xref: /openbmc/linux/fs/btrfs/bio.c (revision 51722b99)
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 "check-integrity.h"
14 #include "dev-replace.h"
15 #include "rcu-string.h"
16 #include "zoned.h"
17 #include "file-item.h"
18 
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;
23 
24 struct btrfs_failed_bio {
25 	struct btrfs_bio *bbio;
26 	int num_copies;
27 	atomic_t repair_count;
28 };
29 
30 /* Is this a data path I/O that needs storage layer checksum and repair? */
is_data_bbio(struct btrfs_bio * bbio)31 static inline bool is_data_bbio(struct btrfs_bio *bbio)
32 {
33 	return bbio->inode && is_data_inode(&bbio->inode->vfs_inode);
34 }
35 
bbio_has_ordered_extent(struct btrfs_bio * bbio)36 static bool bbio_has_ordered_extent(struct btrfs_bio *bbio)
37 {
38 	return is_data_bbio(bbio) && btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE;
39 }
40 
41 /*
42  * Initialize a btrfs_bio structure.  This skips the embedded bio itself as it
43  * is already initialized by the block layer.
44  */
btrfs_bio_init(struct btrfs_bio * bbio,struct btrfs_fs_info * fs_info,btrfs_bio_end_io_t end_io,void * private)45 void btrfs_bio_init(struct btrfs_bio *bbio, struct btrfs_fs_info *fs_info,
46 		    btrfs_bio_end_io_t end_io, void *private)
47 {
48 	memset(bbio, 0, offsetof(struct btrfs_bio, bio));
49 	bbio->fs_info = fs_info;
50 	bbio->end_io = end_io;
51 	bbio->private = private;
52 	atomic_set(&bbio->pending_ios, 1);
53 }
54 
55 /*
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_bio.
58  *
59  * Just like the underlying bio_alloc_bioset it will not fail as it is backed by
60  * a mempool.
61  */
btrfs_bio_alloc(unsigned int nr_vecs,blk_opf_t opf,struct btrfs_fs_info * fs_info,btrfs_bio_end_io_t end_io,void * private)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)
65 {
66 	struct btrfs_bio *bbio;
67 	struct bio *bio;
68 
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);
72 	return bbio;
73 }
74 
btrfs_split_bio(struct btrfs_fs_info * fs_info,struct btrfs_bio * orig_bbio,u64 map_length,bool use_append)75 static struct btrfs_bio *btrfs_split_bio(struct btrfs_fs_info *fs_info,
76 					 struct btrfs_bio *orig_bbio,
77 					 u64 map_length, bool use_append)
78 {
79 	struct btrfs_bio *bbio;
80 	struct bio *bio;
81 
82 	if (use_append) {
83 		unsigned int nr_segs;
84 
85 		bio = bio_split_rw(&orig_bbio->bio, &fs_info->limits, &nr_segs,
86 				   &btrfs_clone_bioset, map_length);
87 	} else {
88 		bio = bio_split(&orig_bbio->bio, map_length >> SECTOR_SHIFT,
89 				GFP_NOFS, &btrfs_clone_bioset);
90 	}
91 	bbio = btrfs_bio(bio);
92 	btrfs_bio_init(bbio, fs_info, NULL, orig_bbio);
93 	bbio->inode = orig_bbio->inode;
94 	bbio->file_offset = orig_bbio->file_offset;
95 	orig_bbio->file_offset += map_length;
96 	if (bbio_has_ordered_extent(bbio)) {
97 		refcount_inc(&orig_bbio->ordered->refs);
98 		bbio->ordered = orig_bbio->ordered;
99 	}
100 	atomic_inc(&orig_bbio->pending_ios);
101 	return bbio;
102 }
103 
104 /* Free a bio that was never submitted to the underlying device. */
btrfs_cleanup_bio(struct btrfs_bio * bbio)105 static void btrfs_cleanup_bio(struct btrfs_bio *bbio)
106 {
107 	if (bbio_has_ordered_extent(bbio))
108 		btrfs_put_ordered_extent(bbio->ordered);
109 	bio_put(&bbio->bio);
110 }
111 
__btrfs_bio_end_io(struct btrfs_bio * bbio)112 static void __btrfs_bio_end_io(struct btrfs_bio *bbio)
113 {
114 	if (bbio_has_ordered_extent(bbio)) {
115 		struct btrfs_ordered_extent *ordered = bbio->ordered;
116 
117 		bbio->end_io(bbio);
118 		btrfs_put_ordered_extent(ordered);
119 	} else {
120 		bbio->end_io(bbio);
121 	}
122 }
123 
btrfs_bio_end_io(struct btrfs_bio * bbio,blk_status_t status)124 void btrfs_bio_end_io(struct btrfs_bio *bbio, blk_status_t status)
125 {
126 	bbio->bio.bi_status = status;
127 	__btrfs_bio_end_io(bbio);
128 }
129 
130 static void btrfs_orig_write_end_io(struct bio *bio);
131 
btrfs_bbio_propagate_error(struct btrfs_bio * bbio,struct btrfs_bio * orig_bbio)132 static void btrfs_bbio_propagate_error(struct btrfs_bio *bbio,
133 				       struct btrfs_bio *orig_bbio)
134 {
135 	/*
136 	 * For writes we tolerate nr_mirrors - 1 write failures, so we can't
137 	 * just blindly propagate a write failure here.  Instead increment the
138 	 * error count in the original I/O context so that it is guaranteed to
139 	 * be larger than the error tolerance.
