Lines Matching +full:data +full:- +full:mirror
1 // SPDX-License-Identifier: GPL-2.0
12 #include "async-thread.h"
13 #include "dev-replace.h"
15 #include "file-item.h"
16 #include "raid-stripe-tree.h"
29 /* Is this a data path I/O that needs storage layer checksum and repair? */
32 return bbio->inode && is_data_inode(bbio->inode); in is_data_bbio()
37 return is_data_bbio(bbio) && btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE; in bbio_has_ordered_extent()
48 bbio->fs_info = fs_info; in btrfs_bio_init()
49 bbio->end_io = end_io; in btrfs_bio_init()
50 bbio->private = private; in btrfs_bio_init()
51 atomic_set(&bbio->pending_ios, 1); in btrfs_bio_init()
52 WRITE_ONCE(bbio->status, BLK_STS_OK); in btrfs_bio_init()
82 bio = bio_split(&orig_bbio->bio, map_length >> SECTOR_SHIFT, GFP_NOFS, in btrfs_split_bio()
89 bbio->inode = orig_bbio->inode; in btrfs_split_bio()
90 bbio->file_offset = orig_bbio->file_offset; in btrfs_split_bio()
91 orig_bbio->file_offset += map_length; in btrfs_split_bio()
93 refcount_inc(&orig_bbio->ordered->refs); in btrfs_split_bio()
94 bbio->ordered = orig_bbio->ordered; in btrfs_split_bio()
96 atomic_inc(&orig_bbio->pending_ios); in btrfs_split_bio()
102 bbio->bio.bi_status = status; in btrfs_bio_end_io()
103 if (bbio->bio.bi_pool == &btrfs_clone_bioset) { in btrfs_bio_end_io()
104 struct btrfs_bio *orig_bbio = bbio->private; in btrfs_bio_end_io()
108 btrfs_put_ordered_extent(bbio->ordered); in btrfs_bio_end_io()
109 bio_put(&bbio->bio); in btrfs_bio_end_io()
119 cmpxchg(&bbio->status, BLK_STS_OK, status); in btrfs_bio_end_io()
121 if (atomic_dec_and_test(&bbio->pending_ios)) { in btrfs_bio_end_io()
124 bbio->bio.bi_status = READ_ONCE(bbio->status); in btrfs_bio_end_io()
127 struct btrfs_ordered_extent *ordered = bbio->ordered; in btrfs_bio_end_io()
129 bbio->end_io(bbio); in btrfs_bio_end_io()
132 bbio->end_io(bbio); in btrfs_bio_end_io()
139 if (cur_mirror == fbio->num_copies) in next_repair_mirror()
140 return cur_mirror + 1 - fbio->num_copies; in next_repair_mirror()
147 return fbio->num_copies; in prev_repair_mirror()
148 return cur_mirror - 1; in prev_repair_mirror()
153 if (atomic_dec_and_test(&fbio->repair_count)) { in btrfs_repair_done()
154 btrfs_bio_end_io(fbio->bbio, fbio->bbio->bio.bi_status); in btrfs_repair_done()
162 struct btrfs_failed_bio *fbio = repair_bbio->private; in btrfs_end_repair_bio()
163 struct btrfs_inode *inode = repair_bbio->inode; in btrfs_end_repair_bio()
164 struct btrfs_fs_info *fs_info = inode->root->fs_info; in btrfs_end_repair_bio()
165 struct bio_vec *bv = bio_first_bvec_all(&repair_bbio->bio); in btrfs_end_repair_bio()
166 int mirror = repair_bbio->mirror_num; in btrfs_end_repair_bio() local
169 * We can only trigger this for data bio, which doesn't support larger in btrfs_end_repair_bio()
172 ASSERT(folio_order(page_folio(bv->bv_page)) == 0); in btrfs_end_repair_bio()
174 if (repair_bbio->bio.bi_status || in btrfs_end_repair_bio()
176 bio_reset(&repair_bbio->bio, NULL, REQ_OP_READ); in btrfs_end_repair_bio()
177 repair_bbio->bio.bi_iter = repair_bbio->saved_iter; in btrfs_end_repair_bio()
179 mirror = next_repair_mirror(fbio, mirror); in btrfs_end_repair_bio()
180 if (mirror == fbio->bbio->mirror_num) { in btrfs_end_repair_bio()
181 btrfs_debug(fs_info, "no mirror left"); in btrfs_end_repair_bio()
182 fbio->bbio->bio.bi_status = BLK_STS_IOERR; in btrfs_end_repair_bio()
186 btrfs_submit_bbio(repair_bbio, mirror); in btrfs_end_repair_bio()
191 mirror = prev_repair_mirror(fbio, mirror); in btrfs_end_repair_bio()
193 repair_bbio->file_offset, fs_info->sectorsize, in btrfs_end_repair_bio()
194 repair_bbio->saved_iter.bi_sector << SECTOR_SHIFT, in btrfs_end_repair_bio()
195 page_folio(bv->bv_page), bv->bv_offset, mirror); in btrfs_end_repair_bio()
196 } while (mirror != fbio->bbio->mirror_num); in btrfs_end_repair_bio()
200 bio_put(&repair_bbio->bio); in btrfs_end_repair_bio()
204 * Try to kick off a repair read to the next available mirror for a bad sector.
