xref: /linux/fs/btrfs/bio.c (revision 1f20a5769446a1acae67ac9e63d07a594829a789)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  * Copyright (C) 2022 Christoph Hellwig.
5  */
6 
7 #include <linux/bio.h>
8 #include "bio.h"
9 #include "ctree.h"
10 #include "volumes.h"
11 #include "raid56.h"
12 #include "async-thread.h"
13 #include "dev-replace.h"
14 #include "zoned.h"
15 #include "file-item.h"
16 #include "raid-stripe-tree.h"
17 
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;
22 
23 struct btrfs_failed_bio {
24 	struct btrfs_bio *bbio;
25 	int num_copies;
26 	atomic_t repair_count;
27 };
28 
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)
31 {
32 	return bbio->inode && is_data_inode(&bbio->inode->vfs_inode);
33 }
34 
35 static bool bbio_has_ordered_extent(struct btrfs_bio *bbio)
36 {
37 	return is_data_bbio(bbio) && btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE;
38 }
39 
40 /*
41  * Initialize a btrfs_bio structure.  This skips the embedded bio itself as it
42  * is already initialized by the block layer.
43  */
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)
46 {
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 }
53 
54 /*
55  * Allocate a btrfs_bio structure.  The btrfs_bio is the main I/O container for
56  * btrfs, and is used for all I/O submitted through btrfs_submit_bio.
57  *
58  * Just like the underlying bio_alloc_bioset it will not fail as it is backed by
59  * a mempool.
60  */
61 struct btrfs_bio *btrfs_bio_alloc(unsigned int nr_vecs, blk_opf_t opf,
62 				  struct btrfs_fs_info *fs_info,
63 				  btrfs_bio_end_io_t end_io, void *private)
64 {
65 	struct btrfs_bio *bbio;
66 	struct bio *bio;
67 
68 	bio = bio_alloc_bioset(NULL, nr_vecs, opf, GFP_NOFS, &btrfs_bioset);
69 	bbio = btrfs_bio(bio);
70 	btrfs_bio_init(bbio, fs_info, end_io, private);
71 	return bbio;
72 }
73 
74 static struct btrfs_bio *btrfs_split_bio(struct btrfs_fs_info *fs_info,
75 					 struct btrfs_bio *orig_bbio,
76 					 u64 map_length, bool use_append)
77 {
78 	struct btrfs_bio *bbio;
79 	struct bio *bio;
80 
81 	if (use_append) {
82 		unsigned int nr_segs;
83 
84 		bio = bio_split_rw(&orig_bbio->bio, &fs_info->limits, &nr_segs,
85 				   &btrfs_clone_bioset, map_length);
86 	} else {
87 		bio = bio_split(&orig_bbio->bio, map_length >> SECTOR_SHIFT,
88 				GFP_NOFS, &btrfs_clone_bioset);
89 	}
90 	bbio = btrfs_bio(bio);
91 	btrfs_bio_init(bbio, fs_info, NULL, orig_bbio);
92 	bbio->inode = orig_bbio->inode;
93 	bbio->file_offset = orig_bbio->file_offset;
94 	orig_bbio->file_offset += map_length;
95 	if (bbio_has_ordered_extent(bbio)) {
96 		refcount_inc(&orig_bbio->ordered->refs);
97 		bbio->ordered = orig_bbio->ordered;
98 	}
99 	atomic_inc(&orig_bbio->pending_ios);
100 	return bbio;
101 }
102 
103 /* Free a bio that was never submitted to the underlying device. */
104 static void btrfs_cleanup_bio(struct btrfs_bio *bbio)
105 {
106 	if (bbio_has_ordered_extent(bbio))
107 		btrfs_put_ordered_extent(bbio->ordered);
108 	bio_put(&bbio->bio);
109 }
110 
111 static void __btrfs_bio_end_io(struct btrfs_bio *bbio)
112 {
113 	if (bbio_has_ordered_extent(bbio)) {
114 		struct btrfs_ordered_extent *ordered = bbio->ordered;
115 
116 		bbio->end_io(bbio);
117 		btrfs_put_ordered_extent(ordered);
118 	} else {
119 		bbio->end_io(bbio);
120 	}
121 }
122 
123 void btrfs_bio_end_io(struct btrfs_bio *bbio, blk_status_t status)
124 {
125 	bbio->bio.bi_status = status;
126 	__btrfs_bio_end_io(bbio);
127 }
128 
129 static void btrfs_orig_write_end_io(struct bio *bio);
130 
131 static void btrfs_bbio_propagate_error(struct btrfs_bio *bbio,
132 				       struct btrfs_bio *orig_bbio)
133 {
134 	/*
135 	 * For writes we tolerate nr_mirrors - 1 write failures, so we can't
136 	 * just blindly propagate a write failure here.  Instead increment the
137 	 * error count in the original I/O context so that it is guaranteed to
138 	 * be larger than the error tolerance.
