1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
6 #include <linux/swap.h>
8 #include <linux/blkdev.h>
10 #include <linux/iocontext.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/export.h>
15 #include <linux/mempool.h>
16 #include <linux/workqueue.h>
17 #include <linux/cgroup.h>
18 #include <linux/highmem.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/blk-crypto.h>
21 #include <linux/xarray.h>
23 #include <trace/events/block.h>
25 #include "blk-rq-qos.h"
26 #include "blk-cgroup.h"
28 #define ALLOC_CACHE_THRESHOLD 16
29 #define ALLOC_CACHE_SLACK 64
30 #define ALLOC_CACHE_MAX 256
32 struct bio_alloc_cache {
33 struct bio *free_list;
34 struct bio *free_list_irq;
39 static struct biovec_slab {
42 struct kmem_cache *slab;
43 } bvec_slabs[] __read_mostly = {
44 { .nr_vecs = 16, .name = "biovec-16" },
45 { .nr_vecs = 64, .name = "biovec-64" },
46 { .nr_vecs = 128, .name = "biovec-128" },
47 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
50 static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
53 /* smaller bios use inline vecs */
55 return &bvec_slabs[0];
57 return &bvec_slabs[1];
59 return &bvec_slabs[2];
60 case 129 ... BIO_MAX_VECS:
61 return &bvec_slabs[3];
69 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
70 * IO code that does not need private memory pools.
72 struct bio_set fs_bio_set;
73 EXPORT_SYMBOL(fs_bio_set);
76 * Our slab pool management
79 struct kmem_cache *slab;
80 unsigned int slab_ref;
81 unsigned int slab_size;
84 static DEFINE_MUTEX(bio_slab_lock);
85 static DEFINE_XARRAY(bio_slabs);
87 static struct bio_slab *create_bio_slab(unsigned int size)
89 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
94 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
95 bslab->slab = kmem_cache_create(bslab->name, size,
96 ARCH_KMALLOC_MINALIGN,
97 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
102 bslab->slab_size = size;
104 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
107 kmem_cache_destroy(bslab->slab);
114 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
116 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
119 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
121 unsigned int size = bs_bio_slab_size(bs);
122 struct bio_slab *bslab;
124 mutex_lock(&bio_slab_lock);
125 bslab = xa_load(&bio_slabs, size);
129 bslab = create_bio_slab(size);
130 mutex_unlock(&bio_slab_lock);
137 static void bio_put_slab(struct bio_set *bs)
139 struct bio_slab *bslab = NULL;
140 unsigned int slab_size = bs_bio_slab_size(bs);
142 mutex_lock(&bio_slab_lock);
144 bslab = xa_load(&bio_slabs, slab_size);
145 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
148 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
150 WARN_ON(!bslab->slab_ref);
152 if (--bslab->slab_ref)
155 xa_erase(&bio_slabs, slab_size);
157 kmem_cache_destroy(bslab->slab);
161 mutex_unlock(&bio_slab_lock);
164 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
166 BUG_ON(nr_vecs > BIO_MAX_VECS);
168 if (nr_vecs == BIO_MAX_VECS)
169 mempool_free(bv, pool);
170 else if (nr_vecs > BIO_INLINE_VECS)
171 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
175 * Make the first allocation restricted and don't dump info on allocation
176 * failures, since we'll fall back to the mempool in case of failure.
178 static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
180 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
181 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
184 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
187 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
189 if (WARN_ON_ONCE(!bvs))
193 * Upgrade the nr_vecs request to take full advantage of the allocation.
194 * We also rely on this in the bvec_free path.
196 *nr_vecs = bvs->nr_vecs;
199 * Try a slab allocation first for all smaller allocations. If that
200 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
201 * The mempool is sized to handle up to BIO_MAX_VECS entries.
203 if (*nr_vecs < BIO_MAX_VECS) {
206 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
207 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
209 *nr_vecs = BIO_MAX_VECS;
212 return mempool_alloc(pool, gfp_mask);
215 void bio_uninit(struct bio *bio)
217 #ifdef CONFIG_BLK_CGROUP
219 blkg_put(bio->bi_blkg);
223 if (bio_integrity(bio))
224 bio_integrity_free(bio);
226 bio_crypt_free_ctx(bio);
228 EXPORT_SYMBOL(bio_uninit);
230 static void bio_free(struct bio *bio)
232 struct bio_set *bs = bio->bi_pool;
238 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
239 mempool_free(p - bs->front_pad, &bs->bio_pool);
243 * Users of this function have their own bio allocation. Subsequently,
244 * they must remember to pair any call to bio_init() with bio_uninit()
245 * when IO has completed, or when the bio is released.
