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)
532 struct bvec_iter iter;
534 bio_for_each_segment(bv, bio, iter)
537 EXPORT_SYMBOL(zero_fill_bio);
540 * bio_truncate - truncate the bio to small size of @new_size
541 * @bio: the bio to be truncated
542 * @new_size: new size for truncating the bio
545 * Truncate the bio to new size of @new_size. If bio_op(bio) is
546 * REQ_OP_READ, zero the truncated part. This function should only
547 * be used for handling corner cases, such as bio eod.
549 void bio_truncate(struct bio *bio, unsigned new_size)
552 struct bvec_iter iter;
553 unsigned int done = 0;
554 bool truncated = false;
556 if (new_size >= bio->bi_iter.bi_size)
559 if (bio_op(bio) != REQ_OP_READ)
562 bio_for_each_segment(bv, bio, iter) {
563 if (done + bv.bv_len > new_size) {
567 offset = new_size - done;
570 zero_user(bv.bv_page, bv.bv_offset + offset,
579 * Don't touch bvec table here and make it really immutable, since
580 * fs bio user has to retrieve all pages via bio_for_each_segment_all
581 * in its .end_bio() callback.
583 * It is enough to truncate bio by updating .bi_size since we can make
584 * correct bvec with the updated .bi_size for drivers.
586 bio->bi_iter.bi_size = new_size;
590 * guard_bio_eod - truncate a BIO to fit the block device
591 * @bio: bio to truncate
593 * This allows us to do IO even on the odd last sectors of a device, even if the
594 * block size is some multiple of the physical sector size.
596 * We'll just truncate the bio to the size of the device, and clear the end of
597 * the buffer head manually. Truly out-of-range accesses will turn into actual
598 * I/O errors, this only handles the "we need to be able to do I/O at the final
601 void guard_bio_eod(struct bio *bio)
603 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
609 * If the *whole* IO is past the end of the device,
610 * let it through, and the IO layer will turn it into
613 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
616 maxsector -= bio->bi_iter.bi_sector;
617 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
620 bio_truncate(bio, maxsector << 9);
623 #define ALLOC_CACHE_MAX 512
624 #define ALLOC_CACHE_SLACK 64
626 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
632 while ((bio = bio_list_pop(&cache->free_list)) != NULL) {
640 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
644 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
646 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
648 bio_alloc_cache_prune(cache, -1U);
653 static void bio_alloc_cache_destroy(struct bio_set *bs)
660 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
661 for_each_possible_cpu(cpu) {
662 struct bio_alloc_cache *cache;
664 cache = per_cpu_ptr(bs->cache, cpu);
665 bio_alloc_cache_prune(cache, -1U);
667 free_percpu(bs->cache);
672 * bio_put - release a reference to a bio
673 * @bio: bio to release reference to
676 * Put a reference to a &struct bio, either one you have gotten with
677 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
679 void bio_put(struct bio *bio)
681 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
682 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
683 if (!atomic_dec_and_test(&bio->__bi_cnt))
687 if (bio_flagged(bio, BIO_PERCPU_CACHE)) {
688 struct bio_alloc_cache *cache;
691 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
692 bio_list_add_head(&cache->free_list, bio);
693 if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK)
694 bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK);
700 EXPORT_SYMBOL(bio_put);
703 * __bio_clone_fast - clone a bio that shares the original bio's biovec
704 * @bio: destination bio
705 * @bio_src: bio to clone
707 * Clone a &bio. Caller will own the returned bio, but not
708 * the actual data it points to. Reference count of returned
711 * Caller must ensure that @bio_src is not freed before @bio.
