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 struct bio_alloc_cache {
29 struct bio *free_list;
33 static struct biovec_slab {
36 struct kmem_cache *slab;
37 } bvec_slabs[] __read_mostly = {
38 { .nr_vecs = 16, .name = "biovec-16" },
39 { .nr_vecs = 64, .name = "biovec-64" },
40 { .nr_vecs = 128, .name = "biovec-128" },
41 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
44 static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
47 /* smaller bios use inline vecs */
49 return &bvec_slabs[0];
51 return &bvec_slabs[1];
53 return &bvec_slabs[2];
54 case 129 ... BIO_MAX_VECS:
55 return &bvec_slabs[3];
63 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
64 * IO code that does not need private memory pools.
66 struct bio_set fs_bio_set;
67 EXPORT_SYMBOL(fs_bio_set);
70 * Our slab pool management
73 struct kmem_cache *slab;
74 unsigned int slab_ref;
75 unsigned int slab_size;
78 static DEFINE_MUTEX(bio_slab_lock);
79 static DEFINE_XARRAY(bio_slabs);
81 static struct bio_slab *create_bio_slab(unsigned int size)
83 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
88 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
89 bslab->slab = kmem_cache_create(bslab->name, size,
90 ARCH_KMALLOC_MINALIGN,
91 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
96 bslab->slab_size = size;
98 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
101 kmem_cache_destroy(bslab->slab);
108 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
110 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
113 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
115 unsigned int size = bs_bio_slab_size(bs);
116 struct bio_slab *bslab;
118 mutex_lock(&bio_slab_lock);
119 bslab = xa_load(&bio_slabs, size);
123 bslab = create_bio_slab(size);
124 mutex_unlock(&bio_slab_lock);
131 static void bio_put_slab(struct bio_set *bs)
133 struct bio_slab *bslab = NULL;
134 unsigned int slab_size = bs_bio_slab_size(bs);
136 mutex_lock(&bio_slab_lock);
138 bslab = xa_load(&bio_slabs, slab_size);
139 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
144 WARN_ON(!bslab->slab_ref);
146 if (--bslab->slab_ref)
149 xa_erase(&bio_slabs, slab_size);
151 kmem_cache_destroy(bslab->slab);
155 mutex_unlock(&bio_slab_lock);
158 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
160 BUG_ON(nr_vecs > BIO_MAX_VECS);
162 if (nr_vecs == BIO_MAX_VECS)
163 mempool_free(bv, pool);
164 else if (nr_vecs > BIO_INLINE_VECS)
165 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
169 * Make the first allocation restricted and don't dump info on allocation
170 * failures, since we'll fall back to the mempool in case of failure.
172 static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
174 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
175 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
178 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
181 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
183 if (WARN_ON_ONCE(!bvs))
187 * Upgrade the nr_vecs request to take full advantage of the allocation.
188 * We also rely on this in the bvec_free path.
190 *nr_vecs = bvs->nr_vecs;
193 * Try a slab allocation first for all smaller allocations. If that
194 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
195 * The mempool is sized to handle up to BIO_MAX_VECS entries.
197 if (*nr_vecs < BIO_MAX_VECS) {
200 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
201 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
203 *nr_vecs = BIO_MAX_VECS;
206 return mempool_alloc(pool, gfp_mask);
209 void bio_uninit(struct bio *bio)
211 #ifdef CONFIG_BLK_CGROUP
213 blkg_put(bio->bi_blkg);
217 if (bio_integrity(bio))
218 bio_integrity_free(bio);
220 bio_crypt_free_ctx(bio);
222 EXPORT_SYMBOL(bio_uninit);
224 static void bio_free(struct bio *bio)
226 struct bio_set *bs = bio->bi_pool;
232 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
233 mempool_free(p - bs->front_pad, &bs->bio_pool);
237 * Users of this function have their own bio allocation. Subsequently,
238 * they must remember to pair any call to bio_init() with bio_uninit()
239 * when IO has completed, or when the bio is released.
241 void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
242 unsigned short max_vecs, blk_opf_t opf)
250 bio->bi_iter.bi_sector = 0;
251 bio->bi_iter.bi_size = 0;
252 bio->bi_iter.bi_idx = 0;
253 bio->bi_iter.bi_bvec_done = 0;
254 bio->bi_end_io = NULL;
255 bio->bi_private = NULL;
256 #ifdef CONFIG_BLK_CGROUP
258 bio->bi_issue.value = 0;
260 bio_associate_blkg(bio);
261 #ifdef CONFIG_BLK_CGROUP_IOCOST
262 bio->bi_iocost_cost = 0;
265 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
266 bio->bi_crypt_context = NULL;
268 #ifdef CONFIG_BLK_DEV_INTEGRITY
269 bio->bi_integrity = NULL;
273 atomic_set(&bio->__bi_remaining, 1);
274 atomic_set(&bio->__bi_cnt, 1);
275 bio->bi_cookie = BLK_QC_T_NONE;
277 bio->bi_max_vecs = max_vecs;
278 bio->bi_io_vec = table;
281 EXPORT_SYMBOL(bio_init);
284 * bio_reset - reinitialize a bio
286 * @bdev: block device to use the bio for
287 * @opf: operation and flags for bio
290 * After calling bio_reset(), @bio will be in the same state as a freshly
291 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
292 * preserved are the ones that are initialized by bio_alloc_bioset(). See
293 * comment in struct bio.
