blk-mq: fix two misuses on RQF_USE_SCHED
[linux-block.git] / block / bio.c
CommitLineData
8c16567d 1// SPDX-License-Identifier: GPL-2.0
1da177e4 2/*
0fe23479 3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
1da177e4
LT
4 */
5#include <linux/mm.h>
6#include <linux/swap.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
a27bb332 9#include <linux/uio.h>
852c788f 10#include <linux/iocontext.h>
1da177e4
LT
11#include <linux/slab.h>
12#include <linux/init.h>
13#include <linux/kernel.h>
630d9c47 14#include <linux/export.h>
1da177e4
LT
15#include <linux/mempool.h>
16#include <linux/workqueue.h>
852c788f 17#include <linux/cgroup.h>
b4c5875d 18#include <linux/highmem.h>
de6a78b6 19#include <linux/sched/sysctl.h>
a892c8d5 20#include <linux/blk-crypto.h>
49d1ec85 21#include <linux/xarray.h>
1da177e4 22
55782138 23#include <trace/events/block.h>
9e234eea 24#include "blk.h"
67b42d0b 25#include "blk-rq-qos.h"
672fdcf0 26#include "blk-cgroup.h"
0bfc2455 27
b99182c5 28#define ALLOC_CACHE_THRESHOLD 16
42b2b2fb 29#define ALLOC_CACHE_MAX 256
b99182c5 30
be4d234d 31struct bio_alloc_cache {
fcade2ce 32 struct bio *free_list;
b99182c5 33 struct bio *free_list_irq;
be4d234d 34 unsigned int nr;
b99182c5 35 unsigned int nr_irq;
be4d234d
JA
36};
37
de76fd89 38static struct biovec_slab {
6ac0b715
CH
39 int nr_vecs;
40 char *name;
41 struct kmem_cache *slab;
de76fd89
CH
42} bvec_slabs[] __read_mostly = {
43 { .nr_vecs = 16, .name = "biovec-16" },
44 { .nr_vecs = 64, .name = "biovec-64" },
45 { .nr_vecs = 128, .name = "biovec-128" },
a8affc03 46 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
1da177e4 47};
6ac0b715 48
7a800a20
CH
49static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
50{
51 switch (nr_vecs) {
52 /* smaller bios use inline vecs */
53 case 5 ... 16:
54 return &bvec_slabs[0];
55 case 17 ... 64:
56 return &bvec_slabs[1];
57 case 65 ... 128:
58 return &bvec_slabs[2];
a8affc03 59 case 129 ... BIO_MAX_VECS:
7a800a20
CH
60 return &bvec_slabs[3];
61 default:
62 BUG();
63 return NULL;
64 }
65}
1da177e4 66
1da177e4
LT
67/*
68 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
69 * IO code that does not need private memory pools.
70 */
f4f8154a 71struct bio_set fs_bio_set;
3f86a82a 72EXPORT_SYMBOL(fs_bio_set);
1da177e4 73
bb799ca0
JA
74/*
75 * Our slab pool management
76 */
77struct bio_slab {
78 struct kmem_cache *slab;
79 unsigned int slab_ref;
80 unsigned int slab_size;
81 char name[8];
82};
83static DEFINE_MUTEX(bio_slab_lock);
49d1ec85 84static DEFINE_XARRAY(bio_slabs);
bb799ca0 85
49d1ec85 86static struct bio_slab *create_bio_slab(unsigned int size)
bb799ca0 87{
49d1ec85 88 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
bb799ca0 89
49d1ec85
ML
90 if (!bslab)
91 return NULL;
bb799ca0 92
49d1ec85
ML
93 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
94 bslab->slab = kmem_cache_create(bslab->name, size,
1a7e76e4
CH
95 ARCH_KMALLOC_MINALIGN,
96 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
49d1ec85
ML
97 if (!bslab->slab)
98 goto fail_alloc_slab;
bb799ca0 99
49d1ec85
ML
100 bslab->slab_ref = 1;
101 bslab->slab_size = size;
bb799ca0 102
49d1ec85
ML
103 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
104 return bslab;
bb799ca0 105
49d1ec85 106 kmem_cache_destroy(bslab->slab);
bb799ca0 107
49d1ec85
ML
108fail_alloc_slab:
109 kfree(bslab);
110 return NULL;
111}
bb799ca0 112
49d1ec85
ML
113static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
114{
9f180e31 115 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
49d1ec85 116}
bb799ca0 117
49d1ec85
ML
118static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
119{
120 unsigned int size = bs_bio_slab_size(bs);
121 struct bio_slab *bslab;
bb799ca0 122
49d1ec85
ML
123 mutex_lock(&bio_slab_lock);
124 bslab = xa_load(&bio_slabs, size);
125 if (bslab)
126 bslab->slab_ref++;
127 else
128 bslab = create_bio_slab(size);
bb799ca0 129 mutex_unlock(&bio_slab_lock);
49d1ec85
ML
130
131 if (bslab)
132 return bslab->slab;
133 return NULL;
bb799ca0
JA
134}
135
136static void bio_put_slab(struct bio_set *bs)
137{
138 struct bio_slab *bslab = NULL;
49d1ec85 139 unsigned int slab_size = bs_bio_slab_size(bs);
bb799ca0
JA
140
141 mutex_lock(&bio_slab_lock);
142
49d1ec85 143 bslab = xa_load(&bio_slabs, slab_size);
bb799ca0
JA
144 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 goto out;
146
49d1ec85
ML
147 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
148
bb799ca0
JA
149 WARN_ON(!bslab->slab_ref);
150
151 if (--bslab->slab_ref)
152 goto out;
153
49d1ec85
ML
154 xa_erase(&bio_slabs, slab_size);
155
bb799ca0 156 kmem_cache_destroy(bslab->slab);
49d1ec85 157 kfree(bslab);
bb799ca0
JA
158
159out:
160 mutex_unlock(&bio_slab_lock);
161}
162
7a800a20 163void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
7ba1ba12 164{
9e8c0d0d 165 BUG_ON(nr_vecs > BIO_MAX_VECS);
ed996a52 166
a8affc03 167 if (nr_vecs == BIO_MAX_VECS)
9f060e22 168 mempool_free(bv, pool);
7a800a20
CH
169 else if (nr_vecs > BIO_INLINE_VECS)
170 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
bb799ca0 171}
bb799ca0 172
f2c3eb9b
CH
173/*
174 * Make the first allocation restricted and don't dump info on allocation
175 * failures, since we'll fall back to the mempool in case of failure.
176 */
177static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
178{
179 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
180 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
bb799ca0
JA
181}
182
7a800a20
CH
183struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
184 gfp_t gfp_mask)
1da177e4 185{
7a800a20 186 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
1da177e4 187
7a800a20 188 if (WARN_ON_ONCE(!bvs))
7ff9345f 189 return NULL;
7ff9345f
JA
190
191 /*
7a800a20
CH
192 * Upgrade the nr_vecs request to take full advantage of the allocation.
193 * We also rely on this in the bvec_free path.
7ff9345f 194 */
7a800a20 195 *nr_vecs = bvs->nr_vecs;
7ff9345f 196
7ff9345f 197 /*
f007a3d6
CH
198 * Try a slab allocation first for all smaller allocations. If that
199 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
a8affc03 200 * The mempool is sized to handle up to BIO_MAX_VECS entries.
7ff9345f 201 */
a8affc03 202 if (*nr_vecs < BIO_MAX_VECS) {
f007a3d6 203 struct bio_vec *bvl;
1da177e4 204
f2c3eb9b 205 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
7a800a20 206 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
f007a3d6 207 return bvl;
a8affc03 208 *nr_vecs = BIO_MAX_VECS;
7ff9345f
JA
209 }
210
f007a3d6 211 return mempool_alloc(pool, gfp_mask);
1da177e4
LT
212}
213
9ae3b3f5 214void bio_uninit(struct bio *bio)
1da177e4 215{
db9819c7
CH
216#ifdef CONFIG_BLK_CGROUP
217 if (bio->bi_blkg) {
218 blkg_put(bio->bi_blkg);
219 bio->bi_blkg = NULL;
220 }
221#endif
ece841ab
JT
222 if (bio_integrity(bio))
223 bio_integrity_free(bio);
a892c8d5
ST
224
225 bio_crypt_free_ctx(bio);
4254bba1 226}
9ae3b3f5 227EXPORT_SYMBOL(bio_uninit);
7ba1ba12 228
4254bba1
KO
229static void bio_free(struct bio *bio)
230{
231 struct bio_set *bs = bio->bi_pool;
066ff571 232 void *p = bio;
4254bba1 233
066ff571 234 WARN_ON_ONCE(!bs);
4254bba1 235
066ff571
CH
236 bio_uninit(bio);
237 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
238 mempool_free(p - bs->front_pad, &bs->bio_pool);
3676347a
PO
239}
240
9ae3b3f5
JA
241/*
242 * Users of this function have their own bio allocation. Subsequently,
243 * they must remember to pair any call to bio_init() with bio_uninit()
244 * when IO has completed, or when the bio is released.
