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/blk-crypto.h>
20 #include <linux/xarray.h>
22 #include <trace/events/block.h>
24 #include "blk-rq-qos.h"
25 #include "blk-cgroup.h"
27 #define ALLOC_CACHE_THRESHOLD 16
28 #define ALLOC_CACHE_MAX 256
30 struct bio_alloc_cache {
31 struct bio *free_list;
32 struct bio *free_list_irq;
37 static struct biovec_slab {
40 struct kmem_cache *slab;
41 } bvec_slabs[] __read_mostly = {
42 { .nr_vecs = 16, .name = "biovec-16" },
43 { .nr_vecs = 64, .name = "biovec-64" },
44 { .nr_vecs = 128, .name = "biovec-128" },
45 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
48 static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
51 /* smaller bios use inline vecs */
53 return &bvec_slabs[0];
55 return &bvec_slabs[1];
57 return &bvec_slabs[2];
58 case 129 ... BIO_MAX_VECS:
59 return &bvec_slabs[3];
67 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
68 * IO code that does not need private memory pools.
70 struct bio_set fs_bio_set;
71 EXPORT_SYMBOL(fs_bio_set);
74 * Our slab pool management
77 struct kmem_cache *slab;
78 unsigned int slab_ref;
79 unsigned int slab_size;
82 static DEFINE_MUTEX(bio_slab_lock);
83 static DEFINE_XARRAY(bio_slabs);
85 static struct bio_slab *create_bio_slab(unsigned int size)
87 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
92 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
93 bslab->slab = kmem_cache_create(bslab->name, size,
94 ARCH_KMALLOC_MINALIGN,
95 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
100 bslab->slab_size = size;
102 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
105 kmem_cache_destroy(bslab->slab);
112 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
114 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
117 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
119 unsigned int size = bs_bio_slab_size(bs);
120 struct bio_slab *bslab;
122 mutex_lock(&bio_slab_lock);
123 bslab = xa_load(&bio_slabs, size);
127 bslab = create_bio_slab(size);
128 mutex_unlock(&bio_slab_lock);
135 static void bio_put_slab(struct bio_set *bs)
137 struct bio_slab *bslab = NULL;
138 unsigned int slab_size = bs_bio_slab_size(bs);
140 mutex_lock(&bio_slab_lock);
142 bslab = xa_load(&bio_slabs, slab_size);
143 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
146 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
148 WARN_ON(!bslab->slab_ref);
150 if (--bslab->slab_ref)
153 xa_erase(&bio_slabs, slab_size);
155 kmem_cache_destroy(bslab->slab);
159 mutex_unlock(&bio_slab_lock);
162 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
164 BUG_ON(nr_vecs > BIO_MAX_VECS);
166 if (nr_vecs == BIO_MAX_VECS)
167 mempool_free(bv, pool);
168 else if (nr_vecs > BIO_INLINE_VECS)
169 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
173 * Make the first allocation restricted and don't dump info on allocation
174 * failures, since we'll fall back to the mempool in case of failure.
176 static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
178 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
179 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
182 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
185 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
187 if (WARN_ON_ONCE(!bvs))
191 * Upgrade the nr_vecs request to take full advantage of the allocation.
192 * We also rely on this in the bvec_free path.
194 *nr_vecs = bvs->nr_vecs;
197 * Try a slab allocation first for all smaller allocations. If that
198 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
199 * The mempool is sized to handle up to BIO_MAX_VECS entries.
201 if (*nr_vecs < BIO_MAX_VECS) {
204 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
205 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
207 *nr_vecs = BIO_MAX_VECS;
210 return mempool_alloc(pool, gfp_mask);
213 void bio_uninit(struct bio *bio)
215 #ifdef CONFIG_BLK_CGROUP
217 blkg_put(bio->bi_blkg);
221 if (bio_integrity(bio))
222 bio_integrity_free(bio);
224 bio_crypt_free_ctx(bio);
226 EXPORT_SYMBOL(bio_uninit);
228 static void bio_free(struct bio *bio)
230 struct bio_set *bs = bio->bi_pool;
236 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
237 mempool_free(p - bs->front_pad, &bs->bio_pool);
241 * Users of this function have their own bio allocation. Subsequently,
242 * they must remember to pair any call to bio_init() with bio_uninit()
243 * when IO has completed, or when the bio is released.
245 void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
246 unsigned short max_vecs, blk_opf_t opf)
254 bio->bi_iter.bi_sector = 0;
255 bio->bi_iter.bi_size = 0;
256 bio->bi_iter.bi_idx = 0;
257 bio->bi_iter.bi_bvec_done = 0;
258 bio->bi_end_io = NULL;
259 bio->bi_private = NULL;
260 #ifdef CONFIG_BLK_CGROUP
262 bio->bi_issue.value = 0;
264 bio_associate_blkg(bio);
265 #ifdef CONFIG_BLK_CGROUP_IOCOST
266 bio->bi_iocost_cost = 0;
269 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
270 bio->bi_crypt_context = NULL;
272 #ifdef CONFIG_BLK_DEV_INTEGRITY
273 bio->bi_integrity = NULL;
277 atomic_set(&bio->__bi_remaining, 1);
278 atomic_set(&bio->__bi_cnt, 1);
279 bio->bi_cookie = BLK_QC_T_NONE;
281 bio->bi_max_vecs = max_vecs;
282 bio->bi_io_vec = table;
285 EXPORT_SYMBOL(bio_init);
288 * bio_reset - reinitialize a bio
290 * @bdev: block device to use the bio for
291 * @opf: operation and flags for bio
294 * After calling bio_reset(), @bio will be in the same state as a freshly
295 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
296 * preserved are the ones that are initialized by bio_alloc_bioset(). See
297 * comment in struct bio.