295 void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
298 memset(bio, 0, BIO_RESET_BYTES);
299 atomic_set(&bio->__bi_remaining, 1);
302 bio_associate_blkg(bio);
305 EXPORT_SYMBOL(bio_reset);
307 static struct bio *__bio_chain_endio(struct bio *bio)
309 struct bio *parent = bio->bi_private;
311 if (bio->bi_status && !parent->bi_status)
312 parent->bi_status = bio->bi_status;
317 static void bio_chain_endio(struct bio *bio)
319 bio_endio(__bio_chain_endio(bio));
323 * bio_chain - chain bio completions
324 * @bio: the target bio
325 * @parent: the parent bio of @bio
327 * The caller won't have a bi_end_io called when @bio completes - instead,
328 * @parent's bi_end_io won't be called until both @parent and @bio have
329 * completed; the chained bio will also be freed when it completes.
331 * The caller must not set bi_private or bi_end_io in @bio.
333 void bio_chain(struct bio *bio, struct bio *parent)
335 BUG_ON(bio->bi_private || bio->bi_end_io);
337 bio->bi_private = parent;
338 bio->bi_end_io = bio_chain_endio;
339 bio_inc_remaining(parent);
341 EXPORT_SYMBOL(bio_chain);
343 struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
344 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
346 struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
355 EXPORT_SYMBOL_GPL(blk_next_bio);
357 static void bio_alloc_rescue(struct work_struct *work)
359 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
363 spin_lock(&bs->rescue_lock);
364 bio = bio_list_pop(&bs->rescue_list);
365 spin_unlock(&bs->rescue_lock);
370 submit_bio_noacct(bio);
374 static void punt_bios_to_rescuer(struct bio_set *bs)
376 struct bio_list punt, nopunt;
379 if (WARN_ON_ONCE(!bs->rescue_workqueue))
382 * In order to guarantee forward progress we must punt only bios that
383 * were allocated from this bio_set; otherwise, if there was a bio on
384 * there for a stacking driver higher up in the stack, processing it
385 * could require allocating bios from this bio_set, and doing that from
386 * our own rescuer would be bad.
388 * Since bio lists are singly linked, pop them all instead of trying to
389 * remove from the middle of the list:
392 bio_list_init(&punt);
393 bio_list_init(&nopunt);
395 while ((bio = bio_list_pop(¤t->bio_list[0])))
396 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
397 current->bio_list[0] = nopunt;
399 bio_list_init(&nopunt);
400 while ((bio = bio_list_pop(¤t->bio_list[1])))
401 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
402 current->bio_list[1] = nopunt;
404 spin_lock(&bs->rescue_lock);
405 bio_list_merge(&bs->rescue_list, &punt);
406 spin_unlock(&bs->rescue_lock);
408 queue_work(bs->rescue_workqueue, &bs->rescue_work);
411 static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
412 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
415 struct bio_alloc_cache *cache;
418 cache = per_cpu_ptr(bs->cache, get_cpu());
419 if (!cache->free_list) {
423 bio = cache->free_list;
424 cache->free_list = bio->bi_next;
428 bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
434 * bio_alloc_bioset - allocate a bio for I/O
435 * @bdev: block device to allocate the bio for (can be %NULL)
436 * @nr_vecs: number of bvecs to pre-allocate
437 * @opf: operation and flags for bio
438 * @gfp_mask: the GFP_* mask given to the slab allocator
439 * @bs: the bio_set to allocate from.
441 * Allocate a bio from the mempools in @bs.
443 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
444 * allocate a bio. This is due to the mempool guarantees. To make this work,
445 * callers must never allocate more than 1 bio at a time from the general pool.
446 * Callers that need to allocate more than 1 bio must always submit the
447 * previously allocated bio for IO before attempting to allocate a new one.
448 * Failure to do so can cause deadlocks under memory pressure.
450 * Note that when running under submit_bio_noacct() (i.e. any block driver),
451 * bios are not submitted until after you return - see the code in
452 * submit_bio_noacct() that converts recursion into iteration, to prevent
455 * This would normally mean allocating multiple bios under submit_bio_noacct()
456 * would be susceptible to deadlocks, but we have
457 * deadlock avoidance code that resubmits any blocked bios from a rescuer
460 * However, we do not guarantee forward progress for allocations from other
461 * mempools. Doing multiple allocations from the same mempool under
462 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
463 * for per bio allocations.
465 * If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done from process
466 * context, not hard/soft IRQ.
468 * Returns: Pointer to new bio on success, NULL on failure.
470 struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
471 blk_opf_t opf, gfp_t gfp_mask,
474 gfp_t saved_gfp = gfp_mask;
478 /* should not use nobvec bioset for nr_vecs > 0 */
479 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
482 if (opf & REQ_ALLOC_CACHE) {
483 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
484 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
489 * No cached bio available, bio returned below marked with
490 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
493 opf &= ~REQ_ALLOC_CACHE;
498 * submit_bio_noacct() converts recursion to iteration; this means if
499 * we're running beneath it, any bios we allocate and submit will not be
500 * submitted (and thus freed) until after we return.