713 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
715 WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs);
718 * most users will be overriding ->bi_bdev with a new target,
719 * so we don't set nor calculate new physical/hw segment counts here
721 bio->bi_bdev = bio_src->bi_bdev;
722 bio_set_flag(bio, BIO_CLONED);
723 if (bio_flagged(bio_src, BIO_THROTTLED))
724 bio_set_flag(bio, BIO_THROTTLED);
725 if (bio_flagged(bio_src, BIO_REMAPPED))
726 bio_set_flag(bio, BIO_REMAPPED);
727 bio->bi_opf = bio_src->bi_opf;
728 bio->bi_ioprio = bio_src->bi_ioprio;
729 bio->bi_write_hint = bio_src->bi_write_hint;
730 bio->bi_iter = bio_src->bi_iter;
731 bio->bi_io_vec = bio_src->bi_io_vec;
733 bio_clone_blkg_association(bio, bio_src);
734 blkcg_bio_issue_init(bio);
736 EXPORT_SYMBOL(__bio_clone_fast);
739 * bio_clone_fast - clone a bio that shares the original bio's biovec
741 * @gfp_mask: allocation priority
742 * @bs: bio_set to allocate from
744 * Like __bio_clone_fast, only also allocates the returned bio
746 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
750 b = bio_alloc_bioset(gfp_mask, 0, bs);
754 __bio_clone_fast(b, bio);
756 if (bio_crypt_clone(b, bio, gfp_mask) < 0)
759 if (bio_integrity(bio) &&
760 bio_integrity_clone(b, bio, gfp_mask) < 0)
769 EXPORT_SYMBOL(bio_clone_fast);
771 const char *bio_devname(struct bio *bio, char *buf)
773 return bdevname(bio->bi_bdev, buf);
775 EXPORT_SYMBOL(bio_devname);
777 static inline bool page_is_mergeable(const struct bio_vec *bv,
778 struct page *page, unsigned int len, unsigned int off,
781 size_t bv_end = bv->bv_offset + bv->bv_len;
782 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
783 phys_addr_t page_addr = page_to_phys(page);
785 if (vec_end_addr + 1 != page_addr + off)
787 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
790 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
793 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
797 * Try to merge a page into a segment, while obeying the hardware segment
798 * size limit. This is not for normal read/write bios, but for passthrough
799 * or Zone Append operations that we can't split.
801 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
802 struct page *page, unsigned len,
803 unsigned offset, bool *same_page)
805 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
806 unsigned long mask = queue_segment_boundary(q);
807 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
808 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
810 if ((addr1 | mask) != (addr2 | mask))
812 if (bv->bv_len + len > queue_max_segment_size(q))
814 return __bio_try_merge_page(bio, page, len, offset, same_page);
818 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
819 * @q: the target queue
820 * @bio: destination bio
822 * @len: vec entry length
823 * @offset: vec entry offset
824 * @max_sectors: maximum number of sectors that can be added
825 * @same_page: return if the segment has been merged inside the same page
827 * Add a page to a bio while respecting the hardware max_sectors, max_segment
828 * and gap limitations.
830 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
831 struct page *page, unsigned int len, unsigned int offset,
832 unsigned int max_sectors, bool *same_page)
834 struct bio_vec *bvec;
836 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
839 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
842 if (bio->bi_vcnt > 0) {
843 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
847 * If the queue doesn't support SG gaps and adding this segment
848 * would create a gap, disallow it.
850 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
851 if (bvec_gap_to_prev(q, bvec, offset))
855 if (bio_full(bio, len))
858 if (bio->bi_vcnt >= queue_max_segments(q))
861 bvec = &bio->bi_io_vec[bio->bi_vcnt];
862 bvec->bv_page = page;
864 bvec->bv_offset = offset;
866 bio->bi_iter.bi_size += len;
871 * bio_add_pc_page - attempt to add page to passthrough bio
872 * @q: the target queue
873 * @bio: destination bio
875 * @len: vec entry length
876 * @offset: vec entry offset
878 * Attempt to add a page to the bio_vec maplist. This can fail for a
879 * number of reasons, such as the bio being full or target block device
880 * limitations. The target block device must allow bio's up to PAGE_SIZE,
881 * so it is always possible to add a single page to an empty bio.
883 * This should only be used by passthrough bios.
885 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
886 struct page *page, unsigned int len, unsigned int offset)
888 bool same_page = false;
889 return bio_add_hw_page(q, bio, page, len, offset,
890 queue_max_hw_sectors(q), &same_page);
892 EXPORT_SYMBOL(bio_add_pc_page);
895 * bio_add_zone_append_page - attempt to add page to zone-append bio
896 * @bio: destination bio
898 * @len: vec entry length
899 * @offset: vec entry offset
901 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
902 * for a zone-append request. This can fail for a number of reasons, such as the
903 * bio being full or the target block device is not a zoned block device or
904 * other limitations of the target block device. The target block device must
905 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
908 * Returns: number of bytes added to the bio, or 0 in case of a failure.