247 void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
248 unsigned short max_vecs, blk_opf_t opf)
256 bio->bi_iter.bi_sector = 0;
257 bio->bi_iter.bi_size = 0;
258 bio->bi_iter.bi_idx = 0;
259 bio->bi_iter.bi_bvec_done = 0;
260 bio->bi_end_io = NULL;
261 bio->bi_private = NULL;
262 #ifdef CONFIG_BLK_CGROUP
264 bio->bi_issue.value = 0;
266 bio_associate_blkg(bio);
267 #ifdef CONFIG_BLK_CGROUP_IOCOST
268 bio->bi_iocost_cost = 0;
271 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
272 bio->bi_crypt_context = NULL;
274 #ifdef CONFIG_BLK_DEV_INTEGRITY
275 bio->bi_integrity = NULL;
279 atomic_set(&bio->__bi_remaining, 1);
280 atomic_set(&bio->__bi_cnt, 1);
281 bio->bi_cookie = BLK_QC_T_NONE;
283 bio->bi_max_vecs = max_vecs;
284 bio->bi_io_vec = table;
287 EXPORT_SYMBOL(bio_init);
290 * bio_reset - reinitialize a bio
292 * @bdev: block device to use the bio for
293 * @opf: operation and flags for bio
296 * After calling bio_reset(), @bio will be in the same state as a freshly
297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
298 * preserved are the ones that are initialized by bio_alloc_bioset(). See
299 * comment in struct bio.
301 void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
304 memset(bio, 0, BIO_RESET_BYTES);
305 atomic_set(&bio->__bi_remaining, 1);
308 bio_associate_blkg(bio);
311 EXPORT_SYMBOL(bio_reset);
313 static struct bio *__bio_chain_endio(struct bio *bio)
315 struct bio *parent = bio->bi_private;
317 if (bio->bi_status && !parent->bi_status)
318 parent->bi_status = bio->bi_status;
323 static void bio_chain_endio(struct bio *bio)
325 bio_endio(__bio_chain_endio(bio));
329 * bio_chain - chain bio completions
330 * @bio: the target bio
331 * @parent: the parent bio of @bio
333 * The caller won't have a bi_end_io called when @bio completes - instead,
334 * @parent's bi_end_io won't be called until both @parent and @bio have
335 * completed; the chained bio will also be freed when it completes.
337 * The caller must not set bi_private or bi_end_io in @bio.
339 void bio_chain(struct bio *bio, struct bio *parent)
341 BUG_ON(bio->bi_private || bio->bi_end_io);
343 bio->bi_private = parent;
344 bio->bi_end_io = bio_chain_endio;
345 bio_inc_remaining(parent);
347 EXPORT_SYMBOL(bio_chain);
349 struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
350 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
352 struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
361 EXPORT_SYMBOL_GPL(blk_next_bio);
363 static void bio_alloc_rescue(struct work_struct *work)
365 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
369 spin_lock(&bs->rescue_lock);
370 bio = bio_list_pop(&bs->rescue_list);
371 spin_unlock(&bs->rescue_lock);
376 submit_bio_noacct(bio);
380 static void punt_bios_to_rescuer(struct bio_set *bs)
382 struct bio_list punt, nopunt;
385 if (WARN_ON_ONCE(!bs->rescue_workqueue))
388 * In order to guarantee forward progress we must punt only bios that
389 * were allocated from this bio_set; otherwise, if there was a bio on
390 * there for a stacking driver higher up in the stack, processing it
391 * could require allocating bios from this bio_set, and doing that from
392 * our own rescuer would be bad.
394 * Since bio lists are singly linked, pop them all instead of trying to
395 * remove from the middle of the list:
398 bio_list_init(&punt);
399 bio_list_init(&nopunt);
401 while ((bio = bio_list_pop(¤t->bio_list[0])))
402 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
403 current->bio_list[0] = nopunt;
405 bio_list_init(&nopunt);
406 while ((bio = bio_list_pop(¤t->bio_list[1])))
407 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
408 current->bio_list[1] = nopunt;
410 spin_lock(&bs->rescue_lock);
411 bio_list_merge(&bs->rescue_list, &punt);
412 spin_unlock(&bs->rescue_lock);
414 queue_work(bs->rescue_workqueue, &bs->rescue_work);
417 static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
421 /* cache->free_list must be empty */
422 if (WARN_ON_ONCE(cache->free_list))
425 local_irq_save(flags);
426 cache->free_list = cache->free_list_irq;
427 cache->free_list_irq = NULL;
428 cache->nr += cache->nr_irq;
430 local_irq_restore(flags);
433 static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
434 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
437 struct bio_alloc_cache *cache;
440 cache = per_cpu_ptr(bs->cache, get_cpu());
441 if (!cache->free_list) {
442 if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
443 bio_alloc_irq_cache_splice(cache);
444 if (!cache->free_list) {
449 bio = cache->free_list;
450 cache->free_list = bio->bi_next;
454 bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
460 * bio_alloc_bioset - allocate a bio for I/O
461 * @bdev: block device to allocate the bio for (can be %NULL)
462 * @nr_vecs: number of bvecs to pre-allocate
463 * @opf: operation and flags for bio
464 * @gfp_mask: the GFP_* mask given to the slab allocator
465 * @bs: the bio_set to allocate from.
467 * Allocate a bio from the mempools in @bs.
