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 static struct biovec_slab {
31 struct kmem_cache *slab;
32 } bvec_slabs[] __read_mostly = {
33 { .nr_vecs = 16, .name = "biovec-16" },
34 { .nr_vecs = 64, .name = "biovec-64" },
35 { .nr_vecs = 128, .name = "biovec-128" },
36 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
39 static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
42 /* smaller bios use inline vecs */
44 return &bvec_slabs[0];
46 return &bvec_slabs[1];
48 return &bvec_slabs[2];
49 case 129 ... BIO_MAX_VECS:
50 return &bvec_slabs[3];
58 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
59 * IO code that does not need private memory pools.
61 struct bio_set fs_bio_set;
62 EXPORT_SYMBOL(fs_bio_set);
65 * Our slab pool management
68 struct kmem_cache *slab;
69 unsigned int slab_ref;
70 unsigned int slab_size;
73 static DEFINE_MUTEX(bio_slab_lock);
74 static DEFINE_XARRAY(bio_slabs);
76 static struct bio_slab *create_bio_slab(unsigned int size)
78 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
83 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
84 bslab->slab = kmem_cache_create(bslab->name, size,
85 ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL);
90 bslab->slab_size = size;
92 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
95 kmem_cache_destroy(bslab->slab);
102 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
104 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
107 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
109 unsigned int size = bs_bio_slab_size(bs);
110 struct bio_slab *bslab;
112 mutex_lock(&bio_slab_lock);
113 bslab = xa_load(&bio_slabs, size);
117 bslab = create_bio_slab(size);
118 mutex_unlock(&bio_slab_lock);
125 static void bio_put_slab(struct bio_set *bs)
127 struct bio_slab *bslab = NULL;
128 unsigned int slab_size = bs_bio_slab_size(bs);
130 mutex_lock(&bio_slab_lock);
132 bslab = xa_load(&bio_slabs, slab_size);
133 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
136 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
138 WARN_ON(!bslab->slab_ref);
140 if (--bslab->slab_ref)
143 xa_erase(&bio_slabs, slab_size);
145 kmem_cache_destroy(bslab->slab);
149 mutex_unlock(&bio_slab_lock);
152 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
154 BIO_BUG_ON(nr_vecs > BIO_MAX_VECS);
156 if (nr_vecs == BIO_MAX_VECS)
157 mempool_free(bv, pool);
158 else if (nr_vecs > BIO_INLINE_VECS)
159 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
163 * Make the first allocation restricted and don't dump info on allocation
164 * failures, since we'll fall back to the mempool in case of failure.
166 static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
168 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
169 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
172 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
175 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
177 if (WARN_ON_ONCE(!bvs))
181 * Upgrade the nr_vecs request to take full advantage of the allocation.
182 * We also rely on this in the bvec_free path.
184 *nr_vecs = bvs->nr_vecs;
187 * Try a slab allocation first for all smaller allocations. If that
188 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
189 * The mempool is sized to handle up to BIO_MAX_VECS entries.
191 if (*nr_vecs < BIO_MAX_VECS) {
194 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
195 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
197 *nr_vecs = BIO_MAX_VECS;
200 return mempool_alloc(pool, gfp_mask);
203 void bio_uninit(struct bio *bio)
205 #ifdef CONFIG_BLK_CGROUP
207 blkg_put(bio->bi_blkg);
211 if (bio_integrity(bio))
212 bio_integrity_free(bio);
214 bio_crypt_free_ctx(bio);
216 EXPORT_SYMBOL(bio_uninit);
218 static void bio_free(struct bio *bio)
220 struct bio_set *bs = bio->bi_pool;
226 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
229 * If we have front padding, adjust the bio pointer before freeing
234 mempool_free(p, &bs->bio_pool);
236 /* Bio was allocated by bio_kmalloc() */
242 * Users of this function have their own bio allocation. Subsequently,
243 * they must remember to pair any call to bio_init() with bio_uninit()
244 * when IO has completed, or when the bio is released.
246 void bio_init(struct bio *bio, struct bio_vec *table,
247 unsigned short max_vecs)
254 bio->bi_write_hint = 0;
256 bio->bi_iter.bi_sector = 0;
257 bio->bi_iter.bi_size = 0;
258 bio->bi_iter.bi_idx = 0;
259 bio->bi_iter.bi_bvec_done = 0;
260 bio->bi_end_io = NULL;
261 bio->bi_private = NULL;
262 #ifdef CONFIG_BLK_CGROUP
264 bio->bi_issue.value = 0;
265 #ifdef CONFIG_BLK_CGROUP_IOCOST
266 bio->bi_iocost_cost = 0;
269 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
270 bio->bi_crypt_context = NULL;
272 #ifdef CONFIG_BLK_DEV_INTEGRITY
273 bio->bi_integrity = NULL;
277 atomic_set(&bio->__bi_remaining, 1);
278 atomic_set(&bio->__bi_cnt, 1);
280 bio->bi_max_vecs = max_vecs;
281 bio->bi_io_vec = table;
284 EXPORT_SYMBOL(bio_init);
287 * bio_reset - reinitialize a bio
291 * After calling bio_reset(), @bio will be in the same state as a freshly
292 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
293 * preserved are the ones that are initialized by bio_alloc_bioset(). See
294 * comment in struct bio.
