NULL, fbio);
258 repair_bbio->inode = failed_bbio->inode;
259 repair_bbio->file_offset = failed_bbio->file_offset + bio_offset;
261 mirror = next_repair_mirror(fbio, failed_bbio->mirror_num);
262 btrfs_debug(fs_info, "submitting repair read to mirror %d", mirror);
263 btrfs_submit_bbio(repair_bbio, mirror);
267 static void btrfs_check_read_bio(struct btrfs_bio *bbio, struct btrfs_device *dev)
269 struct btrfs_inode *inode = bbio->inode;
270 struct btrfs_fs_info *fs_info = inode->root->fs_info;
271 u32 sectorsize = fs_info->sectorsize;
272 struct bvec_iter *iter = &bbio->saved_iter;
273 blk_status_t status = bbio->bio.bi_status;
274 struct btrfs_failed_bio *fbio = NULL;
277 /* Read-repair requires the inode field to be set by the submitter. */
281 * Hand off repair bios to the repair code as there is no upper level
282 * submitter for them.
284 if (bbio->bio.bi_pool == &btrfs_repair_bioset) {
285 btrfs_end_repair_bio(bbio, dev);
289 /* Clear the I/O error. A failed repair will reset it. */
290 bbio->bio.bi_status = BLK_STS_OK;
292 while (iter->bi_size) {
293 struct bio_vec bv = bio_iter_iovec(&bbio->bio, *iter);
295 bv.bv_len = min(bv.bv_len, sectorsize);
296 if (status || !btrfs_data_csum_ok(bbio, dev, offset, &bv))
297 fbio = repair_one_sector(bbio, offset, &bv, fbio);
299 bio_advance_iter_single(&bbio->bio, iter, sectorsize);
300 offset += sectorsize;
303 if (bbio->csum != bbio->csum_inline)
307 btrfs_repair_done(fbio);
309 btrfs_bio_end_io(bbio, bbio->bio.bi_status);
312 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev)
314 if (!dev || !dev->bdev)
316 if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET)
319 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
320 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
321 else if (!(bio->bi_opf & REQ_RAHEAD))
322 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
323 if (bio->bi_opf & REQ_PREFLUSH)
324 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS);
327 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info,
330 if (bio->bi_opf & REQ_META)
331 return fs_info->endio_meta_workers;
332 return fs_info->endio_workers;
335 static void btrfs_end_bio_work(struct work_struct *work)
337 struct btrfs_bio *bbio = container_of(work, struct btrfs_bio, end_io_work);
339 /* Metadata reads are checked and repaired by the submitter. */
340 if (is_data_bbio(bbio))
341 btrfs_check_read_bio(bbio, bbio->bio.bi_private);
343 btrfs_bio_end_io(bbio, bbio->bio.bi_status);
346 static void btrfs_simple_end_io(struct bio *bio)
348 struct btrfs_bio *bbio = btrfs_bio(bio);
349 struct btrfs_device *dev = bio->bi_private;
350 struct btrfs_fs_info *fs_info = bbio->fs_info;
352 btrfs_bio_counter_dec(fs_info);
355 btrfs_log_dev_io_error(bio, dev);
357 if (bio_op(bio) == REQ_OP_READ) {
358 INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work);
359 queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work);
361 if (bio_is_zone_append(bio) && !bio->bi_status)
362 btrfs_record_physical_zoned(bbio);
363 btrfs_bio_end_io(bbio, bbio->bio.bi_status);
367 static void btrfs_raid56_end_io(struct bio *bio)
369 struct btrfs_io_context *bioc = bio->bi_private;
370 struct btrfs_bio *bbio = btrfs_bio(bio);
372 btrfs_bio_counter_dec(bioc->fs_info);
373 bbio->mirror_num = bioc->mirror_num;
374 if (bio_op(bio) == REQ_OP_READ && is_data_bbio(bbio))
375 btrfs_check_read_bio(bbio, NULL);
377 btrfs_bio_end_io(bbio, bbio->bio.bi_status);
379 btrfs_put_bioc(bioc);
382 static void btrfs_orig_write_end_io(struct bio *bio)
384 struct btrfs_io_stripe *stripe = bio->bi_private;
385 struct btrfs_io_context *bioc = stripe->bioc;
386 struct btrfs_bio *bbio = btrfs_bio(bio);
388 btrfs_bio_counter_dec(bioc->fs_info);
390 if (bio->bi_status) {
391 atomic_inc(&bioc->error);
392 btrfs_log_dev_io_error(bio, stripe->dev);
396 * Only send an error to the higher layers if it is beyond the tolerance
399 if (atomic_read(&bioc->error) > bioc->max_errors)
400 bio->bi_status = BLK_STS_IOERR;