910 int bio_add_zone_append_page(struct bio *bio, struct page *page,
911 unsigned int len, unsigned int offset)
913 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
914 bool same_page = false;
916 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
919 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
922 return bio_add_hw_page(q, bio, page, len, offset,
923 queue_max_zone_append_sectors(q), &same_page);
925 EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
928 * __bio_try_merge_page - try appending data to an existing bvec.
929 * @bio: destination bio
930 * @page: start page to add
931 * @len: length of the data to add
932 * @off: offset of the data relative to @page
933 * @same_page: return if the segment has been merged inside the same page
935 * Try to add the data at @page + @off to the last bvec of @bio. This is a
936 * useful optimisation for file systems with a block size smaller than the
939 * Warn if (@len, @off) crosses pages in case that @same_page is true.
941 * Return %true on success or %false on failure.
943 bool __bio_try_merge_page(struct bio *bio, struct page *page,
944 unsigned int len, unsigned int off, bool *same_page)
946 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
949 if (bio->bi_vcnt > 0) {
950 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
952 if (page_is_mergeable(bv, page, len, off, same_page)) {
953 if (bio->bi_iter.bi_size > UINT_MAX - len) {
958 bio->bi_iter.bi_size += len;
964 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
967 * __bio_add_page - add page(s) to a bio in a new segment
968 * @bio: destination bio
969 * @page: start page to add
970 * @len: length of the data to add, may cross pages
971 * @off: offset of the data relative to @page, may cross pages
973 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
974 * that @bio has space for another bvec.
976 void __bio_add_page(struct bio *bio, struct page *page,
977 unsigned int len, unsigned int off)
979 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
981 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
982 WARN_ON_ONCE(bio_full(bio, len));
988 bio->bi_iter.bi_size += len;
991 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
992 bio_set_flag(bio, BIO_WORKINGSET);
994 EXPORT_SYMBOL_GPL(__bio_add_page);
997 * bio_add_page - attempt to add page(s) to bio
998 * @bio: destination bio
999 * @page: start page to add
1000 * @len: vec entry length, may cross pages
1001 * @offset: vec entry offset relative to @page, may cross pages
1003 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1004 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1006 int bio_add_page(struct bio *bio, struct page *page,
1007 unsigned int len, unsigned int offset)
1009 bool same_page = false;
1011 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1012 if (bio_full(bio, len))
1014 __bio_add_page(bio, page, len, offset);
1018 EXPORT_SYMBOL(bio_add_page);
1020 void bio_release_pages(struct bio *bio, bool mark_dirty)
1022 struct bvec_iter_all iter_all;
1023 struct bio_vec *bvec;
1025 if (bio_flagged(bio, BIO_NO_PAGE_REF))
1028 bio_for_each_segment_all(bvec, bio, iter_all) {
1029 if (mark_dirty && !PageCompound(bvec->bv_page))
1030 set_page_dirty_lock(bvec->bv_page);
1031 put_page(bvec->bv_page);
1034 EXPORT_SYMBOL_GPL(bio_release_pages);
1036 static void __bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1038 WARN_ON_ONCE(bio->bi_max_vecs);
1040 bio->bi_vcnt = iter->nr_segs;
1041 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1042 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1043 bio->bi_iter.bi_size = iter->count;
1044 bio_set_flag(bio, BIO_NO_PAGE_REF);
1045 bio_set_flag(bio, BIO_CLONED);
1048 static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1050 __bio_iov_bvec_set(bio, iter);
1051 iov_iter_advance(iter, iter->count);
1055 static int bio_iov_bvec_set_append(struct bio *bio, struct iov_iter *iter)
1057 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1058 struct iov_iter i = *iter;
1060 iov_iter_truncate(&i, queue_max_zone_append_sectors(q) << 9);
1061 __bio_iov_bvec_set(bio, &i);
1062 iov_iter_advance(iter, i.count);
1066 static void bio_put_pages(struct page **pages, size_t size, size_t off)
1068 size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE);
1070 for (i = 0; i < nr; i++)
1074 static int bio_iov_add_page(struct bio *bio, struct page *page,
1075 unsigned int len, unsigned int offset)
1077 bool same_page = false;
1079 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1080 if (WARN_ON_ONCE(bio_full(bio, len)))
1082 __bio_add_page(bio, page, len, offset);
1091 static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1092 unsigned int len, unsigned int offset)
1094 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1095 bool same_page = false;
1097 if (bio_add_hw_page(q, bio, page, len, offset,
1098 queue_max_zone_append_sectors(q), &same_page) != len)
1105 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1108 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1109 * @bio: bio to add pages to
1110 * @iter: iov iterator describing the region to be mapped
1112 * Pins pages from *iter and appends them to @bio's bvec array. The
1113 * pages will have to be released using put_page() when done.