469 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
470 * allocate a bio. This is due to the mempool guarantees. To make this work,
471 * callers must never allocate more than 1 bio at a time from the general pool.
472 * Callers that need to allocate more than 1 bio must always submit the
473 * previously allocated bio for IO before attempting to allocate a new one.
474 * Failure to do so can cause deadlocks under memory pressure.
476 * Note that when running under submit_bio_noacct() (i.e. any block driver),
477 * bios are not submitted until after you return - see the code in
478 * submit_bio_noacct() that converts recursion into iteration, to prevent
481 * This would normally mean allocating multiple bios under submit_bio_noacct()
482 * would be susceptible to deadlocks, but we have
483 * deadlock avoidance code that resubmits any blocked bios from a rescuer
486 * However, we do not guarantee forward progress for allocations from other
487 * mempools. Doing multiple allocations from the same mempool under
488 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
489 * for per bio allocations.
491 * Returns: Pointer to new bio on success, NULL on failure.
493 struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
494 blk_opf_t opf, gfp_t gfp_mask,
497 gfp_t saved_gfp = gfp_mask;
501 /* should not use nobvec bioset for nr_vecs > 0 */
502 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
505 if (opf & REQ_ALLOC_CACHE) {
506 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
507 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
512 * No cached bio available, bio returned below marked with
513 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
516 opf &= ~REQ_ALLOC_CACHE;
521 * submit_bio_noacct() converts recursion to iteration; this means if
522 * we're running beneath it, any bios we allocate and submit will not be
523 * submitted (and thus freed) until after we return.
525 * This exposes us to a potential deadlock if we allocate multiple bios
526 * from the same bio_set() while running underneath submit_bio_noacct().
527 * If we were to allocate multiple bios (say a stacking block driver
528 * that was splitting bios), we would deadlock if we exhausted the
531 * We solve this, and guarantee forward progress, with a rescuer
532 * workqueue per bio_set. If we go to allocate and there are bios on
533 * current->bio_list, we first try the allocation without
534 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
535 * blocking to the rescuer workqueue before we retry with the original
538 if (current->bio_list &&
539 (!bio_list_empty(¤t->bio_list[0]) ||
540 !bio_list_empty(¤t->bio_list[1])) &&
541 bs->rescue_workqueue)
542 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
544 p = mempool_alloc(&bs->bio_pool, gfp_mask);
545 if (!p && gfp_mask != saved_gfp) {
546 punt_bios_to_rescuer(bs);
547 gfp_mask = saved_gfp;
548 p = mempool_alloc(&bs->bio_pool, gfp_mask);
552 if (!mempool_is_saturated(&bs->bio_pool))
553 opf &= ~REQ_ALLOC_CACHE;
555 bio = p + bs->front_pad;
556 if (nr_vecs > BIO_INLINE_VECS) {
557 struct bio_vec *bvl = NULL;
559 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
560 if (!bvl && gfp_mask != saved_gfp) {
561 punt_bios_to_rescuer(bs);
562 gfp_mask = saved_gfp;
563 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
568 bio_init(bio, bdev, bvl, nr_vecs, opf);
569 } else if (nr_vecs) {
570 bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
572 bio_init(bio, bdev, NULL, 0, opf);
579 mempool_free(p, &bs->bio_pool);
582 EXPORT_SYMBOL(bio_alloc_bioset);
585 * bio_kmalloc - kmalloc a bio
586 * @nr_vecs: number of bio_vecs to allocate
587 * @gfp_mask: the GFP_* mask given to the slab allocator
589 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
590 * using bio_init() before use. To free a bio returned from this function use
591 * kfree() after calling bio_uninit(). A bio returned from this function can
592 * be reused by calling bio_uninit() before calling bio_init() again.
594 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
595 * function are not backed by a mempool can fail. Do not use this function
596 * for allocations in the file system I/O path.
598 * Returns: Pointer to new bio on success, NULL on failure.
600 struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
604 if (nr_vecs > UIO_MAXIOV)
606 return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
608 EXPORT_SYMBOL(bio_kmalloc);
610 void zero_fill_bio(struct bio *bio)
613 struct bvec_iter iter;
615 bio_for_each_segment(bv, bio, iter)
618 EXPORT_SYMBOL(zero_fill_bio);
621 * bio_truncate - truncate the bio to small size of @new_size
622 * @bio: the bio to be truncated
623 * @new_size: new size for truncating the bio
626 * Truncate the bio to new size of @new_size. If bio_op(bio) is
627 * REQ_OP_READ, zero the truncated part. This function should only
628 * be used for handling corner cases, such as bio eod.
630 static void bio_truncate(struct bio *bio, unsigned new_size)
633 struct bvec_iter iter;
634 unsigned int done = 0;
635 bool truncated = false;
637 if (new_size >= bio->bi_iter.bi_size)
640 if (bio_op(bio) != REQ_OP_READ)
643 bio_for_each_segment(bv, bio, iter) {
644 if (done + bv.bv_len > new_size) {
648 offset = new_size - done;
651 zero_user(bv.bv_page, bv.bv_offset + offset,
660 * Don't touch bvec table here and make it really immutable, since
661 * fs bio user has to retrieve all pages via bio_for_each_segment_all
662 * in its .end_bio() callback.
