2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
31 #include <scsi/sg.h> /* for struct sg_iovec */
33 #include <trace/events/block.h>
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
39 #define BIO_INLINE_VECS 4
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
46 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
47 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
48 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
56 struct bio_set *fs_bio_set;
57 EXPORT_SYMBOL(fs_bio_set);
60 * Our slab pool management
63 struct kmem_cache *slab;
64 unsigned int slab_ref;
65 unsigned int slab_size;
68 static DEFINE_MUTEX(bio_slab_lock);
69 static struct bio_slab *bio_slabs;
70 static unsigned int bio_slab_nr, bio_slab_max;
72 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
74 unsigned int sz = sizeof(struct bio) + extra_size;
75 struct kmem_cache *slab = NULL;
76 struct bio_slab *bslab, *new_bio_slabs;
77 unsigned int new_bio_slab_max;
78 unsigned int i, entry = -1;
80 mutex_lock(&bio_slab_lock);
83 while (i < bio_slab_nr) {
84 bslab = &bio_slabs[i];
86 if (!bslab->slab && entry == -1)
88 else if (bslab->slab_size == sz) {
99 if (bio_slab_nr == bio_slab_max && entry == -1) {
100 new_bio_slab_max = bio_slab_max << 1;
101 new_bio_slabs = krealloc(bio_slabs,
102 new_bio_slab_max * sizeof(struct bio_slab),
106 bio_slab_max = new_bio_slab_max;
107 bio_slabs = new_bio_slabs;
110 entry = bio_slab_nr++;
112 bslab = &bio_slabs[entry];
114 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
115 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
116 SLAB_HWCACHE_ALIGN, NULL);
122 bslab->slab_size = sz;
124 mutex_unlock(&bio_slab_lock);
128 static void bio_put_slab(struct bio_set *bs)
130 struct bio_slab *bslab = NULL;
133 mutex_lock(&bio_slab_lock);
135 for (i = 0; i < bio_slab_nr; i++) {
136 if (bs->bio_slab == bio_slabs[i].slab) {
137 bslab = &bio_slabs[i];
142 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 WARN_ON(!bslab->slab_ref);
147 if (--bslab->slab_ref)
150 kmem_cache_destroy(bslab->slab);
154 mutex_unlock(&bio_slab_lock);
157 unsigned int bvec_nr_vecs(unsigned short idx)
159 return bvec_slabs[idx].nr_vecs;
162 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
164 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
166 if (idx == BIOVEC_MAX_IDX)
167 mempool_free(bv, pool);
169 struct biovec_slab *bvs = bvec_slabs + idx;
171 kmem_cache_free(bvs->slab, bv);
175 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
181 * see comment near bvec_array define!
199 case 129 ... BIO_MAX_PAGES:
207 * idx now points to the pool we want to allocate from. only the
208 * 1-vec entry pool is mempool backed.
210 if (*idx == BIOVEC_MAX_IDX) {
212 bvl = mempool_alloc(pool, gfp_mask);
214 struct biovec_slab *bvs = bvec_slabs + *idx;
215 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
218 * Make this allocation restricted and don't dump info on
219 * allocation failures, since we'll fallback to the mempool
220 * in case of failure.
222 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
225 * Try a slab allocation. If this fails and __GFP_WAIT
226 * is set, retry with the 1-entry mempool
228 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
229 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
230 *idx = BIOVEC_MAX_IDX;
238 static void __bio_free(struct bio *bio)
240 bio_disassociate_task(bio);
242 if (bio_integrity(bio))
243 bio_integrity_free(bio);
246 static void bio_free(struct bio *bio)
248 struct bio_set *bs = bio->bi_pool;
254 if (bio_flagged(bio, BIO_OWNS_VEC))
255 bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
258 * If we have front padding, adjust the bio pointer before freeing
263 mempool_free(p, bs->bio_pool);
265 /* Bio was allocated by bio_kmalloc() */
270 void bio_init(struct bio *bio)
272 memset(bio, 0, sizeof(*bio));
273 bio->bi_flags = 1 << BIO_UPTODATE;
274 atomic_set(&bio->bi_remaining, 1);
275 atomic_set(&bio->bi_cnt, 1);
277 EXPORT_SYMBOL(bio_init);
280 * bio_reset - reinitialize a bio
284 * After calling bio_reset(), @bio will be in the same state as a freshly
285 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
286 * preserved are the ones that are initialized by bio_alloc_bioset(). See
287 * comment in struct bio.
289 void bio_reset(struct bio *bio)
291 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
295 memset(bio, 0, BIO_RESET_BYTES);
296 bio->bi_flags = flags|(1 << BIO_UPTODATE);
297 atomic_set(&bio->bi_remaining, 1);
299 EXPORT_SYMBOL(bio_reset);
301 static void bio_chain_endio(struct bio *bio, int error)
303 bio_endio(bio->bi_private, error);
308 * bio_chain - chain bio completions
309 * @bio: the target bio
310 * @parent: the @bio's parent bio
312 * The caller won't have a bi_end_io called when @bio completes - instead,
313 * @parent's bi_end_io won't be called until both @parent and @bio have
314 * completed; the chained bio will also be freed when it completes.
316 * The caller must not set bi_private or bi_end_io in @bio.
