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/blk-cgroup.h>
19 #include <linux/highmem.h>
20 #include <linux/sched/sysctl.h>
21 #include <linux/blk-crypto.h>
22 #include <linux/xarray.h>
24 #include <trace/events/block.h>
26 #include "blk-rq-qos.h"
28 struct bio_alloc_cache {
29 struct bio_list 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, SLAB_HWCACHE_ALIGN, NULL);
95 bslab->slab_size = size;
97 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
100 kmem_cache_destroy(bslab->slab);
107 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
109 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
112 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
114 unsigned int size = bs_bio_slab_size(bs);
115 struct bio_slab *bslab;
117 mutex_lock(&bio_slab_lock);
118 bslab = xa_load(&bio_slabs, size);
122 bslab = create_bio_slab(size);
123 mutex_unlock(&bio_slab_lock);
130 static void bio_put_slab(struct bio_set *bs)
132 struct bio_slab *bslab = NULL;
133 unsigned int slab_size = bs_bio_slab_size(bs);
135 mutex_lock(&bio_slab_lock);
137 bslab = xa_load(&bio_slabs, slab_size);
138 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
141 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
143 WARN_ON(!bslab->slab_ref);
145 if (--bslab->slab_ref)
148 xa_erase(&bio_slabs, slab_size);
150 kmem_cache_destroy(bslab->slab);
154 mutex_unlock(&bio_slab_lock);
157 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
159 BIO_BUG_ON(nr_vecs > BIO_MAX_VECS);
161 if (nr_vecs == BIO_MAX_VECS)
162 mempool_free(bv, pool);
163 else if (nr_vecs > BIO_INLINE_VECS)
164 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
168 * Make the first allocation restricted and don't dump info on allocation
169 * failures, since we'll fall back to the mempool in case of failure.
171 static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
173 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
174 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
177 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
180 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
182 if (WARN_ON_ONCE(!bvs))
186 * Upgrade the nr_vecs request to take full advantage of the allocation.
187 * We also rely on this in the bvec_free path.
189 *nr_vecs = bvs->nr_vecs;
192 * Try a slab allocation first for all smaller allocations. If that
193 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
194 * The mempool is sized to handle up to BIO_MAX_VECS entries.
196 if (*nr_vecs < BIO_MAX_VECS) {
199 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
200 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
202 *nr_vecs = BIO_MAX_VECS;
205 return mempool_alloc(pool, gfp_mask);
208 void bio_uninit(struct bio *bio)
210 #ifdef CONFIG_BLK_CGROUP
212 blkg_put(bio->bi_blkg);
216 if (bio_integrity(bio))
217 bio_integrity_free(bio);
219 bio_crypt_free_ctx(bio);
221 EXPORT_SYMBOL(bio_uninit);
223 static void bio_free(struct bio *bio)
225 struct bio_set *bs = bio->bi_pool;
231 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
234 * If we have front padding, adjust the bio pointer before freeing
239 mempool_free(p, &bs->bio_pool);
241 /* Bio was allocated by bio_kmalloc() */
247 * Users of this function have their own bio allocation. Subsequently,
248 * they must remember to pair any call to bio_init() with bio_uninit()
249 * when IO has completed, or when the bio is released.
251 void bio_init(struct bio *bio, struct bio_vec *table,
252 unsigned short max_vecs)
259 bio->bi_write_hint = 0;
261 bio->bi_iter.bi_sector = 0;
262 bio->bi_iter.bi_size = 0;
263 bio->bi_iter.bi_idx = 0;
264 bio->bi_iter.bi_bvec_done = 0;
265 bio->bi_end_io = NULL;
266 bio->bi_private = NULL;
267 #ifdef CONFIG_BLK_CGROUP
269 bio->bi_issue.value = 0;
270 #ifdef CONFIG_BLK_CGROUP_IOCOST
271 bio->bi_iocost_cost = 0;
274 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
275 bio->bi_crypt_context = NULL;
277 #ifdef CONFIG_BLK_DEV_INTEGRITY
278 bio->bi_integrity = NULL;
282 atomic_set(&bio->__bi_remaining, 1);
283 atomic_set(&bio->__bi_cnt, 1);
285 bio->bi_max_vecs = max_vecs;
286 bio->bi_io_vec = table;
289 EXPORT_SYMBOL(bio_init);
292 * bio_reset - reinitialize a bio
296 * After calling bio_reset(), @bio will be in the same state as a freshly
297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
298 * preserved are the ones that are initialized by bio_alloc_bioset(). See
299 * comment in struct bio.
301 void bio_reset(struct bio *bio)
304 memset(bio, 0, BIO_RESET_BYTES);
305 atomic_set(&bio->__bi_remaining, 1);
307 EXPORT_SYMBOL(bio_reset);
309 static struct bio *__bio_chain_endio(struct bio *bio)
311 struct bio *parent = bio->bi_private;
313 if (bio->bi_status && !parent->bi_status)
314 parent->bi_status = bio->bi_status;
319 static void bio_chain_endio(struct bio *bio)
321 bio_endio(__bio_chain_endio(bio));
325 * bio_chain - chain bio completions
326 * @bio: the target bio
327 * @parent: the parent bio of @bio
329 * The caller won't have a bi_end_io called when @bio completes - instead,
330 * @parent's bi_end_io won't be called until both @parent and @bio have
331 * completed; the chained bio will also be freed when it completes.