664 * It is enough to truncate bio by updating .bi_size since we can make
665 * correct bvec with the updated .bi_size for drivers.
667 bio->bi_iter.bi_size = new_size;
671 * guard_bio_eod - truncate a BIO to fit the block device
672 * @bio: bio to truncate
674 * This allows us to do IO even on the odd last sectors of a device, even if the
675 * block size is some multiple of the physical sector size.
677 * We'll just truncate the bio to the size of the device, and clear the end of
678 * the buffer head manually. Truly out-of-range accesses will turn into actual
679 * I/O errors, this only handles the "we need to be able to do I/O at the final
682 void guard_bio_eod(struct bio *bio)
684 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
690 * If the *whole* IO is past the end of the device,
691 * let it through, and the IO layer will turn it into
694 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
697 maxsector -= bio->bi_iter.bi_sector;
698 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
701 bio_truncate(bio, maxsector << 9);
704 static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
710 while ((bio = cache->free_list) != NULL) {
711 cache->free_list = bio->bi_next;
720 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
723 nr -= __bio_alloc_cache_prune(cache, nr);
724 if (!READ_ONCE(cache->free_list)) {
725 bio_alloc_irq_cache_splice(cache);
726 __bio_alloc_cache_prune(cache, nr);
730 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
734 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
736 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
738 bio_alloc_cache_prune(cache, -1U);
743 static void bio_alloc_cache_destroy(struct bio_set *bs)
750 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
751 for_each_possible_cpu(cpu) {
752 struct bio_alloc_cache *cache;
754 cache = per_cpu_ptr(bs->cache, cpu);
755 bio_alloc_cache_prune(cache, -1U);
757 free_percpu(bs->cache);
761 static inline void bio_put_percpu_cache(struct bio *bio)
763 struct bio_alloc_cache *cache;
765 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
766 if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) {
774 if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
775 bio->bi_next = cache->free_list;
776 cache->free_list = bio;
781 local_irq_save(flags);
782 bio->bi_next = cache->free_list_irq;
783 cache->free_list_irq = bio;
785 local_irq_restore(flags);
791 * bio_put - release a reference to a bio
792 * @bio: bio to release reference to
795 * Put a reference to a &struct bio, either one you have gotten with
796 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
798 void bio_put(struct bio *bio)
800 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
801 BUG_ON(!atomic_read(&bio->__bi_cnt));
802 if (!atomic_dec_and_test(&bio->__bi_cnt))
805 if (bio->bi_opf & REQ_ALLOC_CACHE)
806 bio_put_percpu_cache(bio);
810 EXPORT_SYMBOL(bio_put);
812 static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
814 bio_set_flag(bio, BIO_CLONED);
815 bio->bi_ioprio = bio_src->bi_ioprio;
816 bio->bi_iter = bio_src->bi_iter;
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);
825 if (bio_crypt_clone(bio, bio_src, gfp) < 0)
827 if (bio_integrity(bio_src) &&
828 bio_integrity_clone(bio, bio_src, gfp) < 0)
834 * bio_alloc_clone - clone a bio that shares the original bio's biovec
835 * @bdev: block_device to clone onto
836 * @bio_src: bio to clone from
837 * @gfp: allocation priority
838 * @bs: bio_set to allocate from
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.
843 * The caller must ensure that the return bio is not freed before @bio_src.
845 struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
846 gfp_t gfp, struct bio_set *bs)
850 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
854 if (__bio_clone(bio, bio_src, gfp) < 0) {
858 bio->bi_io_vec = bio_src->bi_io_vec;
862 EXPORT_SYMBOL(bio_alloc_clone);
865 * bio_init_clone - clone a bio that shares the original bio's biovec
866 * @bdev: block_device to clone onto
867 * @bio: bio to clone into
868 * @bio_src: bio to clone from
869 * @gfp: allocation priority
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.
874 * The caller must ensure that @bio_src is not freed before @bio.
876 int bio_init_clone(struct block_device *bdev, struct bio *bio,
877 struct bio *bio_src, gfp_t gfp)
881 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
882 ret = __bio_clone(bio, bio_src, gfp);
887 EXPORT_SYMBOL(bio_init_clone);
890 * bio_full - check if the bio is full
892 * @len: length of one segment to be added
894 * Return true if @bio is full and one segment with @len bytes can't be
895 * added to the bio, otherwise return false
897 static inline bool bio_full(struct bio *bio, unsigned len)
899 if (bio->bi_vcnt >= bio->bi_max_vecs)
901 if (bio->bi_iter.bi_size > UINT_MAX - len)
906 static inline bool page_is_mergeable(const struct bio_vec *bv,
907 struct page *page, unsigned int len, unsigned int off,
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;
912 phys_addr_t page_addr = page_to_phys(page);
914 if (vec_end_addr + 1 != page_addr + off)
916 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
918 if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
921 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
924 else if (IS_ENABLED(CONFIG_KMSAN))
926 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
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
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
941 * Warn if (@len, @off) crosses pages in case that @same_page is true.