318 void bio_chain(struct bio *bio, struct bio *parent)
320 BUG_ON(bio->bi_private || bio->bi_end_io);
322 bio->bi_private = parent;
323 bio->bi_end_io = bio_chain_endio;
324 atomic_inc(&parent->bi_remaining);
326 EXPORT_SYMBOL(bio_chain);
328 static void bio_alloc_rescue(struct work_struct *work)
330 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
334 spin_lock(&bs->rescue_lock);
335 bio = bio_list_pop(&bs->rescue_list);
336 spin_unlock(&bs->rescue_lock);
341 generic_make_request(bio);
345 static void punt_bios_to_rescuer(struct bio_set *bs)
347 struct bio_list punt, nopunt;
351 * In order to guarantee forward progress we must punt only bios that
352 * were allocated from this bio_set; otherwise, if there was a bio on
353 * there for a stacking driver higher up in the stack, processing it
354 * could require allocating bios from this bio_set, and doing that from
355 * our own rescuer would be bad.
357 * Since bio lists are singly linked, pop them all instead of trying to
358 * remove from the middle of the list:
361 bio_list_init(&punt);
362 bio_list_init(&nopunt);
364 while ((bio = bio_list_pop(current->bio_list)))
365 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
367 *current->bio_list = nopunt;
369 spin_lock(&bs->rescue_lock);
370 bio_list_merge(&bs->rescue_list, &punt);
371 spin_unlock(&bs->rescue_lock);
373 queue_work(bs->rescue_workqueue, &bs->rescue_work);
377 * bio_alloc_bioset - allocate a bio for I/O
378 * @gfp_mask: the GFP_ mask given to the slab allocator
379 * @nr_iovecs: number of iovecs to pre-allocate
380 * @bs: the bio_set to allocate from.
383 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
384 * backed by the @bs's mempool.
386 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
387 * able to allocate a bio. This is due to the mempool guarantees. To make this
388 * work, callers must never allocate more than 1 bio at a time from this pool.
389 * Callers that need to allocate more than 1 bio must always submit the
390 * previously allocated bio for IO before attempting to allocate a new one.
391 * Failure to do so can cause deadlocks under memory pressure.
393 * Note that when running under generic_make_request() (i.e. any block
394 * driver), bios are not submitted until after you return - see the code in
395 * generic_make_request() that converts recursion into iteration, to prevent
398 * This would normally mean allocating multiple bios under
399 * generic_make_request() would be susceptible to deadlocks, but we have
400 * deadlock avoidance code that resubmits any blocked bios from a rescuer
403 * However, we do not guarantee forward progress for allocations from other
404 * mempools. Doing multiple allocations from the same mempool under
405 * generic_make_request() should be avoided - instead, use bio_set's front_pad
406 * for per bio allocations.
409 * Pointer to new bio on success, NULL on failure.
411 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
413 gfp_t saved_gfp = gfp_mask;
415 unsigned inline_vecs;
416 unsigned long idx = BIO_POOL_NONE;
417 struct bio_vec *bvl = NULL;
422 if (nr_iovecs > UIO_MAXIOV)
425 p = kmalloc(sizeof(struct bio) +
426 nr_iovecs * sizeof(struct bio_vec),
429 inline_vecs = nr_iovecs;
431 /* should not use nobvec bioset for nr_iovecs > 0 */
432 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
435 * generic_make_request() converts recursion to iteration; this
436 * means if we're running beneath it, any bios we allocate and
437 * submit will not be submitted (and thus freed) until after we
440 * This exposes us to a potential deadlock if we allocate
441 * multiple bios from the same bio_set() while running
442 * underneath generic_make_request(). If we were to allocate
443 * multiple bios (say a stacking block driver that was splitting
444 * bios), we would deadlock if we exhausted the mempool's
447 * We solve this, and guarantee forward progress, with a rescuer
448 * workqueue per bio_set. If we go to allocate and there are
449 * bios on current->bio_list, we first try the allocation
450 * without __GFP_WAIT; if that fails, we punt those bios we
451 * would be blocking to the rescuer workqueue before we retry
452 * with the original gfp_flags.
455 if (current->bio_list && !bio_list_empty(current->bio_list))
456 gfp_mask &= ~__GFP_WAIT;
458 p = mempool_alloc(bs->bio_pool, gfp_mask);
459 if (!p && gfp_mask != saved_gfp) {
460 punt_bios_to_rescuer(bs);
461 gfp_mask = saved_gfp;
462 p = mempool_alloc(bs->bio_pool, gfp_mask);
465 front_pad = bs->front_pad;
466 inline_vecs = BIO_INLINE_VECS;
475 if (nr_iovecs > inline_vecs) {
476 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
477 if (!bvl && gfp_mask != saved_gfp) {
478 punt_bios_to_rescuer(bs);
479 gfp_mask = saved_gfp;
480 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
486 bio->bi_flags |= 1 << BIO_OWNS_VEC;
487 } else if (nr_iovecs) {
488 bvl = bio->bi_inline_vecs;
492 bio->bi_flags |= idx << BIO_POOL_OFFSET;
493 bio->bi_max_vecs = nr_iovecs;
494 bio->bi_io_vec = bvl;
498 mempool_free(p, bs->bio_pool);
501 EXPORT_SYMBOL(bio_alloc_bioset);
503 void zero_fill_bio(struct bio *bio)
507 struct bvec_iter iter;
509 bio_for_each_segment(bv, bio, iter) {
510 char *data = bvec_kmap_irq(&bv, &flags);
511 memset(data, 0, bv.bv_len);
512 flush_dcache_page(bv.bv_page);
513 bvec_kunmap_irq(data, &flags);
516 EXPORT_SYMBOL(zero_fill_bio);
519 * bio_put - release a reference to a bio
520 * @bio: bio to release reference to
523 * Put a reference to a &struct bio, either one you have gotten with
524 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
526 void bio_put(struct bio *bio)
528 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
533 if (atomic_dec_and_test(&bio->bi_cnt))
536 EXPORT_SYMBOL(bio_put);
538 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
540 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
541 blk_recount_segments(q, bio);
543 return bio->bi_phys_segments;
545 EXPORT_SYMBOL(bio_phys_segments);
548 * __bio_clone_fast - clone a bio that shares the original bio's biovec
549 * @bio: destination bio
550 * @bio_src: bio to clone
552 * Clone a &bio. Caller will own the returned bio, but not
553 * the actual data it points to. Reference count of returned
556 * Caller must ensure that @bio_src is not freed before @bio.
