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>
32 #include <trace/events/block.h>
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
38 #define BIO_INLINE_VECS 4
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
45 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
47 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
55 struct bio_set *fs_bio_set;
56 EXPORT_SYMBOL(fs_bio_set);
59 * Our slab pool management
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
67 static DEFINE_MUTEX(bio_slab_lock);
68 static struct bio_slab *bio_slabs;
69 static unsigned int bio_slab_nr, bio_slab_max;
71 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
75 struct bio_slab *bslab, *new_bio_slabs;
76 unsigned int new_bio_slab_max;
77 unsigned int i, entry = -1;
79 mutex_lock(&bio_slab_lock);
82 while (i < bio_slab_nr) {
83 bslab = &bio_slabs[i];
85 if (!bslab->slab && entry == -1)
87 else if (bslab->slab_size == sz) {
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
99 new_bio_slab_max = bio_slab_max << 1;
100 new_bio_slabs = krealloc(bio_slabs,
101 new_bio_slab_max * sizeof(struct bio_slab),
105 bio_slab_max = new_bio_slab_max;
106 bio_slabs = new_bio_slabs;
109 entry = bio_slab_nr++;
111 bslab = &bio_slabs[entry];
113 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
114 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
115 SLAB_HWCACHE_ALIGN, NULL);
121 bslab->slab_size = sz;
123 mutex_unlock(&bio_slab_lock);
127 static void bio_put_slab(struct bio_set *bs)
129 struct bio_slab *bslab = NULL;
132 mutex_lock(&bio_slab_lock);
134 for (i = 0; i < bio_slab_nr; i++) {
135 if (bs->bio_slab == bio_slabs[i].slab) {
136 bslab = &bio_slabs[i];
141 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
144 WARN_ON(!bslab->slab_ref);
146 if (--bslab->slab_ref)
149 kmem_cache_destroy(bslab->slab);
153 mutex_unlock(&bio_slab_lock);
156 unsigned int bvec_nr_vecs(unsigned short idx)
158 return bvec_slabs[idx].nr_vecs;
161 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
163 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
165 if (idx == BIOVEC_MAX_IDX)
166 mempool_free(bv, pool);
168 struct biovec_slab *bvs = bvec_slabs + idx;
170 kmem_cache_free(bvs->slab, bv);
174 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
180 * see comment near bvec_array define!
198 case 129 ... BIO_MAX_PAGES:
206 * idx now points to the pool we want to allocate from. only the
207 * 1-vec entry pool is mempool backed.
209 if (*idx == BIOVEC_MAX_IDX) {
211 bvl = mempool_alloc(pool, gfp_mask);
213 struct biovec_slab *bvs = bvec_slabs + *idx;
214 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
217 * Make this allocation restricted and don't dump info on
218 * allocation failures, since we'll fallback to the mempool
219 * in case of failure.
221 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
224 * Try a slab allocation. If this fails and __GFP_WAIT
225 * is set, retry with the 1-entry mempool
227 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
228 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
229 *idx = BIOVEC_MAX_IDX;
237 static void __bio_free(struct bio *bio)
239 bio_disassociate_task(bio);
241 if (bio_integrity(bio))
242 bio_integrity_free(bio);
245 static void bio_free(struct bio *bio)
247 struct bio_set *bs = bio->bi_pool;
253 if (bio_flagged(bio, BIO_OWNS_VEC))
254 bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
257 * If we have front padding, adjust the bio pointer before freeing
262 mempool_free(p, bs->bio_pool);
264 /* Bio was allocated by bio_kmalloc() */
269 void bio_init(struct bio *bio)
271 memset(bio, 0, sizeof(*bio));
272 atomic_set(&bio->__bi_remaining, 1);
273 atomic_set(&bio->__bi_cnt, 1);
275 EXPORT_SYMBOL(bio_init);
278 * bio_reset - reinitialize a bio
282 * After calling bio_reset(), @bio will be in the same state as a freshly
283 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
284 * preserved are the ones that are initialized by bio_alloc_bioset(). See
285 * comment in struct bio.
287 void bio_reset(struct bio *bio)
289 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
293 memset(bio, 0, BIO_RESET_BYTES);
294 bio->bi_flags = flags;
295 atomic_set(&bio->__bi_remaining, 1);
297 EXPORT_SYMBOL(bio_reset);
299 static void bio_chain_endio(struct bio *bio)
301 struct bio *parent = bio->bi_private;
303 parent->bi_error = bio->bi_error;
309 * Increment chain count for the bio. Make sure the CHAIN flag update
310 * is visible before the raised count.
312 static inline void bio_inc_remaining(struct bio *bio)
314 bio->bi_flags |= (1 << BIO_CHAIN);
315 smp_mb__before_atomic();
316 atomic_inc(&bio->__bi_remaining);
320 * bio_chain - chain bio completions
321 * @bio: the target bio
322 * @parent: the @bio's parent 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 generic_make_request(bio);
357 static void punt_bios_to_rescuer(struct bio_set *bs)
359 struct bio_list punt, nopunt;
363 * In order to guarantee forward progress we must punt only bios that
364 * were allocated from this bio_set; otherwise, if there was a bio on
365 * there for a stacking driver higher up in the stack, processing it
366 * could require allocating bios from this bio_set, and doing that from
367 * our own rescuer would be bad.