943 * Return %true on success or %false on failure.
945 static bool __bio_try_merge_page(struct bio *bio, struct page *page,
946 unsigned int len, unsigned int off, bool *same_page)
948 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
951 if (bio->bi_vcnt > 0) {
952 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
954 if (page_is_mergeable(bv, page, len, off, same_page)) {
955 if (bio->bi_iter.bi_size > UINT_MAX - len) {
960 bio->bi_iter.bi_size += len;
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.
972 static 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)
976 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
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;
981 if ((addr1 | mask) != (addr2 | mask))
983 if (bv->bv_len + len > queue_max_segment_size(q))
985 return __bio_try_merge_page(bio, page, len, offset, same_page);
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
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
998 * Add a page to a bio while respecting the hardware max_sectors, max_segment
999 * and gap limitations.
1001 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
1002 struct page *page, unsigned int len, unsigned int offset,
1003 unsigned int max_sectors, bool *same_page)
1005 struct bio_vec *bvec;
1007 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1010 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
1013 if (bio->bi_vcnt > 0) {
1014 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
1018 * If the queue doesn't support SG gaps and adding this segment
1019 * would create a gap, disallow it.
1021 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
1022 if (bvec_gap_to_prev(&q->limits, bvec, offset))
1026 if (bio_full(bio, len))
1029 if (bio->bi_vcnt >= queue_max_segments(q))
1032 bvec = &bio->bi_io_vec[bio->bi_vcnt];
1033 bvec->bv_page = page;
1035 bvec->bv_offset = offset;
1037 bio->bi_iter.bi_size += len;
1042 * bio_add_pc_page - attempt to add page to passthrough bio
1043 * @q: the target queue
1044 * @bio: destination bio
1045 * @page: page to add
1046 * @len: vec entry length
1047 * @offset: vec entry offset
1049 * Attempt to add a page to the bio_vec maplist. This can fail for a
1050 * number of reasons, such as the bio being full or target block device
1051 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1052 * so it is always possible to add a single page to an empty bio.
1054 * This should only be used by passthrough bios.
1056 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1057 struct page *page, unsigned int len, unsigned int offset)
1059 bool same_page = false;
1060 return bio_add_hw_page(q, bio, page, len, offset,
1061 queue_max_hw_sectors(q), &same_page);
1063 EXPORT_SYMBOL(bio_add_pc_page);
1066 * bio_add_zone_append_page - attempt to add page to zone-append bio
1067 * @bio: destination bio
1068 * @page: page to add
1069 * @len: vec entry length
1070 * @offset: vec entry offset
1072 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1073 * for a zone-append request. This can fail for a number of reasons, such as the
1074 * bio being full or the target block device is not a zoned block device or
1075 * other limitations of the target block device. The target block device must
1076 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1079 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1081 int bio_add_zone_append_page(struct bio *bio, struct page *page,
1082 unsigned int len, unsigned int offset)
1084 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1085 bool same_page = false;
1087 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1090 if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1093 return bio_add_hw_page(q, bio, page, len, offset,
1094 queue_max_zone_append_sectors(q), &same_page);
1096 EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1099 * __bio_add_page - add page(s) to a bio in a new segment
1100 * @bio: destination bio
1101 * @page: start page to add
1102 * @len: length of the data to add, may cross pages
1103 * @off: offset of the data relative to @page, may cross pages
1105 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1106 * that @bio has space for another bvec.
1108 void __bio_add_page(struct bio *bio, struct page *page,
1109 unsigned int len, unsigned int off)
1111 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
1113 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1114 WARN_ON_ONCE(bio_full(bio, len));
1117 bv->bv_offset = off;
1120 bio->bi_iter.bi_size += len;
1123 EXPORT_SYMBOL_GPL(__bio_add_page);
1126 * bio_add_page - attempt to add page(s) to bio
1127 * @bio: destination bio
1128 * @page: start page to add
1129 * @len: vec entry length, may cross pages
1130 * @offset: vec entry offset relative to @page, may cross pages
1132 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1133 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1135 int bio_add_page(struct bio *bio, struct page *page,
1136 unsigned int len, unsigned int offset)
1138 bool same_page = false;
1140 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1141 if (bio_full(bio, len))
1143 __bio_add_page(bio, page, len, offset);
1147 EXPORT_SYMBOL(bio_add_page);
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.
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.
1161 * Return: Whether the addition was successful.