558 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
560 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
563 * most users will be overriding ->bi_bdev with a new target,
564 * so we don't set nor calculate new physical/hw segment counts here
566 bio->bi_bdev = bio_src->bi_bdev;
567 bio->bi_flags |= 1 << BIO_CLONED;
568 bio->bi_rw = bio_src->bi_rw;
569 bio->bi_iter = bio_src->bi_iter;
570 bio->bi_io_vec = bio_src->bi_io_vec;
572 EXPORT_SYMBOL(__bio_clone_fast);
575 * bio_clone_fast - clone a bio that shares the original bio's biovec
577 * @gfp_mask: allocation priority
578 * @bs: bio_set to allocate from
580 * Like __bio_clone_fast, only also allocates the returned bio
582 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
586 b = bio_alloc_bioset(gfp_mask, 0, bs);
590 __bio_clone_fast(b, bio);
592 if (bio_integrity(bio)) {
595 ret = bio_integrity_clone(b, bio, gfp_mask);
605 EXPORT_SYMBOL(bio_clone_fast);
608 * bio_clone_bioset - clone a bio
609 * @bio_src: bio to clone
610 * @gfp_mask: allocation priority
611 * @bs: bio_set to allocate from
613 * Clone bio. Caller will own the returned bio, but not the actual data it
614 * points to. Reference count of returned bio will be one.
616 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
619 struct bvec_iter iter;
624 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
625 * bio_src->bi_io_vec to bio->bi_io_vec.
627 * We can't do that anymore, because:
629 * - The point of cloning the biovec is to produce a bio with a biovec
630 * the caller can modify: bi_idx and bi_bvec_done should be 0.
632 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
633 * we tried to clone the whole thing bio_alloc_bioset() would fail.
634 * But the clone should succeed as long as the number of biovecs we
635 * actually need to allocate is fewer than BIO_MAX_PAGES.
637 * - Lastly, bi_vcnt should not be looked at or relied upon by code
638 * that does not own the bio - reason being drivers don't use it for
639 * iterating over the biovec anymore, so expecting it to be kept up
640 * to date (i.e. for clones that share the parent biovec) is just
641 * asking for trouble and would force extra work on
642 * __bio_clone_fast() anyways.
645 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
649 bio->bi_bdev = bio_src->bi_bdev;
650 bio->bi_rw = bio_src->bi_rw;
651 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
652 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
654 if (bio->bi_rw & REQ_DISCARD)
655 goto integrity_clone;
657 if (bio->bi_rw & REQ_WRITE_SAME) {
658 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
659 goto integrity_clone;
662 bio_for_each_segment(bv, bio_src, iter)
663 bio->bi_io_vec[bio->bi_vcnt++] = bv;
666 if (bio_integrity(bio_src)) {
669 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
678 EXPORT_SYMBOL(bio_clone_bioset);
681 * bio_get_nr_vecs - return approx number of vecs
684 * Return the approximate number of pages we can send to this target.
685 * There's no guarantee that you will be able to fit this number of pages
686 * into a bio, it does not account for dynamic restrictions that vary
689 int bio_get_nr_vecs(struct block_device *bdev)
691 struct request_queue *q = bdev_get_queue(bdev);
694 nr_pages = min_t(unsigned,
695 queue_max_segments(q),
696 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
698 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
701 EXPORT_SYMBOL(bio_get_nr_vecs);
703 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
704 *page, unsigned int len, unsigned int offset,
705 unsigned int max_sectors)
707 int retried_segments = 0;
708 struct bio_vec *bvec;
711 * cloned bio must not modify vec list
713 if (unlikely(bio_flagged(bio, BIO_CLONED)))
716 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
720 * For filesystems with a blocksize smaller than the pagesize
721 * we will often be called with the same page as last time and
722 * a consecutive offset. Optimize this special case.
724 if (bio->bi_vcnt > 0) {
725 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
727 if (page == prev->bv_page &&
728 offset == prev->bv_offset + prev->bv_len) {
729 unsigned int prev_bv_len = prev->bv_len;
732 if (q->merge_bvec_fn) {
733 struct bvec_merge_data bvm = {
734 /* prev_bvec is already charged in
735 bi_size, discharge it in order to
736 simulate merging updated prev_bvec
738 .bi_bdev = bio->bi_bdev,
739 .bi_sector = bio->bi_iter.bi_sector,
740 .bi_size = bio->bi_iter.bi_size -
745 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
751 bio->bi_iter.bi_size += len;
756 * If the queue doesn't support SG gaps and adding this
757 * offset would create a gap, disallow it.
759 if (q->queue_flags & (1 << QUEUE_FLAG_SG_GAPS) &&
760 bvec_gap_to_prev(prev, offset))
764 if (bio->bi_vcnt >= bio->bi_max_vecs)
768 * setup the new entry, we might clear it again later if we
769 * cannot add the page
771 bvec = &bio->bi_io_vec[bio->bi_vcnt];
772 bvec->bv_page = page;
774 bvec->bv_offset = offset;
776 bio->bi_phys_segments++;
777 bio->bi_iter.bi_size += len;
780 * Perform a recount if the number of segments is greater
781 * than queue_max_segments(q).