296 void bio_reset(struct bio *bio)
299 memset(bio, 0, BIO_RESET_BYTES);
300 atomic_set(&bio->__bi_remaining, 1);
302 EXPORT_SYMBOL(bio_reset);
304 static struct bio *__bio_chain_endio(struct bio *bio)
306 struct bio *parent = bio->bi_private;
308 if (bio->bi_status && !parent->bi_status)
309 parent->bi_status = bio->bi_status;
314 static void bio_chain_endio(struct bio *bio)
316 bio_endio(__bio_chain_endio(bio));
320 * bio_chain - chain bio completions
321 * @bio: the target bio
322 * @parent: the parent bio of @bio
324 * The caller won't have a bi_end_io called when @bio completes - instead,
325 * @parent's bi_end_io won't be called until both @parent and @bio have
326 * completed; the chained bio will also be freed when it completes.
328 * The caller must not set bi_private or bi_end_io in @bio.
330 void bio_chain(struct bio *bio, struct bio *parent)
332 BUG_ON(bio->bi_private || bio->bi_end_io);
334 bio->bi_private = parent;
335 bio->bi_end_io = bio_chain_endio;
336 bio_inc_remaining(parent);
338 EXPORT_SYMBOL(bio_chain);
340 static void bio_alloc_rescue(struct work_struct *work)
342 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
346 spin_lock(&bs->rescue_lock);
347 bio = bio_list_pop(&bs->rescue_list);
348 spin_unlock(&bs->rescue_lock);
353 submit_bio_noacct(bio);
357 static void punt_bios_to_rescuer(struct bio_set *bs)
359 struct bio_list punt, nopunt;
362 if (WARN_ON_ONCE(!bs->rescue_workqueue))
365 * In order to guarantee forward progress we must punt only bios that
366 * were allocated from this bio_set; otherwise, if there was a bio on
367 * there for a stacking driver higher up in the stack, processing it
368 * could require allocating bios from this bio_set, and doing that from
369 * our own rescuer would be bad.
371 * Since bio lists are singly linked, pop them all instead of trying to
372 * remove from the middle of the list:
375 bio_list_init(&punt);
376 bio_list_init(&nopunt);
378 while ((bio = bio_list_pop(¤t->bio_list[0])))
379 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
380 current->bio_list[0] = nopunt;
382 bio_list_init(&nopunt);
383 while ((bio = bio_list_pop(¤t->bio_list[1])))
384 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
385 current->bio_list[1] = nopunt;
387 spin_lock(&bs->rescue_lock);
388 bio_list_merge(&bs->rescue_list, &punt);
389 spin_unlock(&bs->rescue_lock);
391 queue_work(bs->rescue_workqueue, &bs->rescue_work);
395 * bio_alloc_bioset - allocate a bio for I/O
396 * @gfp_mask: the GFP_* mask given to the slab allocator
397 * @nr_iovecs: number of iovecs to pre-allocate
398 * @bs: the bio_set to allocate from.
400 * Allocate a bio from the mempools in @bs.
402 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
403 * allocate a bio. This is due to the mempool guarantees. To make this work,
404 * callers must never allocate more than 1 bio at a time from the general pool.
405 * Callers that need to allocate more than 1 bio must always submit the
406 * previously allocated bio for IO before attempting to allocate a new one.
407 * Failure to do so can cause deadlocks under memory pressure.
409 * Note that when running under submit_bio_noacct() (i.e. any block driver),
410 * bios are not submitted until after you return - see the code in
411 * submit_bio_noacct() that converts recursion into iteration, to prevent
414 * This would normally mean allocating multiple bios under submit_bio_noacct()
415 * would be susceptible to deadlocks, but we have
416 * deadlock avoidance code that resubmits any blocked bios from a rescuer
419 * However, we do not guarantee forward progress for allocations from other
420 * mempools. Doing multiple allocations from the same mempool under
421 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
422 * for per bio allocations.
424 * Returns: Pointer to new bio on success, NULL on failure.
426 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs,
429 gfp_t saved_gfp = gfp_mask;
433 /* should not use nobvec bioset for nr_iovecs > 0 */
434 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0))
438 * submit_bio_noacct() converts recursion to iteration; this means if
439 * we're running beneath it, any bios we allocate and submit will not be
440 * submitted (and thus freed) until after we return.