1163 bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1166 if (len > UINT_MAX || off > UINT_MAX)
1168 return bio_add_page(bio, &folio->page, len, off) > 0;
1171 void __bio_release_pages(struct bio *bio, bool mark_dirty)
1173 struct bvec_iter_all iter_all;
1174 struct bio_vec *bvec;
1176 bio_for_each_segment_all(bvec, bio, iter_all) {
1177 if (mark_dirty && !PageCompound(bvec->bv_page))
1178 set_page_dirty_lock(bvec->bv_page);
1179 put_page(bvec->bv_page);
1182 EXPORT_SYMBOL_GPL(__bio_release_pages);
1184 void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1186 size_t size = iov_iter_count(iter);
1188 WARN_ON_ONCE(bio->bi_max_vecs);
1190 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1191 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1192 size_t max_sectors = queue_max_zone_append_sectors(q);
1194 size = min(size, max_sectors << SECTOR_SHIFT);
1197 bio->bi_vcnt = iter->nr_segs;
1198 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1199 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1200 bio->bi_iter.bi_size = size;
1201 bio_set_flag(bio, BIO_NO_PAGE_REF);
1202 bio_set_flag(bio, BIO_CLONED);
1205 static int bio_iov_add_page(struct bio *bio, struct page *page,
1206 unsigned int len, unsigned int offset)
1208 bool same_page = false;
1210 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1211 __bio_add_page(bio, page, len, offset);
1220 static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1221 unsigned int len, unsigned int offset)
1223 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1224 bool same_page = false;
1226 if (bio_add_hw_page(q, bio, page, len, offset,
1227 queue_max_zone_append_sectors(q), &same_page) != len)
1234 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1237 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1238 * @bio: bio to add pages to
1239 * @iter: iov iterator describing the region to be mapped
1241 * Pins pages from *iter and appends them to @bio's bvec array. The
1242 * pages will have to be released using put_page() when done.
1243 * For multi-segment *iter, this function only adds pages from the
1244 * next non-empty segment of the iov iterator.
1246 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1248 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1249 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1250 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1251 struct page **pages = (struct page **)bv;
1252 unsigned int gup_flags = 0;
1254 unsigned len, i = 0;
1255 size_t offset, trim;
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.
1263 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1264 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1266 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1267 gup_flags |= FOLL_PCI_P2PDMA;
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.
1276 size = iov_iter_get_pages(iter, pages,
1277 UINT_MAX - bio->bi_iter.bi_size,
1278 nr_pages, &offset, gup_flags);
1279 if (unlikely(size <= 0))
1280 return size ? size : -EFAULT;
1282 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1284 trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1285 iov_iter_revert(iter, trim);
1288 if (unlikely(!size)) {
1293 for (left = size, i = 0; left > 0; left -= len, i++) {
1294 struct page *page = pages[i];
1296 len = min_t(size_t, PAGE_SIZE - offset, left);
1297 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1298 ret = bio_iov_add_zone_append_page(bio, page, len,
1303 bio_iov_add_page(bio, page, len, offset);
1308 iov_iter_revert(iter, left);
1310 while (i < nr_pages)
1311 put_page(pages[i++]);
1317 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1318 * @bio: bio to add pages to
1319 * @iter: iov iterator describing the region to be added
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
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.
1331 * The function tries, but does not guarantee, to pin as many pages as
1332 * fit into the bio, or are requested in @iter, whatever is smaller. If
1333 * MM encounters an error pinning the requested pages, it stops. Error
1334 * is returned only if 0 pages could be pinned.
1336 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1340 if (iov_iter_is_bvec(iter)) {
1341 bio_iov_bvec_set(bio, iter);
1342 iov_iter_advance(iter, bio->bi_iter.bi_size);
1347 ret = __bio_iov_iter_get_pages(bio, iter);
1348 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1350 return bio->bi_vcnt ? 0 : ret;
1352 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1354 static void submit_bio_wait_endio(struct bio *bio)
1356 complete(bio->bi_private);
1360 * submit_bio_wait - submit a bio, and wait until it completes
1361 * @bio: The &struct bio which describes the I/O
1363 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1364 * bio_endio() on failure.
1366 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1367 * result in bio reference to be consumed. The caller must drop the reference
1370 int submit_bio_wait(struct bio *bio)
1372 DECLARE_COMPLETION_ONSTACK_MAP(done,
1373 bio->bi_bdev->bd_disk->lockdep_map);
1374 unsigned long hang_check;
1376 bio->bi_private = &done;
1377 bio->bi_end_io = submit_bio_wait_endio;
1378 bio->bi_opf |= REQ_SYNC;
1381 /* Prevent hang_check timer from firing at us during very long I/O */
1382 hang_check = sysctl_hung_task_timeout_secs;
1384 while (!wait_for_completion_io_timeout(&done,
1385 hang_check * (HZ/2)))
1388 wait_for_completion_io(&done);
1390 return blk_status_to_errno(bio->bi_status);
1392 EXPORT_SYMBOL(submit_bio_wait);
1394 void __bio_advance(struct bio *bio, unsigned bytes)
1396 if (bio_integrity(bio))
1397 bio_integrity_advance(bio, bytes);
1399 bio_crypt_advance(bio, bytes);
1400 bio_advance_iter(bio, &bio->bi_iter, bytes);
1402 EXPORT_SYMBOL(__bio_advance);
1405 * bio_copy_data - copy contents of data buffers from one bio to another
1407 * @dst: destination bio
1409 * Stops when it reaches the end of either @src or @dst - that is, copies
1410 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1412 void bio_copy_data(struct bio *dst, struct bio *src)
1414 struct bvec_iter src_iter = src->bi_iter;
1415 struct bvec_iter dst_iter = dst->bi_iter;
1417 while (src_iter.bi_size && dst_iter.bi_size) {
1418 struct bio_vec src_bv = bio_iter_iovec(src, src_iter);
1419 struct bio_vec dst_bv = bio_iter_iovec(dst, dst_iter);
1420 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1421 void *src_buf = bvec_kmap_local(&src_bv);
1422 void *dst_buf = bvec_kmap_local(&dst_bv);
1424 memcpy(dst_buf, src_buf, bytes);
1426 kunmap_local(dst_buf);
1427 kunmap_local(src_buf);
1429 bio_advance_iter_single(src, &src_iter, bytes);
1430 bio_advance_iter_single(dst, &dst_iter, bytes);
1433 EXPORT_SYMBOL(bio_copy_data);
1435 void bio_free_pages(struct bio *bio)
1437 struct bio_vec *bvec;
1438 struct bvec_iter_all iter_all;
1440 bio_for_each_segment_all(bvec, bio, iter_all)
1441 __free_page(bvec->bv_page);
1443 EXPORT_SYMBOL(bio_free_pages);
1446 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1447 * for performing direct-IO in BIOs.