139 	 */
140 	if (bbio->bio.bi_end_io == &btrfs_orig_write_end_io) {
141 		struct btrfs_io_stripe *orig_stripe = orig_bbio->bio.bi_private;
142 		struct btrfs_io_context *orig_bioc = orig_stripe->bioc;
143 
144 		atomic_add(orig_bioc->max_errors, &orig_bioc->error);
145 	} else {
146 		orig_bbio->bio.bi_status = bbio->bio.bi_status;
147 	}
148 }
149 
150 static void btrfs_orig_bbio_end_io(struct btrfs_bio *bbio)
151 {
152 	if (bbio->bio.bi_pool == &btrfs_clone_bioset) {
153 		struct btrfs_bio *orig_bbio = bbio->private;
154 
155 		if (bbio->bio.bi_status)
156 			btrfs_bbio_propagate_error(bbio, orig_bbio);
157 		btrfs_cleanup_bio(bbio);
158 		bbio = orig_bbio;
159 	}
160 
161 	if (atomic_dec_and_test(&bbio->pending_ios))
162 		__btrfs_bio_end_io(bbio);
163 }
164 
165 static int next_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
166 {
167 	if (cur_mirror == fbio->num_copies)
168 		return cur_mirror + 1 - fbio->num_copies;
169 	return cur_mirror + 1;
170 }
171 
172 static int prev_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
173 {
174 	if (cur_mirror == 1)
175 		return fbio->num_copies;
176 	return cur_mirror - 1;
177 }
178 
179 static void btrfs_repair_done(struct btrfs_failed_bio *fbio)
180 {
181 	if (atomic_dec_and_test(&fbio->repair_count)) {
182 		btrfs_orig_bbio_end_io(fbio->bbio);
183 		mempool_free(fbio, &btrfs_failed_bio_pool);
184 	}
185 }
186 
187 static void btrfs_end_repair_bio(struct btrfs_bio *repair_bbio,
188 				 struct btrfs_device *dev)
189 {
190 	struct btrfs_failed_bio *fbio = repair_bbio->private;
191 	struct btrfs_inode *inode = repair_bbio->inode;
192 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
193 	struct bio_vec *bv = bio_first_bvec_all(&repair_bbio->bio);
194 	int mirror = repair_bbio->mirror_num;
195 
196 	/*
197 	 * We can only trigger this for data bio, which doesn't support larger
198 	 * folios yet.
199 	 */
200 	ASSERT(folio_order(page_folio(bv->bv_page)) == 0);
201 
202 	if (repair_bbio->bio.bi_status ||
203 	    !btrfs_data_csum_ok(repair_bbio, dev, 0, bv)) {
204 		bio_reset(&repair_bbio->bio, NULL, REQ_OP_READ);
205 		repair_bbio->bio.bi_iter = repair_bbio->saved_iter;
206 
207 		mirror = next_repair_mirror(fbio, mirror);
208 		if (mirror == fbio->bbio->mirror_num) {
209 			btrfs_debug(fs_info, "no mirror left");
210 			fbio->bbio->bio.bi_status = BLK_STS_IOERR;
211 			goto done;
212 		}
213 
214 		btrfs_submit_bio(repair_bbio, mirror);
215 		return;
216 	}
217 
218 	do {
219 		mirror = prev_repair_mirror(fbio, mirror);
220 		btrfs_repair_io_failure(fs_info, btrfs_ino(inode),
221 				  repair_bbio->file_offset, fs_info->sectorsize,
222 				  repair_bbio->saved_iter.bi_sector << SECTOR_SHIFT,
223 				  page_folio(bv->bv_page), bv->bv_offset, mirror);
224 	} while (mirror != fbio->bbio->mirror_num);
225 
226 done:
227 	btrfs_repair_done(fbio);
228 	bio_put(&repair_bbio->bio);
229 }
230 
231 /*
232  * Try to kick off a repair read to the next available mirror for a bad sector.