1114 * For multi-segment *iter, this function only adds pages from the
1115 * next non-empty segment of the iov iterator.
1117 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1119 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1120 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1121 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1122 struct page **pages = (struct page **)bv;
1129 * Move page array up in the allocated memory for the bio vecs as far as
1130 * possible so that we can start filling biovecs from the beginning
1131 * without overwriting the temporary page array.
1133 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1134 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1136 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1137 if (unlikely(size <= 0))
1138 return size ? size : -EFAULT;
1140 for (left = size, i = 0; left > 0; left -= len, i++) {
1141 struct page *page = pages[i];
1143 len = min_t(size_t, PAGE_SIZE - offset, left);
1144 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1145 ret = bio_iov_add_zone_append_page(bio, page, len,
1148 ret = bio_iov_add_page(bio, page, len, offset);
1151 bio_put_pages(pages + i, left, offset);
1157 iov_iter_advance(iter, size - left);
1162 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1163 * @bio: bio to add pages to
1164 * @iter: iov iterator describing the region to be added
1166 * This takes either an iterator pointing to user memory, or one pointing to
1167 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1168 * map them into the kernel. On IO completion, the caller should put those
1169 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1170 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1171 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1172 * completed by a call to ->ki_complete() or returns with an error other than
1173 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1174 * on IO completion. If it isn't, then pages should be released.
1176 * The function tries, but does not guarantee, to pin as many pages as
1177 * fit into the bio, or are requested in @iter, whatever is smaller. If
1178 * MM encounters an error pinning the requested pages, it stops. Error
1179 * is returned only if 0 pages could be pinned.
1181 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1182 * responsible for setting BIO_WORKINGSET if necessary.
1184 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1188 if (iov_iter_is_bvec(iter)) {
1189 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1190 return bio_iov_bvec_set_append(bio, iter);
1191 return bio_iov_bvec_set(bio, iter);
1195 ret = __bio_iov_iter_get_pages(bio, iter);
1196 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1198 /* don't account direct I/O as memory stall */
1199 bio_clear_flag(bio, BIO_WORKINGSET);
1200 return bio->bi_vcnt ? 0 : ret;
1202 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1204 static void submit_bio_wait_endio(struct bio *bio)
1206 complete(bio->bi_private);
1210 * submit_bio_wait - submit a bio, and wait until it completes
1211 * @bio: The &struct bio which describes the I/O
1213 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1214 * bio_endio() on failure.
1216 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1217 * result in bio reference to be consumed. The caller must drop the reference
1220 int submit_bio_wait(struct bio *bio)
1222 DECLARE_COMPLETION_ONSTACK_MAP(done,
1223 bio->bi_bdev->bd_disk->lockdep_map);
1224 unsigned long hang_check;
1226 bio->bi_private = &done;
1227 bio->bi_end_io = submit_bio_wait_endio;
1228 bio->bi_opf |= REQ_SYNC;
1231 /* Prevent hang_check timer from firing at us during very long I/O */
1232 hang_check = sysctl_hung_task_timeout_secs;
1234 while (!wait_for_completion_io_timeout(&done,
1235 hang_check * (HZ/2)))
1238 wait_for_completion_io(&done);
1240 return blk_status_to_errno(bio->bi_status);
1242 EXPORT_SYMBOL(submit_bio_wait);
1245 * bio_advance - increment/complete a bio by some number of bytes
1246 * @bio: bio to advance
1247 * @bytes: number of bytes to complete
1249 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1250 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1251 * be updated on the last bvec as well.