442 * This exposes us to a potential deadlock if we allocate multiple bios
443 * from the same bio_set() while running underneath submit_bio_noacct().
444 * If we were to allocate multiple bios (say a stacking block driver
445 * that was splitting bios), we would deadlock if we exhausted the
448 * We solve this, and guarantee forward progress, with a rescuer
449 * workqueue per bio_set. If we go to allocate and there are bios on
450 * current->bio_list, we first try the allocation without
451 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
452 * blocking to the rescuer workqueue before we retry with the original
455 if (current->bio_list &&
456 (!bio_list_empty(¤t->bio_list[0]) ||
457 !bio_list_empty(¤t->bio_list[1])) &&
458 bs->rescue_workqueue)
459 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
461 p = mempool_alloc(&bs->bio_pool, gfp_mask);
462 if (!p && gfp_mask != saved_gfp) {
463 punt_bios_to_rescuer(bs);
464 gfp_mask = saved_gfp;
465 p = mempool_alloc(&bs->bio_pool, gfp_mask);
470 bio = p + bs->front_pad;
471 if (nr_iovecs > BIO_INLINE_VECS) {
472 struct bio_vec *bvl = NULL;
474 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
475 if (!bvl && gfp_mask != saved_gfp) {
476 punt_bios_to_rescuer(bs);
477 gfp_mask = saved_gfp;
478 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
483 bio_init(bio, bvl, nr_iovecs);
484 } else if (nr_iovecs) {
485 bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS);
487 bio_init(bio, NULL, 0);
494 mempool_free(p, &bs->bio_pool);
497 EXPORT_SYMBOL(bio_alloc_bioset);
500 * bio_kmalloc - kmalloc a bio for I/O
501 * @gfp_mask: the GFP_* mask given to the slab allocator
502 * @nr_iovecs: number of iovecs to pre-allocate
504 * Use kmalloc to allocate and initialize a bio.
506 * Returns: Pointer to new bio on success, NULL on failure.
508 struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs)
512 if (nr_iovecs > UIO_MAXIOV)
515 bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask);
518 bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs);
522 EXPORT_SYMBOL(bio_kmalloc);
524 void zero_fill_bio(struct bio *bio)
528 struct bvec_iter iter;
530 bio_for_each_segment(bv, bio, iter) {
531 char *data = bvec_kmap_irq(&bv, &flags);
532 memset(data, 0, bv.bv_len);
533 flush_dcache_page(bv.bv_page);
534 bvec_kunmap_irq(data, &flags);
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, offset, bv.bv_len - offset);
578 * Don't touch bvec table here and make it really immutable, since
579 * fs bio user has to retrieve all pages via bio_for_each_segment_all
580 * in its .end_bio() callback.
582 * It is enough to truncate bio by updating .bi_size since we can make
583 * correct bvec with the updated .bi_size for drivers.
585 bio->bi_iter.bi_size = new_size;
589 * guard_bio_eod - truncate a BIO to fit the block device
590 * @bio: bio to truncate
592 * This allows us to do IO even on the odd last sectors of a device, even if the
593 * block size is some multiple of the physical sector size.
595 * We'll just truncate the bio to the size of the device, and clear the end of
596 * the buffer head manually. Truly out-of-range accesses will turn into actual
597 * I/O errors, this only handles the "we need to be able to do I/O at the final
600 void guard_bio_eod(struct bio *bio)
602 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
608 * If the *whole* IO is past the end of the device,
609 * let it through, and the IO layer will turn it into
612 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
615 maxsector -= bio->bi_iter.bi_sector;
616 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
619 bio_truncate(bio, maxsector << 9);
623 * bio_put - release a reference to a bio
624 * @bio: bio to release reference to
627 * Put a reference to a &struct bio, either one you have gotten with
628 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
630 void bio_put(struct bio *bio)
632 if (!bio_flagged(bio, BIO_REFFED))
635 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
640 if (atomic_dec_and_test(&bio->__bi_cnt))
644 EXPORT_SYMBOL(bio_put);
647 * __bio_clone_fast - clone a bio that shares the original bio's biovec
648 * @bio: destination bio
649 * @bio_src: bio to clone
651 * Clone a &bio. Caller will own the returned bio, but not
652 * the actual data it points to. Reference count of returned
655 * Caller must ensure that @bio_src is not freed before @bio.