245 */
49add496 246void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
16458cf3 247 unsigned short max_vecs, blk_opf_t opf)
1da177e4 248{
da521626 249 bio->bi_next = NULL;
49add496
CH
250 bio->bi_bdev = bdev;
251 bio->bi_opf = opf;
da521626
JA
252 bio->bi_flags = 0;
253 bio->bi_ioprio = 0;
da521626
JA
254 bio->bi_status = 0;
255 bio->bi_iter.bi_sector = 0;
256 bio->bi_iter.bi_size = 0;
257 bio->bi_iter.bi_idx = 0;
258 bio->bi_iter.bi_bvec_done = 0;
259 bio->bi_end_io = NULL;
260 bio->bi_private = NULL;
261#ifdef CONFIG_BLK_CGROUP
262 bio->bi_blkg = NULL;
263 bio->bi_issue.value = 0;
49add496
CH
264 if (bdev)
265 bio_associate_blkg(bio);
da521626
JA
266#ifdef CONFIG_BLK_CGROUP_IOCOST
267 bio->bi_iocost_cost = 0;
268#endif
269#endif
270#ifdef CONFIG_BLK_INLINE_ENCRYPTION
271 bio->bi_crypt_context = NULL;
272#endif
273#ifdef CONFIG_BLK_DEV_INTEGRITY
274 bio->bi_integrity = NULL;
275#endif
276 bio->bi_vcnt = 0;
277
c4cf5261 278 atomic_set(&bio->__bi_remaining, 1);
dac56212 279 atomic_set(&bio->__bi_cnt, 1);
3e08773c 280 bio->bi_cookie = BLK_QC_T_NONE;
3a83f467 281
3a83f467 282 bio->bi_max_vecs = max_vecs;
da521626
JA
283 bio->bi_io_vec = table;
284 bio->bi_pool = NULL;
1da177e4 285}
a112a71d 286EXPORT_SYMBOL(bio_init);
1da177e4 287
f44b48c7
KO
288/**
289 * bio_reset - reinitialize a bio
290 * @bio: bio to reset
a7c50c94
CH
291 * @bdev: block device to use the bio for
292 * @opf: operation and flags for bio
f44b48c7
KO
293 *
294 * Description:
295 * After calling bio_reset(), @bio will be in the same state as a freshly
296 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
297 * preserved are the ones that are initialized by bio_alloc_bioset(). See
298 * comment in struct bio.
299 */
16458cf3 300void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
f44b48c7 301{
9ae3b3f5 302 bio_uninit(bio);
f44b48c7 303 memset(bio, 0, BIO_RESET_BYTES);
c4cf5261 304 atomic_set(&bio->__bi_remaining, 1);
a7c50c94 305 bio->bi_bdev = bdev;
78e34374
CH
306 if (bio->bi_bdev)
307 bio_associate_blkg(bio);
a7c50c94 308 bio->bi_opf = opf;
f44b48c7
KO
309}
310EXPORT_SYMBOL(bio_reset);
311
38f8baae 312static struct bio *__bio_chain_endio(struct bio *bio)
196d38bc 313{
4246a0b6
CH
314 struct bio *parent = bio->bi_private;
315
3edf5346 316 if (bio->bi_status && !parent->bi_status)
4e4cbee9 317 parent->bi_status = bio->bi_status;
196d38bc 318 bio_put(bio);
38f8baae
CH
319 return parent;
320}
321
322static void bio_chain_endio(struct bio *bio)
323{
324 bio_endio(__bio_chain_endio(bio));
196d38bc
KO
325}
326
327/**
328 * bio_chain - chain bio completions
1051a902 329 * @bio: the target bio
5b874af6 330 * @parent: the parent bio of @bio
196d38bc
KO
331 *
332 * The caller won't have a bi_end_io called when @bio completes - instead,
333 * @parent's bi_end_io won't be called until both @parent and @bio have
334 * completed; the chained bio will also be freed when it completes.
335 *
336 * The caller must not set bi_private or bi_end_io in @bio.
337 */
338void bio_chain(struct bio *bio, struct bio *parent)
339{
340 BUG_ON(bio->bi_private || bio->bi_end_io);
341
342 bio->bi_private = parent;
343 bio->bi_end_io = bio_chain_endio;
c4cf5261 344 bio_inc_remaining(parent);
196d38bc
KO
345}
346EXPORT_SYMBOL(bio_chain);
347
0a3140ea 348struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
16458cf3 349 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
3b005bf6 350{
07888c66 351 struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
0a3140ea 352
3b005bf6
CH
353 if (bio) {
354 bio_chain(bio, new);
355 submit_bio(bio);
356 }
357
358 return new;
359}
360EXPORT_SYMBOL_GPL(blk_next_bio);
361
df2cb6da
KO
362static void bio_alloc_rescue(struct work_struct *work)
363{
364 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
365 struct bio *bio;
366
367 while (1) {
368 spin_lock(&bs->rescue_lock);
369 bio = bio_list_pop(&bs->rescue_list);
370 spin_unlock(&bs->rescue_lock);
371
372 if (!bio)
373 break;
374
ed00aabd 375 submit_bio_noacct(bio);
df2cb6da
KO
376 }
377}
378
379static void punt_bios_to_rescuer(struct bio_set *bs)
380{
381 struct bio_list punt, nopunt;
382 struct bio *bio;
383
47e0fb46
N
384 if (WARN_ON_ONCE(!bs->rescue_workqueue))
385 return;
df2cb6da
KO
386 /*
387 * In order to guarantee forward progress we must punt only bios that
388 * were allocated from this bio_set; otherwise, if there was a bio on
389 * there for a stacking driver higher up in the stack, processing it
390 * could require allocating bios from this bio_set, and doing that from
391 * our own rescuer would be bad.
392 *
393 * Since bio lists are singly linked, pop them all instead of trying to
394 * remove from the middle of the list:
395 */
396
397 bio_list_init(&punt);
398 bio_list_init(&nopunt);
399
f5fe1b51 400 while ((bio = bio_list_pop(&current->bio_list[0])))
df2cb6da 401 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
f5fe1b51 402 current->bio_list[0] = nopunt;
df2cb6da 403
f5fe1b51
N
404 bio_list_init(&nopunt);
405 while ((bio = bio_list_pop(&current->bio_list[1])))
406 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
407 current->bio_list[1] = nopunt;
df2cb6da
KO
408
409 spin_lock(&bs->rescue_lock);
410 bio_list_merge(&bs->rescue_list, &punt);
411 spin_unlock(&bs->rescue_lock);
412
413 queue_work(bs->rescue_workqueue, &bs->rescue_work);
414}
415
b99182c5
PB
416static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
417{
418 unsigned long flags;
419
420 /* cache->free_list must be empty */
421 if (WARN_ON_ONCE(cache->free_list))
422 return;
423
424 local_irq_save(flags);
425 cache->free_list = cache->free_list_irq;
426 cache->free_list_irq = NULL;
427 cache->nr += cache->nr_irq;
428 cache->nr_irq = 0;
429 local_irq_restore(flags);
430}
431
0df71650 432static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
16458cf3 433 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
0df71650
MS
434 struct bio_set *bs)
435{
436 struct bio_alloc_cache *cache;
437 struct bio *bio;
438
439 cache = per_cpu_ptr(bs->cache, get_cpu());
440 if (!cache->free_list) {
b99182c5
PB
441 if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
442 bio_alloc_irq_cache_splice(cache);
443 if (!cache->free_list) {
444 put_cpu();
445 return NULL;
446 }
0df71650
MS
447 }
448 bio = cache->free_list;
449 cache->free_list = bio->bi_next;
450 cache->nr--;
451 put_cpu();
452
453 bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
454 bio->bi_pool = bs;
455 return bio;
456}
457
1da177e4
LT
458/**
459 * bio_alloc_bioset - allocate a bio for I/O
609be106
CH
460 * @bdev: block device to allocate the bio for (can be %NULL)
461 * @nr_vecs: number of bvecs to pre-allocate
462 * @opf: operation and flags for bio
519c8e9f 463 * @gfp_mask: the GFP_* mask given to the slab allocator
db18efac 464 * @bs: the bio_set to allocate from.