333 * The caller must not set bi_private or bi_end_io in @bio.
335 void bio_chain(struct bio *bio, struct bio *parent)
337 BUG_ON(bio->bi_private || bio->bi_end_io);
339 bio->bi_private = parent;
340 bio->bi_end_io = bio_chain_endio;
341 bio_inc_remaining(parent);
343 EXPORT_SYMBOL(bio_chain);
345 static void bio_alloc_rescue(struct work_struct *work)
347 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
351 spin_lock(&bs->rescue_lock);
352 bio = bio_list_pop(&bs->rescue_list);
353 spin_unlock(&bs->rescue_lock);
358 submit_bio_noacct(bio);
362 static void punt_bios_to_rescuer(struct bio_set *bs)
364 struct bio_list punt, nopunt;
367 if (WARN_ON_ONCE(!bs->rescue_workqueue))
370 * In order to guarantee forward progress we must punt only bios that
371 * were allocated from this bio_set; otherwise, if there was a bio on
372 * there for a stacking driver higher up in the stack, processing it
373 * could require allocating bios from this bio_set, and doing that from
374 * our own rescuer would be bad.
376 * Since bio lists are singly linked, pop them all instead of trying to
377 * remove from the middle of the list:
380 bio_list_init(&punt);
381 bio_list_init(&nopunt);
383 while ((bio = bio_list_pop(¤t->bio_list[0])))
384 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
385 current->bio_list[0] = nopunt;
387 bio_list_init(&nopunt);
388 while ((bio = bio_list_pop(¤t->bio_list[1])))
389 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
390 current->bio_list[1] = nopunt;
392 spin_lock(&bs->rescue_lock);
393 bio_list_merge(&bs->rescue_list, &punt);
394 spin_unlock(&bs->rescue_lock);
396 queue_work(bs->rescue_workqueue, &bs->rescue_work);
400 * bio_alloc_bioset - allocate a bio for I/O
401 * @gfp_mask: the GFP_* mask given to the slab allocator
402 * @nr_iovecs: number of iovecs to pre-allocate
403 * @bs: the bio_set to allocate from.
405 * Allocate a bio from the mempools in @bs.
407 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
408 * allocate a bio. This is due to the mempool guarantees. To make this work,
409 * callers must never allocate more than 1 bio at a time from the general pool.
410 * Callers that need to allocate more than 1 bio must always submit the
411 * previously allocated bio for IO before attempting to allocate a new one.
412 * Failure to do so can cause deadlocks under memory pressure.
414 * Note that when running under submit_bio_noacct() (i.e. any block driver),
415 * bios are not submitted until after you return - see the code in
416 * submit_bio_noacct() that converts recursion into iteration, to prevent
419 * This would normally mean allocating multiple bios under submit_bio_noacct()
420 * would be susceptible to deadlocks, but we have
421 * deadlock avoidance code that resubmits any blocked bios from a rescuer
424 * However, we do not guarantee forward progress for allocations from other
425 * mempools. Doing multiple allocations from the same mempool under
426 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
427 * for per bio allocations.
429 * Returns: Pointer to new bio on success, NULL on failure.
431 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs,
434 gfp_t saved_gfp = gfp_mask;
438 /* should not use nobvec bioset for nr_iovecs > 0 */
439 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0))
443 * submit_bio_noacct() converts recursion to iteration; this means if
444 * we're running beneath it, any bios we allocate and submit will not be
445 * submitted (and thus freed) until after we return.
447 * This exposes us to a potential deadlock if we allocate multiple bios
448 * from the same bio_set() while running underneath submit_bio_noacct().
449 * If we were to allocate multiple bios (say a stacking block driver
450 * that was splitting bios), we would deadlock if we exhausted the
453 * We solve this, and guarantee forward progress, with a rescuer
454 * workqueue per bio_set. If we go to allocate and there are bios on
455 * current->bio_list, we first try the allocation without
456 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
457 * blocking to the rescuer workqueue before we retry with the original
460 if (current->bio_list &&
461 (!bio_list_empty(¤t->bio_list[0]) ||
462 !bio_list_empty(¤t->bio_list[1])) &&
463 bs->rescue_workqueue)
464 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
466 p = mempool_alloc(&bs->bio_pool, gfp_mask);
467 if (!p && gfp_mask != saved_gfp) {
468 punt_bios_to_rescuer(bs);
469 gfp_mask = saved_gfp;
470 p = mempool_alloc(&bs->bio_pool, gfp_mask);
475 bio = p + bs->front_pad;
476 if (nr_iovecs > BIO_INLINE_VECS) {
477 struct bio_vec *bvl = NULL;
479 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
480 if (!bvl && gfp_mask != saved_gfp) {
481 punt_bios_to_rescuer(bs);
482 gfp_mask = saved_gfp;
483 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
488 bio_init(bio, bvl, nr_iovecs);
489 } else if (nr_iovecs) {
490 bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS);
492 bio_init(bio, NULL, 0);
499 mempool_free(p, &bs->bio_pool);
502 EXPORT_SYMBOL(bio_alloc_bioset);
505 * bio_kmalloc - kmalloc a bio for I/O
506 * @gfp_mask: the GFP_* mask given to the slab allocator
507 * @nr_iovecs: number of iovecs to pre-allocate
509 * Use kmalloc to allocate and initialize a bio.