233  *
234  * This primarily tries to recover good data to serve the actual read request,
235  * but also tries to write the good data back to the bad mirror(s) when a
236  * read succeeded to restore the redundancy.
237  */
238 static struct btrfs_failed_bio *repair_one_sector(struct btrfs_bio *failed_bbio,
239 						  u32 bio_offset,
240 						  struct bio_vec *bv,
241 						  struct btrfs_failed_bio *fbio)
242 {
243 	struct btrfs_inode *inode = failed_bbio->inode;
244 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
245 	const u32 sectorsize = fs_info->sectorsize;
246 	const u64 logical = (failed_bbio->saved_iter.bi_sector << SECTOR_SHIFT);
247 	struct btrfs_bio *repair_bbio;
248 	struct bio *repair_bio;
249 	int num_copies;
250 	int mirror;
251 
252 	btrfs_debug(fs_info, "repair read error: read error at %llu",
253 		    failed_bbio->file_offset + bio_offset);
254 
255 	num_copies = btrfs_num_copies(fs_info, logical, sectorsize);
256 	if (num_copies == 1) {
257 		btrfs_debug(fs_info, "no copy to repair from");
258 		failed_bbio->bio.bi_status = BLK_STS_IOERR;
259 		return fbio;
260 	}
261 
262 	if (!fbio) {
263 		fbio = mempool_alloc(&btrfs_failed_bio_pool, GFP_NOFS);
264 		fbio->bbio = failed_bbio;
265 		fbio->num_copies = num_copies;
266 		atomic_set(&fbio->repair_count, 1);
267 	}
268 
269 	atomic_inc(&fbio->repair_count);
270 
271 	repair_bio = bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS,
272 				      &btrfs_repair_bioset);
273 	repair_bio->bi_iter.bi_sector = failed_bbio->saved_iter.bi_sector;
274 	__bio_add_page(repair_bio, bv->bv_page, bv->bv_len, bv->bv_offset);
275 
276 	repair_bbio = btrfs_bio(repair_bio);
277 	btrfs_bio_init(repair_bbio, fs_info, NULL, fbio);
278 	repair_bbio->inode = failed_bbio->inode;
279 	repair_bbio->file_offset = failed_bbio->file_offset + bio_offset;
280 
281 	mirror = next_repair_mirror(fbio, failed_bbio->mirror_num);
282 	btrfs_debug(fs_info, "submitting repair read to mirror %d", mirror);
283 	btrfs_submit_bio(repair_bbio, mirror);
284 	return fbio;
285 }
286 
287 static void btrfs_check_read_bio(struct btrfs_bio *bbio, struct btrfs_device *dev)
288 {
289 	struct btrfs_inode *inode = bbio->inode;
290 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
291 	u32 sectorsize = fs_info->sectorsize;
292 	struct bvec_iter *iter = &bbio->saved_iter;
293 	blk_status_t status = bbio->bio.bi_status;
294 	struct btrfs_failed_bio *fbio = NULL;
295 	u32 offset = 0;
296 
297 	/* Read-repair requires the inode field to be set by the submitter. */
298 	ASSERT(inode);
299 
300 	/*
301 	 * Hand off repair bios to the repair code as there is no upper level
302 	 * submitter for them.