369 * Since bio lists are singly linked, pop them all instead of trying to
370 * remove from the middle of the list:
373 bio_list_init(&punt);
374 bio_list_init(&nopunt);
376 while ((bio = bio_list_pop(current->bio_list)))
377 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
379 *current->bio_list = nopunt;
381 spin_lock(&bs->rescue_lock);
382 bio_list_merge(&bs->rescue_list, &punt);
383 spin_unlock(&bs->rescue_lock);
385 queue_work(bs->rescue_workqueue, &bs->rescue_work);
389 * bio_alloc_bioset - allocate a bio for I/O
390 * @gfp_mask: the GFP_ mask given to the slab allocator
391 * @nr_iovecs: number of iovecs to pre-allocate
392 * @bs: the bio_set to allocate from.
395 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
396 * backed by the @bs's mempool.
398 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
399 * able to allocate a bio. This is due to the mempool guarantees. To make this
400 * work, callers must never allocate more than 1 bio at a time from this pool.
401 * Callers that need to allocate more than 1 bio must always submit the
402 * previously allocated bio for IO before attempting to allocate a new one.
403 * Failure to do so can cause deadlocks under memory pressure.
405 * Note that when running under generic_make_request() (i.e. any block
406 * driver), bios are not submitted until after you return - see the code in
407 * generic_make_request() that converts recursion into iteration, to prevent
410 * This would normally mean allocating multiple bios under
411 * generic_make_request() would be susceptible to deadlocks, but we have
412 * deadlock avoidance code that resubmits any blocked bios from a rescuer
415 * However, we do not guarantee forward progress for allocations from other
416 * mempools. Doing multiple allocations from the same mempool under
417 * generic_make_request() should be avoided - instead, use bio_set's front_pad
418 * for per bio allocations.
421 * Pointer to new bio on success, NULL on failure.
423 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
425 gfp_t saved_gfp = gfp_mask;
427 unsigned inline_vecs;
428 unsigned long idx = BIO_POOL_NONE;
429 struct bio_vec *bvl = NULL;
434 if (nr_iovecs > UIO_MAXIOV)
437 p = kmalloc(sizeof(struct bio) +
438 nr_iovecs * sizeof(struct bio_vec),
441 inline_vecs = nr_iovecs;
443 /* should not use nobvec bioset for nr_iovecs > 0 */
444 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
447 * generic_make_request() converts recursion to iteration; this
448 * means if we're running beneath it, any bios we allocate and
449 * submit will not be submitted (and thus freed) until after we
452 * This exposes us to a potential deadlock if we allocate
453 * multiple bios from the same bio_set() while running
454 * underneath generic_make_request(). If we were to allocate
455 * multiple bios (say a stacking block driver that was splitting
456 * bios), we would deadlock if we exhausted the mempool's
459 * We solve this, and guarantee forward progress, with a rescuer
460 * workqueue per bio_set. If we go to allocate and there are
461 * bios on current->bio_list, we first try the allocation
462 * without __GFP_WAIT; if that fails, we punt those bios we
463 * would be blocking to the rescuer workqueue before we retry
464 * with the original gfp_flags.
467 if (current->bio_list && !bio_list_empty(current->bio_list))
468 gfp_mask &= ~__GFP_WAIT;
470 p = mempool_alloc(bs->bio_pool, gfp_mask);
471 if (!p && gfp_mask != saved_gfp) {
472 punt_bios_to_rescuer(bs);
473 gfp_mask = saved_gfp;
474 p = mempool_alloc(bs->bio_pool, gfp_mask);
477 front_pad = bs->front_pad;
478 inline_vecs = BIO_INLINE_VECS;
487 if (nr_iovecs > inline_vecs) {
488 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
489 if (!bvl && gfp_mask != saved_gfp) {
490 punt_bios_to_rescuer(bs);
491 gfp_mask = saved_gfp;
492 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
498 bio->bi_flags |= 1 << BIO_OWNS_VEC;
499 } else if (nr_iovecs) {
500 bvl = bio->bi_inline_vecs;
504 bio->bi_flags |= idx << BIO_POOL_OFFSET;
505 bio->bi_max_vecs = nr_iovecs;
506 bio->bi_io_vec = bvl;
510 mempool_free(p, bs->bio_pool);
513 EXPORT_SYMBOL(bio_alloc_bioset);
515 void zero_fill_bio(struct bio *bio)
519 struct bvec_iter iter;
521 bio_for_each_segment(bv, bio, iter) {
522 char *data = bvec_kmap_irq(&bv, &flags);
523 memset(data, 0, bv.bv_len);
524 flush_dcache_page(bv.bv_page);
525 bvec_kunmap_irq(data, &flags);
528 EXPORT_SYMBOL(zero_fill_bio);
531 * bio_put - release a reference to a bio
532 * @bio: bio to release reference to
535 * Put a reference to a &struct bio, either one you have gotten with
536 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
538 void bio_put(struct bio *bio)
540 if (!bio_flagged(bio, BIO_REFFED))
543 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
548 if (atomic_dec_and_test(&bio->__bi_cnt))
552 EXPORT_SYMBOL(bio_put);
554 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
556 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
557 blk_recount_segments(q, bio);
559 return bio->bi_phys_segments;
561 EXPORT_SYMBOL(bio_phys_segments);
564 * __bio_clone_fast - clone a bio that shares the original bio's biovec
565 * @bio: destination bio
566 * @bio_src: bio to clone
568 * Clone a &bio. Caller will own the returned bio, but not
569 * the actual data it points to. Reference count of returned
572 * Caller must ensure that @bio_src is not freed before @bio.