657 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
659 WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs);
662 * most users will be overriding ->bi_bdev with a new target,
663 * so we don't set nor calculate new physical/hw segment counts here
665 bio->bi_bdev = bio_src->bi_bdev;
666 bio_set_flag(bio, BIO_CLONED);
667 if (bio_flagged(bio_src, BIO_THROTTLED))
668 bio_set_flag(bio, BIO_THROTTLED);
669 if (bio_flagged(bio_src, BIO_REMAPPED))
670 bio_set_flag(bio, BIO_REMAPPED);
671 bio->bi_opf = bio_src->bi_opf;
672 bio->bi_ioprio = bio_src->bi_ioprio;
673 bio->bi_write_hint = bio_src->bi_write_hint;
674 bio->bi_iter = bio_src->bi_iter;
675 bio->bi_io_vec = bio_src->bi_io_vec;
677 bio_clone_blkg_association(bio, bio_src);
678 blkcg_bio_issue_init(bio);
680 EXPORT_SYMBOL(__bio_clone_fast);
683 * bio_clone_fast - clone a bio that shares the original bio's biovec
685 * @gfp_mask: allocation priority
686 * @bs: bio_set to allocate from
688 * Like __bio_clone_fast, only also allocates the returned bio
690 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
694 b = bio_alloc_bioset(gfp_mask, 0, bs);
698 __bio_clone_fast(b, bio);
700 if (bio_crypt_clone(b, bio, gfp_mask) < 0)
703 if (bio_integrity(bio) &&
704 bio_integrity_clone(b, bio, gfp_mask) < 0)
713 EXPORT_SYMBOL(bio_clone_fast);
715 const char *bio_devname(struct bio *bio, char *buf)
717 return bdevname(bio->bi_bdev, buf);
719 EXPORT_SYMBOL(bio_devname);
721 static inline bool page_is_mergeable(const struct bio_vec *bv,
722 struct page *page, unsigned int len, unsigned int off,
725 size_t bv_end = bv->bv_offset + bv->bv_len;
726 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
727 phys_addr_t page_addr = page_to_phys(page);
729 if (vec_end_addr + 1 != page_addr + off)
731 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
734 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
737 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
741 * Try to merge a page into a segment, while obeying the hardware segment
742 * size limit. This is not for normal read/write bios, but for passthrough
743 * or Zone Append operations that we can't split.
745 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
746 struct page *page, unsigned len,
747 unsigned offset, bool *same_page)
749 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
750 unsigned long mask = queue_segment_boundary(q);
751 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
752 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
754 if ((addr1 | mask) != (addr2 | mask))
756 if (bv->bv_len + len > queue_max_segment_size(q))
758 return __bio_try_merge_page(bio, page, len, offset, same_page);
762 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
763 * @q: the target queue
764 * @bio: destination bio
766 * @len: vec entry length
767 * @offset: vec entry offset
768 * @max_sectors: maximum number of sectors that can be added
769 * @same_page: return if the segment has been merged inside the same page
771 * Add a page to a bio while respecting the hardware max_sectors, max_segment
772 * and gap limitations.
774 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
775 struct page *page, unsigned int len, unsigned int offset,
776 unsigned int max_sectors, bool *same_page)
778 struct bio_vec *bvec;
780 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
783 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
786 if (bio->bi_vcnt > 0) {
787 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
791 * If the queue doesn't support SG gaps and adding this segment
792 * would create a gap, disallow it.
794 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
795 if (bvec_gap_to_prev(q, bvec, offset))
799 if (bio_full(bio, len))
802 if (bio->bi_vcnt >= queue_max_segments(q))
805 bvec = &bio->bi_io_vec[bio->bi_vcnt];
806 bvec->bv_page = page;
808 bvec->bv_offset = offset;
810 bio->bi_iter.bi_size += len;
815 * bio_add_pc_page - attempt to add page to passthrough bio
816 * @q: the target queue
817 * @bio: destination bio
819 * @len: vec entry length
820 * @offset: vec entry offset
822 * Attempt to add a page to the bio_vec maplist. This can fail for a
823 * number of reasons, such as the bio being full or target block device
824 * limitations. The target block device must allow bio's up to PAGE_SIZE,
825 * so it is always possible to add a single page to an empty bio.
827 * This should only be used by passthrough bios.
829 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
830 struct page *page, unsigned int len, unsigned int offset)
832 bool same_page = false;
833 return bio_add_hw_page(q, bio, page, len, offset,
834 queue_max_hw_sectors(q), &same_page);
836 EXPORT_SYMBOL(bio_add_pc_page);
839 * bio_add_zone_append_page - attempt to add page to zone-append bio
840 * @bio: destination bio
842 * @len: vec entry length
843 * @offset: vec entry offset
845 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
846 * for a zone-append request. This can fail for a number of reasons, such as the
847 * bio being full or the target block device is not a zoned block device or
848 * other limitations of the target block device. The target block device must
849 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
852 * Returns: number of bytes added to the bio, or 0 in case of a failure.
854 int bio_add_zone_append_page(struct bio *bio, struct page *page,
855 unsigned int len, unsigned int offset)
857 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
858 bool same_page = false;
860 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
863 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
866 return bio_add_hw_page(q, bio, page, len, offset,
867 queue_max_zone_append_sectors(q), &same_page);
869 EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
872 * __bio_try_merge_page - try appending data to an existing bvec.