299 void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
302 memset(bio, 0, BIO_RESET_BYTES);
303 atomic_set(&bio->__bi_remaining, 1);
306 bio_associate_blkg(bio);
309 EXPORT_SYMBOL(bio_reset);
311 static struct bio *__bio_chain_endio(struct bio *bio)
313 struct bio *parent = bio->bi_private;
315 if (bio->bi_status && !parent->bi_status)
316 parent->bi_status = bio->bi_status;
321 static void bio_chain_endio(struct bio *bio)
323 bio_endio(__bio_chain_endio(bio));
327 * bio_chain - chain bio completions
328 * @bio: the target bio
329 * @parent: the parent bio of @bio
331 * The caller won't have a bi_end_io called when @bio completes - instead,
332 * @parent's bi_end_io won't be called until both @parent and @bio have
333 * completed; the chained bio will also be freed when it completes.
335 * The caller must not set bi_private or bi_end_io in @bio.
337 void bio_chain(struct bio *bio, struct bio *parent)
339 BUG_ON(bio->bi_private || bio->bi_end_io);
341 bio->bi_private = parent;
342 bio->bi_end_io = bio_chain_endio;
343 bio_inc_remaining(parent);
345 EXPORT_SYMBOL(bio_chain);
347 struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
348 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
350 struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
359 EXPORT_SYMBOL_GPL(blk_next_bio);
361 static void bio_alloc_rescue(struct work_struct *work)
363 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
367 spin_lock(&bs->rescue_lock);
368 bio = bio_list_pop(&bs->rescue_list);
369 spin_unlock(&bs->rescue_lock);
374 submit_bio_noacct(bio);
378 static void punt_bios_to_rescuer(struct bio_set *bs)
380 struct bio_list punt, nopunt;
383 if (WARN_ON_ONCE(!bs->rescue_workqueue))
386 * In order to guarantee forward progress we must punt only bios that
387 * were allocated from this bio_set; otherwise, if there was a bio on
388 * there for a stacking driver higher up in the stack, processing it
389 * could require allocating bios from this bio_set, and doing that from
390 * our own rescuer would be bad.
392 * Since bio lists are singly linked, pop them all instead of trying to
393 * remove from the middle of the list:
396 bio_list_init(&punt);
397 bio_list_init(&nopunt);
399 while ((bio = bio_list_pop(¤t->bio_list[0])))
400 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
401 current->bio_list[0] = nopunt;
403 bio_list_init(&nopunt);
404 while ((bio = bio_list_pop(¤t->bio_list[1])))
405 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
406 current->bio_list[1] = nopunt;
408 spin_lock(&bs->rescue_lock);
409 bio_list_merge(&bs->rescue_list, &punt);
410 spin_unlock(&bs->rescue_lock);
412 queue_work(bs->rescue_workqueue, &bs->rescue_work);
415 static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
419 /* cache->free_list must be empty */
420 if (WARN_ON_ONCE(cache->free_list))
423 local_irq_save(flags);
424 cache->free_list = cache->free_list_irq;
425 cache->free_list_irq = NULL;
426 cache->nr += cache->nr_irq;
428 local_irq_restore(flags);
431 static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
432 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
435 struct bio_alloc_cache *cache;
438 cache = per_cpu_ptr(bs->cache, get_cpu());
439 if (!cache->free_list) {
440 if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
441 bio_alloc_irq_cache_splice(cache);
442 if (!cache->free_list) {
447 bio = cache->free_list;
448 cache->free_list = bio->bi_next;
452 bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
458 * bio_alloc_bioset - allocate a bio for I/O
459 * @bdev: block device to allocate the bio for (can be %NULL)
460 * @nr_vecs: number of bvecs to pre-allocate
461 * @opf: operation and flags for bio
462 * @gfp_mask: the GFP_* mask given to the slab allocator
463 * @bs: the bio_set to allocate from.
465 * Allocate a bio from the mempools in @bs.
467 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
468 * allocate a bio. This is due to the mempool guarantees. To make this work,
469 * callers must never allocate more than 1 bio at a time from the general pool.
470 * Callers that need to allocate more than 1 bio must always submit the
471 * previously allocated bio for IO before attempting to allocate a new one.
472 * Failure to do so can cause deadlocks under memory pressure.
474 * Note that when running under submit_bio_noacct() (i.e. any block driver),
475 * bios are not submitted until after you return - see the code in
476 * submit_bio_noacct() that converts recursion into iteration, to prevent
479 * This would normally mean allocating multiple bios under submit_bio_noacct()
480 * would be susceptible to deadlocks, but we have
481 * deadlock avoidance code that resubmits any blocked bios from a rescuer
484 * However, we do not guarantee forward progress for allocations from other
485 * mempools. Doing multiple allocations from the same mempool under
486 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
487 * for per bio allocations.