402 bio->bi_status = BLK_STS_OK;
404 if (bio_is_zone_append(bio) && !bio->bi_status)
405 stripe->physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
407 btrfs_bio_end_io(bbio, bbio->bio.bi_status);
408 btrfs_put_bioc(bioc);
411 static void btrfs_clone_write_end_io(struct bio *bio)
413 struct btrfs_io_stripe *stripe = bio->bi_private;
415 if (bio->bi_status) {
416 atomic_inc(&stripe->bioc->error);
417 btrfs_log_dev_io_error(bio, stripe->dev);
418 } else if (bio_is_zone_append(bio)) {
419 stripe->physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
422 /* Pass on control to the original bio this one was cloned from */
423 bio_endio(stripe->bioc->orig_bio);
427 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio)
429 if (!dev || !dev->bdev ||
430 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
431 (btrfs_op(bio) == BTRFS_MAP_WRITE &&
432 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
437 bio_set_dev(bio, dev->bdev);
440 * For zone append writing, bi_sector must point the beginning of the
443 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
444 u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
445 u64 zone_start = round_down(physical, dev->fs_info->zone_size);
447 ASSERT(btrfs_dev_is_sequential(dev, physical));
448 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
450 btrfs_debug_in_rcu(dev->fs_info,
451 "%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
452 __func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
453 (unsigned long)dev->bdev->bd_dev, btrfs_dev_name(dev),
454 dev->devid, bio->bi_iter.bi_size);
456 if (bio->bi_opf & REQ_BTRFS_CGROUP_PUNT)
457 blkcg_punt_bio_submit(bio);
462 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr)
464 struct bio *orig_bio = bioc->orig_bio, *bio;
466 ASSERT(bio_op(orig_bio) != REQ_OP_READ);
468 /* Reuse the bio embedded into the btrfs_bio for the last mirror */
469 if (dev_nr == bioc->num_stripes - 1) {
471 bio->bi_end_io = btrfs_orig_write_end_io;
473 bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set);
474 bio_inc_remaining(orig_bio);
475 bio->bi_end_io = btrfs_clone_write_end_io;
478 bio->bi_private = &bioc->stripes[dev_nr];
479 bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT;
480 bioc->stripes[dev_nr].bioc = bioc;
481 bioc->size = bio->bi_iter.bi_size;
482 btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio);
485 static void btrfs_submit_bio(struct bio *bio, struct btrfs_io_context *bioc,
486 struct btrfs_io_stripe *smap, int mirror_num)
489 /* Single mirror read/write fast path. */
490 btrfs_bio(bio)->mirror_num = mirror_num;
491 bio->bi_iter.bi_sector = smap->physical >> SECTOR_SHIFT;
492 if (bio_op(bio) != REQ_OP_READ)
493 btrfs_bio(bio)->orig_physical = smap->physical;
494 bio->bi_private = smap->dev;
495 bio->bi_end_io = btrfs_simple_end_io;
496 btrfs_submit_dev_bio(smap->dev, bio);
497 } else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
498 /* Parity RAID write or read recovery. */
499 bio->bi_private = bioc;
500 bio->bi_end_io = btrfs_raid56_end_io;
501 if (bio_op(bio) == REQ_OP_READ)
502 raid56_parity_recover(bio, bioc, mirror_num);
504 raid56_parity_write(bio, bioc);
506 /* Write to multiple mirrors. */
507 int total_devs = bioc->num_stripes;
509 bioc->orig_bio = bio;
510 for (int dev_nr = 0; dev_nr < total_devs; dev_nr++)
511 btrfs_submit_mirrored_bio(bioc, dev_nr);
515 static blk_status_t btrfs_bio_csum(struct btrfs_bio *bbio)
517 if (bbio->bio.bi_opf & REQ_META)
518 return btree_csum_one_bio(bbio);
519 return btrfs_csum_one_bio(bbio);
523 * Async submit bios are used to offload expensive checksumming onto the worker
526 struct async_submit_bio {
527 struct btrfs_bio *bbio;
528 struct btrfs_io_context *bioc;
529 struct btrfs_io_stripe smap;
531 struct btrfs_work work;
535 * In order to insert checksums into the metadata in large chunks, we wait
536 * until bio submission time. All the pages in the bio are checksummed and
537 * sums are attached onto the ordered extent record.