873 * @bio: destination bio
874 * @page: start page to add
875 * @len: length of the data to add
876 * @off: offset of the data relative to @page
877 * @same_page: return if the segment has been merged inside the same page
879 * Try to add the data at @page + @off to the last bvec of @bio. This is a
880 * useful optimisation for file systems with a block size smaller than the
883 * Warn if (@len, @off) crosses pages in case that @same_page is true.
885 * Return %true on success or %false on failure.
887 bool __bio_try_merge_page(struct bio *bio, struct page *page,
888 unsigned int len, unsigned int off, bool *same_page)
890 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
893 if (bio->bi_vcnt > 0) {
894 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
896 if (page_is_mergeable(bv, page, len, off, same_page)) {
897 if (bio->bi_iter.bi_size > UINT_MAX - len) {
902 bio->bi_iter.bi_size += len;
908 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
911 * __bio_add_page - add page(s) to a bio in a new segment
912 * @bio: destination bio
913 * @page: start page to add
914 * @len: length of the data to add, may cross pages
915 * @off: offset of the data relative to @page, may cross pages
917 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
918 * that @bio has space for another bvec.
920 void __bio_add_page(struct bio *bio, struct page *page,
921 unsigned int len, unsigned int off)
923 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
925 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
926 WARN_ON_ONCE(bio_full(bio, len));
932 bio->bi_iter.bi_size += len;
935 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
936 bio_set_flag(bio, BIO_WORKINGSET);
938 EXPORT_SYMBOL_GPL(__bio_add_page);
941 * bio_add_page - attempt to add page(s) to bio
942 * @bio: destination bio
943 * @page: start page to add
944 * @len: vec entry length, may cross pages
945 * @offset: vec entry offset relative to @page, may cross pages
947 * Attempt to add page(s) to the bio_vec maplist. This will only fail
948 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
950 int bio_add_page(struct bio *bio, struct page *page,
951 unsigned int len, unsigned int offset)
953 bool same_page = false;
955 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
956 if (bio_full(bio, len))
958 __bio_add_page(bio, page, len, offset);
962 EXPORT_SYMBOL(bio_add_page);
964 void bio_release_pages(struct bio *bio, bool mark_dirty)
966 struct bvec_iter_all iter_all;
967 struct bio_vec *bvec;
969 if (bio_flagged(bio, BIO_NO_PAGE_REF))
972 bio_for_each_segment_all(bvec, bio, iter_all) {
973 if (mark_dirty && !PageCompound(bvec->bv_page))
974 set_page_dirty_lock(bvec->bv_page);
975 put_page(bvec->bv_page);
978 EXPORT_SYMBOL_GPL(bio_release_pages);
980 static void __bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
982 WARN_ON_ONCE(bio->bi_max_vecs);
984 bio->bi_vcnt = iter->nr_segs;
985 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
986 bio->bi_iter.bi_bvec_done = iter->iov_offset;
987 bio->bi_iter.bi_size = iter->count;
988 bio_set_flag(bio, BIO_NO_PAGE_REF);
989 bio_set_flag(bio, BIO_CLONED);
992 static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
994 __bio_iov_bvec_set(bio, iter);
995 iov_iter_advance(iter, iter->count);
999 static int bio_iov_bvec_set_append(struct bio *bio, struct iov_iter *iter)
1001 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1002 struct iov_iter i = *iter;
1004 iov_iter_truncate(&i, queue_max_zone_append_sectors(q) << 9);
1005 __bio_iov_bvec_set(bio, &i);
1006 iov_iter_advance(iter, i.count);
1010 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1013 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1014 * @bio: bio to add pages to
1015 * @iter: iov iterator describing the region to be mapped
1017 * Pins pages from *iter and appends them to @bio's bvec array. The
1018 * pages will have to be released using put_page() when done.
1019 * For multi-segment *iter, this function only adds pages from the
1020 * next non-empty segment of the iov iterator.
1022 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1024 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1025 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1026 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1027 struct page **pages = (struct page **)bv;
1028 bool same_page = false;
1034 * Move page array up in the allocated memory for the bio vecs as far as
1035 * possible so that we can start filling biovecs from the beginning
1036 * without overwriting the temporary page array.
1038 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1039 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1041 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1042 if (unlikely(size <= 0))
1043 return size ? size : -EFAULT;
1045 for (left = size, i = 0; left > 0; left -= len, i++) {
1046 struct page *page = pages[i];
1048 len = min_t(size_t, PAGE_SIZE - offset, left);
1050 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1054 if (WARN_ON_ONCE(bio_full(bio, len)))
1056 __bio_add_page(bio, page, len, offset);
1061 iov_iter_advance(iter, size);
1065 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
1067 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1068 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1069 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1070 unsigned int max_append_sectors = queue_max_zone_append_sectors(q);
1071 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1072 struct page **pages = (struct page **)bv;
1078 if (WARN_ON_ONCE(!max_append_sectors))
1082 * Move page array up in the allocated memory for the bio vecs as far as
1083 * possible so that we can start filling biovecs from the beginning
1084 * without overwriting the temporary page array.