489 * Returns: Pointer to new bio on success, NULL on failure.
491 struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
492 blk_opf_t opf, gfp_t gfp_mask,
495 gfp_t saved_gfp = gfp_mask;
499 /* should not use nobvec bioset for nr_vecs > 0 */
500 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
503 if (opf & REQ_ALLOC_CACHE) {
504 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
505 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
510 * No cached bio available, bio returned below marked with
511 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
514 opf &= ~REQ_ALLOC_CACHE;
519 * submit_bio_noacct() converts recursion to iteration; this means if
520 * we're running beneath it, any bios we allocate and submit will not be
521 * submitted (and thus freed) until after we return.
523 * This exposes us to a potential deadlock if we allocate multiple bios
524 * from the same bio_set() while running underneath submit_bio_noacct().
525 * If we were to allocate multiple bios (say a stacking block driver
526 * that was splitting bios), we would deadlock if we exhausted the
529 * We solve this, and guarantee forward progress, with a rescuer
530 * workqueue per bio_set. If we go to allocate and there are bios on
531 * current->bio_list, we first try the allocation without
532 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
533 * blocking to the rescuer workqueue before we retry with the original
536 if (current->bio_list &&
537 (!bio_list_empty(¤t->bio_list[0]) ||
538 !bio_list_empty(¤t->bio_list[1])) &&
539 bs->rescue_workqueue)
540 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
542 p = mempool_alloc(&bs->bio_pool, gfp_mask);
543 if (!p && gfp_mask != saved_gfp) {
544 punt_bios_to_rescuer(bs);
545 gfp_mask = saved_gfp;
546 p = mempool_alloc(&bs->bio_pool, gfp_mask);
550 if (!mempool_is_saturated(&bs->bio_pool))
551 opf &= ~REQ_ALLOC_CACHE;
553 bio = p + bs->front_pad;
554 if (nr_vecs > BIO_INLINE_VECS) {
555 struct bio_vec *bvl = NULL;
557 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
558 if (!bvl && gfp_mask != saved_gfp) {
559 punt_bios_to_rescuer(bs);
560 gfp_mask = saved_gfp;
561 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
566 bio_init(bio, bdev, bvl, nr_vecs, opf);
567 } else if (nr_vecs) {
568 bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
570 bio_init(bio, bdev, NULL, 0, opf);
577 mempool_free(p, &bs->bio_pool);
580 EXPORT_SYMBOL(bio_alloc_bioset);
583 * bio_kmalloc - kmalloc a bio
584 * @nr_vecs: number of bio_vecs to allocate
585 * @gfp_mask: the GFP_* mask given to the slab allocator
587 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
588 * using bio_init() before use. To free a bio returned from this function use
589 * kfree() after calling bio_uninit(). A bio returned from this function can
590 * be reused by calling bio_uninit() before calling bio_init() again.
592 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
593 * function are not backed by a mempool can fail. Do not use this function
594 * for allocations in the file system I/O path.
596 * Returns: Pointer to new bio on success, NULL on failure.
598 struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
602 if (nr_vecs > UIO_MAXIOV)
604 return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
606 EXPORT_SYMBOL(bio_kmalloc);
608 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
611 struct bvec_iter iter;
613 __bio_for_each_segment(bv, bio, iter, start)
616 EXPORT_SYMBOL(zero_fill_bio_iter);
619 * bio_truncate - truncate the bio to small size of @new_size
620 * @bio: the bio to be truncated
621 * @new_size: new size for truncating the bio
624 * Truncate the bio to new size of @new_size. If bio_op(bio) is
625 * REQ_OP_READ, zero the truncated part. This function should only
626 * be used for handling corner cases, such as bio eod.
628 static void bio_truncate(struct bio *bio, unsigned new_size)
631 struct bvec_iter iter;
632 unsigned int done = 0;
633 bool truncated = false;
635 if (new_size >= bio->bi_iter.bi_size)
638 if (bio_op(bio) != REQ_OP_READ)
641 bio_for_each_segment(bv, bio, iter) {
642 if (done + bv.bv_len > new_size) {
646 offset = new_size - done;
649 zero_user(bv.bv_page, bv.bv_offset + offset,
658 * Don't touch bvec table here and make it really immutable, since
659 * fs bio user has to retrieve all pages via bio_for_each_segment_all
660 * in its .end_bio() callback.
662 * It is enough to truncate bio by updating .bi_size since we can make
663 * correct bvec with the updated .bi_size for drivers.
665 bio->bi_iter.bi_size = new_size;
669 * guard_bio_eod - truncate a BIO to fit the block device
670 * @bio: bio to truncate
672 * This allows us to do IO even on the odd last sectors of a device, even if the
673 * block size is some multiple of the physical sector size.