574 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
576 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
579 * most users will be overriding ->bi_bdev with a new target,
580 * so we don't set nor calculate new physical/hw segment counts here
582 bio->bi_bdev = bio_src->bi_bdev;
583 bio->bi_flags |= 1 << BIO_CLONED;
584 bio->bi_rw = bio_src->bi_rw;
585 bio->bi_iter = bio_src->bi_iter;
586 bio->bi_io_vec = bio_src->bi_io_vec;
588 EXPORT_SYMBOL(__bio_clone_fast);
591 * bio_clone_fast - clone a bio that shares the original bio's biovec
593 * @gfp_mask: allocation priority
594 * @bs: bio_set to allocate from
596 * Like __bio_clone_fast, only also allocates the returned bio
598 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
602 b = bio_alloc_bioset(gfp_mask, 0, bs);
606 __bio_clone_fast(b, bio);
608 if (bio_integrity(bio)) {
611 ret = bio_integrity_clone(b, bio, gfp_mask);
621 EXPORT_SYMBOL(bio_clone_fast);
624 * bio_clone_bioset - clone a bio
625 * @bio_src: bio to clone
626 * @gfp_mask: allocation priority
627 * @bs: bio_set to allocate from
629 * Clone bio. Caller will own the returned bio, but not the actual data it
630 * points to. Reference count of returned bio will be one.
632 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
635 struct bvec_iter iter;
640 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
641 * bio_src->bi_io_vec to bio->bi_io_vec.
643 * We can't do that anymore, because:
645 * - The point of cloning the biovec is to produce a bio with a biovec
646 * the caller can modify: bi_idx and bi_bvec_done should be 0.
648 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
649 * we tried to clone the whole thing bio_alloc_bioset() would fail.
650 * But the clone should succeed as long as the number of biovecs we
651 * actually need to allocate is fewer than BIO_MAX_PAGES.
653 * - Lastly, bi_vcnt should not be looked at or relied upon by code
654 * that does not own the bio - reason being drivers don't use it for
655 * iterating over the biovec anymore, so expecting it to be kept up
656 * to date (i.e. for clones that share the parent biovec) is just
657 * asking for trouble and would force extra work on
658 * __bio_clone_fast() anyways.
661 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
665 bio->bi_bdev = bio_src->bi_bdev;
666 bio->bi_rw = bio_src->bi_rw;
667 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
668 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
670 if (bio->bi_rw & REQ_DISCARD)
671 goto integrity_clone;
673 if (bio->bi_rw & REQ_WRITE_SAME) {
674 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
675 goto integrity_clone;
678 bio_for_each_segment(bv, bio_src, iter)
679 bio->bi_io_vec[bio->bi_vcnt++] = bv;
682 if (bio_integrity(bio_src)) {
685 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
694 EXPORT_SYMBOL(bio_clone_bioset);
697 * bio_get_nr_vecs - return approx number of vecs
700 * Return the approximate number of pages we can send to this target.
701 * There's no guarantee that you will be able to fit this number of pages
702 * into a bio, it does not account for dynamic restrictions that vary
705 int bio_get_nr_vecs(struct block_device *bdev)
707 struct request_queue *q = bdev_get_queue(bdev);
710 nr_pages = min_t(unsigned,
711 queue_max_segments(q),
712 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
714 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
717 EXPORT_SYMBOL(bio_get_nr_vecs);
719 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
720 *page, unsigned int len, unsigned int offset,
721 unsigned int max_sectors)
723 int retried_segments = 0;
724 struct bio_vec *bvec;
727 * cloned bio must not modify vec list
729 if (unlikely(bio_flagged(bio, BIO_CLONED)))
732 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
736 * For filesystems with a blocksize smaller than the pagesize
737 * we will often be called with the same page as last time and
738 * a consecutive offset. Optimize this special case.
740 if (bio->bi_vcnt > 0) {
741 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
743 if (page == prev->bv_page &&
744 offset == prev->bv_offset + prev->bv_len) {
745 unsigned int prev_bv_len = prev->bv_len;
748 if (q->merge_bvec_fn) {
749 struct bvec_merge_data bvm = {
750 /* prev_bvec is already charged in
751 bi_size, discharge it in order to
752 simulate merging updated prev_bvec
754 .bi_bdev = bio->bi_bdev,
755 .bi_sector = bio->bi_iter.bi_sector,
756 .bi_size = bio->bi_iter.bi_size -
761 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
767 bio->bi_iter.bi_size += len;
772 * If the queue doesn't support SG gaps and adding this
773 * offset would create a gap, disallow it.
775 if (q->queue_flags & (1 << QUEUE_FLAG_SG_GAPS) &&
776 bvec_gap_to_prev(prev, offset))
780 if (bio->bi_vcnt >= bio->bi_max_vecs)
784 * setup the new entry, we might clear it again later if we
785 * cannot add the page
787 bvec = &bio->bi_io_vec[bio->bi_vcnt];
788 bvec->bv_page = page;
790 bvec->bv_offset = offset;
792 bio->bi_phys_segments++;
793 bio->bi_iter.bi_size += len;
796 * Perform a recount if the number of segments is greater
797 * than queue_max_segments(q).