140 	 */
141 	if (bbio->bio.bi_end_io == &btrfs_orig_write_end_io) {
142 		struct btrfs_io_stripe *orig_stripe = orig_bbio->bio.bi_private;
143 		struct btrfs_io_context *orig_bioc = orig_stripe->bioc;
144 
145 		atomic_add(orig_bioc->max_errors, &orig_bioc->error);
146 	} else {
147 		orig_bbio->bio.bi_status = bbio->bio.bi_status;
148 	}
149 }
150 
btrfs_orig_bbio_end_io(struct btrfs_bio * bbio)151 static void btrfs_orig_bbio_end_io(struct btrfs_bio *bbio)
152 {
153 	if (bbio->bio.bi_pool == &btrfs_clone_bioset) {
154 		struct btrfs_bio *orig_bbio = bbio->private;
155 
156 		if (bbio->bio.bi_status)
157 			btrfs_bbio_propagate_error(bbio, orig_bbio);
158 		btrfs_cleanup_bio(bbio);
159 		bbio = orig_bbio;
160 	}
161 
162 	if (atomic_dec_and_test(&bbio->pending_ios))
163 		__btrfs_bio_end_io(bbio);
164 }
165 
next_repair_mirror(struct btrfs_failed_bio * fbio,int cur_mirror)166 static int next_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
167 {
168 	if (cur_mirror == fbio->num_copies)
169 		return cur_mirror + 1 - fbio->num_copies;
170 	return cur_mirror + 1;
171 }
172 
prev_repair_mirror(struct btrfs_failed_bio * fbio,int cur_mirror)173 static int prev_repair_mirror(struct btrfs_failed_bio *fbio, int cur_mirror)
174 {
175 	if (cur_mirror == 1)
176 		return fbio->num_copies;
177 	return cur_mirror - 1;
178 }
179 
btrfs_repair_done(struct btrfs_failed_bio * fbio)180 static void btrfs_repair_done(struct btrfs_failed_bio *fbio)
181 {
182 	if (atomic_dec_and_test(&fbio->repair_count)) {
183 		btrfs_orig_bbio_end_io(fbio->bbio);
184 		mempool_free(fbio, &btrfs_failed_bio_pool);
185 	}
186 }
187 
btrfs_end_repair_bio(struct btrfs_bio * repair_bbio,struct btrfs_device * dev)188 static void btrfs_end_repair_bio(struct btrfs_bio *repair_bbio,
189 				 struct btrfs_device *dev)
190 {
191 	struct btrfs_failed_bio *fbio = repair_bbio->private;
192 	struct btrfs_inode *inode = repair_bbio->inode;
193 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
194 	struct bio_vec *bv = bio_first_bvec_all(&repair_bbio->bio);
195 	int mirror = repair_bbio->mirror_num;
196 
197 	if (repair_bbio->bio.bi_status ||
198 	    !btrfs_data_csum_ok(repair_bbio, dev, 0, bv)) {
199 		bio_reset(&repair_bbio->bio, NULL, REQ_OP_READ);
200 		repair_bbio->bio.bi_iter = repair_bbio->saved_iter;
201 
202 		mirror = next_repair_mirror(fbio, mirror);
203 		if (mirror == fbio->bbio->mirror_num) {
204 			btrfs_debug(fs_info, "no mirror left");
205 			fbio->bbio->bio.bi_status = BLK_STS_IOERR;
206 			goto done;
207 		}
208 
209 		btrfs_submit_bio(repair_bbio, mirror);
210 		return;
211 	}
212 
213 	do {
214 		mirror = prev_repair_mirror(fbio, mirror);
215 		btrfs_repair_io_failure(fs_info, btrfs_ino(inode),
216 				  repair_bbio->file_offset, fs_info->sectorsize,
217 				  repair_bbio->saved_iter.bi_sector << SECTOR_SHIFT,
218 				  bv->bv_page, bv->bv_offset, mirror);
219 	} while (mirror != fbio->bbio->mirror_num);
220 
221 done:
222 	btrfs_repair_done(fbio);
223 	bio_put(&repair_bbio->bio);
224 }
225 
226 /*
227  * Try to kick off a repair read to the next available mirror for a bad sector.