303 	 */
304 	if (bbio->bio.bi_pool == &btrfs_repair_bioset) {
305 		btrfs_end_repair_bio(bbio, dev);
306 		return;
307 	}
308 
309 	/* Clear the I/O error. A failed repair will reset it. */
310 	bbio->bio.bi_status = BLK_STS_OK;
311 
312 	while (iter->bi_size) {
313 		struct bio_vec bv = bio_iter_iovec(&bbio->bio, *iter);
314 
315 		bv.bv_len = min(bv.bv_len, sectorsize);
316 		if (status || !btrfs_data_csum_ok(bbio, dev, offset, &bv))
317 			fbio = repair_one_sector(bbio, offset, &bv, fbio);
318 
319 		bio_advance_iter_single(&bbio->bio, iter, sectorsize);
320 		offset += sectorsize;
321 	}
322 
323 	if (bbio->csum != bbio->csum_inline)
324 		kfree(bbio->csum);
325 
326 	if (fbio)
327 		btrfs_repair_done(fbio);
328 	else
329 		btrfs_orig_bbio_end_io(bbio);
330 }
331 
332 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev)
333 {
334 	if (!dev || !dev->bdev)
335 		return;
336 	if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET)
337 		return;
338 
339 	if (btrfs_op(bio) == BTRFS_MAP_WRITE)
340 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
341 	else if (!(bio->bi_opf & REQ_RAHEAD))
342 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
343 	if (bio->bi_opf & REQ_PREFLUSH)
344 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS);
345 }
346 
347 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info,
348 						struct bio *bio)
349 {
350 	if (bio->bi_opf & REQ_META)
351 		return fs_info->endio_meta_workers;
352 	return fs_info->endio_workers;
353 }
354 
355 static void btrfs_end_bio_work(struct work_struct *work)
356 {
357 	struct btrfs_bio *bbio = container_of(work, struct btrfs_bio, end_io_work);
358 
359 	/* Metadata reads are checked and repaired by the submitter. */
360 	if (is_data_bbio(bbio))
361 		btrfs_check_read_bio(bbio, bbio->bio.bi_private);
362 	else
363 		btrfs_orig_bbio_end_io(bbio);
364 }
365 
366 static void btrfs_simple_end_io(struct bio *bio)
367 {
368 	struct btrfs_bio *bbio = btrfs_bio(bio);
369 	struct btrfs_device *dev = bio->bi_private;
370 	struct btrfs_fs_info *fs_info = bbio->fs_info;
371 
372 	btrfs_bio_counter_dec(fs_info);
373 
374 	if (bio->bi_status)
375 		btrfs_log_dev_io_error(bio, dev);
376 
377 	if (bio_op(bio) == REQ_OP_READ) {
378 		INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work);
379 		queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work);
380 	} else {
381 		if (bio_op(bio) == REQ_OP_ZONE_APPEND && !bio->bi_status)
382 			btrfs_record_physical_zoned(bbio);
383 		btrfs_orig_bbio_end_io(bbio);
384 	}
385 }
386 
387 static void btrfs_raid56_end_io(struct bio *bio)
388 {
389 	struct btrfs_io_context *bioc = bio->bi_private;
390 	struct btrfs_bio *bbio = btrfs_bio(bio);
391 
392 	btrfs_bio_counter_dec(bioc->fs_info);
393 	bbio->mirror_num = bioc->mirror_num;
394 	if (bio_op(bio) == REQ_OP_READ && is_data_bbio(bbio))
395 		btrfs_check_read_bio(bbio, NULL);
396 	else
397 		btrfs_orig_bbio_end_io(bbio);
398 
399 	btrfs_put_bioc(bioc);
400 }
401 
402 static void btrfs_orig_write_end_io(struct bio *bio)
403 {
404 	struct btrfs_io_stripe *stripe = bio->bi_private;
405 	struct btrfs_io_context *bioc = stripe->bioc;
406 	struct btrfs_bio *bbio = btrfs_bio(bio);
407 
408 	btrfs_bio_counter_dec(bioc->fs_info);
409 
410 	if (bio->bi_status) {
411 		atomic_inc(&bioc->error);
412 		btrfs_log_dev_io_error(bio, stripe->dev);
413 	}
414 
415 	/*
416 	 * Only send an error to the higher layers if it is beyond the tolerance
417 	 * threshold.