1da177e4 465 *
3175199a 466 * Allocate a bio from the mempools in @bs.
3f86a82a 467 *
3175199a
CH
468 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
469 * allocate a bio. This is due to the mempool guarantees. To make this work,
470 * callers must never allocate more than 1 bio at a time from the general pool.
471 * Callers that need to allocate more than 1 bio must always submit the
472 * previously allocated bio for IO before attempting to allocate a new one.
473 * Failure to do so can cause deadlocks under memory pressure.
3f86a82a 474 *
3175199a
CH
475 * Note that when running under submit_bio_noacct() (i.e. any block driver),
476 * bios are not submitted until after you return - see the code in
477 * submit_bio_noacct() that converts recursion into iteration, to prevent
478 * stack overflows.
df2cb6da 479 *
3175199a
CH
480 * This would normally mean allocating multiple bios under submit_bio_noacct()
481 * would be susceptible to deadlocks, but we have
482 * deadlock avoidance code that resubmits any blocked bios from a rescuer
483 * thread.
df2cb6da 484 *
3175199a
CH
485 * However, we do not guarantee forward progress for allocations from other
486 * mempools. Doing multiple allocations from the same mempool under
487 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
488 * for per bio allocations.
df2cb6da 489 *
3175199a 490 * Returns: Pointer to new bio on success, NULL on failure.
3f86a82a 491 */
609be106 492struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
16458cf3 493 blk_opf_t opf, gfp_t gfp_mask,
7a88fa19 494 struct bio_set *bs)
1da177e4 495{
df2cb6da 496 gfp_t saved_gfp = gfp_mask;
451a9ebf
TH
497 struct bio *bio;
498 void *p;
499
609be106
CH
500 /* should not use nobvec bioset for nr_vecs > 0 */
501 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
3175199a 502 return NULL;
df2cb6da 503
0df71650
MS
504 if (opf & REQ_ALLOC_CACHE) {
505 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
506 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
507 gfp_mask, bs);
508 if (bio)
509 return bio;
510 /*
511 * No cached bio available, bio returned below marked with
512 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
513 */
514 } else {
515 opf &= ~REQ_ALLOC_CACHE;
516 }
517 }
518
3175199a
CH
519 /*
520 * submit_bio_noacct() converts recursion to iteration; this means if
521 * we're running beneath it, any bios we allocate and submit will not be
522 * submitted (and thus freed) until after we return.
523 *
524 * This exposes us to a potential deadlock if we allocate multiple bios
525 * from the same bio_set() while running underneath submit_bio_noacct().
526 * If we were to allocate multiple bios (say a stacking block driver
527 * that was splitting bios), we would deadlock if we exhausted the
528 * mempool's reserve.
529 *
530 * We solve this, and guarantee forward progress, with a rescuer
531 * workqueue per bio_set. If we go to allocate and there are bios on
532 * current->bio_list, we first try the allocation without
533 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
534 * blocking to the rescuer workqueue before we retry with the original
535 * gfp_flags.
536 */
537 if (current->bio_list &&
538 (!bio_list_empty(&current->bio_list[0]) ||
539 !bio_list_empty(&current->bio_list[1])) &&
540 bs->rescue_workqueue)
541 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
542
543 p = mempool_alloc(&bs->bio_pool, gfp_mask);
544 if (!p && gfp_mask != saved_gfp) {
545 punt_bios_to_rescuer(bs);
546 gfp_mask = saved_gfp;
8aa6ba2f 547 p = mempool_alloc(&bs->bio_pool, gfp_mask);
3f86a82a 548 }
451a9ebf
TH
549 if (unlikely(!p))
550 return NULL;
759aa12f
PB
551 if (!mempool_is_saturated(&bs->bio_pool))
552 opf &= ~REQ_ALLOC_CACHE;
1da177e4 553
3175199a 554 bio = p + bs->front_pad;
609be106 555 if (nr_vecs > BIO_INLINE_VECS) {
3175199a 556 struct bio_vec *bvl = NULL;
34053979 557
609be106 558 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
df2cb6da
KO
559 if (!bvl && gfp_mask != saved_gfp) {
560 punt_bios_to_rescuer(bs);
561 gfp_mask = saved_gfp;
609be106 562 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
df2cb6da 563 }
34053979
IM
564 if (unlikely(!bvl))
565 goto err_free;
a38352e0 566
49add496 567 bio_init(bio, bdev, bvl, nr_vecs, opf);
609be106 568 } else if (nr_vecs) {
49add496 569 bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
3175199a 570 } else {
49add496 571 bio_init(bio, bdev, NULL, 0, opf);
1da177e4 572 }
3f86a82a
KO
573
574 bio->bi_pool = bs;
1da177e4 575 return bio;
34053979
IM
576
577err_free:
8aa6ba2f 578 mempool_free(p, &bs->bio_pool);
34053979 579 return NULL;
1da177e4 580}
a112a71d 581EXPORT_SYMBOL(bio_alloc_bioset);
1da177e4 582
3175199a 583/**
066ff571
CH
584 * bio_kmalloc - kmalloc a bio
585 * @nr_vecs: number of bio_vecs to allocate
3175199a 586 * @gfp_mask: the GFP_* mask given to the slab allocator
3175199a 587 *
066ff571
CH
588 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
589 * using bio_init() before use. To free a bio returned from this function use
590 * kfree() after calling bio_uninit(). A bio returned from this function can
591 * be reused by calling bio_uninit() before calling bio_init() again.
592 *
593 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
340e1347 594 * function are not backed by a mempool can fail. Do not use this function
066ff571 595 * for allocations in the file system I/O path.
3175199a
CH
596 *
597 * Returns: Pointer to new bio on success, NULL on failure.
598 */
066ff571 599struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
3175199a
CH
600{
601 struct bio *bio;
602
066ff571 603 if (nr_vecs > UIO_MAXIOV)
3175199a 604 return NULL;
066ff571 605 return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
3175199a
CH
606}
607EXPORT_SYMBOL(bio_kmalloc);
608
6f822e1b 609void zero_fill_bio(struct bio *bio)
1da177e4 610{
7988613b
KO
611 struct bio_vec bv;
612 struct bvec_iter iter;
1da177e4 613
ab6c340e
CH
614 bio_for_each_segment(bv, bio, iter)
615 memzero_bvec(&bv);
1da177e4 616}
6f822e1b 617EXPORT_SYMBOL(zero_fill_bio);
1da177e4 618
83c9c547
ML
619/**
620 * bio_truncate - truncate the bio to small size of @new_size
621 * @bio: the bio to be truncated
622 * @new_size: new size for truncating the bio
623 *
624 * Description:
625 * Truncate the bio to new size of @new_size. If bio_op(bio) is
626 * REQ_OP_READ, zero the truncated part. This function should only
627 * be used for handling corner cases, such as bio eod.
628 */
4f7ab09a 629static void bio_truncate(struct bio *bio, unsigned new_size)
85a8ce62
ML
630{
631 struct bio_vec bv;
632 struct bvec_iter iter;
633 unsigned int done = 0;
634 bool truncated = false;
635
636 if (new_size >= bio->bi_iter.bi_size)
637 return;
638
83c9c547 639 if (bio_op(bio) != REQ_OP_READ)
85a8ce62
ML
640 goto exit;
641
642 bio_for_each_segment(bv, bio, iter) {
643 if (done + bv.bv_len > new_size) {
644 unsigned offset;
645
646 if (!truncated)
647 offset = new_size - done;
648 else
649 offset = 0;
3ee859e3
OH
650 zero_user(bv.bv_page, bv.bv_offset + offset,
651 bv.bv_len - offset);
85a8ce62
ML
652 truncated = true;
653 }
654 done += bv.bv_len;
655 }
656
657 exit:
658 /*
659 * Don't touch bvec table here and make it really immutable, since
660 * fs bio user has to retrieve all pages via bio_for_each_segment_all
661 * in its .end_bio() callback.
662 *
663 * It is enough to truncate bio by updating .bi_size since we can make
664 * correct bvec with the updated .bi_size for drivers.
665 */
666 bio->bi_iter.bi_size = new_size;
667}
668
29125ed6
CH
669/**
670 * guard_bio_eod - truncate a BIO to fit the block device
671 * @bio: bio to truncate
672 *
673 * This allows us to do IO even on the odd last sectors of a device, even if the
674 * block size is some multiple of the physical sector size.
675 *
676 * We'll just truncate the bio to the size of the device, and clear the end of
677 * the buffer head manually. Truly out-of-range accesses will turn into actual
678 * I/O errors, this only handles the "we need to be able to do I/O at the final
679 * sector" case.