228  *
229  * This primarily tries to recover good data to serve the actual read request,
230  * but also tries to write the good data back to the bad mirror(s) when a
231  * read succeeded to restore the redundancy.
232  */
repair_one_sector(struct btrfs_bio * failed_bbio,u32 bio_offset,struct bio_vec * bv,struct btrfs_failed_bio * fbio)233 static struct btrfs_failed_bio *repair_one_sector(struct btrfs_bio *failed_bbio,
234 						  u32 bio_offset,
235 						  struct bio_vec *bv,
236 						  struct btrfs_failed_bio *fbio)
237 {
238 	struct btrfs_inode *inode = failed_bbio->inode;
239 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
240 	const u32 sectorsize = fs_info->sectorsize;
241 	const u64 logical = (failed_bbio->saved_iter.bi_sector << SECTOR_SHIFT);
242 	struct btrfs_bio *repair_bbio;
243 	struct bio *repair_bio;
244 	int num_copies;
245 	int mirror;
246 
247 	btrfs_debug(fs_info, "repair read error: read error at %llu",
248 		    failed_bbio->file_offset + bio_offset);
249 
250 	num_copies = btrfs_num_copies(fs_info, logical, sectorsize);
251 	if (num_copies == 1) {
252 		btrfs_debug(fs_info, "no copy to repair from");
253 		failed_bbio->bio.bi_status = BLK_STS_IOERR;
254 		return fbio;
255 	}
256 
257 	if (!fbio) {
258 		fbio = mempool_alloc(&btrfs_failed_bio_pool, GFP_NOFS);
259 		fbio->bbio = failed_bbio;
260 		fbio->num_copies = num_copies;
261 		atomic_set(&fbio->repair_count, 1);
262 	}
263 
264 	atomic_inc(&fbio->repair_count);
265 
266 	repair_bio = bio_alloc_bioset(NULL, 1, REQ_OP_READ, GFP_NOFS,
267 				      &btrfs_repair_bioset);
268 	repair_bio->bi_iter.bi_sector = failed_bbio->saved_iter.bi_sector;
269 	__bio_add_page(repair_bio, bv->bv_page, bv->bv_len, bv->bv_offset);
270 
271 	repair_bbio = btrfs_bio(repair_bio);
272 	btrfs_bio_init(repair_bbio, fs_info, NULL, fbio);
273 	repair_bbio->inode = failed_bbio->inode;
274 	repair_bbio->file_offset = failed_bbio->file_offset + bio_offset;
275 
276 	mirror = next_repair_mirror(fbio, failed_bbio->mirror_num);
277 	btrfs_debug(fs_info, "submitting repair read to mirror %d", mirror);
278 	btrfs_submit_bio(repair_bbio, mirror);
279 	return fbio;
280 }
281 
btrfs_check_read_bio(struct btrfs_bio * bbio,struct btrfs_device * dev)282 static void btrfs_check_read_bio(struct btrfs_bio *bbio, struct btrfs_device *dev)
283 {
284 	struct btrfs_inode *inode = bbio->inode;
285 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
286 	u32 sectorsize = fs_info->sectorsize;
287 	struct bvec_iter *iter = &bbio->saved_iter;
288 	blk_status_t status = bbio->bio.bi_status;
289 	struct btrfs_failed_bio *fbio = NULL;
290 	u32 offset = 0;
291 
292 	/* Read-repair requires the inode field to be set by the submitter. */
293 	ASSERT(inode);
294 
295 	/*
296 	 * Hand off repair bios to the repair code as there is no upper level
297 	 * submitter for them.