418 	 */
419 	if (atomic_read(&bioc->error) > bioc->max_errors)
420 		bio->bi_status = BLK_STS_IOERR;
421 	else
422 		bio->bi_status = BLK_STS_OK;
423 
424 	if (bio_op(bio) == REQ_OP_ZONE_APPEND && !bio->bi_status)
425 		stripe->physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
426 
427 	btrfs_orig_bbio_end_io(bbio);
428 	btrfs_put_bioc(bioc);
429 }
430 
431 static void btrfs_clone_write_end_io(struct bio *bio)
432 {
433 	struct btrfs_io_stripe *stripe = bio->bi_private;
434 
435 	if (bio->bi_status) {
436 		atomic_inc(&stripe->bioc->error);
437 		btrfs_log_dev_io_error(bio, stripe->dev);
438 	} else if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
439 		stripe->physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
440 	}
441 
442 	/* Pass on control to the original bio this one was cloned from */
443 	bio_endio(stripe->bioc->orig_bio);
444 	bio_put(bio);
445 }
446 
447 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio)
448 {
449 	if (!dev || !dev->bdev ||
450 	    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
451 	    (btrfs_op(bio) == BTRFS_MAP_WRITE &&
452 	     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
453 		bio_io_error(bio);
454 		return;
455 	}
456 
457 	bio_set_dev(bio, dev->bdev);
458 
459 	/*
460 	 * For zone append writing, bi_sector must point the beginning of the
461 	 * zone
462 	 */
463 	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
464 		u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
465 		u64 zone_start = round_down(physical, dev->fs_info->zone_size);
466 
467 		ASSERT(btrfs_dev_is_sequential(dev, physical));
468 		bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
469 	}
470 	btrfs_debug_in_rcu(dev->fs_info,
471 	"%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
472 		__func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
473 		(unsigned long)dev->bdev->bd_dev, btrfs_dev_name(dev),
474 		dev->devid, bio->bi_iter.bi_size);
475 
476 	if (bio->bi_opf & REQ_BTRFS_CGROUP_PUNT)
477 		blkcg_punt_bio_submit(bio);
478 	else
479 		submit_bio(bio);
480 }
481 
482 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr)
483 {
484 	struct bio *orig_bio = bioc->orig_bio, *bio;
485 
486 	ASSERT(bio_op(orig_bio) != REQ_OP_READ);
487 
488 	/* Reuse the bio embedded into the btrfs_bio for the last mirror */
489 	if (dev_nr == bioc->num_stripes - 1) {
490 		bio = orig_bio;
491 		bio->bi_end_io = btrfs_orig_write_end_io;
492 	} else {
493 		bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set);
494 		bio_inc_remaining(orig_bio);
495 		bio->bi_end_io = btrfs_clone_write_end_io;
496 	}
497 
498 	bio->bi_private = &bioc->stripes[dev_nr];
499 	bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT;
500 	bioc->stripes[dev_nr].bioc = bioc;
501 	bioc->size = bio->bi_iter.bi_size;
502 	btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio);
503 }
504 
505 static void __btrfs_submit_bio(struct bio *bio, struct btrfs_io_context *bioc,
506 			       struct btrfs_io_stripe *smap, int mirror_num)
507 {
508 	if (!bioc) {
509 		/* Single mirror read/write fast path. */
510 		btrfs_bio(bio)->mirror_num = mirror_num;
511 		bio->bi_iter.bi_sector = smap->physical >> SECTOR_SHIFT;
512 		if (bio_op(bio) != REQ_OP_READ)
513 			btrfs_bio(bio)->orig_physical = smap->physical;
514 		bio->bi_private = smap->dev;
515 		bio->bi_end_io = btrfs_simple_end_io;
516 		btrfs_submit_dev_bio(smap->dev, bio);
517 	} else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
518 		/* Parity RAID write or read recovery. */
519 		bio->bi_private = bioc;
520 		bio->bi_end_io = btrfs_raid56_end_io;
521 		if (bio_op(bio) == REQ_OP_READ)
522 			raid56_parity_recover(bio, bioc, mirror_num);
523 		else
524 			raid56_parity_write(bio, bioc);
525 	} else {
526 		/* Write to multiple mirrors. */
527 		int total_devs = bioc->num_stripes;
528 
529 		bioc->orig_bio = bio;
530 		for (int dev_nr = 0; dev_nr < total_devs; dev_nr++)
531 			btrfs_submit_mirrored_bio(bioc, dev_nr);
532 	}
533 }
534 
535 static blk_status_t btrfs_bio_csum(struct btrfs_bio *bbio)
536 {
537 	if (bbio->bio.bi_opf & REQ_META)
538 		return btree_csum_one_bio(bbio);
539 	return btrfs_csum_one_bio(bbio);
540 }
541 
542 /*
543  * Async submit bios are used to offload expensive checksumming onto the worker
544  * threads.
545  */
546 struct async_submit_bio {
547 	struct btrfs_bio *bbio;
548 	struct btrfs_io_context *bioc;
549 	struct btrfs_io_stripe smap;
550 	int mirror_num;
551 	struct btrfs_work work;
552 };
553 
554 /*
555  * In order to insert checksums into the metadata in large chunks, we wait
556  * until bio submission time.   All the pages in the bio are checksummed and
557  * sums are attached onto the ordered extent record.
558  *
559  * At IO completion time the csums attached on the ordered extent record are
560  * inserted into the btree.