680 */
681void guard_bio_eod(struct bio *bio)
682{
309dca30 683 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
29125ed6
CH
684
685 if (!maxsector)
686 return;
687
688 /*
689 * If the *whole* IO is past the end of the device,
690 * let it through, and the IO layer will turn it into
691 * an EIO.
692 */
693 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
694 return;
695
696 maxsector -= bio->bi_iter.bi_sector;
697 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
698 return;
699
700 bio_truncate(bio, maxsector << 9);
701}
702
b99182c5
PB
703static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
704 unsigned int nr)
be4d234d
JA
705{
706 unsigned int i = 0;
707 struct bio *bio;
708
fcade2ce
JA
709 while ((bio = cache->free_list) != NULL) {
710 cache->free_list = bio->bi_next;
be4d234d
JA
711 cache->nr--;
712 bio_free(bio);
713 if (++i == nr)
714 break;
715 }
b99182c5
PB
716 return i;
717}
718
719static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
720 unsigned int nr)
721{
722 nr -= __bio_alloc_cache_prune(cache, nr);
723 if (!READ_ONCE(cache->free_list)) {
724 bio_alloc_irq_cache_splice(cache);
725 __bio_alloc_cache_prune(cache, nr);
726 }
be4d234d
JA
727}
728
729static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
730{
731 struct bio_set *bs;
732
733 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
734 if (bs->cache) {
735 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
736
737 bio_alloc_cache_prune(cache, -1U);
738 }
739 return 0;
740}
741
742static void bio_alloc_cache_destroy(struct bio_set *bs)
743{
744 int cpu;
745
746 if (!bs->cache)
747 return;
748
749 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
750 for_each_possible_cpu(cpu) {
751 struct bio_alloc_cache *cache;
752
753 cache = per_cpu_ptr(bs->cache, cpu);
754 bio_alloc_cache_prune(cache, -1U);
755 }
756 free_percpu(bs->cache);
605f7415 757 bs->cache = NULL;
be4d234d
JA
758}
759
f25cf75a
PB
760static inline void bio_put_percpu_cache(struct bio *bio)
761{
762 struct bio_alloc_cache *cache;
763
764 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
b99182c5
PB
765 if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) {
766 put_cpu();
767 bio_free(bio);
768 return;
769 }
770
f25cf75a
PB
771 bio_uninit(bio);
772
773 if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
774 bio->bi_next = cache->free_list;
11eb695f 775 bio->bi_bdev = NULL;
f25cf75a
PB
776 cache->free_list = bio;
777 cache->nr++;
778 } else {
b99182c5 779 unsigned long flags;
f25cf75a 780
b99182c5
PB
781 local_irq_save(flags);
782 bio->bi_next = cache->free_list_irq;
783 cache->free_list_irq = bio;
784 cache->nr_irq++;
785 local_irq_restore(flags);
786 }
f25cf75a
PB
787 put_cpu();
788}
789
1da177e4
LT
790/**
791 * bio_put - release a reference to a bio
792 * @bio: bio to release reference to
793 *
794 * Description:
795 * Put a reference to a &struct bio, either one you have gotten with
9b10f6a9 796 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
1da177e4
LT
797 **/
798void bio_put(struct bio *bio)
799{
be4d234d 800 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
9e8c0d0d 801 BUG_ON(!atomic_read(&bio->__bi_cnt));
be4d234d
JA
802 if (!atomic_dec_and_test(&bio->__bi_cnt))
803 return;
804 }
f25cf75a
PB
805 if (bio->bi_opf & REQ_ALLOC_CACHE)
806 bio_put_percpu_cache(bio);
807 else
be4d234d 808 bio_free(bio);
1da177e4 809}
a112a71d 810EXPORT_SYMBOL(bio_put);
1da177e4 811
a0e8de79 812static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
59d276fe 813{
b7c44ed9 814 bio_set_flag(bio, BIO_CLONED);
ca474b73 815 bio->bi_ioprio = bio_src->bi_ioprio;
59d276fe 816 bio->bi_iter = bio_src->bi_iter;
20bd723e 817
7ecc56c6
CH
818 if (bio->bi_bdev) {
819 if (bio->bi_bdev == bio_src->bi_bdev &&
820 bio_flagged(bio_src, BIO_REMAPPED))
821 bio_set_flag(bio, BIO_REMAPPED);
822 bio_clone_blkg_association(bio, bio_src);
823 }
56b4b5ab
CH
824
825 if (bio_crypt_clone(bio, bio_src, gfp) < 0)
826 return -ENOMEM;
827 if (bio_integrity(bio_src) &&
828 bio_integrity_clone(bio, bio_src, gfp) < 0)
829 return -ENOMEM;
830 return 0;
59d276fe 831}
59d276fe
KO
832
833/**
abfc426d
CH
834 * bio_alloc_clone - clone a bio that shares the original bio's biovec
835 * @bdev: block_device to clone onto
a0e8de79
CH
836 * @bio_src: bio to clone from
837 * @gfp: allocation priority
838 * @bs: bio_set to allocate from
59d276fe 839 *
a0e8de79
CH
840 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
841 * bio, but not the actual data it points to.
842 *
843 * The caller must ensure that the return bio is not freed before @bio_src.
59d276fe 844 */
abfc426d
CH
845struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
846 gfp_t gfp, struct bio_set *bs)
59d276fe 847{
a0e8de79 848 struct bio *bio;
59d276fe 849
abfc426d 850 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
a0e8de79 851 if (!bio)
59d276fe
KO
852 return NULL;
853
a0e8de79
CH
854 if (__bio_clone(bio, bio_src, gfp) < 0) {
855 bio_put(bio);
56b4b5ab
CH
856 return NULL;
857 }
a0e8de79 858 bio->bi_io_vec = bio_src->bi_io_vec;
59d276fe 859
a0e8de79 860 return bio;
59d276fe 861}
abfc426d 862EXPORT_SYMBOL(bio_alloc_clone);
59d276fe 863
a0e8de79 864/**
abfc426d
CH
865 * bio_init_clone - clone a bio that shares the original bio's biovec
866 * @bdev: block_device to clone onto
a0e8de79
CH
867 * @bio: bio to clone into
868 * @bio_src: bio to clone from
869 * @gfp: allocation priority
870 *
871 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
872 * The caller owns the returned bio, but not the actual data it points to.
873 *
874 * The caller must ensure that @bio_src is not freed before @bio.
875 */
abfc426d
CH
876int bio_init_clone(struct block_device *bdev, struct bio *bio,
877 struct bio *bio_src, gfp_t gfp)
a0e8de79
CH
878{
879 int ret;
880
abfc426d 881 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
a0e8de79
CH
882 ret = __bio_clone(bio, bio_src, gfp);
883 if (ret)
884 bio_uninit(bio);
885 return ret;
886}
abfc426d 887EXPORT_SYMBOL(bio_init_clone);
a0e8de79 888
9a6083be
CH
889/**
890 * bio_full - check if the bio is full
891 * @bio: bio to check
892 * @len: length of one segment to be added
893 *
894 * Return true if @bio is full and one segment with @len bytes can't be
895 * added to the bio, otherwise return false
896 */
897static inline bool bio_full(struct bio *bio, unsigned len)
898{
899 if (bio->bi_vcnt >= bio->bi_max_vecs)
900 return true;
901 if (bio->bi_iter.bi_size > UINT_MAX - len)
902 return true;
903 return false;
904}
905
5919482e
ML
906static inline bool page_is_mergeable(const struct bio_vec *bv,
907 struct page *page, unsigned int len, unsigned int off,
ff896738 908 bool *same_page)
5919482e 909{
d8166519
MWO
910 size_t bv_end = bv->bv_offset + bv->bv_len;
911 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
5919482e
ML
912 phys_addr_t page_addr = page_to_phys(page);
913
914 if (vec_end_addr + 1 != page_addr + off)
915 return false;
916 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
917 return false;
49580e69
LG
918 if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
919 return false;
52d52d1c 920
ff896738 921 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
d8166519
MWO
922 if (*same_page)
923 return true;
11b331f8
AP
924 else if (IS_ENABLED(CONFIG_KMSAN))
925 return false;
d8166519 926 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
5919482e
ML
927}
928
9774b391
CH
929/**
930 * __bio_try_merge_page - try appending data to an existing bvec.