539 * At IO completion time the csums attached on the ordered extent record are
540 * inserted into the btree.
542 static void run_one_async_start(struct btrfs_work *work)
544 struct async_submit_bio *async =
545 container_of(work, struct async_submit_bio, work);
548 ret = btrfs_bio_csum(async->bbio);
550 async->bbio->bio.bi_status = ret;
554 * In order to insert checksums into the metadata in large chunks, we wait
555 * until bio submission time. All the pages in the bio are checksummed and
556 * sums are attached onto the ordered extent record.
558 * At IO completion time the csums attached on the ordered extent record are
559 * inserted into the tree.
561 * If called with @do_free == true, then it will free the work struct.
563 static void run_one_async_done(struct btrfs_work *work, bool do_free)
565 struct async_submit_bio *async =
566 container_of(work, struct async_submit_bio, work);
567 struct bio *bio = &async->bbio->bio;
570 kfree(container_of(work, struct async_submit_bio, work));
574 /* If an error occurred we just want to clean up the bio and move on. */
575 if (bio->bi_status) {
576 btrfs_bio_end_io(async->bbio, async->bbio->bio.bi_status);
581 * All of the bios that pass through here are from async helpers.
582 * Use REQ_BTRFS_CGROUP_PUNT to issue them from the owning cgroup's
583 * context. This changes nothing when cgroups aren't in use.
585 bio->bi_opf |= REQ_BTRFS_CGROUP_PUNT;
586 btrfs_submit_bio(bio, async->bioc, &async->smap, async->mirror_num);
589 static bool should_async_write(struct btrfs_bio *bbio)
591 bool auto_csum_mode = true;
593 #ifdef CONFIG_BTRFS_EXPERIMENTAL
594 struct btrfs_fs_devices *fs_devices = bbio->fs_info->fs_devices;
595 enum btrfs_offload_csum_mode csum_mode = READ_ONCE(fs_devices->offload_csum_mode);
597 if (csum_mode == BTRFS_OFFLOAD_CSUM_FORCE_OFF)
600 auto_csum_mode = (csum_mode == BTRFS_OFFLOAD_CSUM_AUTO);
603 /* Submit synchronously if the checksum implementation is fast. */
604 if (auto_csum_mode && test_bit(BTRFS_FS_CSUM_IMPL_FAST, &bbio->fs_info->flags))
608 * Try to defer the submission to a workqueue to parallelize the
609 * checksum calculation unless the I/O is issued synchronously.
611 if (op_is_sync(bbio->bio.bi_opf))
614 /* Zoned devices require I/O to be submitted in order. */
615 if ((bbio->bio.bi_opf & REQ_META) && btrfs_is_zoned(bbio->fs_info))
622 * Submit bio to an async queue.
624 * Return true if the work has been successfully submitted, else false.
626 static bool btrfs_wq_submit_bio(struct btrfs_bio *bbio,
627 struct btrfs_io_context *bioc,
628 struct btrfs_io_stripe *smap, int mirror_num)
630 struct btrfs_fs_info *fs_info = bbio->fs_info;
631 struct async_submit_bio *async;
633 async = kmalloc(sizeof(*async), GFP_NOFS);
640 async->mirror_num = mirror_num;
642 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done);
643 btrfs_queue_work(fs_info->workers, &async->work);
647 static u64 btrfs_append_map_length(struct btrfs_bio *bbio, u64 map_length)
649 unsigned int nr_segs;
652 map_length = min(map_length, bbio->fs_info->max_zone_append_size);
653 sector_offset = bio_split_rw_at(&bbio->bio, &bbio->fs_info->limits,
654 &nr_segs, map_length);
657 * bio_split_rw_at() could split at a size smaller than our
658 * sectorsize and thus cause unaligned I/Os. Fix that by
659 * always rounding down to the nearest boundary.