298 	 */
299 	if (bbio->bio.bi_pool == &btrfs_repair_bioset) {
300 		btrfs_end_repair_bio(bbio, dev);
301 		return;
302 	}
303 
304 	/* Clear the I/O error. A failed repair will reset it. */
305 	bbio->bio.bi_status = BLK_STS_OK;
306 
307 	while (iter->bi_size) {
308 		struct bio_vec bv = bio_iter_iovec(&bbio->bio, *iter);
309 
310 		bv.bv_len = min(bv.bv_len, sectorsize);
311 		if (status || !btrfs_data_csum_ok(bbio, dev, offset, &bv))
312 			fbio = repair_one_sector(bbio, offset, &bv, fbio);
313 
314 		bio_advance_iter_single(&bbio->bio, iter, sectorsize);
315 		offset += sectorsize;
316 	}
317 
318 	if (bbio->csum != bbio->csum_inline)
319 		kfree(bbio->csum);
320 
321 	if (fbio)
322 		btrfs_repair_done(fbio);
323 	else
324 		btrfs_orig_bbio_end_io(bbio);
325 }
326 
btrfs_log_dev_io_error(struct bio * bio,struct btrfs_device * dev)327 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev)
328 {
329 	if (!dev || !dev->bdev)
330 		return;
331 	if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET)
332 		return;
333 
334 	if (btrfs_op(bio) == BTRFS_MAP_WRITE)
335 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
336 	else if (!(bio->bi_opf & REQ_RAHEAD))
337 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
338 	if (bio->bi_opf & REQ_PREFLUSH)
339 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS);
340 }
341 
btrfs_end_io_wq(struct btrfs_fs_info * fs_info,struct bio * bio)342 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info,
343 						struct bio *bio)
344 {
345 	if (bio->bi_opf & REQ_META)
346 		return fs_info->endio_meta_workers;
347 	return fs_info->endio_workers;
348 }
349 
btrfs_end_bio_work(struct work_struct * work)350 static void btrfs_end_bio_work(struct work_struct *work)
351 {
352 	struct btrfs_bio *bbio = container_of(work, struct btrfs_bio, end_io_work);
353 
354 	/* Metadata reads are checked and repaired by the submitter. */
355 	if (is_data_bbio(bbio))
356 		btrfs_check_read_bio(bbio, bbio->bio.bi_private);
357 	else
358 		btrfs_orig_bbio_end_io(bbio);
359 }
360 
btrfs_simple_end_io(struct bio * bio)361 static void btrfs_simple_end_io(struct bio *bio)
362 {
363 	struct btrfs_bio *bbio = btrfs_bio(bio);
364 	struct btrfs_device *dev = bio->bi_private;
365 	struct btrfs_fs_info *fs_info = bbio->fs_info;
366 
367 	btrfs_bio_counter_dec(fs_info);
368 
369 	if (bio->bi_status)
370 		btrfs_log_dev_io_error(bio, dev);
371 
372 	if (bio_op(bio) == REQ_OP_READ) {
373 		INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work);
374 		queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work);
375 	} else {
376 		if (bio_op(bio) == REQ_OP_ZONE_APPEND && !bio->bi_status)
377 			btrfs_record_physical_zoned(bbio);
378 		btrfs_orig_bbio_end_io(bbio);
379 	}
380 }
381 
btrfs_raid56_end_io(struct bio * bio)382 static void btrfs_raid56_end_io(struct bio *bio)
383 {
384 	struct btrfs_io_context *bioc = bio->bi_private;
385 	struct btrfs_bio *bbio = btrfs_bio(bio);
386 
387 	btrfs_bio_counter_dec(bioc->fs_info);
388 	bbio->mirror_num = bioc->mirror_num;
389 	if (bio_op(bio) == REQ_OP_READ && is_data_bbio(bbio))
390 		btrfs_check_read_bio(bbio, NULL);
391 	else
392 		btrfs_orig_bbio_end_io(bbio);
393 
394 	btrfs_put_bioc(bioc);
395 }
396 
btrfs_orig_write_end_io(struct bio * bio)397 static void btrfs_orig_write_end_io(struct bio *bio)
398 {
399 	struct btrfs_io_stripe *stripe = bio->bi_private;
400 	struct btrfs_io_context *bioc = stripe->bioc;
401 	struct btrfs_bio *bbio = btrfs_bio(bio);
402 
403 	btrfs_bio_counter_dec(bioc->fs_info);
404 
405 	if (bio->bi_status) {
406 		atomic_inc(&bioc->error);
407 		btrfs_log_dev_io_error(bio, stripe->dev);
408 	}
409 
410 	/*
411 	 * Only send an error to the higher layers if it is beyond the tolerance
412 	 * threshold.