931 * @bio: destination bio
932 * @page: start page to add
933 * @len: length of the data to add
934 * @off: offset of the data relative to @page
935 * @same_page: return if the segment has been merged inside the same page
936 *
937 * Try to add the data at @page + @off to the last bvec of @bio. This is a
938 * useful optimisation for file systems with a block size smaller than the
939 * page size.
940 *
941 * Warn if (@len, @off) crosses pages in case that @same_page is true.
942 *
943 * Return %true on success or %false on failure.
944 */
945static bool __bio_try_merge_page(struct bio *bio, struct page *page,
946 unsigned int len, unsigned int off, bool *same_page)
947{
948 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
949 return false;
950
951 if (bio->bi_vcnt > 0) {
952 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
953
954 if (page_is_mergeable(bv, page, len, off, same_page)) {
955 if (bio->bi_iter.bi_size > UINT_MAX - len) {
956 *same_page = false;
957 return false;
958 }
959 bv->bv_len += len;
960 bio->bi_iter.bi_size += len;
961 return true;
962 }
963 }
964 return false;
965}
966
e4581105
CH
967/*
968 * Try to merge a page into a segment, while obeying the hardware segment
969 * size limit. This is not for normal read/write bios, but for passthrough
970 * or Zone Append operations that we can't split.
971 */
972static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
973 struct page *page, unsigned len,
974 unsigned offset, bool *same_page)
489fbbcb 975{
384209cd 976 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
489fbbcb
ML
977 unsigned long mask = queue_segment_boundary(q);
978 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
979 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
980
981 if ((addr1 | mask) != (addr2 | mask))
982 return false;
489fbbcb
ML
983 if (bv->bv_len + len > queue_max_segment_size(q))
984 return false;
384209cd 985 return __bio_try_merge_page(bio, page, len, offset, same_page);
489fbbcb
ML
986}
987
1da177e4 988/**
e4581105
CH
989 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
990 * @q: the target queue
991 * @bio: destination bio
992 * @page: page to add
993 * @len: vec entry length
994 * @offset: vec entry offset
995 * @max_sectors: maximum number of sectors that can be added
996 * @same_page: return if the segment has been merged inside the same page
c66a14d0 997 *
e4581105
CH
998 * Add a page to a bio while respecting the hardware max_sectors, max_segment
999 * and gap limitations.
1da177e4 1000 */
e4581105 1001int bio_add_hw_page(struct request_queue *q, struct bio *bio,
19047087 1002 struct page *page, unsigned int len, unsigned int offset,
e4581105 1003 unsigned int max_sectors, bool *same_page)
1da177e4 1004{
1da177e4
LT
1005 struct bio_vec *bvec;
1006
e4581105 1007 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1da177e4
LT
1008 return 0;
1009
e4581105 1010 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
1da177e4
LT
1011 return 0;
1012
80cfd548 1013 if (bio->bi_vcnt > 0) {
e4581105 1014 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
384209cd 1015 return len;
320ea869
CH
1016
1017 /*
1018 * If the queue doesn't support SG gaps and adding this segment
1019 * would create a gap, disallow it.
1020 */
384209cd 1021 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
c55ddd90 1022 if (bvec_gap_to_prev(&q->limits, bvec, offset))
320ea869 1023 return 0;
80cfd548
JA
1024 }
1025
79d08f89 1026 if (bio_full(bio, len))
1da177e4
LT
1027 return 0;
1028
14ccb66b 1029 if (bio->bi_vcnt >= queue_max_segments(q))
489fbbcb
ML
1030 return 0;
1031
d58cdfae 1032 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
fcbf6a08 1033 bio->bi_vcnt++;
dcdca753 1034 bio->bi_iter.bi_size += len;
1da177e4
LT
1035 return len;
1036}
19047087 1037
e4581105
CH
1038/**
1039 * bio_add_pc_page - attempt to add page to passthrough bio
1040 * @q: the target queue
1041 * @bio: destination bio
1042 * @page: page to add
1043 * @len: vec entry length
1044 * @offset: vec entry offset
1045 *
1046 * Attempt to add a page to the bio_vec maplist. This can fail for a
1047 * number of reasons, such as the bio being full or target block device
1048 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1049 * so it is always possible to add a single page to an empty bio.
1050 *
1051 * This should only be used by passthrough bios.
1052 */
19047087
ML
1053int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1054 struct page *page, unsigned int len, unsigned int offset)
1055{
d1916c86 1056 bool same_page = false;
e4581105
CH
1057 return bio_add_hw_page(q, bio, page, len, offset,
1058 queue_max_hw_sectors(q), &same_page);
19047087 1059}
a112a71d 1060EXPORT_SYMBOL(bio_add_pc_page);
6e68af66 1061
ae29333f
JT
1062/**
1063 * bio_add_zone_append_page - attempt to add page to zone-append bio
1064 * @bio: destination bio
1065 * @page: page to add
1066 * @len: vec entry length
1067 * @offset: vec entry offset
1068 *
1069 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1070 * for a zone-append request. This can fail for a number of reasons, such as the
1071 * bio being full or the target block device is not a zoned block device or
1072 * other limitations of the target block device. The target block device must
1073 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1074 * to an empty bio.
1075 *
1076 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1077 */
1078int bio_add_zone_append_page(struct bio *bio, struct page *page,
1079 unsigned int len, unsigned int offset)
1080{
3caee463 1081 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
ae29333f
JT
1082 bool same_page = false;
1083
1084 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1085 return 0;
1086
edd1dbc8 1087 if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
ae29333f
JT
1088 return 0;
1089
1090 return bio_add_hw_page(q, bio, page, len, offset,
1091 queue_max_zone_append_sectors(q), &same_page);
1092}
1093EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1094
0aa69fd3 1095/**
551879a4 1096 * __bio_add_page - add page(s) to a bio in a new segment
0aa69fd3 1097 * @bio: destination bio
551879a4
ML
1098 * @page: start page to add
1099 * @len: length of the data to add, may cross pages
1100 * @off: offset of the data relative to @page, may cross pages
0aa69fd3
CH
1101 *
1102 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1103 * that @bio has space for another bvec.
1104 */
1105void __bio_add_page(struct bio *bio, struct page *page,
1106 unsigned int len, unsigned int off)
1107{
0aa69fd3 1108 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
79d08f89 1109 WARN_ON_ONCE(bio_full(bio, len));
0aa69fd3 1110
d58cdfae 1111 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
c66a14d0 1112 bio->bi_iter.bi_size += len;
0aa69fd3
CH
1113 bio->bi_vcnt++;
1114}
1115EXPORT_SYMBOL_GPL(__bio_add_page);
1116
1117/**
551879a4 1118 * bio_add_page - attempt to add page(s) to bio
0aa69fd3 1119 * @bio: destination bio
551879a4
ML
1120 * @page: start page to add
1121 * @len: vec entry length, may cross pages
1122 * @offset: vec entry offset relative to @page, may cross pages
0aa69fd3 1123 *
551879a4 1124 * Attempt to add page(s) to the bio_vec maplist. This will only fail
0aa69fd3
CH
1125 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1126 */
1127int bio_add_page(struct bio *bio, struct page *page,
1128 unsigned int len, unsigned int offset)
1129{
ff896738
CH
1130 bool same_page = false;
1131
1132 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
79d08f89 1133 if (bio_full(bio, len))
0aa69fd3
CH
1134 return 0;
1135 __bio_add_page(bio, page, len, offset);
1136 }
c66a14d0 1137 return len;
1da177e4 1138}
a112a71d 1139EXPORT_SYMBOL(bio_add_page);
1da177e4 1140
7a150f1e
JT
1141void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1142 size_t off)
1143{
1144 WARN_ON_ONCE(len > UINT_MAX);
1145 WARN_ON_ONCE(off > UINT_MAX);
1146 __bio_add_page(bio, &folio->page, len, off);
1147}
1148
85f5a74c
MWO
1149/**
1150 * bio_add_folio - Attempt to add part of a folio to a bio.
1151 * @bio: BIO to add to.
1152 * @folio: Folio to add.
1153 * @len: How many bytes from the folio to add.
1154 * @off: First byte in this folio to add.
1155 *
1156 * Filesystems that use folios can call this function instead of calling
1157 * bio_add_page() for each page in the folio. If @off is bigger than
1158 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1159 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1160 *
1161 * Return: Whether the addition was successful.