511 * Returns: Pointer to new bio on success, NULL on failure.
513 struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs)
517 if (nr_iovecs > UIO_MAXIOV)
520 bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask);
523 bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs);
527 EXPORT_SYMBOL(bio_kmalloc);
529 void zero_fill_bio(struct bio *bio)
533 struct bvec_iter iter;
535 bio_for_each_segment(bv, bio, iter) {
536 char *data = bvec_kmap_irq(&bv, &flags);
537 memset(data, 0, bv.bv_len);
538 flush_dcache_page(bv.bv_page);
539 bvec_kunmap_irq(data, &flags);
542 EXPORT_SYMBOL(zero_fill_bio);
545 * bio_truncate - truncate the bio to small size of @new_size
546 * @bio: the bio to be truncated
547 * @new_size: new size for truncating the bio
550 * Truncate the bio to new size of @new_size. If bio_op(bio) is
551 * REQ_OP_READ, zero the truncated part. This function should only
552 * be used for handling corner cases, such as bio eod.
554 void bio_truncate(struct bio *bio, unsigned new_size)
557 struct bvec_iter iter;
558 unsigned int done = 0;
559 bool truncated = false;
561 if (new_size >= bio->bi_iter.bi_size)
564 if (bio_op(bio) != REQ_OP_READ)
567 bio_for_each_segment(bv, bio, iter) {
568 if (done + bv.bv_len > new_size) {
572 offset = new_size - done;
575 zero_user(bv.bv_page, offset, bv.bv_len - offset);
583 * Don't touch bvec table here and make it really immutable, since
584 * fs bio user has to retrieve all pages via bio_for_each_segment_all
585 * in its .end_bio() callback.
587 * It is enough to truncate bio by updating .bi_size since we can make
588 * correct bvec with the updated .bi_size for drivers.
590 bio->bi_iter.bi_size = new_size;
594 * guard_bio_eod - truncate a BIO to fit the block device
595 * @bio: bio to truncate
597 * This allows us to do IO even on the odd last sectors of a device, even if the
598 * block size is some multiple of the physical sector size.
600 * We'll just truncate the bio to the size of the device, and clear the end of
601 * the buffer head manually. Truly out-of-range accesses will turn into actual
602 * I/O errors, this only handles the "we need to be able to do I/O at the final
605 void guard_bio_eod(struct bio *bio)
607 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
613 * If the *whole* IO is past the end of the device,
614 * let it through, and the IO layer will turn it into
617 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
620 maxsector -= bio->bi_iter.bi_sector;
621 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
624 bio_truncate(bio, maxsector << 9);
627 #define ALLOC_CACHE_MAX 512
628 #define ALLOC_CACHE_SLACK 64
630 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
636 while ((bio = bio_list_pop(&cache->free_list)) != NULL) {
644 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
648 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
650 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
652 bio_alloc_cache_prune(cache, -1U);
657 static void bio_alloc_cache_destroy(struct bio_set *bs)
664 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
665 for_each_possible_cpu(cpu) {
666 struct bio_alloc_cache *cache;
668 cache = per_cpu_ptr(bs->cache, cpu);
669 bio_alloc_cache_prune(cache, -1U);
671 free_percpu(bs->cache);
675 * bio_put - release a reference to a bio
676 * @bio: bio to release reference to
679 * Put a reference to a &struct bio, either one you have gotten with
680 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
682 void bio_put(struct bio *bio)
684 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
685 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
686 if (!atomic_dec_and_test(&bio->__bi_cnt))
690 if (bio_flagged(bio, BIO_PERCPU_CACHE)) {
691 struct bio_alloc_cache *cache;
694 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
695 bio_list_add_head(&cache->free_list, bio);
696 if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK)
697 bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK);
703 EXPORT_SYMBOL(bio_put);
706 * __bio_clone_fast - clone a bio that shares the original bio's biovec
707 * @bio: destination bio
708 * @bio_src: bio to clone
710 * Clone a &bio. Caller will own the returned bio, but not
711 * the actual data it points to. Reference count of returned
714 * Caller must ensure that @bio_src is not freed before @bio.