413 	 */
414 	if (atomic_read(&bioc->error) > bioc->max_errors)
415 		bio->bi_status = BLK_STS_IOERR;
416 	else
417 		bio->bi_status = BLK_STS_OK;
418 
419 	btrfs_orig_bbio_end_io(bbio);
420 	btrfs_put_bioc(bioc);
421 }
422 
btrfs_clone_write_end_io(struct bio * bio)423 static void btrfs_clone_write_end_io(struct bio *bio)
424 {
425 	struct btrfs_io_stripe *stripe = bio->bi_private;
426 
427 	if (bio->bi_status) {
428 		atomic_inc(&stripe->bioc->error);
429 		btrfs_log_dev_io_error(bio, stripe->dev);
430 	}
431 
432 	/* Pass on control to the original bio this one was cloned from */
433 	bio_endio(stripe->bioc->orig_bio);
434 	bio_put(bio);
435 }
436 
btrfs_submit_dev_bio(struct btrfs_device * dev,struct bio * bio)437 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio)
438 {
439 	if (!dev || !dev->bdev ||
440 	    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
441 	    (btrfs_op(bio) == BTRFS_MAP_WRITE &&
442 	     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
443 		bio_io_error(bio);
444 		return;
445 	}
446 
447 	bio_set_dev(bio, dev->bdev);
448 
449 	/*
450 	 * For zone append writing, bi_sector must point the beginning of the
451 	 * zone
452 	 */
453 	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
454 		u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
455 		u64 zone_start = round_down(physical, dev->fs_info->zone_size);
456 
457 		ASSERT(btrfs_dev_is_sequential(dev, physical));
458 		bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
459 	}
460 	btrfs_debug_in_rcu(dev->fs_info,
461 	"%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
462 		__func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
463 		(unsigned long)dev->bdev->bd_dev, btrfs_dev_name(dev),
464 		dev->devid, bio->bi_iter.bi_size);
465 
466 	btrfsic_check_bio(bio);
467 
468 	if (bio->bi_opf & REQ_BTRFS_CGROUP_PUNT)
469 		blkcg_punt_bio_submit(bio);
470 	else
471 		submit_bio(bio);
472 }
473 
btrfs_submit_mirrored_bio(struct btrfs_io_context * bioc,int dev_nr)474 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr)
475 {
476 	struct bio *orig_bio = bioc->orig_bio, *bio;
477 
478 	ASSERT(bio_op(orig_bio) != REQ_OP_READ);
479 
480 	/* Reuse the bio embedded into the btrfs_bio for the last mirror */
481 	if (dev_nr == bioc->num_stripes - 1) {
482 		bio = orig_bio;
483 		bio->bi_end_io = btrfs_orig_write_end_io;
484 	} else {
485 		bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set);
486 		bio_inc_remaining(orig_bio);
487 		bio->bi_end_io = btrfs_clone_write_end_io;
488 	}
489 
490 	bio->bi_private = &bioc->stripes[dev_nr];
491 	bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT;
492 	bioc->stripes[dev_nr].bioc = bioc;
493 	btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio);
494 }
495 
__btrfs_submit_bio(struct bio * bio,struct btrfs_io_context * bioc,struct btrfs_io_stripe * smap,int mirror_num)496 static void __btrfs_submit_bio(struct bio *bio, struct btrfs_io_context *bioc,
497 			       struct btrfs_io_stripe *smap, int mirror_num)
498 {
499 	if (!bioc) {
500 		/* Single mirror read/write fast path. */
501 		btrfs_bio(bio)->mirror_num = mirror_num;
502 		bio->bi_iter.bi_sector = smap->physical >> SECTOR_SHIFT;
503 		if (bio_op(bio) != REQ_OP_READ)
504 			btrfs_bio(bio)->orig_physical = smap->physical;
505 		bio->bi_private = smap->dev;
506 		bio->bi_end_io = btrfs_simple_end_io;
507 		btrfs_submit_dev_bio(smap->dev, bio);
508 	} else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
509 		/* Parity RAID write or read recovery. */
510 		bio->bi_private = bioc;
511 		bio->bi_end_io = btrfs_raid56_end_io;
512 		if (bio_op(bio) == REQ_OP_READ)
513 			raid56_parity_recover(bio, bioc, mirror_num);
514 		else
515 			raid56_parity_write(bio, bioc);
516 	} else {
517 		/* Write to multiple mirrors. */
518 		int total_devs = bioc->num_stripes;
519 
520 		bioc->orig_bio = bio;
521 		for (int dev_nr = 0; dev_nr < total_devs; dev_nr++)
522 			btrfs_submit_mirrored_bio(bioc, dev_nr);
523 	}
524 }
525 
btrfs_bio_csum(struct btrfs_bio * bbio)526 static blk_status_t btrfs_bio_csum(struct btrfs_bio *bbio)
527 {
528 	if (bbio->bio.bi_opf & REQ_META)
529 		return btree_csum_one_bio(bbio);
530 	return btrfs_csum_one_bio(bbio);
531 }
532 
533 /*
534  * Async submit bios are used to offload expensive checksumming onto the worker
535  * threads.