784 while (bio->bi_phys_segments > queue_max_segments(q)) {
786 if (retried_segments)
789 retried_segments = 1;
790 blk_recount_segments(q, bio);
794 * if queue has other restrictions (eg varying max sector size
795 * depending on offset), it can specify a merge_bvec_fn in the
796 * queue to get further control
798 if (q->merge_bvec_fn) {
799 struct bvec_merge_data bvm = {
800 .bi_bdev = bio->bi_bdev,
801 .bi_sector = bio->bi_iter.bi_sector,
802 .bi_size = bio->bi_iter.bi_size - len,
807 * merge_bvec_fn() returns number of bytes it can accept
810 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len)
814 /* If we may be able to merge these biovecs, force a recount */
815 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
816 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
822 bvec->bv_page = NULL;
826 bio->bi_iter.bi_size -= len;
827 blk_recount_segments(q, bio);
832 * bio_add_pc_page - attempt to add page to bio
833 * @q: the target queue
834 * @bio: destination bio
836 * @len: vec entry length
837 * @offset: vec entry offset
839 * Attempt to add a page to the bio_vec maplist. This can fail for a
840 * number of reasons, such as the bio being full or target block device
841 * limitations. The target block device must allow bio's up to PAGE_SIZE,
842 * so it is always possible to add a single page to an empty bio.
844 * This should only be used by REQ_PC bios.
846 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
847 unsigned int len, unsigned int offset)
849 return __bio_add_page(q, bio, page, len, offset,
850 queue_max_hw_sectors(q));
852 EXPORT_SYMBOL(bio_add_pc_page);
855 * bio_add_page - attempt to add page to bio
856 * @bio: destination bio
858 * @len: vec entry length
859 * @offset: vec entry offset
861 * Attempt to add a page to the bio_vec maplist. This can fail for a
862 * number of reasons, such as the bio being full or target block device
863 * limitations. The target block device must allow bio's up to PAGE_SIZE,
864 * so it is always possible to add a single page to an empty bio.
866 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
869 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
870 unsigned int max_sectors;
872 max_sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
873 if ((max_sectors < (len >> 9)) && !bio->bi_iter.bi_size)
874 max_sectors = len >> 9;
876 return __bio_add_page(q, bio, page, len, offset, max_sectors);
878 EXPORT_SYMBOL(bio_add_page);
880 struct submit_bio_ret {
881 struct completion event;
885 static void submit_bio_wait_endio(struct bio *bio, int error)
887 struct submit_bio_ret *ret = bio->bi_private;
890 complete(&ret->event);
894 * submit_bio_wait - submit a bio, and wait until it completes
895 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
896 * @bio: The &struct bio which describes the I/O
898 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
899 * bio_endio() on failure.
901 int submit_bio_wait(int rw, struct bio *bio)
903 struct submit_bio_ret ret;
906 init_completion(&ret.event);
907 bio->bi_private = &ret;
908 bio->bi_end_io = submit_bio_wait_endio;
910 wait_for_completion(&ret.event);
914 EXPORT_SYMBOL(submit_bio_wait);
917 * bio_advance - increment/complete a bio by some number of bytes
918 * @bio: bio to advance
919 * @bytes: number of bytes to complete
921 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
922 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
923 * be updated on the last bvec as well.
925 * @bio will then represent the remaining, uncompleted portion of the io.
927 void bio_advance(struct bio *bio, unsigned bytes)
929 if (bio_integrity(bio))
930 bio_integrity_advance(bio, bytes);
932 bio_advance_iter(bio, &bio->bi_iter, bytes);
934 EXPORT_SYMBOL(bio_advance);
937 * bio_alloc_pages - allocates a single page for each bvec in a bio
938 * @bio: bio to allocate pages for
939 * @gfp_mask: flags for allocation
941 * Allocates pages up to @bio->bi_vcnt.
943 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
946 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
951 bio_for_each_segment_all(bv, bio, i) {
952 bv->bv_page = alloc_page(gfp_mask);
954 while (--bv >= bio->bi_io_vec)
955 __free_page(bv->bv_page);
962 EXPORT_SYMBOL(bio_alloc_pages);
965 * bio_copy_data - copy contents of data buffers from one chain of bios to
967 * @src: source bio list
968 * @dst: destination bio list
970 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
971 * @src and @dst as linked lists of bios.