800 while (bio->bi_phys_segments > queue_max_segments(q)) {
802 if (retried_segments)
805 retried_segments = 1;
806 blk_recount_segments(q, bio);
810 * if queue has other restrictions (eg varying max sector size
811 * depending on offset), it can specify a merge_bvec_fn in the
812 * queue to get further control
814 if (q->merge_bvec_fn) {
815 struct bvec_merge_data bvm = {
816 .bi_bdev = bio->bi_bdev,
817 .bi_sector = bio->bi_iter.bi_sector,
818 .bi_size = bio->bi_iter.bi_size - len,
823 * merge_bvec_fn() returns number of bytes it can accept
826 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len)
830 /* If we may be able to merge these biovecs, force a recount */
831 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
832 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
838 bvec->bv_page = NULL;
842 bio->bi_iter.bi_size -= len;
843 blk_recount_segments(q, bio);
848 * bio_add_pc_page - attempt to add page to bio
849 * @q: the target queue
850 * @bio: destination bio
852 * @len: vec entry length
853 * @offset: vec entry offset
855 * Attempt to add a page to the bio_vec maplist. This can fail for a
856 * number of reasons, such as the bio being full or target block device
857 * limitations. The target block device must allow bio's up to PAGE_SIZE,
858 * so it is always possible to add a single page to an empty bio.
860 * This should only be used by REQ_PC bios.
862 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
863 unsigned int len, unsigned int offset)
865 return __bio_add_page(q, bio, page, len, offset,
866 queue_max_hw_sectors(q));
868 EXPORT_SYMBOL(bio_add_pc_page);
871 * bio_add_page - attempt to add page to bio
872 * @bio: destination bio
874 * @len: vec entry length
875 * @offset: vec entry offset
877 * Attempt to add a page to the bio_vec maplist. This can fail for a
878 * number of reasons, such as the bio being full or target block device
879 * limitations. The target block device must allow bio's up to PAGE_SIZE,
880 * so it is always possible to add a single page to an empty bio.
882 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
885 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
886 unsigned int max_sectors;
888 max_sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
889 if ((max_sectors < (len >> 9)) && !bio->bi_iter.bi_size)
890 max_sectors = len >> 9;
892 return __bio_add_page(q, bio, page, len, offset, max_sectors);
894 EXPORT_SYMBOL(bio_add_page);
896 struct submit_bio_ret {
897 struct completion event;
901 static void submit_bio_wait_endio(struct bio *bio)
903 struct submit_bio_ret *ret = bio->bi_private;
905 ret->error = bio->bi_error;
906 complete(&ret->event);
910 * submit_bio_wait - submit a bio, and wait until it completes
911 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
912 * @bio: The &struct bio which describes the I/O
914 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
915 * bio_endio() on failure.
917 int submit_bio_wait(int rw, struct bio *bio)
919 struct submit_bio_ret ret;
922 init_completion(&ret.event);
923 bio->bi_private = &ret;
924 bio->bi_end_io = submit_bio_wait_endio;
926 wait_for_completion(&ret.event);
930 EXPORT_SYMBOL(submit_bio_wait);
933 * bio_advance - increment/complete a bio by some number of bytes
934 * @bio: bio to advance
935 * @bytes: number of bytes to complete
937 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
938 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
939 * be updated on the last bvec as well.
941 * @bio will then represent the remaining, uncompleted portion of the io.
943 void bio_advance(struct bio *bio, unsigned bytes)
945 if (bio_integrity(bio))
946 bio_integrity_advance(bio, bytes);
948 bio_advance_iter(bio, &bio->bi_iter, bytes);
950 EXPORT_SYMBOL(bio_advance);
953 * bio_alloc_pages - allocates a single page for each bvec in a bio
954 * @bio: bio to allocate pages for
955 * @gfp_mask: flags for allocation
957 * Allocates pages up to @bio->bi_vcnt.
959 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
962 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
967 bio_for_each_segment_all(bv, bio, i) {
968 bv->bv_page = alloc_page(gfp_mask);
970 while (--bv >= bio->bi_io_vec)
971 __free_page(bv->bv_page);
978 EXPORT_SYMBOL(bio_alloc_pages);
981 * bio_copy_data - copy contents of data buffers from one chain of bios to
983 * @src: source bio list
984 * @dst: destination bio list
986 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
987 * @src and @dst as linked lists of bios.
989 * Stops when it reaches the end of either @src or @dst - that is, copies
990 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
992 void bio_copy_data(struct bio *dst, struct bio *src)
994 struct bvec_iter src_iter, dst_iter;
995 struct bio_vec src_bv, dst_bv;
999 src_iter = src->bi_iter;
1000 dst_iter = dst->bi_iter;
1003 if (!src_iter.bi_size) {
1008 src_iter = src->bi_iter;
1011 if (!dst_iter.bi_size) {
1016 dst_iter = dst->bi_iter;
1019 src_bv = bio_iter_iovec(src, src_iter);
1020 dst_bv = bio_iter_iovec(dst, dst_iter);
1022 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1024 src_p = kmap_atomic(src_bv.bv_page);
1025 dst_p = kmap_atomic(dst_bv.bv_page);
1027 memcpy(dst_p + dst_bv.bv_offset,
1028 src_p + src_bv.bv_offset,
1031 kunmap_atomic(dst_p);
1032 kunmap_atomic(src_p);
1034 bio_advance_iter(src, &src_iter, bytes);
1035 bio_advance_iter(dst, &dst_iter, bytes);
1038 EXPORT_SYMBOL(bio_copy_data);
1040 struct bio_map_data {
1042 struct iov_iter iter;
1046 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
1049 if (iov_count > UIO_MAXIOV)
1052 return kmalloc(sizeof(struct bio_map_data) +
1053 sizeof(struct iovec) * iov_count, gfp_mask);
1057 * bio_copy_from_iter - copy all pages from iov_iter to bio
1058 * @bio: The &struct bio which describes the I/O as destination
1059 * @iter: iov_iter as source
1061 * Copy all pages from iov_iter to bio.