561  */
562 static void run_one_async_start(struct btrfs_work *work)
563 {
564 	struct async_submit_bio *async =
565 		container_of(work, struct async_submit_bio, work);
566 	blk_status_t ret;
567 
568 	ret = btrfs_bio_csum(async->bbio);
569 	if (ret)
570 		async->bbio->bio.bi_status = ret;
571 }
572 
573 /*
574  * In order to insert checksums into the metadata in large chunks, we wait
575  * until bio submission time.   All the pages in the bio are checksummed and
576  * sums are attached onto the ordered extent record.
577  *
578  * At IO completion time the csums attached on the ordered extent record are
579  * inserted into the tree.
580  *
581  * If called with @do_free == true, then it will free the work struct.
582  */
583 static void run_one_async_done(struct btrfs_work *work, bool do_free)
584 {
585 	struct async_submit_bio *async =
586 		container_of(work, struct async_submit_bio, work);
587 	struct bio *bio = &async->bbio->bio;
588 
589 	if (do_free) {
590 		kfree(container_of(work, struct async_submit_bio, work));
591 		return;
592 	}
593 
594 	/* If an error occurred we just want to clean up the bio and move on. */
595 	if (bio->bi_status) {
596 		btrfs_orig_bbio_end_io(async->bbio);
597 		return;
598 	}
599 
600 	/*
601 	 * All of the bios that pass through here are from async helpers.
602 	 * Use REQ_BTRFS_CGROUP_PUNT to issue them from the owning cgroup's
603 	 * context.  This changes nothing when cgroups aren't in use.
604 	 */
605 	bio->bi_opf |= REQ_BTRFS_CGROUP_PUNT;
606 	__btrfs_submit_bio(bio, async->bioc, &async->smap, async->mirror_num);
607 }
608 
609 static bool should_async_write(struct btrfs_bio *bbio)
610 {
611 	bool auto_csum_mode = true;
612 
613 #ifdef CONFIG_BTRFS_DEBUG
614 	struct btrfs_fs_devices *fs_devices = bbio->fs_info->fs_devices;
615 	enum btrfs_offload_csum_mode csum_mode = READ_ONCE(fs_devices->offload_csum_mode);
616 
617 	if (csum_mode == BTRFS_OFFLOAD_CSUM_FORCE_OFF)
618 		return false;
619 
620 	auto_csum_mode = (csum_mode == BTRFS_OFFLOAD_CSUM_AUTO);
621 #endif
622 
623 	/* Submit synchronously if the checksum implementation is fast. */
624 	if (auto_csum_mode && test_bit(BTRFS_FS_CSUM_IMPL_FAST, &bbio->fs_info->flags))
625 		return false;
626 
627 	/*
628 	 * Try to defer the submission to a workqueue to parallelize the
629 	 * checksum calculation unless the I/O is issued synchronously.
630 	 */
631 	if (op_is_sync(bbio->bio.bi_opf))
632 		return false;
633 
634 	/* Zoned devices require I/O to be submitted in order. */
635 	if ((bbio->bio.bi_opf & REQ_META) && btrfs_is_zoned(bbio->fs_info))
636 		return false;
637 
638 	return true;
639 }
640 
641 /*
642  * Submit bio to an async queue.
643  *
644  * Return true if the work has been successfully submitted, else false.
645  */
646 static bool btrfs_wq_submit_bio(struct btrfs_bio *bbio,
647 				struct btrfs_io_context *bioc,
648 				struct btrfs_io_stripe *smap, int mirror_num)
649 {
650 	struct btrfs_fs_info *fs_info = bbio->fs_info;
651 	struct async_submit_bio *async;
652 
653 	async = kmalloc(sizeof(*async), GFP_NOFS);
654 	if (!async)
655 		return false;
656 
657 	async->bbio = bbio;
658 	async->bioc = bioc;
659 	async->smap = *smap;
660 	async->mirror_num = mirror_num;
661 
662 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done);
663 	btrfs_queue_work(fs_info->workers, &async->work);
664 	return true;
665 }
666 
667 static bool btrfs_submit_chunk(struct btrfs_bio *bbio, int mirror_num)
668 {
669 	struct btrfs_inode *inode = bbio->inode;
670 	struct btrfs_fs_info *fs_info = bbio->fs_info;
671 	struct btrfs_bio *orig_bbio = bbio;
672 	struct bio *bio = &bbio->bio;
673 	u64 logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
674 	u64 length = bio->bi_iter.bi_size;
675 	u64 map_length = length;
676 	bool use_append = btrfs_use_zone_append(bbio);
677 	struct btrfs_io_context *bioc = NULL;
678 	struct btrfs_io_stripe smap;
679 	blk_status_t ret;
680 	int error;
681 
682 	smap.is_scrub = !bbio->inode;
683 
684 	btrfs_bio_counter_inc_blocked(fs_info);
685 	error = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
686 				&bioc, &smap, &mirror_num);
687 	if (error) {
688 		ret = errno_to_blk_status(error);
689 		goto fail;
690 	}
691 
692 	map_length = min(map_length, length);
693 	if (use_append)
694 		map_length = min(map_length, fs_info->max_zone_append_size);
695 
696 	if (map_length < length) {
697 		bbio = btrfs_split_bio(fs_info, bbio, map_length, use_append);
698 		bio = &bbio->bio;
699 	}
700 
701 	/*
702 	 * Save the iter for the end_io handler and preload the checksums for
703 	 * data reads.