716 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
718 WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs);
721 * most users will be overriding ->bi_bdev with a new target,
722 * so we don't set nor calculate new physical/hw segment counts here
724 bio->bi_bdev = bio_src->bi_bdev;
725 bio_set_flag(bio, BIO_CLONED);
726 if (bio_flagged(bio_src, BIO_THROTTLED))
727 bio_set_flag(bio, BIO_THROTTLED);
728 if (bio_flagged(bio_src, BIO_REMAPPED))
729 bio_set_flag(bio, BIO_REMAPPED);
730 bio->bi_opf = bio_src->bi_opf;
731 bio->bi_ioprio = bio_src->bi_ioprio;
732 bio->bi_write_hint = bio_src->bi_write_hint;
733 bio->bi_iter = bio_src->bi_iter;
734 bio->bi_io_vec = bio_src->bi_io_vec;
736 bio_clone_blkg_association(bio, bio_src);
737 blkcg_bio_issue_init(bio);
739 EXPORT_SYMBOL(__bio_clone_fast);
742 * bio_clone_fast - clone a bio that shares the original bio's biovec
744 * @gfp_mask: allocation priority
745 * @bs: bio_set to allocate from
747 * Like __bio_clone_fast, only also allocates the returned bio
749 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
753 b = bio_alloc_bioset(gfp_mask, 0, bs);
757 __bio_clone_fast(b, bio);
759 if (bio_crypt_clone(b, bio, gfp_mask) < 0)
762 if (bio_integrity(bio) &&
763 bio_integrity_clone(b, bio, gfp_mask) < 0)
772 EXPORT_SYMBOL(bio_clone_fast);
774 const char *bio_devname(struct bio *bio, char *buf)
776 return bdevname(bio->bi_bdev, buf);
778 EXPORT_SYMBOL(bio_devname);
780 static inline bool page_is_mergeable(const struct bio_vec *bv,
781 struct page *page, unsigned int len, unsigned int off,
784 size_t bv_end = bv->bv_offset + bv->bv_len;
785 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
786 phys_addr_t page_addr = page_to_phys(page);
788 if (vec_end_addr + 1 != page_addr + off)
790 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
793 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
796 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
800 * Try to merge a page into a segment, while obeying the hardware segment
801 * size limit. This is not for normal read/write bios, but for passthrough
802 * or Zone Append operations that we can't split.
804 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
805 struct page *page, unsigned len,
806 unsigned offset, bool *same_page)
808 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
809 unsigned long mask = queue_segment_boundary(q);
810 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
811 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
813 if ((addr1 | mask) != (addr2 | mask))
815 if (bv->bv_len + len > queue_max_segment_size(q))
817 return __bio_try_merge_page(bio, page, len, offset, same_page);
821 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
822 * @q: the target queue
823 * @bio: destination bio
825 * @len: vec entry length
826 * @offset: vec entry offset
827 * @max_sectors: maximum number of sectors that can be added
828 * @same_page: return if the segment has been merged inside the same page
830 * Add a page to a bio while respecting the hardware max_sectors, max_segment
831 * and gap limitations.
833 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
834 struct page *page, unsigned int len, unsigned int offset,
835 unsigned int max_sectors, bool *same_page)
837 struct bio_vec *bvec;
839 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
842 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
845 if (bio->bi_vcnt > 0) {
846 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
850 * If the queue doesn't support SG gaps and adding this segment
851 * would create a gap, disallow it.
853 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
854 if (bvec_gap_to_prev(q, bvec, offset))
858 if (bio_full(bio, len))
861 if (bio->bi_vcnt >= queue_max_segments(q))
864 bvec = &bio->bi_io_vec[bio->bi_vcnt];
865 bvec->bv_page = page;
867 bvec->bv_offset = offset;
869 bio->bi_iter.bi_size += len;
874 * bio_add_pc_page - attempt to add page to passthrough bio
875 * @q: the target queue
876 * @bio: destination bio
878 * @len: vec entry length
879 * @offset: vec entry offset
881 * Attempt to add a page to the bio_vec maplist. This can fail for a
882 * number of reasons, such as the bio being full or target block device
883 * limitations. The target block device must allow bio's up to PAGE_SIZE,
884 * so it is always possible to add a single page to an empty bio.
886 * This should only be used by passthrough bios.
888 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
889 struct page *page, unsigned int len, unsigned int offset)
891 bool same_page = false;
892 return bio_add_hw_page(q, bio, page, len, offset,
893 queue_max_hw_sectors(q), &same_page);
895 EXPORT_SYMBOL(bio_add_pc_page);
898 * bio_add_zone_append_page - attempt to add page to zone-append bio
899 * @bio: destination bio
901 * @len: vec entry length
902 * @offset: vec entry offset
904 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
905 * for a zone-append request. This can fail for a number of reasons, such as the
906 * bio being full or the target block device is not a zoned block device or
907 * other limitations of the target block device. The target block device must
908 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
911 * Returns: number of bytes added to the bio, or 0 in case of a failure.
913 int bio_add_zone_append_page(struct bio *bio, struct page *page,
914 unsigned int len, unsigned int offset)
916 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
917 bool same_page = false;
919 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
922 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
925 return bio_add_hw_page(q, bio, page, len, offset,
926 queue_max_zone_append_sectors(q), &same_page);
928 EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
931 * __bio_try_merge_page - try appending data to an existing bvec.