973 * Stops when it reaches the end of either @src or @dst - that is, copies
974 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
976 void bio_copy_data(struct bio *dst, struct bio *src)
978 struct bvec_iter src_iter, dst_iter;
979 struct bio_vec src_bv, dst_bv;
983 src_iter = src->bi_iter;
984 dst_iter = dst->bi_iter;
987 if (!src_iter.bi_size) {
992 src_iter = src->bi_iter;
995 if (!dst_iter.bi_size) {
1000 dst_iter = dst->bi_iter;
1003 src_bv = bio_iter_iovec(src, src_iter);
1004 dst_bv = bio_iter_iovec(dst, dst_iter);
1006 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1008 src_p = kmap_atomic(src_bv.bv_page);
1009 dst_p = kmap_atomic(dst_bv.bv_page);
1011 memcpy(dst_p + dst_bv.bv_offset,
1012 src_p + src_bv.bv_offset,
1015 kunmap_atomic(dst_p);
1016 kunmap_atomic(src_p);
1018 bio_advance_iter(src, &src_iter, bytes);
1019 bio_advance_iter(dst, &dst_iter, bytes);
1022 EXPORT_SYMBOL(bio_copy_data);
1024 struct bio_map_data {
1027 struct sg_iovec sgvecs[];
1030 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
1031 const struct sg_iovec *iov, int iov_count,
1034 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
1035 bmd->nr_sgvecs = iov_count;
1036 bmd->is_our_pages = is_our_pages;
1037 bio->bi_private = bmd;
1040 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
1043 if (iov_count > UIO_MAXIOV)
1046 return kmalloc(sizeof(struct bio_map_data) +
1047 sizeof(struct sg_iovec) * iov_count, gfp_mask);
1050 static int __bio_copy_iov(struct bio *bio, const struct sg_iovec *iov, int iov_count,
1051 int to_user, int from_user, int do_free_page)
1054 struct bio_vec *bvec;
1056 unsigned int iov_off = 0;
1058 bio_for_each_segment_all(bvec, bio, i) {
1059 char *bv_addr = page_address(bvec->bv_page);
1060 unsigned int bv_len = bvec->bv_len;
1062 while (bv_len && iov_idx < iov_count) {
1064 char __user *iov_addr;
1066 bytes = min_t(unsigned int,
1067 iov[iov_idx].iov_len - iov_off, bv_len);
1068 iov_addr = iov[iov_idx].iov_base + iov_off;
1072 ret = copy_to_user(iov_addr, bv_addr,
1076 ret = copy_from_user(bv_addr, iov_addr,
1088 if (iov[iov_idx].iov_len == iov_off) {
1095 __free_page(bvec->bv_page);
1102 * bio_uncopy_user - finish previously mapped bio
1103 * @bio: bio being terminated
1105 * Free pages allocated from bio_copy_user_iov() and write back data
1106 * to user space in case of a read.
1108 int bio_uncopy_user(struct bio *bio)
1110 struct bio_map_data *bmd = bio->bi_private;
1111 struct bio_vec *bvec;
1114 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1116 * if we're in a workqueue, the request is orphaned, so
1117 * don't copy into a random user address space, just free.
1120 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs,
1121 bio_data_dir(bio) == READ,
1122 0, bmd->is_our_pages);
1123 else if (bmd->is_our_pages)
1124 bio_for_each_segment_all(bvec, bio, i)
1125 __free_page(bvec->bv_page);
1131 EXPORT_SYMBOL(bio_uncopy_user);
1134 * bio_copy_user_iov - copy user data to bio
1135 * @q: destination block queue
1136 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1138 * @iov_count: number of elements in the iovec
1139 * @write_to_vm: bool indicating writing to pages or not
1140 * @gfp_mask: memory allocation flags
1142 * Prepares and returns a bio for indirect user io, bouncing data
1143 * to/from kernel pages as necessary. Must be paired with
1144 * call bio_uncopy_user() on io completion.
1146 struct bio *bio_copy_user_iov(struct request_queue *q,
1147 struct rq_map_data *map_data,
1148 const struct sg_iovec *iov, int iov_count,
1149 int write_to_vm, gfp_t gfp_mask)
1151 struct bio_map_data *bmd;
1152 struct bio_vec *bvec;
1157 unsigned int len = 0;
1158 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
1160 for (i = 0; i < iov_count; i++) {
1161 unsigned long uaddr;
1163 unsigned long start;
1165 uaddr = (unsigned long)iov[i].iov_base;
1166 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1167 start = uaddr >> PAGE_SHIFT;
1173 return ERR_PTR(-EINVAL);
1175 nr_pages += end - start;
1176 len += iov[i].iov_len;
1182 bmd = bio_alloc_map_data(iov_count, gfp_mask);
1184 return ERR_PTR(-ENOMEM);
1187 bio = bio_kmalloc(gfp_mask, nr_pages);
1192 bio->bi_rw |= REQ_WRITE;
1197 nr_pages = 1 << map_data->page_order;
1198 i = map_data->offset / PAGE_SIZE;
1201 unsigned int bytes = PAGE_SIZE;
1209 if (i == map_data->nr_entries * nr_pages) {
1214 page = map_data->pages[i / nr_pages];
1215 page += (i % nr_pages);
1219 page = alloc_page(q->bounce_gfp | gfp_mask);
1226 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1239 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
1240 (map_data && map_data->from_user)) {
1241 ret = __bio_copy_iov(bio, iov, iov_count, 0, 1, 0);
1246 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
1250 bio_for_each_segment_all(bvec, bio, i)
1251 __free_page(bvec->bv_page);
1256 return ERR_PTR(ret);
1259 static struct bio *__bio_map_user_iov(struct request_queue *q,
1260 struct block_device *bdev,
1261 const struct sg_iovec *iov, int iov_count,
1262 int write_to_vm, gfp_t gfp_mask)
1266 struct page **pages;
1271 for (i = 0; i < iov_count; i++) {
1272 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1273 unsigned long len = iov[i].iov_len;
1274 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1275 unsigned long start = uaddr >> PAGE_SHIFT;
1281 return ERR_PTR(-EINVAL);
1283 nr_pages += end - start;
1285 * buffer must be aligned to at least hardsector size for now
1287 if (uaddr & queue_dma_alignment(q))
1288 return ERR_PTR(-EINVAL);
1292 return ERR_PTR(-EINVAL);
1294 bio = bio_kmalloc(gfp_mask, nr_pages);
1296 return ERR_PTR(-ENOMEM);
1299 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1303 for (i = 0; i < iov_count; i++) {
1304 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1305 unsigned long len = iov[i].iov_len;
1306 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1307 unsigned long start = uaddr >> PAGE_SHIFT;
1308 const int local_nr_pages = end - start;
1309 const int page_limit = cur_page + local_nr_pages;
1311 ret = get_user_pages_fast(uaddr, local_nr_pages,
1312 write_to_vm, &pages[cur_page]);
1313 if (ret < local_nr_pages) {
1318 offset = uaddr & ~PAGE_MASK;
1319 for (j = cur_page; j < page_limit; j++) {
1320 unsigned int bytes = PAGE_SIZE - offset;
1331 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1341 * release the pages we didn't map into the bio, if any
1343 while (j < page_limit)
1344 page_cache_release(pages[j++]);
1350 * set data direction, and check if mapped pages need bouncing
1353 bio->bi_rw |= REQ_WRITE;
1355 bio->bi_bdev = bdev;
1356 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1360 for (i = 0; i < nr_pages; i++) {
1363 page_cache_release(pages[i]);
1368 return ERR_PTR(ret);
1372 * bio_map_user_iov - map user sg_iovec table into bio
1373 * @q: the struct request_queue for the bio
1374 * @bdev: destination block device
1376 * @iov_count: number of elements in the iovec
1377 * @write_to_vm: bool indicating writing to pages or not
1378 * @gfp_mask: memory allocation flags
1380 * Map the user space address into a bio suitable for io to a block
1381 * device. Returns an error pointer in case of error.