704 	 */
705 	if (bio_op(bio) == REQ_OP_READ && is_data_bbio(bbio)) {
706 		bbio->saved_iter = bio->bi_iter;
707 		ret = btrfs_lookup_bio_sums(bbio);
708 		if (ret)
709 			goto fail_put_bio;
710 	}
711 
712 	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
713 		if (use_append) {
714 			bio->bi_opf &= ~REQ_OP_WRITE;
715 			bio->bi_opf |= REQ_OP_ZONE_APPEND;
716 		}
717 
718 		if (is_data_bbio(bbio) && bioc &&
719 		    btrfs_need_stripe_tree_update(bioc->fs_info, bioc->map_type)) {
720 			/*
721 			 * No locking for the list update, as we only add to
722 			 * the list in the I/O submission path, and list
723 			 * iteration only happens in the completion path, which
724 			 * can't happen until after the last submission.
725 			 */
726 			btrfs_get_bioc(bioc);
727 			list_add_tail(&bioc->rst_ordered_entry, &bbio->ordered->bioc_list);
728 		}
729 
730 		/*
731 		 * Csum items for reloc roots have already been cloned at this
732 		 * point, so they are handled as part of the no-checksum case.
733 		 */
734 		if (inode && !(inode->flags & BTRFS_INODE_NODATASUM) &&
735 		    !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state) &&
736 		    !btrfs_is_data_reloc_root(inode->root)) {
737 			if (should_async_write(bbio) &&
738 			    btrfs_wq_submit_bio(bbio, bioc, &smap, mirror_num))
739 				goto done;
740 
741 			ret = btrfs_bio_csum(bbio);
742 			if (ret)
743 				goto fail_put_bio;
744 		} else if (use_append) {
745 			ret = btrfs_alloc_dummy_sum(bbio);
746 			if (ret)
747 				goto fail_put_bio;
748 		}
749 	}
750 
751 	__btrfs_submit_bio(bio, bioc, &smap, mirror_num);
752 done:
753 	return map_length == length;
754 
755 fail_put_bio:
756 	if (map_length < length)
757 		btrfs_cleanup_bio(bbio);
758 fail:
759 	btrfs_bio_counter_dec(fs_info);
760 	btrfs_bio_end_io(orig_bbio, ret);
761 	/* Do not submit another chunk */
762 	return true;
763 }
764 
765 void btrfs_submit_bio(struct btrfs_bio *bbio, int mirror_num)
766 {
767 	/* If bbio->inode is not populated, its file_offset must be 0. */
768 	ASSERT(bbio->inode || bbio->file_offset == 0);
769 
770 	while (!btrfs_submit_chunk(bbio, mirror_num))
771 		;
772 }
773 
774 /*
775  * Submit a repair write.
776  *
777  * This bypasses btrfs_submit_bio deliberately, as that writes all copies in a
778  * RAID setup.  Here we only want to write the one bad copy, so we do the
779  * mapping ourselves and submit the bio directly.
780  *
781  * The I/O is issued synchronously to block the repair read completion from
782  * freeing the bio.
783  */
784 int btrfs_repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
785 			    u64 length, u64 logical, struct folio *folio,
786 			    unsigned int folio_offset, int mirror_num)
787 {
788 	struct btrfs_io_stripe smap = { 0 };
789 	struct bio_vec bvec;
790 	struct bio bio;
791 	int ret = 0;
792 
793 	ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
794 	BUG_ON(!mirror_num);
795 
796 	if (btrfs_repair_one_zone(fs_info, logical))
797 		return 0;
798 
799 	/*
800 	 * Avoid races with device replace and make sure our bioc has devices
801 	 * associated to its stripes that don't go away while we are doing the
802 	 * read repair operation.