502 * This exposes us to a potential deadlock if we allocate multiple bios
503 * from the same bio_set() while running underneath submit_bio_noacct().
504 * If we were to allocate multiple bios (say a stacking block driver
505 * that was splitting bios), we would deadlock if we exhausted the
508 * We solve this, and guarantee forward progress, with a rescuer
509 * workqueue per bio_set. If we go to allocate and there are bios on
510 * current->bio_list, we first try the allocation without
511 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
512 * blocking to the rescuer workqueue before we retry with the original
515 if (current->bio_list &&
516 (!bio_list_empty(¤t->bio_list[0]) ||
517 !bio_list_empty(¤t->bio_list[1])) &&
518 bs->rescue_workqueue)
519 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
521 p = mempool_alloc(&bs->bio_pool, gfp_mask);
522 if (!p && gfp_mask != saved_gfp) {
523 punt_bios_to_rescuer(bs);
524 gfp_mask = saved_gfp;
525 p = mempool_alloc(&bs->bio_pool, gfp_mask);
529 if (!mempool_is_saturated(&bs->bio_pool))
530 opf &= ~REQ_ALLOC_CACHE;
532 bio = p + bs->front_pad;
533 if (nr_vecs > BIO_INLINE_VECS) {
534 struct bio_vec *bvl = NULL;
536 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
537 if (!bvl && gfp_mask != saved_gfp) {
538 punt_bios_to_rescuer(bs);
539 gfp_mask = saved_gfp;
540 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
545 bio_init(bio, bdev, bvl, nr_vecs, opf);
546 } else if (nr_vecs) {
547 bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
549 bio_init(bio, bdev, NULL, 0, opf);
556 mempool_free(p, &bs->bio_pool);
559 EXPORT_SYMBOL(bio_alloc_bioset);
562 * bio_kmalloc - kmalloc a bio
563 * @nr_vecs: number of bio_vecs to allocate
564 * @gfp_mask: the GFP_* mask given to the slab allocator
566 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
567 * using bio_init() before use. To free a bio returned from this function use
568 * kfree() after calling bio_uninit(). A bio returned from this function can
569 * be reused by calling bio_uninit() before calling bio_init() again.
571 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
572 * function are not backed by a mempool can fail. Do not use this function
573 * for allocations in the file system I/O path.
575 * Returns: Pointer to new bio on success, NULL on failure.
577 struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
581 if (nr_vecs > UIO_MAXIOV)
583 return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
585 EXPORT_SYMBOL(bio_kmalloc);
587 void zero_fill_bio(struct bio *bio)
590 struct bvec_iter iter;
592 bio_for_each_segment(bv, bio, iter)
595 EXPORT_SYMBOL(zero_fill_bio);
598 * bio_truncate - truncate the bio to small size of @new_size
599 * @bio: the bio to be truncated
600 * @new_size: new size for truncating the bio
603 * Truncate the bio to new size of @new_size. If bio_op(bio) is
604 * REQ_OP_READ, zero the truncated part. This function should only
605 * be used for handling corner cases, such as bio eod.
607 static void bio_truncate(struct bio *bio, unsigned new_size)
610 struct bvec_iter iter;
611 unsigned int done = 0;
612 bool truncated = false;
614 if (new_size >= bio->bi_iter.bi_size)
617 if (bio_op(bio) != REQ_OP_READ)
620 bio_for_each_segment(bv, bio, iter) {
621 if (done + bv.bv_len > new_size) {
625 offset = new_size - done;
628 zero_user(bv.bv_page, bv.bv_offset + offset,
637 * Don't touch bvec table here and make it really immutable, since
638 * fs bio user has to retrieve all pages via bio_for_each_segment_all
639 * in its .end_bio() callback.
641 * It is enough to truncate bio by updating .bi_size since we can make
642 * correct bvec with the updated .bi_size for drivers.
644 bio->bi_iter.bi_size = new_size;
648 * guard_bio_eod - truncate a BIO to fit the block device
649 * @bio: bio to truncate
651 * This allows us to do IO even on the odd last sectors of a device, even if the
652 * block size is some multiple of the physical sector size.