206 * This primarily tries to recover good data to serve the actual read request,
207 * but also tries to write the good data back to the bad mirror(s) when a
215 struct btrfs_inode *inode = failed_bbio->inode; in repair_one_sector()
216 struct btrfs_fs_info *fs_info = inode->root->fs_info; in repair_one_sector()
217 const u32 sectorsize = fs_info->sectorsize; in repair_one_sector()
218 const u64 logical = (failed_bbio->saved_iter.bi_sector << SECTOR_SHIFT); in repair_one_sector()
222 int mirror; in repair_one_sector() local
225 failed_bbio->file_offset + bio_offset); in repair_one_sector()
230 failed_bbio->bio.bi_status = BLK_STS_IOERR; in repair_one_sector()
236 fbio->bbio = failed_bbio; in repair_one_sector()
237 fbio->num_copies = num_copies; in repair_one_sector()
238 atomic_set(&fbio->repair_count, 1); in repair_one_sector()
241 atomic_inc(&fbio->repair_count); in repair_one_sector()
245 repair_bio->bi_iter.bi_sector = failed_bbio->saved_iter.bi_sector; in repair_one_sector()
246 __bio_add_page(repair_bio, bv->bv_page, bv->bv_len, bv->bv_offset); in repair_one_sector()
250 repair_bbio->inode = failed_bbio->inode; in repair_one_sector()
251 repair_bbio->file_offset = failed_bbio->file_offset + bio_offset; in repair_one_sector()
253 mirror = next_repair_mirror(fbio, failed_bbio->mirror_num); in repair_one_sector()
254 btrfs_debug(fs_info, "submitting repair read to mirror %d", mirror); in repair_one_sector()
255 btrfs_submit_bbio(repair_bbio, mirror); in repair_one_sector()
261 struct btrfs_inode *inode = bbio->inode; in btrfs_check_read_bio()
262 struct btrfs_fs_info *fs_info = inode->root->fs_info; in btrfs_check_read_bio()
263 u32 sectorsize = fs_info->sectorsize; in btrfs_check_read_bio()
264 struct bvec_iter *iter = &bbio->saved_iter; in btrfs_check_read_bio()
265 blk_status_t status = bbio->bio.bi_status; in btrfs_check_read_bio()
269 /* Read-repair requires the inode field to be set by the submitter. */ in btrfs_check_read_bio()
276 if (bbio->bio.bi_pool == &btrfs_repair_bioset) { in btrfs_check_read_bio()
282 bbio->bio.bi_status = BLK_STS_OK; in btrfs_check_read_bio()
284 while (iter->bi_size) { in btrfs_check_read_bio()
285 struct bio_vec bv = bio_iter_iovec(&bbio->bio, *iter); in btrfs_check_read_bio()
291 bio_advance_iter_single(&bbio->bio, iter, sectorsize); in btrfs_check_read_bio()
295 if (bbio->csum != bbio->csum_inline) in btrfs_check_read_bio()
296 kfree(bbio->csum); in btrfs_check_read_bio()
301 btrfs_bio_end_io(bbio, bbio->bio.bi_status); in btrfs_check_read_bio()
306 if (!dev || !dev->bdev) in btrfs_log_dev_io_error()
308 if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET) in btrfs_log_dev_io_error()
313 else if (!(bio->bi_opf & REQ_RAHEAD)) in btrfs_log_dev_io_error()
315 if (bio->bi_opf & REQ_PREFLUSH) in btrfs_log_dev_io_error()
322 if (bio->bi_opf & REQ_META) in btrfs_end_io_wq()
323 return fs_info->endio_meta_workers; in btrfs_end_io_wq()
324 return fs_info->endio_workers; in btrfs_end_io_wq()
333 btrfs_check_read_bio(bbio, bbio->bio.bi_private); in btrfs_end_bio_work()
335 btrfs_bio_end_io(bbio, bbio->bio.bi_status); in btrfs_end_bio_work()
341 struct btrfs_device *dev = bio->bi_private; in btrfs_simple_end_io()