1086 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1087 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1089 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1090 if (unlikely(size <= 0))
1091 return size ? size : -EFAULT;
1093 for (left = size, i = 0; left > 0; left -= len, i++) {
1094 struct page *page = pages[i];
1095 bool same_page = false;
1097 len = min_t(size_t, PAGE_SIZE - offset, left);
1098 if (bio_add_hw_page(q, bio, page, len, offset,
1099 max_append_sectors, &same_page) != len) {
1108 iov_iter_advance(iter, size - left);
1113 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1114 * @bio: bio to add pages to
1115 * @iter: iov iterator describing the region to be added
1117 * This takes either an iterator pointing to user memory, or one pointing to
1118 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1119 * map them into the kernel. On IO completion, the caller should put those
1120 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1121 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1122 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1123 * completed by a call to ->ki_complete() or returns with an error other than
1124 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1125 * on IO completion. If it isn't, then pages should be released.
1127 * The function tries, but does not guarantee, to pin as many pages as
1128 * fit into the bio, or are requested in @iter, whatever is smaller. If
1129 * MM encounters an error pinning the requested pages, it stops. Error
1130 * is returned only if 0 pages could be pinned.
1132 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1133 * responsible for setting BIO_WORKINGSET if necessary.
1135 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1139 if (iov_iter_is_bvec(iter)) {
1140 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1141 return bio_iov_bvec_set_append(bio, iter);
1142 return bio_iov_bvec_set(bio, iter);
1146 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1147 ret = __bio_iov_append_get_pages(bio, iter);
1149 ret = __bio_iov_iter_get_pages(bio, iter);
1150 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1152 /* don't account direct I/O as memory stall */
1153 bio_clear_flag(bio, BIO_WORKINGSET);
1154 return bio->bi_vcnt ? 0 : ret;
1156 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1158 static void submit_bio_wait_endio(struct bio *bio)
1160 complete(bio->bi_private);
1164 * submit_bio_wait - submit a bio, and wait until it completes
1165 * @bio: The &struct bio which describes the I/O
1167 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1168 * bio_endio() on failure.
1170 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1171 * result in bio reference to be consumed. The caller must drop the reference
1174 int submit_bio_wait(struct bio *bio)
1176 DECLARE_COMPLETION_ONSTACK_MAP(done,
1177 bio->bi_bdev->bd_disk->lockdep_map);
1178 unsigned long hang_check;
1180 bio->bi_private = &done;
1181 bio->bi_end_io = submit_bio_wait_endio;
1182 bio->bi_opf |= REQ_SYNC;
1185 /* Prevent hang_check timer from firing at us during very long I/O */
1186 hang_check = sysctl_hung_task_timeout_secs;
1188 while (!wait_for_completion_io_timeout(&done,
1189 hang_check * (HZ/2)))
1192 wait_for_completion_io(&done);
1194 return blk_status_to_errno(bio->bi_status);
1196 EXPORT_SYMBOL(submit_bio_wait);
1199 * bio_advance - increment/complete a bio by some number of bytes
1200 * @bio: bio to advance
1201 * @bytes: number of bytes to complete
1203 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1204 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1205 * be updated on the last bvec as well.
1207 * @bio will then represent the remaining, uncompleted portion of the io.
1209 void bio_advance(struct bio *bio, unsigned bytes)
1211 if (bio_integrity(bio))
1212 bio_integrity_advance(bio, bytes);
1214 bio_crypt_advance(bio, bytes);
1215 bio_advance_iter(bio, &bio->bi_iter, bytes);
1217 EXPORT_SYMBOL(bio_advance);
1219 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1220 struct bio *src, struct bvec_iter *src_iter)
1222 struct bio_vec src_bv, dst_bv;
1223 void *src_p, *dst_p;
1226 while (src_iter->bi_size && dst_iter->bi_size) {
1227 src_bv = bio_iter_iovec(src, *src_iter);
1228 dst_bv = bio_iter_iovec(dst, *dst_iter);
1230 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1232 src_p = kmap_atomic(src_bv.bv_page);
1233 dst_p = kmap_atomic(dst_bv.bv_page);
1235 memcpy(dst_p + dst_bv.bv_offset,
1236 src_p + src_bv.bv_offset,
1239 kunmap_atomic(dst_p);
1240 kunmap_atomic(src_p);
1242 flush_dcache_page(dst_bv.bv_page);
1244 bio_advance_iter_single(src, src_iter, bytes);
1245 bio_advance_iter_single(dst, dst_iter, bytes);
1248 EXPORT_SYMBOL(bio_copy_data_iter);
1251 * bio_copy_data - copy contents of data buffers from one bio to another
1253 * @dst: destination bio
1255 * Stops when it reaches the end of either @src or @dst - that is, copies
1256 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1258 void bio_copy_data(struct bio *dst, struct bio *src)
1260 struct bvec_iter src_iter = src->bi_iter;
1261 struct bvec_iter dst_iter = dst->bi_iter;
1263 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1265 EXPORT_SYMBOL(bio_copy_data);
1267 void bio_free_pages(struct bio *bio)
1269 struct bio_vec *bvec;
1270 struct bvec_iter_all iter_all;
1272 bio_for_each_segment_all(bvec, bio, iter_all)
1273 __free_page(bvec->bv_page);
1275 EXPORT_SYMBOL(bio_free_pages);
1278 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1279 * for performing direct-IO in BIOs.