654 * We'll just truncate the bio to the size of the device, and clear the end of
655 * the buffer head manually. Truly out-of-range accesses will turn into actual
656 * I/O errors, this only handles the "we need to be able to do I/O at the final
659 void guard_bio_eod(struct bio *bio)
661 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
667 * If the *whole* IO is past the end of the device,
668 * let it through, and the IO layer will turn it into
671 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
674 maxsector -= bio->bi_iter.bi_sector;
675 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
678 bio_truncate(bio, maxsector << 9);
681 #define ALLOC_CACHE_MAX 512
682 #define ALLOC_CACHE_SLACK 64
684 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
690 while ((bio = cache->free_list) != NULL) {
691 cache->free_list = bio->bi_next;
699 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
703 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
705 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
707 bio_alloc_cache_prune(cache, -1U);
712 static void bio_alloc_cache_destroy(struct bio_set *bs)
719 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
720 for_each_possible_cpu(cpu) {
721 struct bio_alloc_cache *cache;
723 cache = per_cpu_ptr(bs->cache, cpu);
724 bio_alloc_cache_prune(cache, -1U);
726 free_percpu(bs->cache);
730 static inline void bio_put_percpu_cache(struct bio *bio)
732 struct bio_alloc_cache *cache;
734 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
737 if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
738 bio->bi_next = cache->free_list;
739 cache->free_list = bio;
747 if (cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK)
748 bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK);
753 * bio_put - release a reference to a bio
754 * @bio: bio to release reference to
757 * Put a reference to a &struct bio, either one you have gotten with
758 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
760 void bio_put(struct bio *bio)
762 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
763 BUG_ON(!atomic_read(&bio->__bi_cnt));
764 if (!atomic_dec_and_test(&bio->__bi_cnt))
767 if (bio->bi_opf & REQ_ALLOC_CACHE)
768 bio_put_percpu_cache(bio);
772 EXPORT_SYMBOL(bio_put);
774 static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
776 bio_set_flag(bio, BIO_CLONED);
777 bio->bi_ioprio = bio_src->bi_ioprio;
778 bio->bi_iter = bio_src->bi_iter;
781 if (bio->bi_bdev == bio_src->bi_bdev &&
782 bio_flagged(bio_src, BIO_REMAPPED))
783 bio_set_flag(bio, BIO_REMAPPED);
784 bio_clone_blkg_association(bio, bio_src);
787 if (bio_crypt_clone(bio, bio_src, gfp) < 0)
789 if (bio_integrity(bio_src) &&
790 bio_integrity_clone(bio, bio_src, gfp) < 0)
796 * bio_alloc_clone - clone a bio that shares the original bio's biovec
797 * @bdev: block_device to clone onto
798 * @bio_src: bio to clone from
799 * @gfp: allocation priority
800 * @bs: bio_set to allocate from
802 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
803 * bio, but not the actual data it points to.
805 * The caller must ensure that the return bio is not freed before @bio_src.
807 struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
808 gfp_t gfp, struct bio_set *bs)
812 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
816 if (__bio_clone(bio, bio_src, gfp) < 0) {
820 bio->bi_io_vec = bio_src->bi_io_vec;
824 EXPORT_SYMBOL(bio_alloc_clone);
827 * bio_init_clone - clone a bio that shares the original bio's biovec
828 * @bdev: block_device to clone onto
829 * @bio: bio to clone into
830 * @bio_src: bio to clone from
831 * @gfp: allocation priority
833 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
834 * The caller owns the returned bio, but not the actual data it points to.
836 * The caller must ensure that @bio_src is not freed before @bio.
838 int bio_init_clone(struct block_device *bdev, struct bio *bio,
839 struct bio *bio_src, gfp_t gfp)
843 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
844 ret = __bio_clone(bio, bio_src, gfp);
849 EXPORT_SYMBOL(bio_init_clone);
852 * bio_full - check if the bio is full
854 * @len: length of one segment to be added
856 * Return true if @bio is full and one segment with @len bytes can't be
857 * added to the bio, otherwise return false
859 static inline bool bio_full(struct bio *bio, unsigned len)
861 if (bio->bi_vcnt >= bio->bi_max_vecs)
863 if (bio->bi_iter.bi_size > UINT_MAX - len)
868 static inline bool page_is_mergeable(const struct bio_vec *bv,
869 struct page *page, unsigned int len, unsigned int off,
872 size_t bv_end = bv->bv_offset + bv->bv_len;
873 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
874 phys_addr_t page_addr = page_to_phys(page);
876 if (vec_end_addr + 1 != page_addr + off)
878 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
880 if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
883 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
886 else if (IS_ENABLED(CONFIG_KMSAN))
888 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
892 * __bio_try_merge_page - try appending data to an existing bvec.
893 * @bio: destination bio
894 * @page: start page to add
895 * @len: length of the data to add
896 * @off: offset of the data relative to @page
897 * @same_page: return if the segment has been merged inside the same page
899 * Try to add the data at @page + @off to the last bvec of @bio. This is a
900 * useful optimisation for file systems with a block size smaller than the
903 * Warn if (@len, @off) crosses pages in case that @same_page is true.
905 * Return %true on success or %false on failure.
907 static bool __bio_try_merge_page(struct bio *bio, struct page *page,
908 unsigned int len, unsigned int off, bool *same_page)
910 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
913 if (bio->bi_vcnt > 0) {
914 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
916 if (page_is_mergeable(bv, page, len, off, same_page)) {
917 if (bio->bi_iter.bi_size > UINT_MAX - len) {
922 bio->bi_iter.bi_size += len;
930 * Try to merge a page into a segment, while obeying the hardware segment
931 * size limit. This is not for normal read/write bios, but for passthrough
932 * or Zone Append operations that we can't split.