661 return ALIGN_DOWN(sector_offset << SECTOR_SHIFT, bbio->fs_info->sectorsize);
666 static bool btrfs_submit_chunk(struct btrfs_bio *bbio, int mirror_num)
668 struct btrfs_inode *inode = bbio->inode;
669 struct btrfs_fs_info *fs_info = bbio->fs_info;
670 struct bio *bio = &bbio->bio;
671 u64 logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
672 u64 length = bio->bi_iter.bi_size;
673 u64 map_length = length;
674 bool use_append = btrfs_use_zone_append(bbio);
675 struct btrfs_io_context *bioc = NULL;
676 struct btrfs_io_stripe smap;
680 if (!bbio->inode || btrfs_is_data_reloc_root(inode->root))
681 smap.rst_search_commit_root = true;
683 smap.rst_search_commit_root = false;
685 btrfs_bio_counter_inc_blocked(fs_info);
686 error = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
687 &bioc, &smap, &mirror_num);
689 ret = errno_to_blk_status(error);
690 btrfs_bio_counter_dec(fs_info);
694 map_length = min(map_length, length);
696 map_length = btrfs_append_map_length(bbio, map_length);
698 if (map_length < length) {
699 struct btrfs_bio *split;
701 split = btrfs_split_bio(fs_info, bbio, map_length);
703 ret = errno_to_blk_status(PTR_ERR(split));
704 btrfs_bio_counter_dec(fs_info);
712 * Save the iter for the end_io handler and preload the checksums for
715 if (bio_op(bio) == REQ_OP_READ && is_data_bbio(bbio)) {
716 bbio->saved_iter = bio->bi_iter;
717 ret = btrfs_lookup_bio_sums(bbio);
722 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
724 bio->bi_opf &= ~REQ_OP_WRITE;
725 bio->bi_opf |= REQ_OP_ZONE_APPEND;
728 if (is_data_bbio(bbio) && bioc &&
729 btrfs_need_stripe_tree_update(bioc->fs_info, bioc->map_type)) {
731 * No locking for the list update, as we only add to
732 * the list in the I/O submission path, and list
733 * iteration only happens in the completion path, which
734 * can't happen until after the last submission.
736 btrfs_get_bioc(bioc);
737 list_add_tail(&bioc->rst_ordered_entry, &bbio->ordered->bioc_list);
741 * Csum items for reloc roots have already been cloned at this
742 * point, so they are handled as part of the no-checksum case.
744 if (inode && !(inode->flags & BTRFS_INODE_NODATASUM) &&
745 !test_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state) &&
746 !btrfs_is_data_reloc_root(inode->root)) {
747 if (should_async_write(bbio) &&
748 btrfs_wq_submit_bio(bbio, bioc, &smap, mirror_num))
751 ret = btrfs_bio_csum(bbio);
754 } else if (use_append ||
755 (btrfs_is_zoned(fs_info) && inode &&
756 inode->flags & BTRFS_INODE_NODATASUM)) {
757 ret = btrfs_alloc_dummy_sum(bbio);
763 btrfs_submit_bio(bio, bioc, &smap, mirror_num);
765 return map_length == length;
768 btrfs_bio_counter_dec(fs_info);
770 * We have split the original bbio, now we have to end both the current
771 * @bbio and remaining one, as the remaining one will never be submitted.