342 struct btrfs_fs_info *fs_info = bbio->fs_info; in btrfs_simple_end_io()
346 if (bio->bi_status) in btrfs_simple_end_io()
350 INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work); in btrfs_simple_end_io()
351 queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work); in btrfs_simple_end_io()
353 if (bio_is_zone_append(bio) && !bio->bi_status) in btrfs_simple_end_io()
355 btrfs_bio_end_io(bbio, bbio->bio.bi_status); in btrfs_simple_end_io()
361 struct btrfs_io_context *bioc = bio->bi_private; in btrfs_raid56_end_io()
364 btrfs_bio_counter_dec(bioc->fs_info); in btrfs_raid56_end_io()
365 bbio->mirror_num = bioc->mirror_num; in btrfs_raid56_end_io()
369 btrfs_bio_end_io(bbio, bbio->bio.bi_status); in btrfs_raid56_end_io()
376 struct btrfs_io_stripe *stripe = bio->bi_private; in btrfs_orig_write_end_io()
377 struct btrfs_io_context *bioc = stripe->bioc; in btrfs_orig_write_end_io()
380 btrfs_bio_counter_dec(bioc->fs_info); in btrfs_orig_write_end_io()
382 if (bio->bi_status) { in btrfs_orig_write_end_io()
383 atomic_inc(&bioc->error); in btrfs_orig_write_end_io()
384 btrfs_log_dev_io_error(bio, stripe->dev); in btrfs_orig_write_end_io()
391 if (atomic_read(&bioc->error) > bioc->max_errors) in btrfs_orig_write_end_io()
392 bio->bi_status = BLK_STS_IOERR; in btrfs_orig_write_end_io()
394 bio->bi_status = BLK_STS_OK; in btrfs_orig_write_end_io()
396 if (bio_is_zone_append(bio) && !bio->bi_status) in btrfs_orig_write_end_io()
397 stripe->physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; in btrfs_orig_write_end_io()
399 btrfs_bio_end_io(bbio, bbio->bio.bi_status); in btrfs_orig_write_end_io()
405 struct btrfs_io_stripe *stripe = bio->bi_private; in btrfs_clone_write_end_io()
407 if (bio->bi_status) { in btrfs_clone_write_end_io()
408 atomic_inc(&stripe->bioc->error); in btrfs_clone_write_end_io()
409 btrfs_log_dev_io_error(bio, stripe->dev); in btrfs_clone_write_end_io()
411 stripe->physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; in btrfs_clone_write_end_io()
415 bio_endio(stripe->bioc->orig_bio); in btrfs_clone_write_end_io()
421 if (!dev || !dev->bdev || in btrfs_submit_dev_bio()
422 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || in btrfs_submit_dev_bio()
424 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { in btrfs_submit_dev_bio()
429 bio_set_dev(bio, dev->bdev); in btrfs_submit_dev_bio()
436 u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT; in btrfs_submit_dev_bio()
437 u64 zone_start = round_down(physical, dev->fs_info->zone_size); in btrfs_submit_dev_bio()
440 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; in btrfs_submit_dev_bio()
442 btrfs_debug_in_rcu(dev->fs_info, in btrfs_submit_dev_bio()
444 __func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, in btrfs_submit_dev_bio()
445 (unsigned long)dev->bdev->bd_dev, btrfs_dev_name(dev), in btrfs_submit_dev_bio()
446 dev->devid, bio->bi_iter.bi_size); in btrfs_submit_dev_bio()
452 if (dev->fs_devices->collect_fs_stats && bio_op(bio) == REQ_OP_READ && dev->fs_info) in btrfs_submit_dev_bio()
453 percpu_counter_add(&dev->fs_info->stats_read_blocks, in btrfs_submit_dev_bio()