1383 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1384 const struct sg_iovec *iov, int iov_count,
1385 int write_to_vm, gfp_t gfp_mask)
1389 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1395 * subtle -- if __bio_map_user() ended up bouncing a bio,
1396 * it would normally disappear when its bi_end_io is run.
1397 * however, we need it for the unmap, so grab an extra
1405 static void __bio_unmap_user(struct bio *bio)
1407 struct bio_vec *bvec;
1411 * make sure we dirty pages we wrote to
1413 bio_for_each_segment_all(bvec, bio, i) {
1414 if (bio_data_dir(bio) == READ)
1415 set_page_dirty_lock(bvec->bv_page);
1417 page_cache_release(bvec->bv_page);
1424 * bio_unmap_user - unmap a bio
1425 * @bio: the bio being unmapped
1427 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1428 * a process context.
1430 * bio_unmap_user() may sleep.
1432 void bio_unmap_user(struct bio *bio)
1434 __bio_unmap_user(bio);
1437 EXPORT_SYMBOL(bio_unmap_user);
1439 static void bio_map_kern_endio(struct bio *bio, int err)
1444 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1445 unsigned int len, gfp_t gfp_mask)
1447 unsigned long kaddr = (unsigned long)data;
1448 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1449 unsigned long start = kaddr >> PAGE_SHIFT;
1450 const int nr_pages = end - start;
1454 bio = bio_kmalloc(gfp_mask, nr_pages);
1456 return ERR_PTR(-ENOMEM);
1458 offset = offset_in_page(kaddr);
1459 for (i = 0; i < nr_pages; i++) {
1460 unsigned int bytes = PAGE_SIZE - offset;
1468 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1477 bio->bi_end_io = bio_map_kern_endio;
1482 * bio_map_kern - map kernel address into bio
1483 * @q: the struct request_queue for the bio
1484 * @data: pointer to buffer to map
1485 * @len: length in bytes
1486 * @gfp_mask: allocation flags for bio allocation
1488 * Map the kernel address into a bio suitable for io to a block
1489 * device. Returns an error pointer in case of error.
1491 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1496 bio = __bio_map_kern(q, data, len, gfp_mask);
1500 if (bio->bi_iter.bi_size == len)
1504 * Don't support partial mappings.
1507 return ERR_PTR(-EINVAL);
1509 EXPORT_SYMBOL(bio_map_kern);
1511 static void bio_copy_kern_endio(struct bio *bio, int err)
1513 struct bio_vec *bvec;
1514 const int read = bio_data_dir(bio) == READ;
1515 char *p = bio->bi_private;
1518 bio_for_each_segment_all(bvec, bio, i) {
1519 char *addr = page_address(bvec->bv_page);
1522 memcpy(p, addr, bvec->bv_len);
1524 __free_page(bvec->bv_page);
1532 * bio_copy_kern - copy kernel address into bio
1533 * @q: the struct request_queue for the bio
1534 * @data: pointer to buffer to copy
1535 * @len: length in bytes
1536 * @gfp_mask: allocation flags for bio and page allocation
1537 * @reading: data direction is READ
1539 * copy the kernel address into a bio suitable for io to a block
1540 * device. Returns an error pointer in case of error.
1542 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1543 gfp_t gfp_mask, int reading)
1545 unsigned long kaddr = (unsigned long)data;
1546 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1547 unsigned long start = kaddr >> PAGE_SHIFT;
1548 struct bio_vec *bvec;
1551 int nr_pages = 0, i;
1557 return ERR_PTR(-EINVAL);
1559 nr_pages = end - start;
1560 bio = bio_kmalloc(gfp_mask, nr_pages);
1562 return ERR_PTR(-ENOMEM);
1566 unsigned int bytes = PAGE_SIZE;
1571 page = alloc_page(q->bounce_gfp | gfp_mask);
1576 memcpy(page_address(page), p, bytes);
1578 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1586 bio->bi_rw |= REQ_WRITE;
1588 bio->bi_private = data;
1589 bio->bi_end_io = bio_copy_kern_endio;
1593 bio_for_each_segment_all(bvec, bio, i)
1594 __free_page(bvec->bv_page);
1596 return ERR_PTR(-ENOMEM);
1598 EXPORT_SYMBOL(bio_copy_kern);
1601 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1602 * for performing direct-IO in BIOs.