932 * @bio: destination bio
933 * @page: start page to add
934 * @len: length of the data to add
935 * @off: offset of the data relative to @page
936 * @same_page: return if the segment has been merged inside the same page
938 * Try to add the data at @page + @off to the last bvec of @bio. This is a
939 * useful optimisation for file systems with a block size smaller than the
942 * Warn if (@len, @off) crosses pages in case that @same_page is true.
944 * Return %true on success or %false on failure.
946 bool __bio_try_merge_page(struct bio *bio, struct page *page,
947 unsigned int len, unsigned int off, bool *same_page)
949 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
952 if (bio->bi_vcnt > 0) {
953 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
955 if (page_is_mergeable(bv, page, len, off, same_page)) {
956 if (bio->bi_iter.bi_size > UINT_MAX - len) {
961 bio->bi_iter.bi_size += len;
967 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
970 * __bio_add_page - add page(s) to a bio in a new segment
971 * @bio: destination bio
972 * @page: start page to add
973 * @len: length of the data to add, may cross pages
974 * @off: offset of the data relative to @page, may cross pages
976 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
977 * that @bio has space for another bvec.
979 void __bio_add_page(struct bio *bio, struct page *page,
980 unsigned int len, unsigned int off)
982 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
984 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
985 WARN_ON_ONCE(bio_full(bio, len));
991 bio->bi_iter.bi_size += len;
994 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
995 bio_set_flag(bio, BIO_WORKINGSET);
997 EXPORT_SYMBOL_GPL(__bio_add_page);
1000 * bio_add_page - attempt to add page(s) to bio
1001 * @bio: destination bio
1002 * @page: start page to add
1003 * @len: vec entry length, may cross pages
1004 * @offset: vec entry offset relative to @page, may cross pages
1006 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1007 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1009 int bio_add_page(struct bio *bio, struct page *page,
1010 unsigned int len, unsigned int offset)
1012 bool same_page = false;
1014 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1015 if (bio_full(bio, len))
1017 __bio_add_page(bio, page, len, offset);
1021 EXPORT_SYMBOL(bio_add_page);
1023 void bio_release_pages(struct bio *bio, bool mark_dirty)
1025 struct bvec_iter_all iter_all;
1026 struct bio_vec *bvec;
1028 if (bio_flagged(bio, BIO_NO_PAGE_REF))
1031 bio_for_each_segment_all(bvec, bio, iter_all) {
1032 if (mark_dirty && !PageCompound(bvec->bv_page))
1033 set_page_dirty_lock(bvec->bv_page);
1034 put_page(bvec->bv_page);
1037 EXPORT_SYMBOL_GPL(bio_release_pages);
1039 static void __bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1041 WARN_ON_ONCE(bio->bi_max_vecs);
1043 bio->bi_vcnt = iter->nr_segs;
1044 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1045 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1046 bio->bi_iter.bi_size = iter->count;
1047 bio_set_flag(bio, BIO_NO_PAGE_REF);
1048 bio_set_flag(bio, BIO_CLONED);
1051 static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1053 __bio_iov_bvec_set(bio, iter);
1054 iov_iter_advance(iter, iter->count);
1058 static int bio_iov_bvec_set_append(struct bio *bio, struct iov_iter *iter)
1060 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1061 struct iov_iter i = *iter;
1063 iov_iter_truncate(&i, queue_max_zone_append_sectors(q) << 9);
1064 __bio_iov_bvec_set(bio, &i);
1065 iov_iter_advance(iter, i.count);
1069 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1072 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1073 * @bio: bio to add pages to
1074 * @iter: iov iterator describing the region to be mapped
1076 * Pins pages from *iter and appends them to @bio's bvec array. The
1077 * pages will have to be released using put_page() when done.
1078 * For multi-segment *iter, this function only adds pages from the
1079 * next non-empty segment of the iov iterator.
1081 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1083 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1084 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1085 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1086 struct page **pages = (struct page **)bv;
1087 bool same_page = false;
1093 * Move page array up in the allocated memory for the bio vecs as far as
1094 * possible so that we can start filling biovecs from the beginning
1095 * without overwriting the temporary page array.
1097 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1098 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1100 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1101 if (unlikely(size <= 0))
1102 return size ? size : -EFAULT;
1104 for (left = size, i = 0; left > 0; left -= len, i++) {
1105 struct page *page = pages[i];
1107 len = min_t(size_t, PAGE_SIZE - offset, left);
1109 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1113 if (WARN_ON_ONCE(bio_full(bio, len)))
1115 __bio_add_page(bio, page, len, offset);
1120 iov_iter_advance(iter, size);
1124 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
1126 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1127 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1128 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1129 unsigned int max_append_sectors = queue_max_zone_append_sectors(q);
1130 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1131 struct page **pages = (struct page **)bv;
1137 if (WARN_ON_ONCE(!max_append_sectors))
1141 * Move page array up in the allocated memory for the bio vecs as far as
1142 * possible so that we can start filling biovecs from the beginning
1143 * without overwriting the temporary page array.