1449 * The problem is that we cannot run set_page_dirty() from interrupt context
1450 * because the required locks are not interrupt-safe. So what we can do is to
1451 * mark the pages dirty _before_ performing IO. And in interrupt context,
1452 * check that the pages are still dirty. If so, fine. If not, redirty them
1453 * in process context.
1455 * We special-case compound pages here: normally this means reads into hugetlb
1456 * pages. The logic in here doesn't really work right for compound pages
1457 * because the VM does not uniformly chase down the head page in all cases.
1458 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1459 * handle them at all. So we skip compound pages here at an early stage.
1461 * Note that this code is very hard to test under normal circumstances because
1462 * direct-io pins the pages with get_user_pages(). This makes
1463 * is_page_cache_freeable return false, and the VM will not clean the pages.
1464 * But other code (eg, flusher threads) could clean the pages if they are mapped
1467 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1468 * deferred bio dirtying paths.
1472 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1474 void bio_set_pages_dirty(struct bio *bio)
1476 struct bio_vec *bvec;
1477 struct bvec_iter_all iter_all;
1479 bio_for_each_segment_all(bvec, bio, iter_all) {
1480 if (!PageCompound(bvec->bv_page))
1481 set_page_dirty_lock(bvec->bv_page);
1486 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1487 * If they are, then fine. If, however, some pages are clean then they must
1488 * have been written out during the direct-IO read. So we take another ref on
1489 * the BIO and re-dirty the pages in process context.
1491 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1492 * here on. It will run one put_page() against each page and will run one
1493 * bio_put() against the BIO.
1496 static void bio_dirty_fn(struct work_struct *work);
1498 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1499 static DEFINE_SPINLOCK(bio_dirty_lock);
1500 static struct bio *bio_dirty_list;
1503 * This runs in process context
1505 static void bio_dirty_fn(struct work_struct *work)
1507 struct bio *bio, *next;
1509 spin_lock_irq(&bio_dirty_lock);
1510 next = bio_dirty_list;
1511 bio_dirty_list = NULL;
1512 spin_unlock_irq(&bio_dirty_lock);
1514 while ((bio = next) != NULL) {
1515 next = bio->bi_private;
1517 bio_release_pages(bio, true);
1522 void bio_check_pages_dirty(struct bio *bio)
1524 struct bio_vec *bvec;
1525 unsigned long flags;
1526 struct bvec_iter_all iter_all;
1528 bio_for_each_segment_all(bvec, bio, iter_all) {
1529 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1533 bio_release_pages(bio, false);
1537 spin_lock_irqsave(&bio_dirty_lock, flags);
1538 bio->bi_private = bio_dirty_list;
1539 bio_dirty_list = bio;
1540 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1541 schedule_work(&bio_dirty_work);
1544 static inline bool bio_remaining_done(struct bio *bio)
1547 * If we're not chaining, then ->__bi_remaining is always 1 and
1548 * we always end io on the first invocation.
1550 if (!bio_flagged(bio, BIO_CHAIN))
1553 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1555 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1556 bio_clear_flag(bio, BIO_CHAIN);
1564 * bio_endio - end I/O on a bio
1568 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1569 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1570 * bio unless they own it and thus know that it has an end_io function.
1572 * bio_endio() can be called several times on a bio that has been chained
1573 * using bio_chain(). The ->bi_end_io() function will only be called the
1576 void bio_endio(struct bio *bio)
1579 if (!bio_remaining_done(bio))
1581 if (!bio_integrity_endio(bio))
1584 rq_qos_done_bio(bio);
1586 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1587 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1588 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1592 * Need to have a real endio function for chained bios, otherwise
1593 * various corner cases will break (like stacking block devices that
1594 * save/restore bi_end_io) - however, we want to avoid unbounded
1595 * recursion and blowing the stack. Tail call optimization would
1596 * handle this, but compiling with frame pointers also disables
1597 * gcc's sibling call optimization.