675 * We'll just truncate the bio to the size of the device, and clear the end of
676 * the buffer head manually. Truly out-of-range accesses will turn into actual
677 * I/O errors, this only handles the "we need to be able to do I/O at the final
680 void guard_bio_eod(struct bio *bio)
682 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
688 * If the *whole* IO is past the end of the device,
689 * let it through, and the IO layer will turn it into
692 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
695 maxsector -= bio->bi_iter.bi_sector;
696 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
699 bio_truncate(bio, maxsector << 9);
702 static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
708 while ((bio = cache->free_list) != NULL) {
709 cache->free_list = bio->bi_next;
718 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
721 nr -= __bio_alloc_cache_prune(cache, nr);
722 if (!READ_ONCE(cache->free_list)) {
723 bio_alloc_irq_cache_splice(cache);
724 __bio_alloc_cache_prune(cache, nr);
728 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
732 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
734 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
736 bio_alloc_cache_prune(cache, -1U);
741 static void bio_alloc_cache_destroy(struct bio_set *bs)
748 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
749 for_each_possible_cpu(cpu) {
750 struct bio_alloc_cache *cache;
752 cache = per_cpu_ptr(bs->cache, cpu);
753 bio_alloc_cache_prune(cache, -1U);
755 free_percpu(bs->cache);
759 static inline void bio_put_percpu_cache(struct bio *bio)
761 struct bio_alloc_cache *cache;
763 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
764 if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX)
769 bio->bi_next = cache->free_list;
770 /* Not necessary but helps not to iopoll already freed bios */
772 cache->free_list = bio;
774 } else if (in_hardirq()) {
775 lockdep_assert_irqs_disabled();
778 bio->bi_next = cache->free_list_irq;
779 cache->free_list_irq = bio;
792 * bio_put - release a reference to a bio
793 * @bio: bio to release reference to
796 * Put a reference to a &struct bio, either one you have gotten with
797 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
799 void bio_put(struct bio *bio)
801 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
802 BUG_ON(!atomic_read(&bio->__bi_cnt));
803 if (!atomic_dec_and_test(&bio->__bi_cnt))
806 if (bio->bi_opf & REQ_ALLOC_CACHE)
807 bio_put_percpu_cache(bio);
811 EXPORT_SYMBOL(bio_put);
813 static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
815 bio_set_flag(bio, BIO_CLONED);
816 bio->bi_ioprio = bio_src->bi_ioprio;
817 bio->bi_iter = bio_src->bi_iter;
820 if (bio->bi_bdev == bio_src->bi_bdev &&
821 bio_flagged(bio_src, BIO_REMAPPED))
822 bio_set_flag(bio, BIO_REMAPPED);
823 bio_clone_blkg_association(bio, bio_src);
826 if (bio_crypt_clone(bio, bio_src, gfp) < 0)
828 if (bio_integrity(bio_src) &&
829 bio_integrity_clone(bio, bio_src, gfp) < 0)
835 * bio_alloc_clone - clone a bio that shares the original bio's biovec
836 * @bdev: block_device to clone onto
837 * @bio_src: bio to clone from
838 * @gfp: allocation priority
839 * @bs: bio_set to allocate from
841 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
842 * bio, but not the actual data it points to.
844 * The caller must ensure that the return bio is not freed before @bio_src.
846 struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
847 gfp_t gfp, struct bio_set *bs)
851 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
855 if (__bio_clone(bio, bio_src, gfp) < 0) {
859 bio->bi_io_vec = bio_src->bi_io_vec;
863 EXPORT_SYMBOL(bio_alloc_clone);
866 * bio_init_clone - clone a bio that shares the original bio's biovec
867 * @bdev: block_device to clone onto
868 * @bio: bio to clone into
869 * @bio_src: bio to clone from
870 * @gfp: allocation priority
872 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
873 * The caller owns the returned bio, but not the actual data it points to.
875 * The caller must ensure that @bio_src is not freed before @bio.
877 int bio_init_clone(struct block_device *bdev, struct bio *bio,
878 struct bio *bio_src, gfp_t gfp)
882 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
883 ret = __bio_clone(bio, bio_src, gfp);
888 EXPORT_SYMBOL(bio_init_clone);
891 * bio_full - check if the bio is full
893 * @len: length of one segment to be added
895 * Return true if @bio is full and one segment with @len bytes can't be
896 * added to the bio, otherwise return false
898 static inline bool bio_full(struct bio *bio, unsigned len)
900 if (bio->bi_vcnt >= bio->bi_max_vecs)
902 if (bio->bi_iter.bi_size > UINT_MAX - len)
907 static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page,
908 unsigned int len, unsigned int off, bool *same_page)
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);
923 if (IS_ENABLED(CONFIG_KMSAN))
925 if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE)
934 * Try to merge a page into a segment, while obeying the hardware segment
935 * size limit. This is not for normal read/write bios, but for passthrough
936 * or Zone Append operations that we can't split.
938 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
939 struct page *page, unsigned len, unsigned offset,
942 unsigned long mask = queue_segment_boundary(q);
943 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
944 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
946 if ((addr1 | mask) != (addr2 | mask))
948 if (len > queue_max_segment_size(q) - bv->bv_len)
950 return bvec_try_merge_page(bv, page, len, offset, same_page);
954 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
955 * @q: the target queue
956 * @bio: destination bio
958 * @len: vec entry length
959 * @offset: vec entry offset
960 * @max_sectors: maximum number of sectors that can be added
961 * @same_page: return if the segment has been merged inside the same page
963 * Add a page to a bio while respecting the hardware max_sectors, max_segment
964 * and gap limitations.