1145 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1146 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1148 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1149 if (unlikely(size <= 0))
1150 return size ? size : -EFAULT;
1152 for (left = size, i = 0; left > 0; left -= len, i++) {
1153 struct page *page = pages[i];
1154 bool same_page = false;
1156 len = min_t(size_t, PAGE_SIZE - offset, left);
1157 if (bio_add_hw_page(q, bio, page, len, offset,
1158 max_append_sectors, &same_page) != len) {
1167 iov_iter_advance(iter, size - left);
1172 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1173 * @bio: bio to add pages to
1174 * @iter: iov iterator describing the region to be added
1176 * This takes either an iterator pointing to user memory, or one pointing to
1177 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1178 * map them into the kernel. On IO completion, the caller should put those
1179 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1180 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1181 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1182 * completed by a call to ->ki_complete() or returns with an error other than
1183 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1184 * on IO completion. If it isn't, then pages should be released.
1186 * The function tries, but does not guarantee, to pin as many pages as
1187 * fit into the bio, or are requested in @iter, whatever is smaller. If
1188 * MM encounters an error pinning the requested pages, it stops. Error
1189 * is returned only if 0 pages could be pinned.
1191 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1192 * responsible for setting BIO_WORKINGSET if necessary.
1194 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1198 if (iov_iter_is_bvec(iter)) {
1199 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1200 return bio_iov_bvec_set_append(bio, iter);
1201 return bio_iov_bvec_set(bio, iter);
1205 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1206 ret = __bio_iov_append_get_pages(bio, iter);
1208 ret = __bio_iov_iter_get_pages(bio, iter);
1209 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1211 /* don't account direct I/O as memory stall */
1212 bio_clear_flag(bio, BIO_WORKINGSET);
1213 return bio->bi_vcnt ? 0 : ret;
1215 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1217 static void submit_bio_wait_endio(struct bio *bio)
1219 complete(bio->bi_private);
1223 * submit_bio_wait - submit a bio, and wait until it completes
1224 * @bio: The &struct bio which describes the I/O
1226 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1227 * bio_endio() on failure.
1229 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1230 * result in bio reference to be consumed. The caller must drop the reference
1233 int submit_bio_wait(struct bio *bio)
1235 DECLARE_COMPLETION_ONSTACK_MAP(done,
1236 bio->bi_bdev->bd_disk->lockdep_map);
1237 unsigned long hang_check;
1239 bio->bi_private = &done;
1240 bio->bi_end_io = submit_bio_wait_endio;
1241 bio->bi_opf |= REQ_SYNC;
1244 /* Prevent hang_check timer from firing at us during very long I/O */
1245 hang_check = sysctl_hung_task_timeout_secs;
1247 while (!wait_for_completion_io_timeout(&done,
1248 hang_check * (HZ/2)))
1251 wait_for_completion_io(&done);
1253 return blk_status_to_errno(bio->bi_status);
1255 EXPORT_SYMBOL(submit_bio_wait);
1258 * bio_advance - increment/complete a bio by some number of bytes
1259 * @bio: bio to advance
1260 * @bytes: number of bytes to complete
1262 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1263 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1264 * be updated on the last bvec as well.
1266 * @bio will then represent the remaining, uncompleted portion of the io.
1268 void bio_advance(struct bio *bio, unsigned bytes)
1270 if (bio_integrity(bio))
1271 bio_integrity_advance(bio, bytes);
1273 bio_crypt_advance(bio, bytes);
1274 bio_advance_iter(bio, &bio->bi_iter, bytes);
1276 EXPORT_SYMBOL(bio_advance);
1278 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1279 struct bio *src, struct bvec_iter *src_iter)
1281 struct bio_vec src_bv, dst_bv;
1282 void *src_p, *dst_p;
1285 while (src_iter->bi_size && dst_iter->bi_size) {
1286 src_bv = bio_iter_iovec(src, *src_iter);
1287 dst_bv = bio_iter_iovec(dst, *dst_iter);
1289 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1291 src_p = kmap_atomic(src_bv.bv_page);
1292 dst_p = kmap_atomic(dst_bv.bv_page);
1294 memcpy(dst_p + dst_bv.bv_offset,
1295 src_p + src_bv.bv_offset,
1298 kunmap_atomic(dst_p);
1299 kunmap_atomic(src_p);
1301 flush_dcache_page(dst_bv.bv_page);
1303 bio_advance_iter_single(src, src_iter, bytes);
1304 bio_advance_iter_single(dst, dst_iter, bytes);
1307 EXPORT_SYMBOL(bio_copy_data_iter);
1310 * bio_copy_data - copy contents of data buffers from one bio to another
1312 * @dst: destination bio
1314 * Stops when it reaches the end of either @src or @dst - that is, copies
1315 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1317 void bio_copy_data(struct bio *dst, struct bio *src)
1319 struct bvec_iter src_iter = src->bi_iter;
1320 struct bvec_iter dst_iter = dst->bi_iter;
1322 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1324 EXPORT_SYMBOL(bio_copy_data);
1326 void bio_free_pages(struct bio *bio)
1328 struct bio_vec *bvec;
1329 struct bvec_iter_all iter_all;
1331 bio_for_each_segment_all(bvec, bio, iter_all)
1332 __free_page(bvec->bv_page);
1334 EXPORT_SYMBOL(bio_free_pages);
1337 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1338 * for performing direct-IO in BIOs.