1604 * The problem is that we cannot run set_page_dirty() from interrupt context
1605 * because the required locks are not interrupt-safe. So what we can do is to
1606 * mark the pages dirty _before_ performing IO. And in interrupt context,
1607 * check that the pages are still dirty. If so, fine. If not, redirty them
1608 * in process context.
1610 * We special-case compound pages here: normally this means reads into hugetlb
1611 * pages. The logic in here doesn't really work right for compound pages
1612 * because the VM does not uniformly chase down the head page in all cases.
1613 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1614 * handle them at all. So we skip compound pages here at an early stage.
1616 * Note that this code is very hard to test under normal circumstances because
1617 * direct-io pins the pages with get_user_pages(). This makes
1618 * is_page_cache_freeable return false, and the VM will not clean the pages.
1619 * But other code (eg, flusher threads) could clean the pages if they are mapped
1622 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1623 * deferred bio dirtying paths.
1627 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1629 void bio_set_pages_dirty(struct bio *bio)
1631 struct bio_vec *bvec;
1634 bio_for_each_segment_all(bvec, bio, i) {
1635 struct page *page = bvec->bv_page;
1637 if (page && !PageCompound(page))
1638 set_page_dirty_lock(page);
1642 static void bio_release_pages(struct bio *bio)
1644 struct bio_vec *bvec;
1647 bio_for_each_segment_all(bvec, bio, i) {
1648 struct page *page = bvec->bv_page;
1656 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1657 * If they are, then fine. If, however, some pages are clean then they must
1658 * have been written out during the direct-IO read. So we take another ref on
1659 * the BIO and the offending pages and re-dirty the pages in process context.
1661 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1662 * here on. It will run one page_cache_release() against each page and will
1663 * run one bio_put() against the BIO.
1666 static void bio_dirty_fn(struct work_struct *work);
1668 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1669 static DEFINE_SPINLOCK(bio_dirty_lock);
1670 static struct bio *bio_dirty_list;
1673 * This runs in process context
1675 static void bio_dirty_fn(struct work_struct *work)
1677 unsigned long flags;
1680 spin_lock_irqsave(&bio_dirty_lock, flags);
1681 bio = bio_dirty_list;
1682 bio_dirty_list = NULL;
1683 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1686 struct bio *next = bio->bi_private;
1688 bio_set_pages_dirty(bio);
1689 bio_release_pages(bio);
1695 void bio_check_pages_dirty(struct bio *bio)
1697 struct bio_vec *bvec;
1698 int nr_clean_pages = 0;
1701 bio_for_each_segment_all(bvec, bio, i) {
1702 struct page *page = bvec->bv_page;
1704 if (PageDirty(page) || PageCompound(page)) {
1705 page_cache_release(page);
1706 bvec->bv_page = NULL;
1712 if (nr_clean_pages) {
1713 unsigned long flags;
1715 spin_lock_irqsave(&bio_dirty_lock, flags);
1716 bio->bi_private = bio_dirty_list;
1717 bio_dirty_list = bio;
1718 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1719 schedule_work(&bio_dirty_work);
1725 void generic_start_io_acct(int rw, unsigned long sectors,
1726 struct hd_struct *part)
1728 int cpu = part_stat_lock();
1730 part_round_stats(cpu, part);
1731 part_stat_inc(cpu, part, ios[rw]);
1732 part_stat_add(cpu, part, sectors[rw], sectors);
1733 part_inc_in_flight(part, rw);
1737 EXPORT_SYMBOL(generic_start_io_acct);
1739 void generic_end_io_acct(int rw, struct hd_struct *part,
1740 unsigned long start_time)
1742 unsigned long duration = jiffies - start_time;
1743 int cpu = part_stat_lock();
1745 part_stat_add(cpu, part, ticks[rw], duration);
1746 part_round_stats(cpu, part);
1747 part_dec_in_flight(part, rw);
1751 EXPORT_SYMBOL(generic_end_io_acct);
1753 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1754 void bio_flush_dcache_pages(struct bio *bi)
1756 struct bio_vec bvec;
1757 struct bvec_iter iter;
1759 bio_for_each_segment(bvec, bi, iter)
1760 flush_dcache_page(bvec.bv_page);
1762 EXPORT_SYMBOL(bio_flush_dcache_pages);
1766 * bio_endio - end I/O on a bio
1768 * @error: error, if any
1771 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1772 * preferred way to end I/O on a bio, it takes care of clearing
1773 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1774 * established -Exxxx (-EIO, for instance) error values in case
1775 * something went wrong. No one should call bi_end_io() directly on a
1776 * bio unless they own it and thus know that it has an end_io
1779 void bio_endio(struct bio *bio, int error)
1782 BUG_ON(atomic_read(&bio->bi_remaining) <= 0);
1785 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1786 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1789 if (!atomic_dec_and_test(&bio->bi_remaining))
1793 * Need to have a real endio function for chained bios,
1794 * otherwise various corner cases will break (like stacking
1795 * block devices that save/restore bi_end_io) - however, we want
1796 * to avoid unbounded recursion and blowing the stack. Tail call
1797 * optimization would handle this, but compiling with frame
1798 * pointers also disables gcc's sibling call optimization.
1800 if (bio->bi_end_io == bio_chain_endio) {
1801 struct bio *parent = bio->bi_private;
1806 bio->bi_end_io(bio, error);
1811 EXPORT_SYMBOL(bio_endio);
1814 * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining
1816 * @error: error, if any
1818 * For code that has saved and restored bi_end_io; thing hard before using this
1819 * function, probably you should've cloned the entire bio.