966 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
967 struct page *page, unsigned int len, unsigned int offset,
968 unsigned int max_sectors, bool *same_page)
970 unsigned int max_size = max_sectors << SECTOR_SHIFT;
972 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
975 len = min3(len, max_size, queue_max_segment_size(q));
976 if (len > max_size - bio->bi_iter.bi_size)
979 if (bio->bi_vcnt > 0) {
980 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
982 if (bvec_try_merge_hw_page(q, bv, page, len, offset,
984 bio->bi_iter.bi_size += len;
989 min(bio->bi_max_vecs, queue_max_segments(q)))
993 * If the queue doesn't support SG gaps and adding this segment
994 * would create a gap, disallow it.
996 if (bvec_gap_to_prev(&q->limits, bv, offset))
1000 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
1002 bio->bi_iter.bi_size += len;
1007 * bio_add_pc_page - attempt to add page to passthrough bio
1008 * @q: the target queue
1009 * @bio: destination bio
1010 * @page: page to add
1011 * @len: vec entry length
1012 * @offset: vec entry offset
1014 * Attempt to add a page to the bio_vec maplist. This can fail for a
1015 * number of reasons, such as the bio being full or target block device
1016 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1017 * so it is always possible to add a single page to an empty bio.
1019 * This should only be used by passthrough bios.
1021 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1022 struct page *page, unsigned int len, unsigned int offset)
1024 bool same_page = false;
1025 return bio_add_hw_page(q, bio, page, len, offset,
1026 queue_max_hw_sectors(q), &same_page);
1028 EXPORT_SYMBOL(bio_add_pc_page);
1031 * bio_add_zone_append_page - attempt to add page to zone-append bio
1032 * @bio: destination bio
1033 * @page: page to add
1034 * @len: vec entry length
1035 * @offset: vec entry offset
1037 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1038 * for a zone-append request. This can fail for a number of reasons, such as the
1039 * bio being full or the target block device is not a zoned block device or
1040 * other limitations of the target block device. The target block device must
1041 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1044 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1046 int bio_add_zone_append_page(struct bio *bio, struct page *page,
1047 unsigned int len, unsigned int offset)
1049 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1050 bool same_page = false;
1052 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1055 if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1058 return bio_add_hw_page(q, bio, page, len, offset,
1059 queue_max_zone_append_sectors(q), &same_page);
1061 EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1064 * __bio_add_page - add page(s) to a bio in a new segment
1065 * @bio: destination bio
1066 * @page: start page to add
1067 * @len: length of the data to add, may cross pages
1068 * @off: offset of the data relative to @page, may cross pages
1070 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1071 * that @bio has space for another bvec.
1073 void __bio_add_page(struct bio *bio, struct page *page,
1074 unsigned int len, unsigned int off)
1076 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1077 WARN_ON_ONCE(bio_full(bio, len));
1079 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
1080 bio->bi_iter.bi_size += len;
1083 EXPORT_SYMBOL_GPL(__bio_add_page);
1086 * bio_add_page - attempt to add page(s) to bio
1087 * @bio: destination bio
1088 * @page: start page to add
1089 * @len: vec entry length, may cross pages
1090 * @offset: vec entry offset relative to @page, may cross pages
1092 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1093 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1095 int bio_add_page(struct bio *bio, struct page *page,
1096 unsigned int len, unsigned int offset)
1098 bool same_page = false;
1100 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1102 if (bio->bi_iter.bi_size > UINT_MAX - len)
1105 if (bio->bi_vcnt > 0 &&
1106 bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1107 page, len, offset, &same_page)) {
1108 bio->bi_iter.bi_size += len;
1112 if (bio->bi_vcnt >= bio->bi_max_vecs)
1114 __bio_add_page(bio, page, len, offset);
1117 EXPORT_SYMBOL(bio_add_page);
1119 void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1122 WARN_ON_ONCE(len > UINT_MAX);
1123 WARN_ON_ONCE(off > UINT_MAX);
1124 __bio_add_page(bio, &folio->page, len, off);
1128 * bio_add_folio - Attempt to add part of a folio to a bio.
1129 * @bio: BIO to add to.
1130 * @folio: Folio to add.
1131 * @len: How many bytes from the folio to add.
1132 * @off: First byte in this folio to add.
1134 * Filesystems that use folios can call this function instead of calling
1135 * bio_add_page() for each page in the folio. If @off is bigger than
1136 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1137 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1139 * Return: Whether the addition was successful.