1340 * The problem is that we cannot run set_page_dirty() from interrupt context
1341 * because the required locks are not interrupt-safe. So what we can do is to
1342 * mark the pages dirty _before_ performing IO. And in interrupt context,
1343 * check that the pages are still dirty. If so, fine. If not, redirty them
1344 * in process context.
1346 * We special-case compound pages here: normally this means reads into hugetlb
1347 * pages. The logic in here doesn't really work right for compound pages
1348 * because the VM does not uniformly chase down the head page in all cases.
1349 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1350 * handle them at all. So we skip compound pages here at an early stage.
1352 * Note that this code is very hard to test under normal circumstances because
1353 * direct-io pins the pages with get_user_pages(). This makes
1354 * is_page_cache_freeable return false, and the VM will not clean the pages.
1355 * But other code (eg, flusher threads) could clean the pages if they are mapped
1358 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1359 * deferred bio dirtying paths.
1363 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1365 void bio_set_pages_dirty(struct bio *bio)
1367 struct bio_vec *bvec;
1368 struct bvec_iter_all iter_all;
1370 bio_for_each_segment_all(bvec, bio, iter_all) {
1371 if (!PageCompound(bvec->bv_page))
1372 set_page_dirty_lock(bvec->bv_page);
1377 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1378 * If they are, then fine. If, however, some pages are clean then they must
1379 * have been written out during the direct-IO read. So we take another ref on
1380 * the BIO and re-dirty the pages in process context.
1382 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1383 * here on. It will run one put_page() against each page and will run one
1384 * bio_put() against the BIO.
1387 static void bio_dirty_fn(struct work_struct *work);
1389 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1390 static DEFINE_SPINLOCK(bio_dirty_lock);
1391 static struct bio *bio_dirty_list;
1394 * This runs in process context
1396 static void bio_dirty_fn(struct work_struct *work)
1398 struct bio *bio, *next;
1400 spin_lock_irq(&bio_dirty_lock);
1401 next = bio_dirty_list;
1402 bio_dirty_list = NULL;
1403 spin_unlock_irq(&bio_dirty_lock);
1405 while ((bio = next) != NULL) {
1406 next = bio->bi_private;
1408 bio_release_pages(bio, true);
1413 void bio_check_pages_dirty(struct bio *bio)
1415 struct bio_vec *bvec;
1416 unsigned long flags;
1417 struct bvec_iter_all iter_all;
1419 bio_for_each_segment_all(bvec, bio, iter_all) {
1420 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1424 bio_release_pages(bio, false);
1428 spin_lock_irqsave(&bio_dirty_lock, flags);
1429 bio->bi_private = bio_dirty_list;
1430 bio_dirty_list = bio;
1431 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1432 schedule_work(&bio_dirty_work);
1435 static inline bool bio_remaining_done(struct bio *bio)
1438 * If we're not chaining, then ->__bi_remaining is always 1 and
1439 * we always end io on the first invocation.
1441 if (!bio_flagged(bio, BIO_CHAIN))
1444 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1446 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1447 bio_clear_flag(bio, BIO_CHAIN);
1455 * bio_endio - end I/O on a bio
1459 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1460 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1461 * bio unless they own it and thus know that it has an end_io function.
1463 * bio_endio() can be called several times on a bio that has been chained
1464 * using bio_chain(). The ->bi_end_io() function will only be called the
1467 void bio_endio(struct bio *bio)
1470 if (!bio_remaining_done(bio))
1472 if (!bio_integrity_endio(bio))
1476 rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio);
1478 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1479 trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio);
1480 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1484 * Need to have a real endio function for chained bios, otherwise
1485 * various corner cases will break (like stacking block devices that
1486 * save/restore bi_end_io) - however, we want to avoid unbounded
1487 * recursion and blowing the stack. Tail call optimization would
1488 * handle this, but compiling with frame pointers also disables
1489 * gcc's sibling call optimization.
1491 if (bio->bi_end_io == bio_chain_endio) {
1492 bio = __bio_chain_endio(bio);
1496 blk_throtl_bio_endio(bio);
1497 /* release cgroup info */
1500 bio->bi_end_io(bio);
1502 EXPORT_SYMBOL(bio_endio);
1505 * bio_split - split a bio
1506 * @bio: bio to split
1507 * @sectors: number of sectors to split from the front of @bio
1509 * @bs: bio set to allocate from
1511 * Allocates and returns a new bio which represents @sectors from the start of
1512 * @bio, and updates @bio to represent the remaining sectors.
1514 * Unless this is a discard request the newly allocated bio will point
1515 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1516 * neither @bio nor @bs are freed before the split bio.