1281 * The problem is that we cannot run set_page_dirty() from interrupt context
1282 * because the required locks are not interrupt-safe. So what we can do is to
1283 * mark the pages dirty _before_ performing IO. And in interrupt context,
1284 * check that the pages are still dirty. If so, fine. If not, redirty them
1285 * in process context.
1287 * We special-case compound pages here: normally this means reads into hugetlb
1288 * pages. The logic in here doesn't really work right for compound pages
1289 * because the VM does not uniformly chase down the head page in all cases.
1290 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1291 * handle them at all. So we skip compound pages here at an early stage.
1293 * Note that this code is very hard to test under normal circumstances because
1294 * direct-io pins the pages with get_user_pages(). This makes
1295 * is_page_cache_freeable return false, and the VM will not clean the pages.
1296 * But other code (eg, flusher threads) could clean the pages if they are mapped
1299 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1300 * deferred bio dirtying paths.
1304 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1306 void bio_set_pages_dirty(struct bio *bio)
1308 struct bio_vec *bvec;
1309 struct bvec_iter_all iter_all;
1311 bio_for_each_segment_all(bvec, bio, iter_all) {
1312 if (!PageCompound(bvec->bv_page))
1313 set_page_dirty_lock(bvec->bv_page);
1318 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1319 * If they are, then fine. If, however, some pages are clean then they must
1320 * have been written out during the direct-IO read. So we take another ref on
1321 * the BIO and re-dirty the pages in process context.
1323 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1324 * here on. It will run one put_page() against each page and will run one
1325 * bio_put() against the BIO.
1328 static void bio_dirty_fn(struct work_struct *work);
1330 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1331 static DEFINE_SPINLOCK(bio_dirty_lock);
1332 static struct bio *bio_dirty_list;
1335 * This runs in process context
1337 static void bio_dirty_fn(struct work_struct *work)
1339 struct bio *bio, *next;
1341 spin_lock_irq(&bio_dirty_lock);
1342 next = bio_dirty_list;
1343 bio_dirty_list = NULL;
1344 spin_unlock_irq(&bio_dirty_lock);
1346 while ((bio = next) != NULL) {
1347 next = bio->bi_private;
1349 bio_release_pages(bio, true);
1354 void bio_check_pages_dirty(struct bio *bio)
1356 struct bio_vec *bvec;
1357 unsigned long flags;
1358 struct bvec_iter_all iter_all;
1360 bio_for_each_segment_all(bvec, bio, iter_all) {
1361 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1365 bio_release_pages(bio, false);
1369 spin_lock_irqsave(&bio_dirty_lock, flags);
1370 bio->bi_private = bio_dirty_list;
1371 bio_dirty_list = bio;
1372 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1373 schedule_work(&bio_dirty_work);
1376 static inline bool bio_remaining_done(struct bio *bio)
1379 * If we're not chaining, then ->__bi_remaining is always 1 and
1380 * we always end io on the first invocation.
1382 if (!bio_flagged(bio, BIO_CHAIN))
1385 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1387 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1388 bio_clear_flag(bio, BIO_CHAIN);
1396 * bio_endio - end I/O on a bio
1400 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1401 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1402 * bio unless they own it and thus know that it has an end_io function.
1404 * bio_endio() can be called several times on a bio that has been chained
1405 * using bio_chain(). The ->bi_end_io() function will only be called the
1408 void bio_endio(struct bio *bio)
1411 if (!bio_remaining_done(bio))
1413 if (!bio_integrity_endio(bio))
1417 rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio);
1419 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1420 trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio);
1421 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1425 * Need to have a real endio function for chained bios, otherwise
1426 * various corner cases will break (like stacking block devices that
1427 * save/restore bi_end_io) - however, we want to avoid unbounded
1428 * recursion and blowing the stack. Tail call optimization would
1429 * handle this, but compiling with frame pointers also disables
1430 * gcc's sibling call optimization.