1599 if (bio->bi_end_io == bio_chain_endio) {
1600 bio = __bio_chain_endio(bio);
1604 blk_throtl_bio_endio(bio);
1605 /* release cgroup info */
1608 bio->bi_end_io(bio);
1610 EXPORT_SYMBOL(bio_endio);
1613 * bio_split - split a bio
1614 * @bio: bio to split
1615 * @sectors: number of sectors to split from the front of @bio
1617 * @bs: bio set to allocate from
1619 * Allocates and returns a new bio which represents @sectors from the start of
1620 * @bio, and updates @bio to represent the remaining sectors.
1622 * Unless this is a discard request the newly allocated bio will point
1623 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1624 * neither @bio nor @bs are freed before the split bio.
1626 struct bio *bio_split(struct bio *bio, int sectors,
1627 gfp_t gfp, struct bio_set *bs)
1631 BUG_ON(sectors <= 0);
1632 BUG_ON(sectors >= bio_sectors(bio));
1634 /* Zone append commands cannot be split */
1635 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1638 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1642 split->bi_iter.bi_size = sectors << 9;
1644 if (bio_integrity(split))
1645 bio_integrity_trim(split);
1647 bio_advance(bio, split->bi_iter.bi_size);
1649 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1650 bio_set_flag(split, BIO_TRACE_COMPLETION);
1654 EXPORT_SYMBOL(bio_split);
1657 * bio_trim - trim a bio
1659 * @offset: number of sectors to trim from the front of @bio
1660 * @size: size we want to trim @bio to, in sectors
1662 * This function is typically used for bios that are cloned and submitted
1663 * to the underlying device in parts.
1665 void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1667 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1668 offset + size > bio_sectors(bio)))
1672 if (offset == 0 && size == bio->bi_iter.bi_size)
1675 bio_advance(bio, offset << 9);
1676 bio->bi_iter.bi_size = size;
1678 if (bio_integrity(bio))
1679 bio_integrity_trim(bio);
1681 EXPORT_SYMBOL_GPL(bio_trim);
1684 * create memory pools for biovec's in a bio_set.
1685 * use the global biovec slabs created for general use.
1687 int biovec_init_pool(mempool_t *pool, int pool_entries)
1689 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1691 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1695 * bioset_exit - exit a bioset initialized with bioset_init()
1697 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1700 void bioset_exit(struct bio_set *bs)
1702 bio_alloc_cache_destroy(bs);
1703 if (bs->rescue_workqueue)
1704 destroy_workqueue(bs->rescue_workqueue);
1705 bs->rescue_workqueue = NULL;
1707 mempool_exit(&bs->bio_pool);
1708 mempool_exit(&bs->bvec_pool);
1710 bioset_integrity_free(bs);
1713 bs->bio_slab = NULL;
1715 EXPORT_SYMBOL(bioset_exit);
1718 * bioset_init - Initialize a bio_set
1719 * @bs: pool to initialize
1720 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1721 * @front_pad: Number of bytes to allocate in front of the returned bio
1722 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1723 * and %BIOSET_NEED_RESCUER
1726 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1727 * to ask for a number of bytes to be allocated in front of the bio.
1728 * Front pad allocation is useful for embedding the bio inside
1729 * another structure, to avoid allocating extra data to go with the bio.
1730 * Note that the bio must be embedded at the END of that structure always,
1731 * or things will break badly.
1732 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1733 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1734 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1735 * to dispatch queued requests when the mempool runs out of space.
1738 int bioset_init(struct bio_set *bs,
1739 unsigned int pool_size,
1740 unsigned int front_pad,
1743 bs->front_pad = front_pad;
1744 if (flags & BIOSET_NEED_BVECS)
1745 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1749 spin_lock_init(&bs->rescue_lock);
1750 bio_list_init(&bs->rescue_list);
1751 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1753 bs->bio_slab = bio_find_or_create_slab(bs);
1757 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1760 if ((flags & BIOSET_NEED_BVECS) &&
1761 biovec_init_pool(&bs->bvec_pool, pool_size))
1764 if (flags & BIOSET_NEED_RESCUER) {
1765 bs->rescue_workqueue = alloc_workqueue("bioset",
1767 if (!bs->rescue_workqueue)
1770 if (flags & BIOSET_PERCPU_CACHE) {
1771 bs->cache = alloc_percpu(struct bio_alloc_cache);
1774 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1782 EXPORT_SYMBOL(bioset_init);
1784 static int __init init_bio(void)
1788 bio_integrity_init();
1790 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1791 struct biovec_slab *bvs = bvec_slabs + i;
1793 bvs->slab = kmem_cache_create(bvs->name,
1794 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1795 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1798 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1801 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1802 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1803 panic("bio: can't allocate bios\n");
1805 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1806 panic("bio: can't create integrity pool\n");
1810 subsys_initcall(init_bio);