1253 * @bio will then represent the remaining, uncompleted portion of the io.
1255 void bio_advance(struct bio *bio, unsigned bytes)
1257 if (bio_integrity(bio))
1258 bio_integrity_advance(bio, bytes);
1260 bio_crypt_advance(bio, bytes);
1261 bio_advance_iter(bio, &bio->bi_iter, bytes);
1263 EXPORT_SYMBOL(bio_advance);
1265 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1266 struct bio *src, struct bvec_iter *src_iter)
1268 while (src_iter->bi_size && dst_iter->bi_size) {
1269 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1270 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1271 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1272 void *src_buf = bvec_kmap_local(&src_bv);
1273 void *dst_buf = bvec_kmap_local(&dst_bv);
1275 memcpy(dst_buf, src_buf, bytes);
1277 kunmap_local(dst_buf);
1278 kunmap_local(src_buf);
1280 bio_advance_iter_single(src, src_iter, bytes);
1281 bio_advance_iter_single(dst, dst_iter, bytes);
1284 EXPORT_SYMBOL(bio_copy_data_iter);
1287 * bio_copy_data - copy contents of data buffers from one bio to another
1289 * @dst: destination bio
1291 * Stops when it reaches the end of either @src or @dst - that is, copies
1292 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1294 void bio_copy_data(struct bio *dst, struct bio *src)
1296 struct bvec_iter src_iter = src->bi_iter;
1297 struct bvec_iter dst_iter = dst->bi_iter;
1299 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1301 EXPORT_SYMBOL(bio_copy_data);
1303 void bio_free_pages(struct bio *bio)
1305 struct bio_vec *bvec;
1306 struct bvec_iter_all iter_all;
1308 bio_for_each_segment_all(bvec, bio, iter_all)
1309 __free_page(bvec->bv_page);
1311 EXPORT_SYMBOL(bio_free_pages);
1314 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1315 * for performing direct-IO in BIOs.
1317 * The problem is that we cannot run set_page_dirty() from interrupt context
1318 * because the required locks are not interrupt-safe. So what we can do is to
1319 * mark the pages dirty _before_ performing IO. And in interrupt context,
1320 * check that the pages are still dirty. If so, fine. If not, redirty them
1321 * in process context.
1323 * We special-case compound pages here: normally this means reads into hugetlb
1324 * pages. The logic in here doesn't really work right for compound pages
1325 * because the VM does not uniformly chase down the head page in all cases.
1326 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1327 * handle them at all. So we skip compound pages here at an early stage.
1329 * Note that this code is very hard to test under normal circumstances because
1330 * direct-io pins the pages with get_user_pages(). This makes
1331 * is_page_cache_freeable return false, and the VM will not clean the pages.
1332 * But other code (eg, flusher threads) could clean the pages if they are mapped
1335 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1336 * deferred bio dirtying paths.
1340 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1342 void bio_set_pages_dirty(struct bio *bio)
1344 struct bio_vec *bvec;
1345 struct bvec_iter_all iter_all;
1347 bio_for_each_segment_all(bvec, bio, iter_all) {
1348 if (!PageCompound(bvec->bv_page))
1349 set_page_dirty_lock(bvec->bv_page);
1354 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1355 * If they are, then fine. If, however, some pages are clean then they must
1356 * have been written out during the direct-IO read. So we take another ref on
1357 * the BIO and re-dirty the pages in process context.
1359 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1360 * here on. It will run one put_page() against each page and will run one
1361 * bio_put() against the BIO.