536  */
537 struct async_submit_bio {
538 	struct btrfs_bio *bbio;
539 	struct btrfs_io_context *bioc;
540 	struct btrfs_io_stripe smap;
541 	int mirror_num;
542 	struct btrfs_work work;
543 };
544 
545 /*
546  * In order to insert checksums into the metadata in large chunks, we wait
547  * until bio submission time.   All the pages in the bio are checksummed and
548  * sums are attached onto the ordered extent record.
549  *
550  * At IO completion time the csums attached on the ordered extent record are
551  * inserted into the btree.
552  */
run_one_async_start(struct btrfs_work * work)553 static void run_one_async_start(struct btrfs_work *work)
554 {
555 	struct async_submit_bio *async =
556 		container_of(work, struct async_submit_bio, work);
557 	blk_status_t ret;
558 
559 	ret = btrfs_bio_csum(async->bbio);
560 	if (ret)
561 		async->bbio->bio.bi_status = ret;
562 }
563 
564 /*
565  * In order to insert checksums into the metadata in large chunks, we wait
566  * until bio submission time.   All the pages in the bio are checksummed and
567  * sums are attached onto the ordered extent record.
568  *
569  * At IO completion time the csums attached on the ordered extent record are
570  * inserted into the tree.
571  */
run_one_async_done(struct btrfs_work * work)572 static void run_one_async_done(struct btrfs_work *work)
573 {
574 	struct async_submit_bio *async =
575 		container_of(work, struct async_submit_bio, work);
576 	struct bio *bio = &async->bbio->bio;
577 
578 	/* If an error occurred we just want to clean up the bio and move on. */
579 	if (bio->bi_status) {
580 		btrfs_orig_bbio_end_io(async->bbio);
581 		return;
582 	}
583 
584 	/*
585 	 * All of the bios that pass through here are from async helpers.
586 	 * Use REQ_BTRFS_CGROUP_PUNT to issue them from the owning cgroup's
587 	 * context.  This changes nothing when cgroups aren't in use.
588 	 */
589 	bio->bi_opf |= REQ_BTRFS_CGROUP_PUNT;
590 	__btrfs_submit_bio(bio, async->bioc, &async->smap, async->mirror_num);
591 }
592 
run_one_async_free(struct btrfs_work * work)593 static void run_one_async_free(struct btrfs_work *work)
594 {
595 	kfree(container_of(work, struct async_submit_bio, work));
596 }
597 
should_async_write(struct btrfs_bio * bbio)598 static bool should_async_write(struct btrfs_bio *bbio)
599 {
600 	/* Submit synchronously if the checksum implementation is fast. */
601 	if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &bbio->fs_info->flags))
602 		return false;
603 
604 	/*
605 	 * Try to defer the submission to a workqueue to parallelize the
606 	 * checksum calculation unless the I/O is issued synchronously.
607 	 */
608 	if (op_is_sync(bbio->bio.bi_opf))
609 		return false;
610 
611 	/* Zoned devices require I/O to be submitted in order. */
612 	if ((bbio->bio.bi_opf & REQ_META) && btrfs_is_zoned(bbio->fs_info))
613 		return false;
614 
615 	return true;
616 }
617 
618 /*
619  * Submit bio to an async queue.
620  *
621  * Return true if the work has been succesfuly submitted, else false.
622  */
btrfs_wq_submit_bio(struct btrfs_bio * bbio,struct btrfs_io_context * bioc,struct btrfs_io_stripe * smap,int mirror_num)623 static bool btrfs_wq_submit_bio(struct btrfs_bio *bbio,
624 				struct btrfs_io_context *bioc,
625 				struct btrfs_io_stripe *smap, int mirror_num)
626 {
627 	struct btrfs_fs_info *fs_info = bbio->fs_info;
628 	struct async_submit_bio *async;
629 
630 	async = kmalloc(sizeof(*async), GFP_NOFS);
631 	if (!async)
632 		return false;
633 
634 	async->bbio = bbio;
635 	async->bioc = bioc;
636 	async->smap = *smap;
637 	async->mirror_num = mirror_num;
638 
639 	btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
640 			run_one_async_free);
641 	btrfs_queue_work(fs_info->workers, &async->work);
642 	return true;
643 }
644 
btrfs_submit_chunk(struct btrfs_bio * bbio,int mirror_num)645 static bool btrfs_submit_chunk(struct btrfs_bio *bbio, int mirror_num)
646 {
647 	struct btrfs_inode *inode = bbio->inode;
648 	struct btrfs_fs_info *fs_info = bbio->fs_info;
649 	struct bio *bio = &bbio->bio;
650 	u64 logical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
651 	u64 length = bio->bi_iter.bi_size;
652 	u64 map_length = length;
653 	bool use_append = btrfs_use_zone_append(bbio);
654 	struct btrfs_io_context *bioc = NULL;
655 	struct btrfs_io_stripe smap;
656 	blk_status_t ret;
657 	int error;
658 
659 	btrfs_bio_counter_inc_blocked(fs_info);
660 	error = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
661 				&bioc, &smap, &mirror_num, 1);
662 	if (error) {
663 		ret = errno_to_blk_status(error);
664 		goto fail;
665 	}
666 
667 	map_length = min(map_length, length);
668 	if (use_append)
669 		map_length = min(map_length, fs_info->max_zone_append_size);
670 
671 	if (map_length < length) {
672 		bbio = btrfs_split_bio(fs_info, bbio, map_length, use_append);
673 		bio = &bbio->bio;
674 	}
675 
676 	/*
677 	 * Save the iter for the end_io handler and preload the checksums for
678 	 * data reads.