1518 struct bio *bio_split(struct bio *bio, int sectors,
1519 gfp_t gfp, struct bio_set *bs)
1523 BUG_ON(sectors <= 0);
1524 BUG_ON(sectors >= bio_sectors(bio));
1526 /* Zone append commands cannot be split */
1527 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1530 split = bio_clone_fast(bio, gfp, bs);
1534 split->bi_iter.bi_size = sectors << 9;
1536 if (bio_integrity(split))
1537 bio_integrity_trim(split);
1539 bio_advance(bio, split->bi_iter.bi_size);
1541 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1542 bio_set_flag(split, BIO_TRACE_COMPLETION);
1546 EXPORT_SYMBOL(bio_split);
1549 * bio_trim - trim a bio
1551 * @offset: number of sectors to trim from the front of @bio
1552 * @size: size we want to trim @bio to, in sectors
1554 void bio_trim(struct bio *bio, int offset, int size)
1556 /* 'bio' is a cloned bio which we need to trim to match
1557 * the given offset and size.
1561 if (offset == 0 && size == bio->bi_iter.bi_size)
1564 bio_advance(bio, offset << 9);
1565 bio->bi_iter.bi_size = size;
1567 if (bio_integrity(bio))
1568 bio_integrity_trim(bio);
1571 EXPORT_SYMBOL_GPL(bio_trim);
1574 * create memory pools for biovec's in a bio_set.
1575 * use the global biovec slabs created for general use.
1577 int biovec_init_pool(mempool_t *pool, int pool_entries)
1579 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1581 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1585 * bioset_exit - exit a bioset initialized with bioset_init()
1587 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1590 void bioset_exit(struct bio_set *bs)
1592 bio_alloc_cache_destroy(bs);
1593 if (bs->rescue_workqueue)
1594 destroy_workqueue(bs->rescue_workqueue);
1595 bs->rescue_workqueue = NULL;
1597 mempool_exit(&bs->bio_pool);
1598 mempool_exit(&bs->bvec_pool);
1600 bioset_integrity_free(bs);
1603 bs->bio_slab = NULL;
1605 EXPORT_SYMBOL(bioset_exit);
1608 * bioset_init - Initialize a bio_set
1609 * @bs: pool to initialize
1610 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1611 * @front_pad: Number of bytes to allocate in front of the returned bio
1612 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1613 * and %BIOSET_NEED_RESCUER
1616 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1617 * to ask for a number of bytes to be allocated in front of the bio.
1618 * Front pad allocation is useful for embedding the bio inside
1619 * another structure, to avoid allocating extra data to go with the bio.
1620 * Note that the bio must be embedded at the END of that structure always,
1621 * or things will break badly.
1622 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1623 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1624 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1625 * dispatch queued requests when the mempool runs out of space.
1628 int bioset_init(struct bio_set *bs,
1629 unsigned int pool_size,
1630 unsigned int front_pad,
1633 bs->front_pad = front_pad;
1634 if (flags & BIOSET_NEED_BVECS)
1635 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1639 spin_lock_init(&bs->rescue_lock);
1640 bio_list_init(&bs->rescue_list);
1641 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1643 bs->bio_slab = bio_find_or_create_slab(bs);
1647 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1650 if ((flags & BIOSET_NEED_BVECS) &&
1651 biovec_init_pool(&bs->bvec_pool, pool_size))
1654 if (flags & BIOSET_NEED_RESCUER) {
1655 bs->rescue_workqueue = alloc_workqueue("bioset",
1657 if (!bs->rescue_workqueue)
1660 if (flags & BIOSET_PERCPU_CACHE) {
1661 bs->cache = alloc_percpu(struct bio_alloc_cache);
1664 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1672 EXPORT_SYMBOL(bioset_init);
1675 * Initialize and setup a new bio_set, based on the settings from
1678 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1683 if (src->bvec_pool.min_nr)
1684 flags |= BIOSET_NEED_BVECS;
1685 if (src->rescue_workqueue)
1686 flags |= BIOSET_NEED_RESCUER;
1688 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1690 EXPORT_SYMBOL(bioset_init_from_src);
1693 * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb
1694 * @kiocb: kiocb describing the IO
1695 * @nr_iovecs: number of iovecs to pre-allocate
1696 * @bs: bio_set to allocate from
1699 * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only
1700 * used to check if we should dip into the per-cpu bio_set allocation
1701 * cache. The allocation uses GFP_KERNEL internally. On return, the
1702 * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio
1703 * MUST be done from process context, not hard/soft IRQ.
1706 struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs,
1709 struct bio_alloc_cache *cache;
1712 if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS)
1713 return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
1715 cache = per_cpu_ptr(bs->cache, get_cpu());
1716 bio = bio_list_pop(&cache->free_list);
1720 bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs);
1722 bio_set_flag(bio, BIO_PERCPU_CACHE);
1726 bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
1727 bio_set_flag(bio, BIO_PERCPU_CACHE);
1730 EXPORT_SYMBOL_GPL(bio_alloc_kiocb);
1732 static int __init init_bio(void)
1736 bio_integrity_init();
1738 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1739 struct biovec_slab *bvs = bvec_slabs + i;
1741 bvs->slab = kmem_cache_create(bvs->name,
1742 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1743 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1746 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1749 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
1750 panic("bio: can't allocate bios\n");
1752 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1753 panic("bio: can't create integrity pool\n");
1757 subsys_initcall(init_bio);