1062 * Returns 0 on success, or error on failure.
1064 static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
1067 struct bio_vec *bvec;
1069 bio_for_each_segment_all(bvec, bio, i) {
1072 ret = copy_page_from_iter(bvec->bv_page,
1077 if (!iov_iter_count(&iter))
1080 if (ret < bvec->bv_len)
1088 * bio_copy_to_iter - copy all pages from bio to iov_iter
1089 * @bio: The &struct bio which describes the I/O as source
1090 * @iter: iov_iter as destination
1092 * Copy all pages from bio to iov_iter.
1093 * Returns 0 on success, or error on failure.
1095 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1098 struct bio_vec *bvec;
1100 bio_for_each_segment_all(bvec, bio, i) {
1103 ret = copy_page_to_iter(bvec->bv_page,
1108 if (!iov_iter_count(&iter))
1111 if (ret < bvec->bv_len)
1118 static void bio_free_pages(struct bio *bio)
1120 struct bio_vec *bvec;
1123 bio_for_each_segment_all(bvec, bio, i)
1124 __free_page(bvec->bv_page);
1128 * bio_uncopy_user - finish previously mapped bio
1129 * @bio: bio being terminated
1131 * Free pages allocated from bio_copy_user_iov() and write back data
1132 * to user space in case of a read.
1134 int bio_uncopy_user(struct bio *bio)
1136 struct bio_map_data *bmd = bio->bi_private;
1139 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1141 * if we're in a workqueue, the request is orphaned, so
1142 * don't copy into a random user address space, just free.
1144 if (current->mm && bio_data_dir(bio) == READ)
1145 ret = bio_copy_to_iter(bio, bmd->iter);
1146 if (bmd->is_our_pages)
1147 bio_free_pages(bio);
1153 EXPORT_SYMBOL(bio_uncopy_user);
1156 * bio_copy_user_iov - copy user data to bio
1157 * @q: destination block queue
1158 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1159 * @iter: iovec iterator
1160 * @gfp_mask: memory allocation flags
1162 * Prepares and returns a bio for indirect user io, bouncing data
1163 * to/from kernel pages as necessary. Must be paired with
1164 * call bio_uncopy_user() on io completion.
1166 struct bio *bio_copy_user_iov(struct request_queue *q,
1167 struct rq_map_data *map_data,
1168 const struct iov_iter *iter,
1171 struct bio_map_data *bmd;
1176 unsigned int len = iter->count;
1177 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
1179 for (i = 0; i < iter->nr_segs; i++) {
1180 unsigned long uaddr;
1182 unsigned long start;
1184 uaddr = (unsigned long) iter->iov[i].iov_base;
1185 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1187 start = uaddr >> PAGE_SHIFT;
1193 return ERR_PTR(-EINVAL);
1195 nr_pages += end - start;
1201 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
1203 return ERR_PTR(-ENOMEM);
1206 * We need to do a deep copy of the iov_iter including the iovecs.
1207 * The caller provided iov might point to an on-stack or otherwise
1210 bmd->is_our_pages = map_data ? 0 : 1;
1211 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1212 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1213 iter->nr_segs, iter->count);
1216 bio = bio_kmalloc(gfp_mask, nr_pages);
1220 if (iter->type & WRITE)
1221 bio->bi_rw |= REQ_WRITE;
1226 nr_pages = 1 << map_data->page_order;
1227 i = map_data->offset / PAGE_SIZE;
1230 unsigned int bytes = PAGE_SIZE;
1238 if (i == map_data->nr_entries * nr_pages) {
1243 page = map_data->pages[i / nr_pages];
1244 page += (i % nr_pages);
1248 page = alloc_page(q->bounce_gfp | gfp_mask);
1255 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1268 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1269 (map_data && map_data->from_user)) {
1270 ret = bio_copy_from_iter(bio, *iter);
1275 bio->bi_private = bmd;
1279 bio_free_pages(bio);
1283 return ERR_PTR(ret);
1287 * bio_map_user_iov - map user iovec into bio
1288 * @q: the struct request_queue for the bio
1289 * @iter: iovec iterator
1290 * @gfp_mask: memory allocation flags
1292 * Map the user space address into a bio suitable for io to a block
1293 * device. Returns an error pointer in case of error.