803 	 */
804 	btrfs_bio_counter_inc_blocked(fs_info);
805 	ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
806 	if (ret < 0)
807 		goto out_counter_dec;
808 
809 	if (!smap.dev->bdev ||
810 	    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &smap.dev->dev_state)) {
811 		ret = -EIO;
812 		goto out_counter_dec;
813 	}
814 
815 	bio_init(&bio, smap.dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
816 	bio.bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT;
817 	ret = bio_add_folio(&bio, folio, length, folio_offset);
818 	ASSERT(ret);
819 	ret = submit_bio_wait(&bio);
820 	if (ret) {
821 		/* try to remap that extent elsewhere? */
822 		btrfs_dev_stat_inc_and_print(smap.dev, BTRFS_DEV_STAT_WRITE_ERRS);
823 		goto out_bio_uninit;
824 	}
825 
826 	btrfs_info_rl_in_rcu(fs_info,
827 		"read error corrected: ino %llu off %llu (dev %s sector %llu)",
828 			     ino, start, btrfs_dev_name(smap.dev),
829 			     smap.physical >> SECTOR_SHIFT);
830 	ret = 0;
831 
832 out_bio_uninit:
833 	bio_uninit(&bio);
834 out_counter_dec:
835 	btrfs_bio_counter_dec(fs_info);
836 	return ret;
837 }
838 
839 /*
840  * Submit a btrfs_bio based repair write.
841  *
842  * If @dev_replace is true, the write would be submitted to dev-replace target.
843  */
844 void btrfs_submit_repair_write(struct btrfs_bio *bbio, int mirror_num, bool dev_replace)
845 {
846 	struct btrfs_fs_info *fs_info = bbio->fs_info;
847 	u64 logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
848 	u64 length = bbio->bio.bi_iter.bi_size;
849 	struct btrfs_io_stripe smap = { 0 };
850 	int ret;
851 
852 	ASSERT(fs_info);
853 	ASSERT(mirror_num > 0);
854 	ASSERT(btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE);
855 	ASSERT(!bbio->inode);
856 
857 	btrfs_bio_counter_inc_blocked(fs_info);
858 	ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
859 	if (ret < 0)
860 		goto fail;
861 
862 	if (dev_replace) {
863 		ASSERT(smap.dev == fs_info->dev_replace.srcdev);
864 		smap.dev = fs_info->dev_replace.tgtdev;
865 	}
866 	__btrfs_submit_bio(&bbio->bio, NULL, &smap, mirror_num);
867 	return;
868 
869 fail:
870 	btrfs_bio_counter_dec(fs_info);
871 	btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
872 }
873 
874 int __init btrfs_bioset_init(void)
875 {
876 	if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
877 			offsetof(struct btrfs_bio, bio),
878 			BIOSET_NEED_BVECS))
879 		return -ENOMEM;
880 	if (bioset_init(&btrfs_clone_bioset, BIO_POOL_SIZE,
881 			offsetof(struct btrfs_bio, bio), 0))
882 		goto out_free_bioset;
883 	if (bioset_init(&btrfs_repair_bioset, BIO_POOL_SIZE,
884 			offsetof(struct btrfs_bio, bio),
885 			BIOSET_NEED_BVECS))
886 		goto out_free_clone_bioset;
887 	if (mempool_init_kmalloc_pool(&btrfs_failed_bio_pool, BIO_POOL_SIZE,
888 				      sizeof(struct btrfs_failed_bio)))
889 		goto out_free_repair_bioset;
890 	return 0;
891 
892 out_free_repair_bioset:
893 	bioset_exit(&btrfs_repair_bioset);
894 out_free_clone_bioset:
895 	bioset_exit(&btrfs_clone_bioset);
896 out_free_bioset:
897 	bioset_exit(&btrfs_bioset);
898 	return -ENOMEM;
899 }
900 
901 void __cold btrfs_bioset_exit(void)
902 {
903 	mempool_exit(&btrfs_failed_bio_pool);
904 	bioset_exit(&btrfs_repair_bioset);
905 	bioset_exit(&btrfs_clone_bioset);
906 	bioset_exit(&btrfs_bioset);
907 }
908