1141 bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1144 if (len > UINT_MAX || off > UINT_MAX)
1146 return bio_add_page(bio, &folio->page, len, off) > 0;
1148 EXPORT_SYMBOL(bio_add_folio);
1150 void __bio_release_pages(struct bio *bio, bool mark_dirty)
1152 struct folio_iter fi;
1154 bio_for_each_folio_all(fi, bio) {
1159 folio_lock(fi.folio);
1160 folio_mark_dirty(fi.folio);
1161 folio_unlock(fi.folio);
1163 page = folio_page(fi.folio, fi.offset / PAGE_SIZE);
1165 bio_release_page(bio, page++);
1167 } while (done < fi.length);
1170 EXPORT_SYMBOL_GPL(__bio_release_pages);
1172 void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1174 size_t size = iov_iter_count(iter);
1176 WARN_ON_ONCE(bio->bi_max_vecs);
1178 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1179 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1180 size_t max_sectors = queue_max_zone_append_sectors(q);
1182 size = min(size, max_sectors << SECTOR_SHIFT);
1185 bio->bi_vcnt = iter->nr_segs;
1186 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1187 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1188 bio->bi_iter.bi_size = size;
1189 bio_set_flag(bio, BIO_CLONED);
1192 static int bio_iov_add_page(struct bio *bio, struct page *page,
1193 unsigned int len, unsigned int offset)
1195 bool same_page = false;
1197 if (WARN_ON_ONCE(bio->bi_iter.bi_size > UINT_MAX - len))
1200 if (bio->bi_vcnt > 0 &&
1201 bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1202 page, len, offset, &same_page)) {
1203 bio->bi_iter.bi_size += len;
1205 bio_release_page(bio, page);
1208 __bio_add_page(bio, page, len, offset);
1212 static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1213 unsigned int len, unsigned int offset)
1215 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1216 bool same_page = false;
1218 if (bio_add_hw_page(q, bio, page, len, offset,
1219 queue_max_zone_append_sectors(q), &same_page) != len)
1222 bio_release_page(bio, page);
1226 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1229 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1230 * @bio: bio to add pages to
1231 * @iter: iov iterator describing the region to be mapped
1233 * Extracts pages from *iter and appends them to @bio's bvec array. The pages
1234 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.
1235 * For a multi-segment *iter, this function only adds pages from the next
1236 * non-empty segment of the iov iterator.
1238 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1240 iov_iter_extraction_t extraction_flags = 0;
1241 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1242 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1243 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1244 struct page **pages = (struct page **)bv;
1246 unsigned len, i = 0;
1251 * Move page array up in the allocated memory for the bio vecs as far as
1252 * possible so that we can start filling biovecs from the beginning
1253 * without overwriting the temporary page array.
1255 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1256 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1258 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1259 extraction_flags |= ITER_ALLOW_P2PDMA;
1262 * Each segment in the iov is required to be a block size multiple.
1263 * However, we may not be able to get the entire segment if it spans
1264 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1265 * result to ensure the bio's total size is correct. The remainder of
1266 * the iov data will be picked up in the next bio iteration.
1268 size = iov_iter_extract_pages(iter, &pages,
1269 UINT_MAX - bio->bi_iter.bi_size,
1270 nr_pages, extraction_flags, &offset);
1271 if (unlikely(size <= 0))
1272 return size ? size : -EFAULT;
1274 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1277 size_t trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1278 iov_iter_revert(iter, trim);
1282 if (unlikely(!size)) {
1287 for (left = size, i = 0; left > 0; left -= len, i++) {
1288 struct page *page = pages[i];
1290 len = min_t(size_t, PAGE_SIZE - offset, left);
1291 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1292 ret = bio_iov_add_zone_append_page(bio, page, len,
1297 bio_iov_add_page(bio, page, len, offset);
1302 iov_iter_revert(iter, left);
1304 while (i < nr_pages)
1305 bio_release_page(bio, pages[i++]);
1311 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1312 * @bio: bio to add pages to
1313 * @iter: iov iterator describing the region to be added
1315 * This takes either an iterator pointing to user memory, or one pointing to
1316 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1317 * map them into the kernel. On IO completion, the caller should put those
1318 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1319 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1320 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1321 * completed by a call to ->ki_complete() or returns with an error other than
1322 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1323 * on IO completion. If it isn't, then pages should be released.
1325 * The function tries, but does not guarantee, to pin as many pages as
1326 * fit into the bio, or are requested in @iter, whatever is smaller. If
1327 * MM encounters an error pinning the requested pages, it stops. Error
1328 * is returned only if 0 pages could be pinned.
1330 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1334 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1337 if (iov_iter_is_bvec(iter)) {
1338 bio_iov_bvec_set(bio, iter);
1339 iov_iter_advance(iter, bio->bi_iter.bi_size);
1343 if (iov_iter_extract_will_pin(iter))
1344 bio_set_flag(bio, BIO_PAGE_PINNED);
1346 ret = __bio_iov_iter_get_pages(bio, iter);
1347 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1349 return bio->bi_vcnt ? 0 : ret;
1351 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1353 static void submit_bio_wait_endio(struct bio *bio)
1355 complete(bio->bi_private);
1359 * submit_bio_wait - submit a bio, and wait until it completes
1360 * @bio: The &struct bio which describes the I/O
1362 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1363 * bio_endio() on failure.