1295 struct bio *bio_map_user_iov(struct request_queue *q,
1296 const struct iov_iter *iter,
1301 struct page **pages;
1308 iov_for_each(iov, i, *iter) {
1309 unsigned long uaddr = (unsigned long) iov.iov_base;
1310 unsigned long len = iov.iov_len;
1311 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1312 unsigned long start = uaddr >> PAGE_SHIFT;
1318 return ERR_PTR(-EINVAL);
1320 nr_pages += end - start;
1322 * buffer must be aligned to at least hardsector size for now
1324 if (uaddr & queue_dma_alignment(q))
1325 return ERR_PTR(-EINVAL);
1329 return ERR_PTR(-EINVAL);
1331 bio = bio_kmalloc(gfp_mask, nr_pages);
1333 return ERR_PTR(-ENOMEM);
1336 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1340 iov_for_each(iov, i, *iter) {
1341 unsigned long uaddr = (unsigned long) iov.iov_base;
1342 unsigned long len = iov.iov_len;
1343 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1344 unsigned long start = uaddr >> PAGE_SHIFT;
1345 const int local_nr_pages = end - start;
1346 const int page_limit = cur_page + local_nr_pages;
1348 ret = get_user_pages_fast(uaddr, local_nr_pages,
1349 (iter->type & WRITE) != WRITE,
1351 if (ret < local_nr_pages) {
1356 offset = uaddr & ~PAGE_MASK;
1357 for (j = cur_page; j < page_limit; j++) {
1358 unsigned int bytes = PAGE_SIZE - offset;
1369 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1379 * release the pages we didn't map into the bio, if any
1381 while (j < page_limit)
1382 page_cache_release(pages[j++]);
1388 * set data direction, and check if mapped pages need bouncing
1390 if (iter->type & WRITE)
1391 bio->bi_rw |= REQ_WRITE;
1393 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1396 * subtle -- if __bio_map_user() ended up bouncing a bio,
1397 * it would normally disappear when its bi_end_io is run.
1398 * however, we need it for the unmap, so grab an extra
1405 for (j = 0; j < nr_pages; j++) {
1408 page_cache_release(pages[j]);
1413 return ERR_PTR(ret);
1416 static void __bio_unmap_user(struct bio *bio)
1418 struct bio_vec *bvec;
1422 * make sure we dirty pages we wrote to
1424 bio_for_each_segment_all(bvec, bio, i) {
1425 if (bio_data_dir(bio) == READ)
1426 set_page_dirty_lock(bvec->bv_page);
1428 page_cache_release(bvec->bv_page);
1435 * bio_unmap_user - unmap a bio
1436 * @bio: the bio being unmapped
1438 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1439 * a process context.
1441 * bio_unmap_user() may sleep.
1443 void bio_unmap_user(struct bio *bio)
1445 __bio_unmap_user(bio);
1448 EXPORT_SYMBOL(bio_unmap_user);
1450 static void bio_map_kern_endio(struct bio *bio)
1456 * bio_map_kern - map kernel address into bio
1457 * @q: the struct request_queue for the bio
1458 * @data: pointer to buffer to map
1459 * @len: length in bytes
1460 * @gfp_mask: allocation flags for bio allocation
1462 * Map the kernel address into a bio suitable for io to a block
1463 * device. Returns an error pointer in case of error.
1465 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1468 unsigned long kaddr = (unsigned long)data;
1469 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1470 unsigned long start = kaddr >> PAGE_SHIFT;
1471 const int nr_pages = end - start;
1475 bio = bio_kmalloc(gfp_mask, nr_pages);
1477 return ERR_PTR(-ENOMEM);
1479 offset = offset_in_page(kaddr);
1480 for (i = 0; i < nr_pages; i++) {
1481 unsigned int bytes = PAGE_SIZE - offset;
1489 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1491 /* we don't support partial mappings */
1493 return ERR_PTR(-EINVAL);
1501 bio->bi_end_io = bio_map_kern_endio;
1504 EXPORT_SYMBOL(bio_map_kern);
1506 static void bio_copy_kern_endio(struct bio *bio)
1508 bio_free_pages(bio);
1512 static void bio_copy_kern_endio_read(struct bio *bio)
1514 char *p = bio->bi_private;
1515 struct bio_vec *bvec;
1518 bio_for_each_segment_all(bvec, bio, i) {
1519 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1523 bio_copy_kern_endio(bio);
1527 * bio_copy_kern - copy kernel address into bio
1528 * @q: the struct request_queue for the bio
1529 * @data: pointer to buffer to copy
1530 * @len: length in bytes
1531 * @gfp_mask: allocation flags for bio and page allocation
1532 * @reading: data direction is READ
1534 * copy the kernel address into a bio suitable for io to a block
1535 * device. Returns an error pointer in case of error.
1537 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1538 gfp_t gfp_mask, int reading)
1540 unsigned long kaddr = (unsigned long)data;
1541 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1542 unsigned long start = kaddr >> PAGE_SHIFT;
1551 return ERR_PTR(-EINVAL);
1553 nr_pages = end - start;
1554 bio = bio_kmalloc(gfp_mask, nr_pages);
1556 return ERR_PTR(-ENOMEM);
1560 unsigned int bytes = PAGE_SIZE;
1565 page = alloc_page(q->bounce_gfp | gfp_mask);
1570 memcpy(page_address(page), p, bytes);
1572 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1580 bio->bi_end_io = bio_copy_kern_endio_read;
1581 bio->bi_private = data;
1583 bio->bi_end_io = bio_copy_kern_endio;
1584 bio->bi_rw |= REQ_WRITE;
1590 bio_free_pages(bio);
1592 return ERR_PTR(-ENOMEM);
1594 EXPORT_SYMBOL(bio_copy_kern);
1597 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1598 * for performing direct-IO in BIOs.