679 	 */
680 	if (bio_op(bio) == REQ_OP_READ && is_data_bbio(bbio)) {
681 		bbio->saved_iter = bio->bi_iter;
682 		ret = btrfs_lookup_bio_sums(bbio);
683 		if (ret)
684 			goto fail;
685 	}
686 
687 	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
688 		if (use_append) {
689 			bio->bi_opf &= ~REQ_OP_WRITE;
690 			bio->bi_opf |= REQ_OP_ZONE_APPEND;
691 		}
692 
693 		/*
694 		 * Csum items for reloc roots have already been cloned at this
695 		 * point, so they are handled as part of the no-checksum case.
696 		 */
697 		if (inode && !(inode->flags & BTRFS_INODE_NODATASUM) &&
698 		    !test_bit(BTRFS_FS_STATE_NO_CSUMS, &fs_info->fs_state) &&
699 		    !btrfs_is_data_reloc_root(inode->root)) {
700 			if (should_async_write(bbio) &&
701 			    btrfs_wq_submit_bio(bbio, bioc, &smap, mirror_num))
702 				goto done;
703 
704 			ret = btrfs_bio_csum(bbio);
705 			if (ret)
706 				goto fail;
707 		} else if (use_append ||
708 			   (btrfs_is_zoned(fs_info) && inode &&
709 			    inode->flags & BTRFS_INODE_NODATASUM)) {
710 			ret = btrfs_alloc_dummy_sum(bbio);
711 			if (ret)
712 				goto fail;
713 		}
714 	}
715 
716 	__btrfs_submit_bio(bio, bioc, &smap, mirror_num);
717 done:
718 	return map_length == length;
719 
720 fail:
721 	btrfs_bio_counter_dec(fs_info);
722 	/*
723 	 * We have split the original bbio, now we have to end both the current
724 	 * @bbio and remaining one, as the remaining one will never be submitted.
725 	 */
726 	if (map_length < length) {
727 		struct btrfs_bio *remaining = bbio->private;
728 
729 		ASSERT(bbio->bio.bi_pool == &btrfs_clone_bioset);
730 		ASSERT(remaining);
731 
732 		remaining->bio.bi_status = ret;
733 		btrfs_orig_bbio_end_io(remaining);
734 	}
735 	bbio->bio.bi_status = ret;
736 	btrfs_orig_bbio_end_io(bbio);
737 	/* Do not submit another chunk */
738 	return true;
739 }
740 
btrfs_submit_bio(struct btrfs_bio * bbio,int mirror_num)741 void btrfs_submit_bio(struct btrfs_bio *bbio, int mirror_num)
742 {
743 	/* If bbio->inode is not populated, its file_offset must be 0. */
744 	ASSERT(bbio->inode || bbio->file_offset == 0);
745 
746 	while (!btrfs_submit_chunk(bbio, mirror_num))
747 		;
748 }
749 
750 /*
751  * Submit a repair write.
752  *
753  * This bypasses btrfs_submit_bio deliberately, as that writes all copies in a
754  * RAID setup.  Here we only want to write the one bad copy, so we do the
755  * mapping ourselves and submit the bio directly.
756  *
757  * The I/O is issued synchronously to block the repair read completion from
758  * freeing the bio.
759  */
btrfs_repair_io_failure(struct btrfs_fs_info * fs_info,u64 ino,u64 start,u64 length,u64 logical,struct page * page,unsigned int pg_offset,int mirror_num)760 int btrfs_repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
761 			    u64 length, u64 logical, struct page *page,
762 			    unsigned int pg_offset, int mirror_num)
763 {
764 	struct btrfs_io_stripe smap = { 0 };
765 	struct bio_vec bvec;
766 	struct bio bio;
767 	int ret = 0;
768 
769 	ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
770 	BUG_ON(!mirror_num);
771 
772 	if (btrfs_repair_one_zone(fs_info, logical))
773 		return 0;
774 
775 	/*
776 	 * Avoid races with device replace and make sure our bioc has devices
777 	 * associated to its stripes that don't go away while we are doing the
778 	 * read repair operation.