1821 void bio_endio_nodec(struct bio *bio, int error)
1823 atomic_inc(&bio->bi_remaining);
1824 bio_endio(bio, error);
1826 EXPORT_SYMBOL(bio_endio_nodec);
1829 * bio_split - split a bio
1830 * @bio: bio to split
1831 * @sectors: number of sectors to split from the front of @bio
1833 * @bs: bio set to allocate from
1835 * Allocates and returns a new bio which represents @sectors from the start of
1836 * @bio, and updates @bio to represent the remaining sectors.
1838 * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's
1839 * responsibility to ensure that @bio is not freed before the split.
1841 struct bio *bio_split(struct bio *bio, int sectors,
1842 gfp_t gfp, struct bio_set *bs)
1844 struct bio *split = NULL;
1846 BUG_ON(sectors <= 0);
1847 BUG_ON(sectors >= bio_sectors(bio));
1849 split = bio_clone_fast(bio, gfp, bs);
1853 split->bi_iter.bi_size = sectors << 9;
1855 if (bio_integrity(split))
1856 bio_integrity_trim(split, 0, sectors);
1858 bio_advance(bio, split->bi_iter.bi_size);
1862 EXPORT_SYMBOL(bio_split);
1865 * bio_trim - trim a bio
1867 * @offset: number of sectors to trim from the front of @bio
1868 * @size: size we want to trim @bio to, in sectors
1870 void bio_trim(struct bio *bio, int offset, int size)
1872 /* 'bio' is a cloned bio which we need to trim to match
1873 * the given offset and size.
1877 if (offset == 0 && size == bio->bi_iter.bi_size)
1880 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1882 bio_advance(bio, offset << 9);
1884 bio->bi_iter.bi_size = size;
1886 EXPORT_SYMBOL_GPL(bio_trim);
1889 * create memory pools for biovec's in a bio_set.
1890 * use the global biovec slabs created for general use.
1892 mempool_t *biovec_create_pool(int pool_entries)
1894 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1896 return mempool_create_slab_pool(pool_entries, bp->slab);
1899 void bioset_free(struct bio_set *bs)
1901 if (bs->rescue_workqueue)
1902 destroy_workqueue(bs->rescue_workqueue);
1905 mempool_destroy(bs->bio_pool);
1908 mempool_destroy(bs->bvec_pool);
1910 bioset_integrity_free(bs);
1915 EXPORT_SYMBOL(bioset_free);
1917 static struct bio_set *__bioset_create(unsigned int pool_size,
1918 unsigned int front_pad,
1919 bool create_bvec_pool)
1921 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1924 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1928 bs->front_pad = front_pad;
1930 spin_lock_init(&bs->rescue_lock);
1931 bio_list_init(&bs->rescue_list);
1932 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1934 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1935 if (!bs->bio_slab) {
1940 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1944 if (create_bvec_pool) {
1945 bs->bvec_pool = biovec_create_pool(pool_size);
1950 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1951 if (!bs->rescue_workqueue)
1961 * bioset_create - Create a bio_set
1962 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1963 * @front_pad: Number of bytes to allocate in front of the returned bio
1966 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1967 * to ask for a number of bytes to be allocated in front of the bio.
1968 * Front pad allocation is useful for embedding the bio inside
1969 * another structure, to avoid allocating extra data to go with the bio.
1970 * Note that the bio must be embedded at the END of that structure always,
1971 * or things will break badly.
1973 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1975 return __bioset_create(pool_size, front_pad, true);
1977 EXPORT_SYMBOL(bioset_create);
1980 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1981 * @pool_size: Number of bio to cache in the mempool
1982 * @front_pad: Number of bytes to allocate in front of the returned bio
1985 * Same functionality as bioset_create() except that mempool is not
1986 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1988 struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1990 return __bioset_create(pool_size, front_pad, false);
1992 EXPORT_SYMBOL(bioset_create_nobvec);
1994 #ifdef CONFIG_BLK_CGROUP
1996 * bio_associate_current - associate a bio with %current
1999 * Associate @bio with %current if it hasn't been associated yet. Block
2000 * layer will treat @bio as if it were issued by %current no matter which
2001 * task actually issues it.
2003 * This function takes an extra reference of @task's io_context and blkcg
2004 * which will be put when @bio is released. The caller must own @bio,
2005 * ensure %current->io_context exists, and is responsible for synchronizing
2006 * calls to this function.
2008 int bio_associate_current(struct bio *bio)
2010 struct io_context *ioc;
2011 struct cgroup_subsys_state *css;
2016 ioc = current->io_context;
2020 /* acquire active ref on @ioc and associate */
2021 get_io_context_active(ioc);
2024 /* associate blkcg if exists */
2026 css = task_css(current, blkio_cgrp_id);
2027 if (css && css_tryget_online(css))
2035 * bio_disassociate_task - undo bio_associate_current()
2038 void bio_disassociate_task(struct bio *bio)
2041 put_io_context(bio->bi_ioc);
2045 css_put(bio->bi_css);
2050 #endif /* CONFIG_BLK_CGROUP */
2052 static void __init biovec_init_slabs(void)
2056 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2058 struct biovec_slab *bvs = bvec_slabs + i;
2060 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2065 size = bvs->nr_vecs * sizeof(struct bio_vec);
2066 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2067 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2071 static int __init init_bio(void)
2075 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2077 panic("bio: can't allocate bios\n");
2079 bio_integrity_init();
2080 biovec_init_slabs();
2082 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2084 panic("bio: can't allocate bios\n");
2086 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2087 panic("bio: can't create integrity pool\n");
2091 subsys_initcall(init_bio);