1600 * The problem is that we cannot run set_page_dirty() from interrupt context
1601 * because the required locks are not interrupt-safe. So what we can do is to
1602 * mark the pages dirty _before_ performing IO. And in interrupt context,
1603 * check that the pages are still dirty. If so, fine. If not, redirty them
1604 * in process context.
1606 * We special-case compound pages here: normally this means reads into hugetlb
1607 * pages. The logic in here doesn't really work right for compound pages
1608 * because the VM does not uniformly chase down the head page in all cases.
1609 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1610 * handle them at all. So we skip compound pages here at an early stage.
1612 * Note that this code is very hard to test under normal circumstances because
1613 * direct-io pins the pages with get_user_pages(). This makes
1614 * is_page_cache_freeable return false, and the VM will not clean the pages.
1615 * But other code (eg, flusher threads) could clean the pages if they are mapped
1618 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1619 * deferred bio dirtying paths.
1623 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1625 void bio_set_pages_dirty(struct bio *bio)
1627 struct bio_vec *bvec;
1630 bio_for_each_segment_all(bvec, bio, i) {
1631 struct page *page = bvec->bv_page;
1633 if (page && !PageCompound(page))
1634 set_page_dirty_lock(page);
1638 static void bio_release_pages(struct bio *bio)
1640 struct bio_vec *bvec;
1643 bio_for_each_segment_all(bvec, bio, i) {
1644 struct page *page = bvec->bv_page;
1652 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1653 * If they are, then fine. If, however, some pages are clean then they must
1654 * have been written out during the direct-IO read. So we take another ref on
1655 * the BIO and the offending pages and re-dirty the pages in process context.
1657 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1658 * here on. It will run one page_cache_release() against each page and will
1659 * run one bio_put() against the BIO.
1662 static void bio_dirty_fn(struct work_struct *work);
1664 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1665 static DEFINE_SPINLOCK(bio_dirty_lock);
1666 static struct bio *bio_dirty_list;
1669 * This runs in process context
1671 static void bio_dirty_fn(struct work_struct *work)
1673 unsigned long flags;
1676 spin_lock_irqsave(&bio_dirty_lock, flags);
1677 bio = bio_dirty_list;
1678 bio_dirty_list = NULL;
1679 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1682 struct bio *next = bio->bi_private;
1684 bio_set_pages_dirty(bio);
1685 bio_release_pages(bio);
1691 void bio_check_pages_dirty(struct bio *bio)
1693 struct bio_vec *bvec;
1694 int nr_clean_pages = 0;
1697 bio_for_each_segment_all(bvec, bio, i) {
1698 struct page *page = bvec->bv_page;
1700 if (PageDirty(page) || PageCompound(page)) {
1701 page_cache_release(page);
1702 bvec->bv_page = NULL;
1708 if (nr_clean_pages) {
1709 unsigned long flags;
1711 spin_lock_irqsave(&bio_dirty_lock, flags);
1712 bio->bi_private = bio_dirty_list;
1713 bio_dirty_list = bio;
1714 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1715 schedule_work(&bio_dirty_work);
1721 void generic_start_io_acct(int rw, unsigned long sectors,
1722 struct hd_struct *part)
1724 int cpu = part_stat_lock();
1726 part_round_stats(cpu, part);
1727 part_stat_inc(cpu, part, ios[rw]);
1728 part_stat_add(cpu, part, sectors[rw], sectors);
1729 part_inc_in_flight(part, rw);
1733 EXPORT_SYMBOL(generic_start_io_acct);
1735 void generic_end_io_acct(int rw, struct hd_struct *part,
1736 unsigned long start_time)
1738 unsigned long duration = jiffies - start_time;
1739 int cpu = part_stat_lock();
1741 part_stat_add(cpu, part, ticks[rw], duration);
1742 part_round_stats(cpu, part);
1743 part_dec_in_flight(part, rw);
1747 EXPORT_SYMBOL(generic_end_io_acct);
1749 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1750 void bio_flush_dcache_pages(struct bio *bi)
1752 struct bio_vec bvec;
1753 struct bvec_iter iter;
1755 bio_for_each_segment(bvec, bi, iter)
1756 flush_dcache_page(bvec.bv_page);
1758 EXPORT_SYMBOL(bio_flush_dcache_pages);
1761 static inline bool bio_remaining_done(struct bio *bio)
1764 * If we're not chaining, then ->__bi_remaining is always 1 and
1765 * we always end io on the first invocation.
1767 if (!bio_flagged(bio, BIO_CHAIN))
1770 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1772 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1773 clear_bit(BIO_CHAIN, &bio->bi_flags);
1781 * bio_endio - end I/O on a bio
1785 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1786 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1787 * bio unless they own it and thus know that it has an end_io function.
1789 void bio_endio(struct bio *bio)
1792 if (unlikely(!bio_remaining_done(bio)))
1796 * Need to have a real endio function for chained bios,
1797 * otherwise various corner cases will break (like stacking
1798 * block devices that save/restore bi_end_io) - however, we want
1799 * to avoid unbounded recursion and blowing the stack. Tail call
1800 * optimization would handle this, but compiling with frame
1801 * pointers also disables gcc's sibling call optimization.