1162 */
1163bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1164 size_t off)
1165{
1166 if (len > UINT_MAX || off > UINT_MAX)
455a844d 1167 return false;
85f5a74c
MWO
1168 return bio_add_page(bio, &folio->page, len, off) > 0;
1169}
cd57b771 1170EXPORT_SYMBOL(bio_add_folio);
85f5a74c 1171
c809084a 1172void __bio_release_pages(struct bio *bio, bool mark_dirty)
7321ecbf
CH
1173{
1174 struct bvec_iter_all iter_all;
1175 struct bio_vec *bvec;
7321ecbf 1176
d241a95f
CH
1177 bio_for_each_segment_all(bvec, bio, iter_all) {
1178 if (mark_dirty && !PageCompound(bvec->bv_page))
1179 set_page_dirty_lock(bvec->bv_page);
fd363244 1180 bio_release_page(bio, bvec->bv_page);
d241a95f 1181 }
7321ecbf 1182}
c809084a 1183EXPORT_SYMBOL_GPL(__bio_release_pages);
7321ecbf 1184
1bb6b810 1185void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
6d0c48ae 1186{
fa5fa8ec
PB
1187 size_t size = iov_iter_count(iter);
1188
7a800a20 1189 WARN_ON_ONCE(bio->bi_max_vecs);
c42bca92 1190
fa5fa8ec
PB
1191 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1192 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1193 size_t max_sectors = queue_max_zone_append_sectors(q);
1194
1195 size = min(size, max_sectors << SECTOR_SHIFT);
1196 }
1197
c42bca92 1198 bio->bi_vcnt = iter->nr_segs;
c42bca92
PB
1199 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1200 bio->bi_iter.bi_bvec_done = iter->iov_offset;
fa5fa8ec 1201 bio->bi_iter.bi_size = size;
977be012 1202 bio_set_flag(bio, BIO_CLONED);
7de55b7d 1203}
c42bca92 1204
c58c0074
KB
1205static int bio_iov_add_page(struct bio *bio, struct page *page,
1206 unsigned int len, unsigned int offset)
1207{
1208 bool same_page = false;
1209
1210 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
c58c0074
KB
1211 __bio_add_page(bio, page, len, offset);
1212 return 0;
1213 }
1214
1215 if (same_page)
a7e689dd 1216 bio_release_page(bio, page);
c58c0074
KB
1217 return 0;
1218}
1219
1220static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1221 unsigned int len, unsigned int offset)
1222{
1223 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1224 bool same_page = false;
1225
1226 if (bio_add_hw_page(q, bio, page, len, offset,
1227 queue_max_zone_append_sectors(q), &same_page) != len)
1228 return -EINVAL;
1229 if (same_page)
a7e689dd 1230 bio_release_page(bio, page);
c58c0074
KB
1231 return 0;
1232}
1233
576ed913
CH
1234#define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1235
2cefe4db 1236/**
17d51b10 1237 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
2cefe4db
KO
1238 * @bio: bio to add pages to
1239 * @iter: iov iterator describing the region to be mapped
1240 *
a7e689dd
DH
1241 * Extracts pages from *iter and appends them to @bio's bvec array. The pages
1242 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.
1243 * For a multi-segment *iter, this function only adds pages from the next
1244 * non-empty segment of the iov iterator.
2cefe4db 1245 */
17d51b10 1246static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
2cefe4db 1247{
f62e52d1 1248 iov_iter_extraction_t extraction_flags = 0;
576ed913
CH
1249 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1250 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
2cefe4db
KO
1251 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1252 struct page **pages = (struct page **)bv;
576ed913 1253 ssize_t size, left;
e97424fd 1254 unsigned len, i = 0;
480cb846 1255 size_t offset, trim;
325347d9 1256 int ret = 0;
576ed913
CH
1257
1258 /*
1259 * Move page array up in the allocated memory for the bio vecs as far as
1260 * possible so that we can start filling biovecs from the beginning
1261 * without overwriting the temporary page array.
c58c0074 1262 */
576ed913
CH
1263 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1264 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
2cefe4db 1265
5e3e3f2e 1266 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
f62e52d1 1267 extraction_flags |= ITER_ALLOW_P2PDMA;
5e3e3f2e 1268
b1a000d3
KB
1269 /*
1270 * Each segment in the iov is required to be a block size multiple.
1271 * However, we may not be able to get the entire segment if it spans
1272 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1273 * result to ensure the bio's total size is correct. The remainder of
1274 * the iov data will be picked up in the next bio iteration.
1275 */
a7e689dd
DH
1276 size = iov_iter_extract_pages(iter, &pages,
1277 UINT_MAX - bio->bi_iter.bi_size,
1278 nr_pages, extraction_flags, &offset);
480cb846
AV
1279 if (unlikely(size <= 0))
1280 return size ? size : -EFAULT;
1281
1282 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1283
1284 trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1285 iov_iter_revert(iter, trim);
1286
1287 size -= trim;
1288 if (unlikely(!size)) {
1289 ret = -EFAULT;
e97424fd
KB
1290 goto out;
1291 }
2cefe4db 1292
576ed913
CH
1293 for (left = size, i = 0; left > 0; left -= len, i++) {
1294 struct page *page = pages[i];
2cefe4db 1295
576ed913 1296 len = min_t(size_t, PAGE_SIZE - offset, left);
34cdb8c8 1297 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
c58c0074
KB
1298 ret = bio_iov_add_zone_append_page(bio, page, len,
1299 offset);
e97424fd 1300 if (ret)
34cdb8c8 1301 break;
34cdb8c8
KB
1302 } else
1303 bio_iov_add_page(bio, page, len, offset);
45691804 1304
576ed913 1305 offset = 0;
2cefe4db
KO
1306 }
1307
480cb846 1308 iov_iter_revert(iter, left);
e97424fd
KB
1309out:
1310 while (i < nr_pages)
a7e689dd 1311 bio_release_page(bio, pages[i++]);
e97424fd 1312
325347d9 1313 return ret;
2cefe4db 1314}
17d51b10
MW
1315
1316/**
6d0c48ae 1317 * bio_iov_iter_get_pages - add user or kernel pages to a bio
17d51b10 1318 * @bio: bio to add pages to
6d0c48ae
JA
1319 * @iter: iov iterator describing the region to be added
1320 *
1321 * This takes either an iterator pointing to user memory, or one pointing to
1322 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1323 * map them into the kernel. On IO completion, the caller should put those
c42bca92
PB
1324 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1325 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1326 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1327 * completed by a call to ->ki_complete() or returns with an error other than
1328 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1329 * on IO completion. If it isn't, then pages should be released.
17d51b10 1330 *
17d51b10 1331 * The function tries, but does not guarantee, to pin as many pages as
5cd3ddc1 1332 * fit into the bio, or are requested in @iter, whatever is smaller. If
6d0c48ae
JA
1333 * MM encounters an error pinning the requested pages, it stops. Error
1334 * is returned only if 0 pages could be pinned.
17d51b10
MW
1335 */
1336int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1337{
c42bca92 1338 int ret = 0;
14eacf12 1339
c42bca92 1340 if (iov_iter_is_bvec(iter)) {
fa5fa8ec
PB
1341 bio_iov_bvec_set(bio, iter);
1342 iov_iter_advance(iter, bio->bi_iter.bi_size);
1343 return 0;
c42bca92 1344 }
17d51b10 1345
a7e689dd
DH
1346 if (iov_iter_extract_will_pin(iter))
1347 bio_set_flag(bio, BIO_PAGE_PINNED);
17d51b10 1348 do {
c58c0074 1349 ret = __bio_iov_iter_get_pages(bio, iter);
79d08f89 1350 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
17d51b10 1351
14eacf12 1352 return bio->bi_vcnt ? 0 : ret;
17d51b10 1353}
29b2a3aa 1354EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
2cefe4db 1355
4246a0b6 1356static void submit_bio_wait_endio(struct bio *bio)
9e882242 1357{
65e53aab 1358 complete(bio->bi_private);
9e882242
KO
1359}
1360
1361/**
1362 * submit_bio_wait - submit a bio, and wait until it completes
9e882242
KO
1363 * @bio: The &struct bio which describes the I/O
1364 *
1365 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1366 * bio_endio() on failure.
3d289d68
JK
1367 *
1368 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1369 * result in bio reference to be consumed. The caller must drop the reference
1370 * on his own.