1365 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1366 * result in bio reference to be consumed. The caller must drop the reference
1369 int submit_bio_wait(struct bio *bio)
1371 DECLARE_COMPLETION_ONSTACK_MAP(done,
1372 bio->bi_bdev->bd_disk->lockdep_map);
1374 bio->bi_private = &done;
1375 bio->bi_end_io = submit_bio_wait_endio;
1376 bio->bi_opf |= REQ_SYNC;
1380 return blk_status_to_errno(bio->bi_status);
1382 EXPORT_SYMBOL(submit_bio_wait);
1384 void __bio_advance(struct bio *bio, unsigned bytes)
1386 if (bio_integrity(bio))
1387 bio_integrity_advance(bio, bytes);
1389 bio_crypt_advance(bio, bytes);
1390 bio_advance_iter(bio, &bio->bi_iter, bytes);
1392 EXPORT_SYMBOL(__bio_advance);
1394 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1395 struct bio *src, struct bvec_iter *src_iter)
1397 while (src_iter->bi_size && dst_iter->bi_size) {
1398 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1399 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1400 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1401 void *src_buf = bvec_kmap_local(&src_bv);
1402 void *dst_buf = bvec_kmap_local(&dst_bv);
1404 memcpy(dst_buf, src_buf, bytes);
1406 kunmap_local(dst_buf);
1407 kunmap_local(src_buf);
1409 bio_advance_iter_single(src, src_iter, bytes);
1410 bio_advance_iter_single(dst, dst_iter, bytes);
1413 EXPORT_SYMBOL(bio_copy_data_iter);
1416 * bio_copy_data - copy contents of data buffers from one bio to another
1418 * @dst: destination bio
1420 * Stops when it reaches the end of either @src or @dst - that is, copies
1421 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1423 void bio_copy_data(struct bio *dst, struct bio *src)
1425 struct bvec_iter src_iter = src->bi_iter;
1426 struct bvec_iter dst_iter = dst->bi_iter;
1428 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1430 EXPORT_SYMBOL(bio_copy_data);
1432 void bio_free_pages(struct bio *bio)
1434 struct bio_vec *bvec;
1435 struct bvec_iter_all iter_all;
1437 bio_for_each_segment_all(bvec, bio, iter_all)
1438 __free_page(bvec->bv_page);
1440 EXPORT_SYMBOL(bio_free_pages);
1443 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1444 * for performing direct-IO in BIOs.
1446 * The problem is that we cannot run folio_mark_dirty() from interrupt context
1447 * because the required locks are not interrupt-safe. So what we can do is to
1448 * mark the pages dirty _before_ performing IO. And in interrupt context,
1449 * check that the pages are still dirty. If so, fine. If not, redirty them
1450 * in process context.
1452 * Note that this code is very hard to test under normal circumstances because
1453 * direct-io pins the pages with get_user_pages(). This makes
1454 * is_page_cache_freeable return false, and the VM will not clean the pages.
1455 * But other code (eg, flusher threads) could clean the pages if they are mapped
1458 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1459 * deferred bio dirtying paths.
1463 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1465 void bio_set_pages_dirty(struct bio *bio)
1467 struct folio_iter fi;
1469 bio_for_each_folio_all(fi, bio) {
1470 folio_lock(fi.folio);
1471 folio_mark_dirty(fi.folio);
1472 folio_unlock(fi.folio);
1475 EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1478 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1479 * If they are, then fine. If, however, some pages are clean then they must
1480 * have been written out during the direct-IO read. So we take another ref on
1481 * the BIO and re-dirty the pages in process context.
1483 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1484 * here on. It will unpin each page and will run one bio_put() against the
1488 static void bio_dirty_fn(struct work_struct *work);
1490 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1491 static DEFINE_SPINLOCK(bio_dirty_lock);
1492 static struct bio *bio_dirty_list;
1495 * This runs in process context
1497 static void bio_dirty_fn(struct work_struct *work)
1499 struct bio *bio, *next;
1501 spin_lock_irq(&bio_dirty_lock);
1502 next = bio_dirty_list;
1503 bio_dirty_list = NULL;
1504 spin_unlock_irq(&bio_dirty_lock);
1506 while ((bio = next) != NULL) {
1507 next = bio->bi_private;
1509 bio_release_pages(bio, true);
1514 void bio_check_pages_dirty(struct bio *bio)
1516 struct folio_iter fi;
1517 unsigned long flags;
1519 bio_for_each_folio_all(fi, bio) {
1520 if (!folio_test_dirty(fi.folio))
1524 bio_release_pages(bio, false);
1528 spin_lock_irqsave(&bio_dirty_lock, flags);
1529 bio->bi_private = bio_dirty_list;
1530 bio_dirty_list = bio;
1531 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1532 schedule_work(&bio_dirty_work);
1534 EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1536 static inline bool bio_remaining_done(struct bio *bio)
1539 * If we're not chaining, then ->__bi_remaining is always 1 and
1540 * we always end io on the first invocation.
1542 if (!bio_flagged(bio, BIO_CHAIN))
1545 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1547 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1548 bio_clear_flag(bio, BIO_CHAIN);
1556 * bio_endio - end I/O on a bio
1560 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1561 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1562 * bio unless they own it and thus know that it has an end_io function.
1564 * bio_endio() can be called several times on a bio that has been chained
1565 * using bio_chain(). The ->bi_end_io() function will only be called the
1568 void bio_endio(struct bio *bio)
1571 if (!bio_remaining_done(bio))
1573 if (!bio_integrity_endio(bio))
1576 rq_qos_done_bio(bio);
1578 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1579 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1580 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1584 * Need to have a real endio function for chained bios, otherwise
1585 * various corner cases will break (like stacking block devices that
1586 * save/restore bi_end_io) - however, we want to avoid unbounded
1587 * recursion and blowing the stack. Tail call optimization would
1588 * handle this, but compiling with frame pointers also disables
1589 * gcc's sibling call optimization.