1364 static void bio_dirty_fn(struct work_struct *work);
1366 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1367 static DEFINE_SPINLOCK(bio_dirty_lock);
1368 static struct bio *bio_dirty_list;
1371 * This runs in process context
1373 static void bio_dirty_fn(struct work_struct *work)
1375 struct bio *bio, *next;
1377 spin_lock_irq(&bio_dirty_lock);
1378 next = bio_dirty_list;
1379 bio_dirty_list = NULL;
1380 spin_unlock_irq(&bio_dirty_lock);
1382 while ((bio = next) != NULL) {
1383 next = bio->bi_private;
1385 bio_release_pages(bio, true);
1390 void bio_check_pages_dirty(struct bio *bio)
1392 struct bio_vec *bvec;
1393 unsigned long flags;
1394 struct bvec_iter_all iter_all;
1396 bio_for_each_segment_all(bvec, bio, iter_all) {
1397 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1401 bio_release_pages(bio, false);
1405 spin_lock_irqsave(&bio_dirty_lock, flags);
1406 bio->bi_private = bio_dirty_list;
1407 bio_dirty_list = bio;
1408 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1409 schedule_work(&bio_dirty_work);
1412 static inline bool bio_remaining_done(struct bio *bio)
1415 * If we're not chaining, then ->__bi_remaining is always 1 and
1416 * we always end io on the first invocation.
1418 if (!bio_flagged(bio, BIO_CHAIN))
1421 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1423 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1424 bio_clear_flag(bio, BIO_CHAIN);
1432 * bio_endio - end I/O on a bio
1436 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1437 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1438 * bio unless they own it and thus know that it has an end_io function.
1440 * bio_endio() can be called several times on a bio that has been chained
1441 * using bio_chain(). The ->bi_end_io() function will only be called the
1444 void bio_endio(struct bio *bio)
1447 if (!bio_remaining_done(bio))
1449 if (!bio_integrity_endio(bio))
1452 rq_qos_done_bio(bio);
1454 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1455 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1456 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1460 * Need to have a real endio function for chained bios, otherwise
1461 * various corner cases will break (like stacking block devices that
1462 * save/restore bi_end_io) - however, we want to avoid unbounded
1463 * recursion and blowing the stack. Tail call optimization would
1464 * handle this, but compiling with frame pointers also disables
1465 * gcc's sibling call optimization.
1467 if (bio->bi_end_io == bio_chain_endio) {
1468 bio = __bio_chain_endio(bio);
1472 blk_throtl_bio_endio(bio);
1473 /* release cgroup info */
1476 bio->bi_end_io(bio);
1478 EXPORT_SYMBOL(bio_endio);
1481 * bio_split - split a bio
1482 * @bio: bio to split
1483 * @sectors: number of sectors to split from the front of @bio
1485 * @bs: bio set to allocate from
1487 * Allocates and returns a new bio which represents @sectors from the start of
1488 * @bio, and updates @bio to represent the remaining sectors.
1490 * Unless this is a discard request the newly allocated bio will point
1491 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1492 * neither @bio nor @bs are freed before the split bio.
1494 struct bio *bio_split(struct bio *bio, int sectors,
1495 gfp_t gfp, struct bio_set *bs)
1499 BUG_ON(sectors <= 0);
1500 BUG_ON(sectors >= bio_sectors(bio));
1502 /* Zone append commands cannot be split */
1503 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1506 split = bio_clone_fast(bio, gfp, bs);
1510 split->bi_iter.bi_size = sectors << 9;
1512 if (bio_integrity(split))
1513 bio_integrity_trim(split);
1515 bio_advance(bio, split->bi_iter.bi_size);
1517 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1518 bio_set_flag(split, BIO_TRACE_COMPLETION);
1522 EXPORT_SYMBOL(bio_split);
1525 * bio_trim - trim a bio
1527 * @offset: number of sectors to trim from the front of @bio
1528 * @size: size we want to trim @bio to, in sectors
1530 * This function is typically used for bios that are cloned and submitted
1531 * to the underlying device in parts.
1533 void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1535 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1536 offset + size > bio_sectors(bio)))
1540 if (offset == 0 && size == bio->bi_iter.bi_size)
1543 bio_advance(bio, offset << 9);
1544 bio->bi_iter.bi_size = size;
1546 if (bio_integrity(bio))
1547 bio_integrity_trim(bio);
1549 EXPORT_SYMBOL_GPL(bio_trim);
1552 * create memory pools for biovec's in a bio_set.