934 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
935 struct page *page, unsigned len,
936 unsigned offset, bool *same_page)
938 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
939 unsigned long mask = queue_segment_boundary(q);
940 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
941 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
943 if ((addr1 | mask) != (addr2 | mask))
945 if (bv->bv_len + len > queue_max_segment_size(q))
947 return __bio_try_merge_page(bio, page, len, offset, same_page);
951 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
952 * @q: the target queue
953 * @bio: destination bio
955 * @len: vec entry length
956 * @offset: vec entry offset
957 * @max_sectors: maximum number of sectors that can be added
958 * @same_page: return if the segment has been merged inside the same page
960 * Add a page to a bio while respecting the hardware max_sectors, max_segment
961 * and gap limitations.
963 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
964 struct page *page, unsigned int len, unsigned int offset,
965 unsigned int max_sectors, bool *same_page)
967 struct bio_vec *bvec;
969 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
972 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
975 if (bio->bi_vcnt > 0) {
976 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
980 * If the queue doesn't support SG gaps and adding this segment
981 * would create a gap, disallow it.
983 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
984 if (bvec_gap_to_prev(&q->limits, bvec, offset))
988 if (bio_full(bio, len))
991 if (bio->bi_vcnt >= queue_max_segments(q))
994 bvec = &bio->bi_io_vec[bio->bi_vcnt];
995 bvec->bv_page = page;
997 bvec->bv_offset = offset;
999 bio->bi_iter.bi_size += len;
1004 * bio_add_pc_page - attempt to add page to passthrough bio
1005 * @q: the target queue
1006 * @bio: destination bio
1007 * @page: page to add
1008 * @len: vec entry length
1009 * @offset: vec entry offset
1011 * Attempt to add a page to the bio_vec maplist. This can fail for a
1012 * number of reasons, such as the bio being full or target block device
1013 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1014 * so it is always possible to add a single page to an empty bio.
1016 * This should only be used by passthrough bios.
1018 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1019 struct page *page, unsigned int len, unsigned int offset)
1021 bool same_page = false;
1022 return bio_add_hw_page(q, bio, page, len, offset,
1023 queue_max_hw_sectors(q), &same_page);
1025 EXPORT_SYMBOL(bio_add_pc_page);
1028 * bio_add_zone_append_page - attempt to add page to zone-append bio
1029 * @bio: destination bio
1030 * @page: page to add
1031 * @len: vec entry length
1032 * @offset: vec entry offset
1034 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1035 * for a zone-append request. This can fail for a number of reasons, such as the
1036 * bio being full or the target block device is not a zoned block device or
1037 * other limitations of the target block device. The target block device must
1038 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1041 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1043 int bio_add_zone_append_page(struct bio *bio, struct page *page,
1044 unsigned int len, unsigned int offset)
1046 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1047 bool same_page = false;
1049 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1052 if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1055 return bio_add_hw_page(q, bio, page, len, offset,
1056 queue_max_zone_append_sectors(q), &same_page);
1058 EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1061 * __bio_add_page - add page(s) to a bio in a new segment
1062 * @bio: destination bio
1063 * @page: start page to add
1064 * @len: length of the data to add, may cross pages
1065 * @off: offset of the data relative to @page, may cross pages
1067 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1068 * that @bio has space for another bvec.
1070 void __bio_add_page(struct bio *bio, struct page *page,
1071 unsigned int len, unsigned int off)
1073 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
1075 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1076 WARN_ON_ONCE(bio_full(bio, len));
1079 bv->bv_offset = off;
1082 bio->bi_iter.bi_size += len;
1085 EXPORT_SYMBOL_GPL(__bio_add_page);
1088 * bio_add_page - attempt to add page(s) to bio
1089 * @bio: destination bio
1090 * @page: start page to add
1091 * @len: vec entry length, may cross pages
1092 * @offset: vec entry offset relative to @page, may cross pages
1094 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1095 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1097 int bio_add_page(struct bio *bio, struct page *page,
1098 unsigned int len, unsigned int offset)
1100 bool same_page = false;
1102 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1103 if (bio_full(bio, len))
1105 __bio_add_page(bio, page, len, offset);
1109 EXPORT_SYMBOL(bio_add_page);
1112 * bio_add_folio - Attempt to add part of a folio to a bio.
1113 * @bio: BIO to add to.
1114 * @folio: Folio to add.
1115 * @len: How many bytes from the folio to add.
1116 * @off: First byte in this folio to add.
1118 * Filesystems that use folios can call this function instead of calling
1119 * bio_add_page() for each page in the folio. If @off is bigger than
1120 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1121 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1123 * Return: Whether the addition was successful.