1432 if (bio->bi_end_io == bio_chain_endio) {
1433 bio = __bio_chain_endio(bio);
1437 blk_throtl_bio_endio(bio);
1438 /* release cgroup info */
1441 bio->bi_end_io(bio);
1443 EXPORT_SYMBOL(bio_endio);
1446 * bio_split - split a bio
1447 * @bio: bio to split
1448 * @sectors: number of sectors to split from the front of @bio
1450 * @bs: bio set to allocate from
1452 * Allocates and returns a new bio which represents @sectors from the start of
1453 * @bio, and updates @bio to represent the remaining sectors.
1455 * Unless this is a discard request the newly allocated bio will point
1456 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1457 * neither @bio nor @bs are freed before the split bio.
1459 struct bio *bio_split(struct bio *bio, int sectors,
1460 gfp_t gfp, struct bio_set *bs)
1464 BUG_ON(sectors <= 0);
1465 BUG_ON(sectors >= bio_sectors(bio));
1467 /* Zone append commands cannot be split */
1468 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1471 split = bio_clone_fast(bio, gfp, bs);
1475 split->bi_iter.bi_size = sectors << 9;
1477 if (bio_integrity(split))
1478 bio_integrity_trim(split);
1480 bio_advance(bio, split->bi_iter.bi_size);
1482 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1483 bio_set_flag(split, BIO_TRACE_COMPLETION);
1487 EXPORT_SYMBOL(bio_split);
1490 * bio_trim - trim a bio
1492 * @offset: number of sectors to trim from the front of @bio
1493 * @size: size we want to trim @bio to, in sectors
1495 void bio_trim(struct bio *bio, int offset, int size)
1497 /* 'bio' is a cloned bio which we need to trim to match
1498 * the given offset and size.
1502 if (offset == 0 && size == bio->bi_iter.bi_size)
1505 bio_advance(bio, offset << 9);
1506 bio->bi_iter.bi_size = size;
1508 if (bio_integrity(bio))
1509 bio_integrity_trim(bio);
1512 EXPORT_SYMBOL_GPL(bio_trim);
1515 * create memory pools for biovec's in a bio_set.
1516 * use the global biovec slabs created for general use.
1518 int biovec_init_pool(mempool_t *pool, int pool_entries)
1520 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1522 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1526 * bioset_exit - exit a bioset initialized with bioset_init()
1528 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1531 void bioset_exit(struct bio_set *bs)
1533 if (bs->rescue_workqueue)
1534 destroy_workqueue(bs->rescue_workqueue);
1535 bs->rescue_workqueue = NULL;
1537 mempool_exit(&bs->bio_pool);
1538 mempool_exit(&bs->bvec_pool);
1540 bioset_integrity_free(bs);
1543 bs->bio_slab = NULL;
1545 EXPORT_SYMBOL(bioset_exit);
1548 * bioset_init - Initialize a bio_set
1549 * @bs: pool to initialize
1550 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1551 * @front_pad: Number of bytes to allocate in front of the returned bio
1552 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1553 * and %BIOSET_NEED_RESCUER
1556 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1557 * to ask for a number of bytes to be allocated in front of the bio.
1558 * Front pad allocation is useful for embedding the bio inside
1559 * another structure, to avoid allocating extra data to go with the bio.
1560 * Note that the bio must be embedded at the END of that structure always,
1561 * or things will break badly.
1562 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1563 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1564 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1565 * dispatch queued requests when the mempool runs out of space.
1568 int bioset_init(struct bio_set *bs,
1569 unsigned int pool_size,
1570 unsigned int front_pad,
1573 bs->front_pad = front_pad;
1574 if (flags & BIOSET_NEED_BVECS)
1575 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1579 spin_lock_init(&bs->rescue_lock);
1580 bio_list_init(&bs->rescue_list);
1581 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1583 bs->bio_slab = bio_find_or_create_slab(bs);
1587 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1590 if ((flags & BIOSET_NEED_BVECS) &&
1591 biovec_init_pool(&bs->bvec_pool, pool_size))
1594 if (!(flags & BIOSET_NEED_RESCUER))
1597 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1598 if (!bs->rescue_workqueue)
1606 EXPORT_SYMBOL(bioset_init);
1609 * Initialize and setup a new bio_set, based on the settings from
1612 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1617 if (src->bvec_pool.min_nr)
1618 flags |= BIOSET_NEED_BVECS;
1619 if (src->rescue_workqueue)
1620 flags |= BIOSET_NEED_RESCUER;
1622 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1624 EXPORT_SYMBOL(bioset_init_from_src);
1626 static int __init init_bio(void)
1630 bio_integrity_init();
1632 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1633 struct biovec_slab *bvs = bvec_slabs + i;
1635 bvs->slab = kmem_cache_create(bvs->name,
1636 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1637 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1640 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
1641 panic("bio: can't allocate bios\n");
1643 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1644 panic("bio: can't create integrity pool\n");
1648 subsys_initcall(init_bio);