1591 if (bio->bi_end_io == bio_chain_endio) {
1592 bio = __bio_chain_endio(bio);
1596 blk_throtl_bio_endio(bio);
1597 /* release cgroup info */
1600 bio->bi_end_io(bio);
1602 EXPORT_SYMBOL(bio_endio);
1605 * bio_split - split a bio
1606 * @bio: bio to split
1607 * @sectors: number of sectors to split from the front of @bio
1609 * @bs: bio set to allocate from
1611 * Allocates and returns a new bio which represents @sectors from the start of
1612 * @bio, and updates @bio to represent the remaining sectors.
1614 * Unless this is a discard request the newly allocated bio will point
1615 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1616 * neither @bio nor @bs are freed before the split bio.
1618 struct bio *bio_split(struct bio *bio, int sectors,
1619 gfp_t gfp, struct bio_set *bs)
1623 BUG_ON(sectors <= 0);
1624 BUG_ON(sectors >= bio_sectors(bio));
1626 /* Zone append commands cannot be split */
1627 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1630 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1634 split->bi_iter.bi_size = sectors << 9;
1636 if (bio_integrity(split))
1637 bio_integrity_trim(split);
1639 bio_advance(bio, split->bi_iter.bi_size);
1641 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1642 bio_set_flag(split, BIO_TRACE_COMPLETION);
1646 EXPORT_SYMBOL(bio_split);
1649 * bio_trim - trim a bio
1651 * @offset: number of sectors to trim from the front of @bio
1652 * @size: size we want to trim @bio to, in sectors
1654 * This function is typically used for bios that are cloned and submitted
1655 * to the underlying device in parts.
1657 void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1659 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1660 offset + size > bio_sectors(bio)))
1664 if (offset == 0 && size == bio->bi_iter.bi_size)
1667 bio_advance(bio, offset << 9);
1668 bio->bi_iter.bi_size = size;
1670 if (bio_integrity(bio))
1671 bio_integrity_trim(bio);
1673 EXPORT_SYMBOL_GPL(bio_trim);
1676 * create memory pools for biovec's in a bio_set.
1677 * use the global biovec slabs created for general use.
1679 int biovec_init_pool(mempool_t *pool, int pool_entries)
1681 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1683 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1687 * bioset_exit - exit a bioset initialized with bioset_init()
1689 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1692 void bioset_exit(struct bio_set *bs)
1694 bio_alloc_cache_destroy(bs);
1695 if (bs->rescue_workqueue)
1696 destroy_workqueue(bs->rescue_workqueue);
1697 bs->rescue_workqueue = NULL;
1699 mempool_exit(&bs->bio_pool);
1700 mempool_exit(&bs->bvec_pool);
1702 bioset_integrity_free(bs);
1705 bs->bio_slab = NULL;
1707 EXPORT_SYMBOL(bioset_exit);
1710 * bioset_init - Initialize a bio_set
1711 * @bs: pool to initialize
1712 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1713 * @front_pad: Number of bytes to allocate in front of the returned bio
1714 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1715 * and %BIOSET_NEED_RESCUER
1718 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1719 * to ask for a number of bytes to be allocated in front of the bio.
1720 * Front pad allocation is useful for embedding the bio inside
1721 * another structure, to avoid allocating extra data to go with the bio.
1722 * Note that the bio must be embedded at the END of that structure always,
1723 * or things will break badly.
1724 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1725 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1726 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1727 * to dispatch queued requests when the mempool runs out of space.
1730 int bioset_init(struct bio_set *bs,
1731 unsigned int pool_size,
1732 unsigned int front_pad,
1735 bs->front_pad = front_pad;
1736 if (flags & BIOSET_NEED_BVECS)
1737 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1741 spin_lock_init(&bs->rescue_lock);
1742 bio_list_init(&bs->rescue_list);
1743 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1745 bs->bio_slab = bio_find_or_create_slab(bs);
1749 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1752 if ((flags & BIOSET_NEED_BVECS) &&
1753 biovec_init_pool(&bs->bvec_pool, pool_size))
1756 if (flags & BIOSET_NEED_RESCUER) {
1757 bs->rescue_workqueue = alloc_workqueue("bioset",
1759 if (!bs->rescue_workqueue)
1762 if (flags & BIOSET_PERCPU_CACHE) {
1763 bs->cache = alloc_percpu(struct bio_alloc_cache);
1766 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1774 EXPORT_SYMBOL(bioset_init);
1776 static int __init init_bio(void)
1780 BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1782 bio_integrity_init();
1784 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1785 struct biovec_slab *bvs = bvec_slabs + i;
1787 bvs->slab = kmem_cache_create(bvs->name,
1788 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1789 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1792 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1795 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1796 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1797 panic("bio: can't allocate bios\n");
1799 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1800 panic("bio: can't create integrity pool\n");
1804 subsys_initcall(init_bio);