1125 bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1128 if (len > UINT_MAX || off > UINT_MAX)
1130 return bio_add_page(bio, &folio->page, len, off) > 0;
1133 void __bio_release_pages(struct bio *bio, bool mark_dirty)
1135 struct bvec_iter_all iter_all;
1136 struct bio_vec *bvec;
1138 bio_for_each_segment_all(bvec, bio, iter_all) {
1139 if (mark_dirty && !PageCompound(bvec->bv_page))
1140 set_page_dirty_lock(bvec->bv_page);
1141 put_page(bvec->bv_page);
1144 EXPORT_SYMBOL_GPL(__bio_release_pages);
1146 void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1148 size_t size = iov_iter_count(iter);
1150 WARN_ON_ONCE(bio->bi_max_vecs);
1152 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1153 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1154 size_t max_sectors = queue_max_zone_append_sectors(q);
1156 size = min(size, max_sectors << SECTOR_SHIFT);
1159 bio->bi_vcnt = iter->nr_segs;
1160 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1161 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1162 bio->bi_iter.bi_size = size;
1163 bio_set_flag(bio, BIO_NO_PAGE_REF);
1164 bio_set_flag(bio, BIO_CLONED);
1167 static int bio_iov_add_page(struct bio *bio, struct page *page,
1168 unsigned int len, unsigned int offset)
1170 bool same_page = false;
1172 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1173 __bio_add_page(bio, page, len, offset);
1182 static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1183 unsigned int len, unsigned int offset)
1185 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1186 bool same_page = false;
1188 if (bio_add_hw_page(q, bio, page, len, offset,
1189 queue_max_zone_append_sectors(q), &same_page) != len)
1196 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1199 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1200 * @bio: bio to add pages to
1201 * @iter: iov iterator describing the region to be mapped
1203 * Pins pages from *iter and appends them to @bio's bvec array. The
1204 * pages will have to be released using put_page() when done.
1205 * For multi-segment *iter, this function only adds pages from the
1206 * next non-empty segment of the iov iterator.
1208 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1210 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1211 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1212 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1213 struct page **pages = (struct page **)bv;
1214 unsigned int gup_flags = 0;
1216 unsigned len, i = 0;
1217 size_t offset, trim;
1221 * Move page array up in the allocated memory for the bio vecs as far as
1222 * possible so that we can start filling biovecs from the beginning
1223 * without overwriting the temporary page array.
1225 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1226 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1228 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1229 gup_flags |= FOLL_PCI_P2PDMA;
1232 * Each segment in the iov is required to be a block size multiple.
1233 * However, we may not be able to get the entire segment if it spans
1234 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1235 * result to ensure the bio's total size is correct. The remainder of
1236 * the iov data will be picked up in the next bio iteration.
1238 size = iov_iter_get_pages(iter, pages,
1239 UINT_MAX - bio->bi_iter.bi_size,
1240 nr_pages, &offset, gup_flags);
1241 if (unlikely(size <= 0))
1242 return size ? size : -EFAULT;
1244 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1246 trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1247 iov_iter_revert(iter, trim);
1250 if (unlikely(!size)) {
1255 for (left = size, i = 0; left > 0; left -= len, i++) {
1256 struct page *page = pages[i];
1258 len = min_t(size_t, PAGE_SIZE - offset, left);
1259 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1260 ret = bio_iov_add_zone_append_page(bio, page, len,
1265 bio_iov_add_page(bio, page, len, offset);
1270 iov_iter_revert(iter, left);
1272 while (i < nr_pages)
1273 put_page(pages[i++]);
1279 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1280 * @bio: bio to add pages to
1281 * @iter: iov iterator describing the region to be added
1283 * This takes either an iterator pointing to user memory, or one pointing to
1284 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1285 * map them into the kernel. On IO completion, the caller should put those
1286 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1287 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1288 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1289 * completed by a call to ->ki_complete() or returns with an error other than
1290 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1291 * on IO completion. If it isn't, then pages should be released.
1293 * The function tries, but does not guarantee, to pin as many pages as
1294 * fit into the bio, or are requested in @iter, whatever is smaller. If
1295 * MM encounters an error pinning the requested pages, it stops. Error
1296 * is returned only if 0 pages could be pinned.
1298 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1302 if (iov_iter_is_bvec(iter)) {
1303 bio_iov_bvec_set(bio, iter);
1304 iov_iter_advance(iter, bio->bi_iter.bi_size);
1309 ret = __bio_iov_iter_get_pages(bio, iter);
1310 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1312 return bio->bi_vcnt ? 0 : ret;
1314 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1316 static void submit_bio_wait_endio(struct bio *bio)
1318 complete(bio->bi_private);
1322 * submit_bio_wait - submit a bio, and wait until it completes
1323 * @bio: The &struct bio which describes the I/O
1325 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1326 * bio_endio() on failure.
