xref: /linux/fs/btrfs/bio.c (revision 811f35ff59b6f99ae272d6f5b96bc9e974f88196)
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 
18 static struct bio_set btrfs_bioset;
19 
20 /*
21  * Initialize a btrfs_bio structure.  This skips the embedded bio itself as it
22  * is already initialized by the block layer.
23  */
24 static inline void btrfs_bio_init(struct btrfs_bio *bbio,
25 				  btrfs_bio_end_io_t end_io, void *private)
26 {
27 	memset(bbio, 0, offsetof(struct btrfs_bio, bio));
28 	bbio->end_io = end_io;
29 	bbio->private = private;
30 }
31 
32 /*
33  * Allocate a btrfs_bio structure.  The btrfs_bio is the main I/O container for
34  * btrfs, and is used for all I/O submitted through btrfs_submit_bio.
35  *
36  * Just like the underlying bio_alloc_bioset it will not fail as it is backed by
37  * a mempool.
38  */
39 struct bio *btrfs_bio_alloc(unsigned int nr_vecs, blk_opf_t opf,
40 			    btrfs_bio_end_io_t end_io, void *private)
41 {
42 	struct bio *bio;
43 
44 	bio = bio_alloc_bioset(NULL, nr_vecs, opf, GFP_NOFS, &btrfs_bioset);
45 	btrfs_bio_init(btrfs_bio(bio), end_io, private);
46 	return bio;
47 }
48 
49 struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size,
50 				    btrfs_bio_end_io_t end_io, void *private)
51 {
52 	struct bio *bio;
53 	struct btrfs_bio *bbio;
54 
55 	ASSERT(offset <= UINT_MAX && size <= UINT_MAX);
56 
57 	bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset);
58 	bbio = btrfs_bio(bio);
59 	btrfs_bio_init(bbio, end_io, private);
60 
61 	bio_trim(bio, offset >> 9, size >> 9);
62 	bbio->iter = bio->bi_iter;
63 	return bio;
64 }
65 
66 static void btrfs_log_dev_io_error(struct bio *bio, struct btrfs_device *dev)
67 {
68 	if (!dev || !dev->bdev)
69 		return;
70 	if (bio->bi_status != BLK_STS_IOERR && bio->bi_status != BLK_STS_TARGET)
71 		return;
72 
73 	if (btrfs_op(bio) == BTRFS_MAP_WRITE)
74 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
75 	if (!(bio->bi_opf & REQ_RAHEAD))
76 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
77 	if (bio->bi_opf & REQ_PREFLUSH)
78 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_FLUSH_ERRS);
79 }
80 
81 static struct workqueue_struct *btrfs_end_io_wq(struct btrfs_fs_info *fs_info,
82 						struct bio *bio)
83 {
84 	if (bio->bi_opf & REQ_META)
85 		return fs_info->endio_meta_workers;
86 	return fs_info->endio_workers;
87 }
88 
89 static void btrfs_end_bio_work(struct work_struct *work)
90 {
91 	struct btrfs_bio *bbio = container_of(work, struct btrfs_bio, end_io_work);
92 
93 	bbio->end_io(bbio);
94 }
95 
96 static void btrfs_simple_end_io(struct bio *bio)
97 {
98 	struct btrfs_fs_info *fs_info = bio->bi_private;
99 	struct btrfs_bio *bbio = btrfs_bio(bio);
100 
101 	btrfs_bio_counter_dec(fs_info);
102 
103 	if (bio->bi_status)
104 		btrfs_log_dev_io_error(bio, bbio->device);
105 
106 	if (bio_op(bio) == REQ_OP_READ) {
107 		INIT_WORK(&bbio->end_io_work, btrfs_end_bio_work);
108 		queue_work(btrfs_end_io_wq(fs_info, bio), &bbio->end_io_work);
109 	} else {
110 		bbio->end_io(bbio);
111 	}
112 }
113 
114 static void btrfs_raid56_end_io(struct bio *bio)
115 {
116 	struct btrfs_io_context *bioc = bio->bi_private;
117 	struct btrfs_bio *bbio = btrfs_bio(bio);
118 
119 	btrfs_bio_counter_dec(bioc->fs_info);
120 	bbio->mirror_num = bioc->mirror_num;
121 	bbio->end_io(bbio);
122 
123 	btrfs_put_bioc(bioc);
124 }
125 
126 static void btrfs_orig_write_end_io(struct bio *bio)
127 {
128 	struct btrfs_io_stripe *stripe = bio->bi_private;
129 	struct btrfs_io_context *bioc = stripe->bioc;
130 	struct btrfs_bio *bbio = btrfs_bio(bio);
131 
132 	btrfs_bio_counter_dec(bioc->fs_info);
133 
134 	if (bio->bi_status) {
135 		atomic_inc(&bioc->error);
136 		btrfs_log_dev_io_error(bio, stripe->dev);
137 	}
138 
139 	/*
140 	 * Only send an error to the higher layers if it is beyond the tolerance
141 	 * threshold.