1553 * use the global biovec slabs created for general use.
1555 int biovec_init_pool(mempool_t *pool, int pool_entries)
1557 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1559 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1563 * bioset_exit - exit a bioset initialized with bioset_init()
1565 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1568 void bioset_exit(struct bio_set *bs)
1570 bio_alloc_cache_destroy(bs);
1571 if (bs->rescue_workqueue)
1572 destroy_workqueue(bs->rescue_workqueue);
1573 bs->rescue_workqueue = NULL;
1575 mempool_exit(&bs->bio_pool);
1576 mempool_exit(&bs->bvec_pool);
1578 bioset_integrity_free(bs);
1581 bs->bio_slab = NULL;
1583 EXPORT_SYMBOL(bioset_exit);
1586 * bioset_init - Initialize a bio_set
1587 * @bs: pool to initialize
1588 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1589 * @front_pad: Number of bytes to allocate in front of the returned bio
1590 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1591 * and %BIOSET_NEED_RESCUER
1594 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1595 * to ask for a number of bytes to be allocated in front of the bio.
1596 * Front pad allocation is useful for embedding the bio inside
1597 * another structure, to avoid allocating extra data to go with the bio.
1598 * Note that the bio must be embedded at the END of that structure always,
1599 * or things will break badly.
1600 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1601 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1602 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1603 * dispatch queued requests when the mempool runs out of space.
1606 int bioset_init(struct bio_set *bs,
1607 unsigned int pool_size,
1608 unsigned int front_pad,
1611 bs->front_pad = front_pad;
1612 if (flags & BIOSET_NEED_BVECS)
1613 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1617 spin_lock_init(&bs->rescue_lock);
1618 bio_list_init(&bs->rescue_list);
1619 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1621 bs->bio_slab = bio_find_or_create_slab(bs);
1625 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1628 if ((flags & BIOSET_NEED_BVECS) &&
1629 biovec_init_pool(&bs->bvec_pool, pool_size))
1632 if (flags & BIOSET_NEED_RESCUER) {
1633 bs->rescue_workqueue = alloc_workqueue("bioset",
1635 if (!bs->rescue_workqueue)
1638 if (flags & BIOSET_PERCPU_CACHE) {
1639 bs->cache = alloc_percpu(struct bio_alloc_cache);
1642 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1650 EXPORT_SYMBOL(bioset_init);
1653 * Initialize and setup a new bio_set, based on the settings from
1656 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1661 if (src->bvec_pool.min_nr)
1662 flags |= BIOSET_NEED_BVECS;
1663 if (src->rescue_workqueue)
1664 flags |= BIOSET_NEED_RESCUER;
1666 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1668 EXPORT_SYMBOL(bioset_init_from_src);
1671 * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb
1672 * @kiocb: kiocb describing the IO
1673 * @nr_vecs: number of iovecs to pre-allocate
1674 * @bs: bio_set to allocate from
1677 * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only
1678 * used to check if we should dip into the per-cpu bio_set allocation
1679 * cache. The allocation uses GFP_KERNEL internally. On return, the
1680 * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio
1681 * MUST be done from process context, not hard/soft IRQ.
1684 struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs,
1687 struct bio_alloc_cache *cache;
1690 if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS)
1691 return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
1693 cache = per_cpu_ptr(bs->cache, get_cpu());
1694 bio = bio_list_pop(&cache->free_list);
1698 bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs);
1700 bio_set_flag(bio, BIO_PERCPU_CACHE);
1704 bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
1705 bio_set_flag(bio, BIO_PERCPU_CACHE);
1708 EXPORT_SYMBOL_GPL(bio_alloc_kiocb);
1710 static int __init init_bio(void)
1714 bio_integrity_init();
1716 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1717 struct biovec_slab *bvs = bvec_slabs + i;
1719 bvs->slab = kmem_cache_create(bvs->name,
1720 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1721 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1724 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1727 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
1728 panic("bio: can't allocate bios\n");
1730 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1731 panic("bio: can't create integrity pool\n");
1735 subsys_initcall(init_bio);