454 bio->bi_iter.bi_size >> dev->fs_info->sectorsize_bits); in btrfs_submit_dev_bio()
456 if (bio->bi_opf & REQ_BTRFS_CGROUP_PUNT) in btrfs_submit_dev_bio()
464 struct bio *orig_bio = bioc->orig_bio, *bio; in btrfs_submit_mirrored_bio()
468 /* Reuse the bio embedded into the btrfs_bio for the last mirror */ in btrfs_submit_mirrored_bio()
469 if (dev_nr == bioc->num_stripes - 1) { in btrfs_submit_mirrored_bio()
471 bio->bi_end_io = btrfs_orig_write_end_io; in btrfs_submit_mirrored_bio()
475 bio->bi_end_io = btrfs_clone_write_end_io; in btrfs_submit_mirrored_bio()
478 bio->bi_private = &bioc->stripes[dev_nr]; in btrfs_submit_mirrored_bio()
479 bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT; in btrfs_submit_mirrored_bio()
480 bioc->stripes[dev_nr].bioc = bioc; in btrfs_submit_mirrored_bio()
481 bioc->size = bio->bi_iter.bi_size; in btrfs_submit_mirrored_bio()
482 btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio); in btrfs_submit_mirrored_bio()
489 /* Single mirror read/write fast path. */ in btrfs_submit_bio()
490 btrfs_bio(bio)->mirror_num = mirror_num; in btrfs_submit_bio()
491 bio->bi_iter.bi_sector = smap->physical >> SECTOR_SHIFT; in btrfs_submit_bio()
493 btrfs_bio(bio)->orig_physical = smap->physical; in btrfs_submit_bio()
494 bio->bi_private = smap->dev; in btrfs_submit_bio()
495 bio->bi_end_io = btrfs_simple_end_io; in btrfs_submit_bio()
496 btrfs_submit_dev_bio(smap->dev, bio); in btrfs_submit_bio()
497 } else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { in btrfs_submit_bio()
499 bio->bi_private = bioc; in btrfs_submit_bio()
500 bio->bi_end_io = btrfs_raid56_end_io; in btrfs_submit_bio()
507 int total_devs = bioc->num_stripes; in btrfs_submit_bio()
509 bioc->orig_bio = bio; in btrfs_submit_bio()
517 if (bbio->bio.bi_opf & REQ_META) in btrfs_bio_csum()
548 ret = btrfs_bio_csum(async->bbio); in run_one_async_start()
550 async->bbio->bio.bi_status = ret; in run_one_async_start()
567 struct bio *bio = &async->bbio->bio; in run_one_async_done()
575 if (bio->bi_status) { in run_one_async_done()
576 btrfs_bio_end_io(async->bbio, bio->bi_status); in run_one_async_done()
585 bio->bi_opf |= REQ_BTRFS_CGROUP_PUNT; in run_one_async_done()
586 btrfs_submit_bio(bio, async->bioc, &async->smap, async->mirror_num); in run_one_async_done()
594 struct btrfs_fs_devices *fs_devices = bbio->fs_info->fs_devices; in should_async_write()
595 enum btrfs_offload_csum_mode csum_mode = READ_ONCE(fs_devices->offload_csum_mode); in should_async_write()
604 if (auto_csum_mode && test_bit(BTRFS_FS_CSUM_IMPL_FAST, &bbio->fs_info->flags)) in should_async_write()
611 if (op_is_sync(bbio->bio.bi_opf)) in should_async_write()
615 if ((bbio->bio.bi_opf & REQ_META) && btrfs_is_zoned(bbio->fs_info)) in should_async_write()
630 struct btrfs_fs_info *fs_info = bbio->fs_info; in btrfs_wq_submit_bio()
637 async->bbio = bbio; in btrfs_wq_submit_bio()
638 async->bioc = bioc; in btrfs_wq_submit_bio()
639 async->smap = *smap; in btrfs_wq_submit_bio()
640 async->mirror_num = mirror_num; in btrfs_wq_submit_bio()
642 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done); in btrfs_wq_submit_bio()
643 btrfs_queue_work(fs_info->workers, &async->work); in btrfs_wq_submit_bio()