142 	 */
143 	if (atomic_read(&bioc->error) > bioc->max_errors)
144 		bio->bi_status = BLK_STS_IOERR;
145 	else
146 		bio->bi_status = BLK_STS_OK;
147 
148 	bbio->end_io(bbio);
149 	btrfs_put_bioc(bioc);
150 }
151 
152 static void btrfs_clone_write_end_io(struct bio *bio)
153 {
154 	struct btrfs_io_stripe *stripe = bio->bi_private;
155 
156 	if (bio->bi_status) {
157 		atomic_inc(&stripe->bioc->error);
158 		btrfs_log_dev_io_error(bio, stripe->dev);
159 	}
160 
161 	/* Pass on control to the original bio this one was cloned from */
162 	bio_endio(stripe->bioc->orig_bio);
163 	bio_put(bio);
164 }
165 
166 static void btrfs_submit_dev_bio(struct btrfs_device *dev, struct bio *bio)
167 {
168 	if (!dev || !dev->bdev ||
169 	    test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
170 	    (btrfs_op(bio) == BTRFS_MAP_WRITE &&
171 	     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
172 		bio_io_error(bio);
173 		return;
174 	}
175 
176 	bio_set_dev(bio, dev->bdev);
177 
178 	/*
179 	 * For zone append writing, bi_sector must point the beginning of the
180 	 * zone
181 	 */
182 	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
183 		u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
184 
185 		if (btrfs_dev_is_sequential(dev, physical)) {
186 			u64 zone_start = round_down(physical,
187 						    dev->fs_info->zone_size);
188 
189 			bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
190 		} else {
191 			bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
192 			bio->bi_opf |= REQ_OP_WRITE;
193 		}
194 	}
195 	btrfs_debug_in_rcu(dev->fs_info,
196 	"%s: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
197 		__func__, bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
198 		(unsigned long)dev->bdev->bd_dev, btrfs_dev_name(dev),
199 		dev->devid, bio->bi_iter.bi_size);
200 
201 	btrfsic_check_bio(bio);
202 	submit_bio(bio);
203 }
204 
205 static void btrfs_submit_mirrored_bio(struct btrfs_io_context *bioc, int dev_nr)
206 {
207 	struct bio *orig_bio = bioc->orig_bio, *bio;
208 
209 	ASSERT(bio_op(orig_bio) != REQ_OP_READ);
210 
211 	/* Reuse the bio embedded into the btrfs_bio for the last mirror */
212 	if (dev_nr == bioc->num_stripes - 1) {
213 		bio = orig_bio;
214 		bio->bi_end_io = btrfs_orig_write_end_io;
215 	} else {
216 		bio = bio_alloc_clone(NULL, orig_bio, GFP_NOFS, &fs_bio_set);
217 		bio_inc_remaining(orig_bio);
218 		bio->bi_end_io = btrfs_clone_write_end_io;
219 	}
220 
221 	bio->bi_private = &bioc->stripes[dev_nr];
222 	bio->bi_iter.bi_sector = bioc->stripes[dev_nr].physical >> SECTOR_SHIFT;
223 	bioc->stripes[dev_nr].bioc = bioc;
224 	btrfs_submit_dev_bio(bioc->stripes[dev_nr].dev, bio);
225 }
226 
227 void btrfs_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio, int mirror_num)
228 {
229 	u64 logical = bio->bi_iter.bi_sector << 9;
230 	u64 length = bio->bi_iter.bi_size;
231 	u64 map_length = length;
232 	struct btrfs_io_context *bioc = NULL;
233 	struct btrfs_io_stripe smap;
234 	int ret;
235 
236 	btrfs_bio_counter_inc_blocked(fs_info);
237 	ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
238 				&bioc, &smap, &mirror_num, 1);
239 	if (ret) {
240 		btrfs_bio_counter_dec(fs_info);
241 		btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret));
242 		return;
243 	}
244 
245 	if (map_length < length) {
246 		btrfs_crit(fs_info,
247 			   "mapping failed logical %llu bio len %llu len %llu",
248 			   logical, length, map_length);
249 		BUG();
250 	}
251 
252 	if (!bioc) {
253 		/* Single mirror read/write fast path */
254 		btrfs_bio(bio)->mirror_num = mirror_num;
255 		btrfs_bio(bio)->device = smap.dev;
256 		bio->bi_iter.bi_sector = smap.physical >> SECTOR_SHIFT;
257 		bio->bi_private = fs_info;
258 		bio->bi_end_io = btrfs_simple_end_io;
259 		btrfs_submit_dev_bio(smap.dev, bio);
260 	} else if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
261 		/* Parity RAID write or read recovery */
262 		bio->bi_private = bioc;
263 		bio->bi_end_io = btrfs_raid56_end_io;
264 		if (bio_op(bio) == REQ_OP_READ)
265 			raid56_parity_recover(bio, bioc, mirror_num);
266 		else
267 			raid56_parity_write(bio, bioc);
268 	} else {
269 		/* Write to multiple mirrors */
270 		int total_devs = bioc->num_stripes;
271 		int dev_nr;
272 
273 		bioc->orig_bio = bio;
274 		for (dev_nr = 0; dev_nr < total_devs; dev_nr++)
275 			btrfs_submit_mirrored_bio(bioc, dev_nr);
276 	}
277 }
278 
279 /*
280  * Submit a repair write.