1328 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1329 * result in bio reference to be consumed. The caller must drop the reference
1332 int submit_bio_wait(struct bio *bio)
1334 DECLARE_COMPLETION_ONSTACK_MAP(done,
1335 bio->bi_bdev->bd_disk->lockdep_map);
1336 unsigned long hang_check;
1338 bio->bi_private = &done;
1339 bio->bi_end_io = submit_bio_wait_endio;
1340 bio->bi_opf |= REQ_SYNC;
1343 /* Prevent hang_check timer from firing at us during very long I/O */
1344 hang_check = sysctl_hung_task_timeout_secs;
1346 while (!wait_for_completion_io_timeout(&done,
1347 hang_check * (HZ/2)))
1350 wait_for_completion_io(&done);
1352 return blk_status_to_errno(bio->bi_status);
1354 EXPORT_SYMBOL(submit_bio_wait);
1356 void __bio_advance(struct bio *bio, unsigned bytes)
1358 if (bio_integrity(bio))
1359 bio_integrity_advance(bio, bytes);
1361 bio_crypt_advance(bio, bytes);
1362 bio_advance_iter(bio, &bio->bi_iter, bytes);
1364 EXPORT_SYMBOL(__bio_advance);
1366 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1367 struct bio *src, struct bvec_iter *src_iter)
1369 while (src_iter->bi_size && dst_iter->bi_size) {
1370 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1371 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1372 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1373 void *src_buf = bvec_kmap_local(&src_bv);
1374 void *dst_buf = bvec_kmap_local(&dst_bv);
1376 memcpy(dst_buf, src_buf, bytes);
1378 kunmap_local(dst_buf);
1379 kunmap_local(src_buf);
1381 bio_advance_iter_single(src, src_iter, bytes);
1382 bio_advance_iter_single(dst, dst_iter, bytes);
1385 EXPORT_SYMBOL(bio_copy_data_iter);
1388 * bio_copy_data - copy contents of data buffers from one bio to another
1390 * @dst: destination bio
1392 * Stops when it reaches the end of either @src or @dst - that is, copies
1393 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1395 void bio_copy_data(struct bio *dst, struct bio *src)
1397 struct bvec_iter src_iter = src->bi_iter;
1398 struct bvec_iter dst_iter = dst->bi_iter;
1400 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1402 EXPORT_SYMBOL(bio_copy_data);
1404 void bio_free_pages(struct bio *bio)
1406 struct bio_vec *bvec;
1407 struct bvec_iter_all iter_all;
1409 bio_for_each_segment_all(bvec, bio, iter_all)
1410 __free_page(bvec->bv_page);
1412 EXPORT_SYMBOL(bio_free_pages);
1415 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1416 * for performing direct-IO in BIOs.
1418 * The problem is that we cannot run set_page_dirty() from interrupt context
1419 * because the required locks are not interrupt-safe. So what we can do is to
1420 * mark the pages dirty _before_ performing IO. And in interrupt context,
1421 * check that the pages are still dirty. If so, fine. If not, redirty them
1422 * in process context.
1424 * We special-case compound pages here: normally this means reads into hugetlb
1425 * pages. The logic in here doesn't really work right for compound pages
1426 * because the VM does not uniformly chase down the head page in all cases.
1427 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1428 * handle them at all. So we skip compound pages here at an early stage.
1430 * Note that this code is very hard to test under normal circumstances because
1431 * direct-io pins the pages with get_user_pages(). This makes
1432 * is_page_cache_freeable return false, and the VM will not clean the pages.
1433 * But other code (eg, flusher threads) could clean the pages if they are mapped
1436 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1437 * deferred bio dirtying paths.
1441 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1443 void bio_set_pages_dirty(struct bio *bio)
1445 struct bio_vec *bvec;
1446 struct bvec_iter_all iter_all;
1448 bio_for_each_segment_all(bvec, bio, iter_all) {
1449 if (!PageCompound(bvec->bv_page))
1450 set_page_dirty_lock(bvec->bv_page);
1455 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1456 * If they are, then fine. If, however, some pages are clean then they must
1457 * have been written out during the direct-IO read. So we take another ref on
1458 * the BIO and re-dirty the pages in process context.
1460 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1461 * here on. It will run one put_page() against each page and will run one
1462 * bio_put() against the BIO.
1465 static void bio_dirty_fn(struct work_struct *work);
1467 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1468 static DEFINE_SPINLOCK(bio_dirty_lock);
1469 static struct bio *bio_dirty_list;
1472 * This runs in process context
1474 static void bio_dirty_fn(struct work_struct *work)
1476 struct bio *bio, *next;
1478 spin_lock_irq(&bio_dirty_lock);
1479 next = bio_dirty_list;
1480 bio_dirty_list = NULL;
1481 spin_unlock_irq(&bio_dirty_lock);
1483 while ((bio = next) != NULL) {
1484 next = bio->bi_private;
1486 bio_release_pages(bio, true);
1491 void bio_check_pages_dirty(struct bio *bio)
1493 struct bio_vec *bvec;
1494 unsigned long flags;
1495 struct bvec_iter_all iter_all;
1497 bio_for_each_segment_all(bvec, bio, iter_all) {
1498 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1502 bio_release_pages(bio, false);
1506 spin_lock_irqsave(&bio_dirty_lock, flags);
1507 bio->bi_private = bio_dirty_list;
1508 bio_dirty_list = bio;
1509 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1510 schedule_work(&bio_dirty_work);
1513 static inline bool bio_remaining_done(struct bio *bio)
1516 * If we're not chaining, then ->__bi_remaining is always 1 and
1517 * we always end io on the first invocation.
1519 if (!bio_flagged(bio, BIO_CHAIN))
1522 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1524 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1525 bio_clear_flag(bio, BIO_CHAIN);
1533 * bio_endio - end I/O on a bio
1537 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1538 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1539 * bio unless they own it and thus know that it has an end_io function.
1541 * bio_endio() can be called several times on a bio that has been chained
1542 * using bio_chain(). The ->bi_end_io() function will only be called the
1545 void bio_endio(struct bio *bio)
1548 if (!bio_remaining_done(bio))
1550 if (!bio_integrity_endio(bio))
1553 rq_qos_done_bio(bio);
1555 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1556 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1557 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1561 * Need to have a real endio function for chained bios, otherwise
1562 * various corner cases will break (like stacking block devices that
1563 * save/restore bi_end_io) - however, we want to avoid unbounded
1564 * recursion and blowing the stack. Tail call optimization would
1565 * handle this, but compiling with frame pointers also disables
1566 * gcc's sibling call optimization.