652 map_length = min(map_length, bbio->fs_info->max_zone_append_size); in btrfs_append_map_length()
653 sector_offset = bio_split_rw_at(&bbio->bio, &bbio->fs_info->limits, in btrfs_append_map_length()
661 return ALIGN_DOWN(sector_offset << SECTOR_SHIFT, bbio->fs_info->sectorsize); in btrfs_append_map_length()
668 struct btrfs_inode *inode = bbio->inode; in btrfs_submit_chunk()
669 struct btrfs_fs_info *fs_info = bbio->fs_info; in btrfs_submit_chunk()
670 struct bio *bio = &bbio->bio; in btrfs_submit_chunk()
671 u64 logical = bio->bi_iter.bi_sector << SECTOR_SHIFT; in btrfs_submit_chunk()
672 u64 length = bio->bi_iter.bi_size; in btrfs_submit_chunk()
680 if (!bbio->inode || btrfs_is_data_reloc_root(inode->root)) in btrfs_submit_chunk()
708 bio = &bbio->bio; in btrfs_submit_chunk()
713 * data reads. in btrfs_submit_chunk()
716 bbio->saved_iter = bio->bi_iter; in btrfs_submit_chunk()
724 bio->bi_opf &= ~REQ_OP_WRITE; in btrfs_submit_chunk()
725 bio->bi_opf |= REQ_OP_ZONE_APPEND; in btrfs_submit_chunk()
728 if (is_data_bbio(bbio) && bioc && bioc->use_rst) { in btrfs_submit_chunk()
736 list_add_tail(&bioc->rst_ordered_entry, &bbio->ordered->bioc_list); in btrfs_submit_chunk()
741 * point, so they are handled as part of the no-checksum case. in btrfs_submit_chunk()
743 if (inode && !(inode->flags & BTRFS_INODE_NODATASUM) && in btrfs_submit_chunk()
744 !test_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state) && in btrfs_submit_chunk()
745 !btrfs_is_data_reloc_root(inode->root)) { in btrfs_submit_chunk()
755 inode->flags & BTRFS_INODE_NODATASUM)) { in btrfs_submit_chunk()
773 struct btrfs_bio *remaining = bbio->private; in btrfs_submit_chunk()
775 ASSERT(bbio->bio.bi_pool == &btrfs_clone_bioset); in btrfs_submit_chunk()
788 /* If bbio->inode is not populated, its file_offset must be 0. */ in btrfs_submit_bbio()
789 ASSERT(bbio->inode || bbio->file_offset == 0); in btrfs_submit_bbio()
814 ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); in btrfs_repair_io_failure()
830 if (!smap.dev->bdev || in btrfs_repair_io_failure()
831 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &smap.dev->dev_state)) { in btrfs_repair_io_failure()
832 ret = -EIO; in btrfs_repair_io_failure()
836 bio_init(&bio, smap.dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC); in btrfs_repair_io_failure()
863 * If @dev_replace is true, the write would be submitted to dev-replace target.
867 struct btrfs_fs_info *fs_info = bbio->fs_info; in btrfs_submit_repair_write()
868 u64 logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT; in btrfs_submit_repair_write()
869 u64 length = bbio->bio.bi_iter.bi_size; in btrfs_submit_repair_write()
875 ASSERT(btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE); in btrfs_submit_repair_write()
876 ASSERT(!bbio->inode); in btrfs_submit_repair_write()
884 ASSERT(smap.dev == fs_info->dev_replace.srcdev); in btrfs_submit_repair_write()
885 smap.dev = fs_info->dev_replace.tgtdev; in btrfs_submit_repair_write()
887 btrfs_submit_bio(&bbio->bio, NULL, &smap, mirror_num); in btrfs_submit_repair_write()
900 return -ENOMEM; in btrfs_bioset_init()
919 return -ENOMEM; in btrfs_bioset_init()