281  *
282  * This bypasses btrfs_submit_bio deliberately, as that writes all copies in a
283  * RAID setup.  Here we only want to write the one bad copy, so we do the
284  * mapping ourselves and submit the bio directly.
285  *
286  * The I/O is issued sychronously to block the repair read completion from
287  * freeing the bio.
288  */
289 int btrfs_repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
290 			    u64 length, u64 logical, struct page *page,
291 			    unsigned int pg_offset, int mirror_num)
292 {
293 	struct btrfs_device *dev;
294 	struct bio_vec bvec;
295 	struct bio bio;
296 	u64 map_length = 0;
297 	u64 sector;
298 	struct btrfs_io_context *bioc = NULL;
299 	int ret = 0;
300 
301 	ASSERT(!(fs_info->sb->s_flags & SB_RDONLY));
302 	BUG_ON(!mirror_num);
303 
304 	if (btrfs_repair_one_zone(fs_info, logical))
305 		return 0;
306 
307 	map_length = length;
308 
309 	/*
310 	 * Avoid races with device replace and make sure our bioc has devices
311 	 * associated to its stripes that don't go away while we are doing the
312 	 * read repair operation.
313 	 */
314 	btrfs_bio_counter_inc_blocked(fs_info);
315 	if (btrfs_is_parity_mirror(fs_info, logical, length)) {
316 		/*
317 		 * Note that we don't use BTRFS_MAP_WRITE because it's supposed
318 		 * to update all raid stripes, but here we just want to correct
319 		 * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad
320 		 * stripe's dev and sector.
321 		 */
322 		ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical,
323 				      &map_length, &bioc, 0);
324 		if (ret)
325 			goto out_counter_dec;
326 		ASSERT(bioc->mirror_num == 1);
327 	} else {
328 		ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
329 				      &map_length, &bioc, mirror_num);
330 		if (ret)
331 			goto out_counter_dec;
332 		/*
333 		 * This happens when dev-replace is also running, and the
334 		 * mirror_num indicates the dev-replace target.
335 		 *
336 		 * In this case, we don't need to do anything, as the read
337 		 * error just means the replace progress hasn't reached our
338 		 * read range, and later replace routine would handle it well.
339 		 */
340 		if (mirror_num != bioc->mirror_num)
341 			goto out_counter_dec;
342 	}
343 
344 	sector = bioc->stripes[bioc->mirror_num - 1].physical >> 9;
345 	dev = bioc->stripes[bioc->mirror_num - 1].dev;
346 	btrfs_put_bioc(bioc);
347 
348 	if (!dev || !dev->bdev ||
349 	    !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) {
350 		ret = -EIO;
351 		goto out_counter_dec;
352 	}
353 
354 	bio_init(&bio, dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC);
355 	bio.bi_iter.bi_sector = sector;
356 	__bio_add_page(&bio, page, length, pg_offset);
357 
358 	btrfsic_check_bio(&bio);
359 	ret = submit_bio_wait(&bio);
360 	if (ret) {
361 		/* try to remap that extent elsewhere? */
362 		btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
363 		goto out_bio_uninit;
364 	}
365 
366 	btrfs_info_rl_in_rcu(fs_info,
367 		"read error corrected: ino %llu off %llu (dev %s sector %llu)",
368 			     ino, start, btrfs_dev_name(dev), sector);
369 	ret = 0;
370 
371 out_bio_uninit:
372 	bio_uninit(&bio);
373 out_counter_dec:
374 	btrfs_bio_counter_dec(fs_info);
375 	return ret;
376 }
377 
378 int __init btrfs_bioset_init(void)
379 {
380 	if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE,
381 			offsetof(struct btrfs_bio, bio),
382 			BIOSET_NEED_BVECS))
383 		return -ENOMEM;
384 	return 0;
385 }
386 
387 void __cold btrfs_bioset_exit(void)
388 {
389 	bioset_exit(&btrfs_bioset);
390 }
391