1568 if (bio->bi_end_io == bio_chain_endio) {
1569 bio = __bio_chain_endio(bio);
1573 blk_throtl_bio_endio(bio);
1574 /* release cgroup info */
1577 bio->bi_end_io(bio);
1579 EXPORT_SYMBOL(bio_endio);
1582 * bio_split - split a bio
1583 * @bio: bio to split
1584 * @sectors: number of sectors to split from the front of @bio
1586 * @bs: bio set to allocate from
1588 * Allocates and returns a new bio which represents @sectors from the start of
1589 * @bio, and updates @bio to represent the remaining sectors.
1591 * Unless this is a discard request the newly allocated bio will point
1592 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1593 * neither @bio nor @bs are freed before the split bio.
1595 struct bio *bio_split(struct bio *bio, int sectors,
1596 gfp_t gfp, struct bio_set *bs)
1600 BUG_ON(sectors <= 0);
1601 BUG_ON(sectors >= bio_sectors(bio));
1603 /* Zone append commands cannot be split */
1604 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1607 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1611 split->bi_iter.bi_size = sectors << 9;
1613 if (bio_integrity(split))
1614 bio_integrity_trim(split);
1616 bio_advance(bio, split->bi_iter.bi_size);
1618 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1619 bio_set_flag(split, BIO_TRACE_COMPLETION);
1623 EXPORT_SYMBOL(bio_split);
1626 * bio_trim - trim a bio
1628 * @offset: number of sectors to trim from the front of @bio
1629 * @size: size we want to trim @bio to, in sectors
1631 * This function is typically used for bios that are cloned and submitted
1632 * to the underlying device in parts.
1634 void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1636 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1637 offset + size > bio_sectors(bio)))
1641 if (offset == 0 && size == bio->bi_iter.bi_size)
1644 bio_advance(bio, offset << 9);
1645 bio->bi_iter.bi_size = size;
1647 if (bio_integrity(bio))
1648 bio_integrity_trim(bio);
1650 EXPORT_SYMBOL_GPL(bio_trim);
1653 * create memory pools for biovec's in a bio_set.
1654 * use the global biovec slabs created for general use.
1656 int biovec_init_pool(mempool_t *pool, int pool_entries)
1658 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1660 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1664 * bioset_exit - exit a bioset initialized with bioset_init()
1666 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1669 void bioset_exit(struct bio_set *bs)
1671 bio_alloc_cache_destroy(bs);
1672 if (bs->rescue_workqueue)
1673 destroy_workqueue(bs->rescue_workqueue);
1674 bs->rescue_workqueue = NULL;
1676 mempool_exit(&bs->bio_pool);
1677 mempool_exit(&bs->bvec_pool);
1679 bioset_integrity_free(bs);
1682 bs->bio_slab = NULL;
1684 EXPORT_SYMBOL(bioset_exit);
1687 * bioset_init - Initialize a bio_set
1688 * @bs: pool to initialize
1689 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1690 * @front_pad: Number of bytes to allocate in front of the returned bio
1691 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1692 * and %BIOSET_NEED_RESCUER
1695 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1696 * to ask for a number of bytes to be allocated in front of the bio.
1697 * Front pad allocation is useful for embedding the bio inside
1698 * another structure, to avoid allocating extra data to go with the bio.
1699 * Note that the bio must be embedded at the END of that structure always,
1700 * or things will break badly.
1701 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1702 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1703 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1704 * to dispatch queued requests when the mempool runs out of space.
1707 int bioset_init(struct bio_set *bs,
1708 unsigned int pool_size,
1709 unsigned int front_pad,
1712 bs->front_pad = front_pad;
1713 if (flags & BIOSET_NEED_BVECS)
1714 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1718 spin_lock_init(&bs->rescue_lock);
1719 bio_list_init(&bs->rescue_list);
1720 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1722 bs->bio_slab = bio_find_or_create_slab(bs);
1726 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1729 if ((flags & BIOSET_NEED_BVECS) &&
1730 biovec_init_pool(&bs->bvec_pool, pool_size))
1733 if (flags & BIOSET_NEED_RESCUER) {
1734 bs->rescue_workqueue = alloc_workqueue("bioset",
1736 if (!bs->rescue_workqueue)
1739 if (flags & BIOSET_PERCPU_CACHE) {
1740 bs->cache = alloc_percpu(struct bio_alloc_cache);
1743 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1751 EXPORT_SYMBOL(bioset_init);
1753 static int __init init_bio(void)
1757 bio_integrity_init();
1759 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1760 struct biovec_slab *bvs = bvec_slabs + i;
1762 bvs->slab = kmem_cache_create(bvs->name,
1763 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1764 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1767 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1770 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1771 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1772 panic("bio: can't allocate bios\n");
1774 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1775 panic("bio: can't create integrity pool\n");
1779 subsys_initcall(init_bio);