773 if (map_length < length) {
774 struct btrfs_bio *remaining = bbio->private;
776 ASSERT(bbio->bio.bi_pool == &btrfs_clone_bioset);
779 btrfs_bio_end_io(remaining, ret);
782 btrfs_bio_end_io(bbio, ret);
783 /* Do not submit another chunk */
787 void btrfs_submit_bbio(struct btrfs_bio *bbio, int mirror_num)
789 /* If bbio->inode is not populated, its file_offset must be 0. */
790 ASSERT(bbio->inode || bbio->file_offset == 0);
792 while (!btrfs_submit_chunk(bbio, mirror_num))
797 * Submit a repair write.
799 * This bypasses btrfs_submit_bbio() deliberately, as that writes all copies in a
800 * RAID setup. Here we only want to write the one bad copy, so we do the
801 * mapping ourselves and submit the bio directly.
803 * The I/O is issued synchronously to block the repair read completion from
806 int btrfs_repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
807 u64 length, u64 logical, struct folio *folio,
808 unsigned int folio_offset, int mirror_num)
810 struct btrfs_io_stripe smap = { 0 };
815 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
818 if (btrfs_repair_one_zone(fs_info, logical))
822 * Avoid races with device replace and make sure our bioc has devices
823 * associated to its stripes that don't go away while we are doing the
824 * read repair operation.
826 btrfs_bio_counter_inc_blocked(fs_info);
827 ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
829 goto out_counter_dec;
831 if (!smap.dev->bdev ||
832 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &smap.dev->dev_state)) {
834 goto out_counter_dec;
837 bio_init(&bio, smap.dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
838 bio.bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT;
839 ret = bio_add_folio(&bio, folio, length, folio_offset);
841 ret = submit_bio_wait(&bio);
843 /* try to remap that extent elsewhere? */
844 btrfs_dev_stat_inc_and_print(smap.dev, BTRFS_DEV_STAT_WRITE_ERRS);
848 btrfs_info_rl_in_rcu(fs_info,
849 "read error corrected: ino %llu off %llu (dev %s sector %llu)",
850 ino, start, btrfs_dev_name(smap.dev),
851 smap.physical >> SECTOR_SHIFT);
857 btrfs_bio_counter_dec(fs_info);
862 * Submit a btrfs_bio based repair write.
864 * If @dev_replace is true, the write would be submitted to dev-replace target.
866 void btrfs_submit_repair_write(struct btrfs_bio *bbio, int mirror_num, bool dev_replace)
868 struct btrfs_fs_info *fs_info = bbio->fs_info;
869 u64 logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
870 u64 length = bbio->bio.bi_iter.bi_size;
871 struct btrfs_io_stripe smap = { 0 };
875 ASSERT(mirror_num > 0);
876 ASSERT(btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE);
877 ASSERT(!bbio->inode);
879 btrfs_bio_counter_inc_blocked(fs_info);
880 ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
885 ASSERT(smap.dev == fs_info->dev_replace.srcdev);
886 smap.dev = fs_info->dev_replace.tgtdev;
888 btrfs_submit_bio(&bbio->bio, NULL, &smap, mirror_num);
892 btrfs_bio_counter_dec(fs_info);
893 btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
896 int __init btrfs_bioset_init(void)
898 if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
899 offsetof(struct btrfs_bio, bio),
902 if (bioset_init(&btrfs_clone_bioset, BIO_POOL_SIZE,
903 offsetof(struct btrfs_bio, bio), 0))
904 goto out_free_bioset;
905 if (bioset_init(&btrfs_repair_bioset, BIO_POOL_SIZE,
906 offsetof(struct btrfs_bio, bio),
908 goto out_free_clone_bioset;
909 if (mempool_init_kmalloc_pool(&btrfs_failed_bio_pool, BIO_POOL_SIZE,
910 sizeof(struct btrfs_failed_bio)))
911 goto out_free_repair_bioset;
914 out_free_repair_bioset:
915 bioset_exit(&btrfs_repair_bioset);
916 out_free_clone_bioset:
917 bioset_exit(&btrfs_clone_bioset);
919 bioset_exit(&btrfs_bioset);
923 void __cold btrfs_bioset_exit(void)
925 mempool_exit(&btrfs_failed_bio_pool);
926 bioset_exit(&btrfs_repair_bioset);
927 bioset_exit(&btrfs_clone_bioset);
928 bioset_exit(&btrfs_bioset);