9e882242 1371 */
4e49ea4a 1372int submit_bio_wait(struct bio *bio)
9e882242 1373{
309dca30
CH
1374 DECLARE_COMPLETION_ONSTACK_MAP(done,
1375 bio->bi_bdev->bd_disk->lockdep_map);
de6a78b6 1376 unsigned long hang_check;
9e882242 1377
65e53aab 1378 bio->bi_private = &done;
9e882242 1379 bio->bi_end_io = submit_bio_wait_endio;
1eff9d32 1380 bio->bi_opf |= REQ_SYNC;
4e49ea4a 1381 submit_bio(bio);
de6a78b6
ML
1382
1383 /* Prevent hang_check timer from firing at us during very long I/O */
1384 hang_check = sysctl_hung_task_timeout_secs;
1385 if (hang_check)
1386 while (!wait_for_completion_io_timeout(&done,
1387 hang_check * (HZ/2)))
1388 ;
1389 else
1390 wait_for_completion_io(&done);
9e882242 1391
65e53aab 1392 return blk_status_to_errno(bio->bi_status);
9e882242
KO
1393}
1394EXPORT_SYMBOL(submit_bio_wait);
1395
d4aa57a1 1396void __bio_advance(struct bio *bio, unsigned bytes)
054bdf64
KO
1397{
1398 if (bio_integrity(bio))
1399 bio_integrity_advance(bio, bytes);
1400
a892c8d5 1401 bio_crypt_advance(bio, bytes);
4550dd6c 1402 bio_advance_iter(bio, &bio->bi_iter, bytes);
054bdf64 1403}
d4aa57a1 1404EXPORT_SYMBOL(__bio_advance);
054bdf64 1405
ee4b4e22
JA
1406void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1407 struct bio *src, struct bvec_iter *src_iter)
1408{
1409 while (src_iter->bi_size && dst_iter->bi_size) {
1410 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1411 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1412 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1413 void *src_buf = bvec_kmap_local(&src_bv);
1414 void *dst_buf = bvec_kmap_local(&dst_bv);
1415
1416 memcpy(dst_buf, src_buf, bytes);
1417
1418 kunmap_local(dst_buf);
1419 kunmap_local(src_buf);
1420
1421 bio_advance_iter_single(src, src_iter, bytes);
1422 bio_advance_iter_single(dst, dst_iter, bytes);
1423 }
1424}
1425EXPORT_SYMBOL(bio_copy_data_iter);
1426
38a72dac 1427/**
45db54d5
KO
1428 * bio_copy_data - copy contents of data buffers from one bio to another
1429 * @src: source bio
1430 * @dst: destination bio
38a72dac
KO
1431 *
1432 * Stops when it reaches the end of either @src or @dst - that is, copies
1433 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1434 */
1435void bio_copy_data(struct bio *dst, struct bio *src)
1436{
45db54d5
KO
1437 struct bvec_iter src_iter = src->bi_iter;
1438 struct bvec_iter dst_iter = dst->bi_iter;
1439
ee4b4e22 1440 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
38a72dac 1441}
16ac3d63
KO
1442EXPORT_SYMBOL(bio_copy_data);
1443
491221f8 1444void bio_free_pages(struct bio *bio)
1dfa0f68
CH
1445{
1446 struct bio_vec *bvec;
6dc4f100 1447 struct bvec_iter_all iter_all;
1dfa0f68 1448
2b070cfe 1449 bio_for_each_segment_all(bvec, bio, iter_all)
1dfa0f68
CH
1450 __free_page(bvec->bv_page);
1451}
491221f8 1452EXPORT_SYMBOL(bio_free_pages);
1dfa0f68 1453
1da177e4
LT
1454/*
1455 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1456 * for performing direct-IO in BIOs.
1457 *
1458 * The problem is that we cannot run set_page_dirty() from interrupt context
1459 * because the required locks are not interrupt-safe. So what we can do is to
1460 * mark the pages dirty _before_ performing IO. And in interrupt context,
1461 * check that the pages are still dirty. If so, fine. If not, redirty them
1462 * in process context.
1463 *
1464 * We special-case compound pages here: normally this means reads into hugetlb
1465 * pages. The logic in here doesn't really work right for compound pages
1466 * because the VM does not uniformly chase down the head page in all cases.
1467 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1468 * handle them at all. So we skip compound pages here at an early stage.
1469 *
1470 * Note that this code is very hard to test under normal circumstances because
1471 * direct-io pins the pages with get_user_pages(). This makes
1472 * is_page_cache_freeable return false, and the VM will not clean the pages.
0d5c3eba 1473 * But other code (eg, flusher threads) could clean the pages if they are mapped
1da177e4
LT
1474 * pagecache.
1475 *
1476 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1477 * deferred bio dirtying paths.
1478 */
1479
1480/*
1481 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1482 */
1483void bio_set_pages_dirty(struct bio *bio)
1484{
cb34e057 1485 struct bio_vec *bvec;
6dc4f100 1486 struct bvec_iter_all iter_all;
1da177e4 1487
2b070cfe 1488 bio_for_each_segment_all(bvec, bio, iter_all) {
3bb50983
CH
1489 if (!PageCompound(bvec->bv_page))
1490 set_page_dirty_lock(bvec->bv_page);
1da177e4
LT
1491 }
1492}
1493
1da177e4
LT
1494/*
1495 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1496 * If they are, then fine. If, however, some pages are clean then they must
1497 * have been written out during the direct-IO read. So we take another ref on
24d5493f 1498 * the BIO and re-dirty the pages in process context.
1da177e4
LT
1499 *
1500 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
fd363244
DH
1501 * here on. It will unpin each page and will run one bio_put() against the
1502 * BIO.
1da177e4
LT
1503 */
1504
65f27f38 1505static void bio_dirty_fn(struct work_struct *work);
1da177e4 1506
65f27f38 1507static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1da177e4
LT
1508static DEFINE_SPINLOCK(bio_dirty_lock);
1509static struct bio *bio_dirty_list;
1510
1511/*
1512 * This runs in process context
1513 */
65f27f38 1514static void bio_dirty_fn(struct work_struct *work)
1da177e4 1515{
24d5493f 1516 struct bio *bio, *next;
1da177e4 1517
24d5493f
CH
1518 spin_lock_irq(&bio_dirty_lock);
1519 next = bio_dirty_list;
1da177e4 1520 bio_dirty_list = NULL;
24d5493f 1521 spin_unlock_irq(&bio_dirty_lock);
1da177e4 1522
24d5493f
CH
1523 while ((bio = next) != NULL) {
1524 next = bio->bi_private;
1da177e4 1525
d241a95f 1526 bio_release_pages(bio, true);
1da177e4 1527 bio_put(bio);
1da177e4
LT
1528 }
1529}
1530
1531void bio_check_pages_dirty(struct bio *bio)
1532{
cb34e057 1533 struct bio_vec *bvec;
24d5493f 1534 unsigned long flags;
6dc4f100 1535 struct bvec_iter_all iter_all;
1da177e4 1536
2b070cfe 1537 bio_for_each_segment_all(bvec, bio, iter_all) {
24d5493f
CH
1538 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1539 goto defer;
1da177e4
LT
1540 }
1541
d241a95f 1542 bio_release_pages(bio, false);
24d5493f
CH
1543 bio_put(bio);
1544 return;
1545defer:
1546 spin_lock_irqsave(&bio_dirty_lock, flags);
1547 bio->bi_private = bio_dirty_list;
1548 bio_dirty_list = bio;
1549 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1550 schedule_work(&bio_dirty_work);
1da177e4
LT
1551}
1552
c4cf5261
JA
1553static inline bool bio_remaining_done(struct bio *bio)
1554{
1555 /*
1556 * If we're not chaining, then ->__bi_remaining is always 1 and
1557 * we always end io on the first invocation.
1558 */
1559 if (!bio_flagged(bio, BIO_CHAIN))
1560 return true;
1561
1562 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1563
326e1dbb 1564 if (atomic_dec_and_test(&bio->__bi_remaining)) {
b7c44ed9 1565 bio_clear_flag(bio, BIO_CHAIN);
c4cf5261 1566 return true;
326e1dbb 1567 }
c4cf5261
JA
1568
1569 return false;
1570}
1571
1da177e4
LT
1572/**
1573 * bio_endio - end I/O on a bio
1574 * @bio: bio
1da177e4
LT
1575 *
1576 * Description:
4246a0b6
CH
1577 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1578 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1579 * bio unless they own it and thus know that it has an end_io function.
fbbaf700
N
1580 *
1581 * bio_endio() can be called several times on a bio that has been chained
1582 * using bio_chain(). The ->bi_end_io() function will only be called the
60b6a7e6 1583 * last time.
1da177e4 1584 **/
4246a0b6 1585void bio_endio(struct bio *bio)
1da177e4 1586{
ba8c6967 1587again:
2b885517 1588 if (!bio_remaining_done(bio))
ba8c6967 1589 return;
7c20f116
CH
1590 if (!bio_integrity_endio(bio))
1591 return;
1da177e4 1592
aa1b46dc 1593 rq_qos_done_bio(bio);
67b42d0b 1594
60b6a7e6 1595 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
3caee463 1596 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
60b6a7e6
EH
1597 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1598 }
1599
ba8c6967
CH
1600 /*
1601 * Need to have a real endio function for chained bios, otherwise
1602 * various corner cases will break (like stacking block devices that
1603 * save/restore bi_end_io) - however, we want to avoid unbounded
1604 * recursion and blowing the stack. Tail call optimization would
1605 * handle this, but compiling with frame pointers also disables
1606 * gcc's sibling call optimization.