779 	 */
780 	btrfs_bio_counter_inc_blocked(fs_info);
781 	ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
782 	if (ret < 0)
783 		goto out_counter_dec;
784 
785 	if (!smap.dev->bdev ||
786 	    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &smap.dev->dev_state)) {
787 		ret = -EIO;
788 		goto out_counter_dec;
789 	}
790 
791 	bio_init(&bio, smap.dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
792 	bio.bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT;
793 	__bio_add_page(&bio, page, length, pg_offset);
794 
795 	btrfsic_check_bio(&bio);
796 	ret = submit_bio_wait(&bio);
797 	if (ret) {
798 		/* try to remap that extent elsewhere? */
799 		btrfs_dev_stat_inc_and_print(smap.dev, BTRFS_DEV_STAT_WRITE_ERRS);
800 		goto out_bio_uninit;
801 	}
802 
803 	btrfs_info_rl_in_rcu(fs_info,
804 		"read error corrected: ino %llu off %llu (dev %s sector %llu)",
805 			     ino, start, btrfs_dev_name(smap.dev),
806 			     smap.physical >> SECTOR_SHIFT);
807 	ret = 0;
808 
809 out_bio_uninit:
810 	bio_uninit(&bio);
811 out_counter_dec:
812 	btrfs_bio_counter_dec(fs_info);
813 	return ret;
814 }
815 
816 /*
817  * Submit a btrfs_bio based repair write.
818  *
819  * If @dev_replace is true, the write would be submitted to dev-replace target.
820  */
btrfs_submit_repair_write(struct btrfs_bio * bbio,int mirror_num,bool dev_replace)821 void btrfs_submit_repair_write(struct btrfs_bio *bbio, int mirror_num, bool dev_replace)
822 {
823 	struct btrfs_fs_info *fs_info = bbio->fs_info;
824 	u64 logical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
825 	u64 length = bbio->bio.bi_iter.bi_size;
826 	struct btrfs_io_stripe smap = { 0 };
827 	int ret;
828 
829 	ASSERT(fs_info);
830 	ASSERT(mirror_num > 0);
831 	ASSERT(btrfs_op(&bbio->bio) == BTRFS_MAP_WRITE);
832 	ASSERT(!bbio->inode);
833 
834 	btrfs_bio_counter_inc_blocked(fs_info);
835 	ret = btrfs_map_repair_block(fs_info, &smap, logical, length, mirror_num);
836 	if (ret < 0)
837 		goto fail;
838 
839 	if (dev_replace) {
840 		ASSERT(smap.dev == fs_info->dev_replace.srcdev);
841 		smap.dev = fs_info->dev_replace.tgtdev;
842 	}
843 	__btrfs_submit_bio(&bbio->bio, NULL, &smap, mirror_num);
844 	return;
845 
846 fail:
847 	btrfs_bio_counter_dec(fs_info);
848 	btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
849 }
850 
btrfs_bioset_init(void)851 int __init btrfs_bioset_init(void)
852 {
853 	if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
854 			offsetof(struct btrfs_bio, bio),
855 			BIOSET_NEED_BVECS))
856 		return -ENOMEM;
857 	if (bioset_init(&btrfs_clone_bioset, BIO_POOL_SIZE,
858 			offsetof(struct btrfs_bio, bio), 0))
859 		goto out_free_bioset;
860 	if (bioset_init(&btrfs_repair_bioset, BIO_POOL_SIZE,
861 			offsetof(struct btrfs_bio, bio),
862 			BIOSET_NEED_BVECS))
863 		goto out_free_clone_bioset;
864 	if (mempool_init_kmalloc_pool(&btrfs_failed_bio_pool, BIO_POOL_SIZE,
865 				      sizeof(struct btrfs_failed_bio)))
866 		goto out_free_repair_bioset;
867 	return 0;
868 
869 out_free_repair_bioset:
870 	bioset_exit(&btrfs_repair_bioset);
871 out_free_clone_bioset:
872 	bioset_exit(&btrfs_clone_bioset);
873 out_free_bioset:
874 	bioset_exit(&btrfs_bioset);
875 	return -ENOMEM;
876 }
877 
btrfs_bioset_exit(void)878 void __cold btrfs_bioset_exit(void)
879 {
880 	mempool_exit(&btrfs_failed_bio_pool);
881 	bioset_exit(&btrfs_repair_bioset);
882 	bioset_exit(&btrfs_clone_bioset);
883 	bioset_exit(&btrfs_bioset);
884 }
885