1607 */
1608 if (bio->bi_end_io == bio_chain_endio) {
1609 bio = __bio_chain_endio(bio);
1610 goto again;
196d38bc 1611 }
ba8c6967 1612
9e234eea 1613 blk_throtl_bio_endio(bio);
b222dd2f
SL
1614 /* release cgroup info */
1615 bio_uninit(bio);
ba8c6967
CH
1616 if (bio->bi_end_io)
1617 bio->bi_end_io(bio);
1da177e4 1618}
a112a71d 1619EXPORT_SYMBOL(bio_endio);
1da177e4 1620
20d0189b
KO
1621/**
1622 * bio_split - split a bio
1623 * @bio: bio to split
1624 * @sectors: number of sectors to split from the front of @bio
1625 * @gfp: gfp mask
1626 * @bs: bio set to allocate from
1627 *
1628 * Allocates and returns a new bio which represents @sectors from the start of
1629 * @bio, and updates @bio to represent the remaining sectors.
1630 *
f3f5da62 1631 * Unless this is a discard request the newly allocated bio will point
dad77584
BVA
1632 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1633 * neither @bio nor @bs are freed before the split bio.
20d0189b
KO
1634 */
1635struct bio *bio_split(struct bio *bio, int sectors,
1636 gfp_t gfp, struct bio_set *bs)
1637{
f341a4d3 1638 struct bio *split;
20d0189b
KO
1639
1640 BUG_ON(sectors <= 0);
1641 BUG_ON(sectors >= bio_sectors(bio));
1642
0512a75b
KB
1643 /* Zone append commands cannot be split */
1644 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1645 return NULL;
1646
abfc426d 1647 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
20d0189b
KO
1648 if (!split)
1649 return NULL;
1650
1651 split->bi_iter.bi_size = sectors << 9;
1652
1653 if (bio_integrity(split))
fbd08e76 1654 bio_integrity_trim(split);
20d0189b
KO
1655
1656 bio_advance(bio, split->bi_iter.bi_size);
1657
fbbaf700 1658 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
20d59023 1659 bio_set_flag(split, BIO_TRACE_COMPLETION);
fbbaf700 1660
20d0189b
KO
1661 return split;
1662}
1663EXPORT_SYMBOL(bio_split);
1664
6678d83f
KO
1665/**
1666 * bio_trim - trim a bio
1667 * @bio: bio to trim
1668 * @offset: number of sectors to trim from the front of @bio
1669 * @size: size we want to trim @bio to, in sectors
e83502ca
CK
1670 *
1671 * This function is typically used for bios that are cloned and submitted
1672 * to the underlying device in parts.
6678d83f 1673 */
e83502ca 1674void bio_trim(struct bio *bio, sector_t offset, sector_t size)
6678d83f 1675{
e83502ca 1676 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
8535c018 1677 offset + size > bio_sectors(bio)))
e83502ca 1678 return;
6678d83f
KO
1679
1680 size <<= 9;
4f024f37 1681 if (offset == 0 && size == bio->bi_iter.bi_size)
6678d83f
KO
1682 return;
1683
6678d83f 1684 bio_advance(bio, offset << 9);
4f024f37 1685 bio->bi_iter.bi_size = size;
376a78ab
DM
1686
1687 if (bio_integrity(bio))
fbd08e76 1688 bio_integrity_trim(bio);
6678d83f
KO
1689}
1690EXPORT_SYMBOL_GPL(bio_trim);
1691
1da177e4
LT
1692/*
1693 * create memory pools for biovec's in a bio_set.
1694 * use the global biovec slabs created for general use.
1695 */
8aa6ba2f 1696int biovec_init_pool(mempool_t *pool, int pool_entries)
1da177e4 1697{
7a800a20 1698 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1da177e4 1699
8aa6ba2f 1700 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1da177e4
LT
1701}
1702
917a38c7
KO
1703/*
1704 * bioset_exit - exit a bioset initialized with bioset_init()
1705 *
1706 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1707 * kzalloc()).
1708 */
1709void bioset_exit(struct bio_set *bs)
1da177e4 1710{
be4d234d 1711 bio_alloc_cache_destroy(bs);
df2cb6da
KO
1712 if (bs->rescue_workqueue)
1713 destroy_workqueue(bs->rescue_workqueue);
917a38c7 1714 bs->rescue_workqueue = NULL;
df2cb6da 1715
8aa6ba2f
KO
1716 mempool_exit(&bs->bio_pool);
1717 mempool_exit(&bs->bvec_pool);
9f060e22 1718
7878cba9 1719 bioset_integrity_free(bs);
917a38c7
KO
1720 if (bs->bio_slab)
1721 bio_put_slab(bs);
1722 bs->bio_slab = NULL;
1723}
1724EXPORT_SYMBOL(bioset_exit);
1da177e4 1725
917a38c7
KO
1726/**
1727 * bioset_init - Initialize a bio_set
dad08527 1728 * @bs: pool to initialize
917a38c7
KO
1729 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1730 * @front_pad: Number of bytes to allocate in front of the returned bio
1731 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1732 * and %BIOSET_NEED_RESCUER
1733 *
dad08527
KO
1734 * Description:
1735 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1736 * to ask for a number of bytes to be allocated in front of the bio.
1737 * Front pad allocation is useful for embedding the bio inside
1738 * another structure, to avoid allocating extra data to go with the bio.
1739 * Note that the bio must be embedded at the END of that structure always,
1740 * or things will break badly.
1741 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
abfc426d
CH
1742 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1743 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1744 * to dispatch queued requests when the mempool runs out of space.
dad08527 1745 *
917a38c7
KO
1746 */
1747int bioset_init(struct bio_set *bs,
1748 unsigned int pool_size,
1749 unsigned int front_pad,
1750 int flags)
1751{
917a38c7 1752 bs->front_pad = front_pad;
9f180e31
ML
1753 if (flags & BIOSET_NEED_BVECS)
1754 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1755 else
1756 bs->back_pad = 0;
917a38c7
KO
1757
1758 spin_lock_init(&bs->rescue_lock);
1759 bio_list_init(&bs->rescue_list);
1760 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1761
49d1ec85 1762 bs->bio_slab = bio_find_or_create_slab(bs);
917a38c7
KO
1763 if (!bs->bio_slab)
1764 return -ENOMEM;
1765
1766 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1767 goto bad;
1768
1769 if ((flags & BIOSET_NEED_BVECS) &&
1770 biovec_init_pool(&bs->bvec_pool, pool_size))
1771 goto bad;
1772
be4d234d
JA
1773 if (flags & BIOSET_NEED_RESCUER) {
1774 bs->rescue_workqueue = alloc_workqueue("bioset",
1775 WQ_MEM_RECLAIM, 0);
1776 if (!bs->rescue_workqueue)
1777 goto bad;
1778 }
1779 if (flags & BIOSET_PERCPU_CACHE) {
1780 bs->cache = alloc_percpu(struct bio_alloc_cache);
1781 if (!bs->cache)
1782 goto bad;
1783 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1784 }
917a38c7
KO
1785
1786 return 0;
1787bad:
1788 bioset_exit(bs);
1789 return -ENOMEM;
1790}
1791EXPORT_SYMBOL(bioset_init);
1792
de76fd89 1793static int __init init_bio(void)
1da177e4
LT
1794{
1795 int i;
1796
a3df2e45
JA
1797 BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1798
7878cba9 1799 bio_integrity_init();
1da177e4 1800
de76fd89
CH
1801 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1802 struct biovec_slab *bvs = bvec_slabs + i;
a7fcd37c 1803
de76fd89
CH
1804 bvs->slab = kmem_cache_create(bvs->name,
1805 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1806 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1da177e4 1807 }
1da177e4 1808
be4d234d
JA
1809 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1810 bio_cpu_dead);
1811
12c5b70c
JA
1812 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1813 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1da177e4
LT
1814 panic("bio: can't allocate bios\n");
1815
f4f8154a 1816 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
a91a2785
MP
1817 panic("bio: can't create integrity pool\n");
1818
1da177e4
LT
1819 return 0;
1820}
1da177e4 1821subsys_initcall(init_bio);