1803 if (bio->bi_end_io == bio_chain_endio) {
1804 struct bio *parent = bio->bi_private;
1805 parent->bi_error = bio->bi_error;
1810 bio->bi_end_io(bio);
1815 EXPORT_SYMBOL(bio_endio);
1818 * bio_split - split a bio
1819 * @bio: bio to split
1820 * @sectors: number of sectors to split from the front of @bio
1822 * @bs: bio set to allocate from
1824 * Allocates and returns a new bio which represents @sectors from the start of
1825 * @bio, and updates @bio to represent the remaining sectors.
1827 * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's
1828 * responsibility to ensure that @bio is not freed before the split.
1830 struct bio *bio_split(struct bio *bio, int sectors,
1831 gfp_t gfp, struct bio_set *bs)
1833 struct bio *split = NULL;
1835 BUG_ON(sectors <= 0);
1836 BUG_ON(sectors >= bio_sectors(bio));
1838 split = bio_clone_fast(bio, gfp, bs);
1842 split->bi_iter.bi_size = sectors << 9;
1844 if (bio_integrity(split))
1845 bio_integrity_trim(split, 0, sectors);
1847 bio_advance(bio, split->bi_iter.bi_size);
1851 EXPORT_SYMBOL(bio_split);
1854 * bio_trim - trim a bio
1856 * @offset: number of sectors to trim from the front of @bio
1857 * @size: size we want to trim @bio to, in sectors
1859 void bio_trim(struct bio *bio, int offset, int size)
1861 /* 'bio' is a cloned bio which we need to trim to match
1862 * the given offset and size.
1866 if (offset == 0 && size == bio->bi_iter.bi_size)
1869 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1871 bio_advance(bio, offset << 9);
1873 bio->bi_iter.bi_size = size;
1875 EXPORT_SYMBOL_GPL(bio_trim);
1878 * create memory pools for biovec's in a bio_set.
1879 * use the global biovec slabs created for general use.
1881 mempool_t *biovec_create_pool(int pool_entries)
1883 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1885 return mempool_create_slab_pool(pool_entries, bp->slab);
1888 void bioset_free(struct bio_set *bs)
1890 if (bs->rescue_workqueue)
1891 destroy_workqueue(bs->rescue_workqueue);
1894 mempool_destroy(bs->bio_pool);
1897 mempool_destroy(bs->bvec_pool);
1899 bioset_integrity_free(bs);
1904 EXPORT_SYMBOL(bioset_free);
1906 static struct bio_set *__bioset_create(unsigned int pool_size,
1907 unsigned int front_pad,
1908 bool create_bvec_pool)
1910 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1913 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1917 bs->front_pad = front_pad;
1919 spin_lock_init(&bs->rescue_lock);
1920 bio_list_init(&bs->rescue_list);
1921 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1923 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1924 if (!bs->bio_slab) {
1929 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1933 if (create_bvec_pool) {
1934 bs->bvec_pool = biovec_create_pool(pool_size);
1939 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1940 if (!bs->rescue_workqueue)
1950 * bioset_create - Create a bio_set
1951 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1952 * @front_pad: Number of bytes to allocate in front of the returned bio
1955 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1956 * to ask for a number of bytes to be allocated in front of the bio.
1957 * Front pad allocation is useful for embedding the bio inside
1958 * another structure, to avoid allocating extra data to go with the bio.
1959 * Note that the bio must be embedded at the END of that structure always,
1960 * or things will break badly.
1962 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1964 return __bioset_create(pool_size, front_pad, true);
1966 EXPORT_SYMBOL(bioset_create);
1969 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1970 * @pool_size: Number of bio to cache in the mempool
1971 * @front_pad: Number of bytes to allocate in front of the returned bio
1974 * Same functionality as bioset_create() except that mempool is not
1975 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1977 struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1979 return __bioset_create(pool_size, front_pad, false);
1981 EXPORT_SYMBOL(bioset_create_nobvec);
1983 #ifdef CONFIG_BLK_CGROUP
1986 * bio_associate_blkcg - associate a bio with the specified blkcg
1988 * @blkcg_css: css of the blkcg to associate
1990 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1991 * treat @bio as if it were issued by a task which belongs to the blkcg.
1993 * This function takes an extra reference of @blkcg_css which will be put
1994 * when @bio is released. The caller must own @bio and is responsible for
1995 * synchronizing calls to this function.
1997 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1999 if (unlikely(bio->bi_css))
2002 bio->bi_css = blkcg_css;
2007 * bio_associate_current - associate a bio with %current
2010 * Associate @bio with %current if it hasn't been associated yet. Block
2011 * layer will treat @bio as if it were issued by %current no matter which
2012 * task actually issues it.
2014 * This function takes an extra reference of @task's io_context and blkcg
2015 * which will be put when @bio is released. The caller must own @bio,
2016 * ensure %current->io_context exists, and is responsible for synchronizing
2017 * calls to this function.
2019 int bio_associate_current(struct bio *bio)
2021 struct io_context *ioc;
2026 ioc = current->io_context;
2030 get_io_context_active(ioc);
2032 bio->bi_css = task_get_css(current, blkio_cgrp_id);
2037 * bio_disassociate_task - undo bio_associate_current()
2040 void bio_disassociate_task(struct bio *bio)
2043 put_io_context(bio->bi_ioc);
2047 css_put(bio->bi_css);
2052 #endif /* CONFIG_BLK_CGROUP */
2054 static void __init biovec_init_slabs(void)
2058 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2060 struct biovec_slab *bvs = bvec_slabs + i;
2062 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2067 size = bvs->nr_vecs * sizeof(struct bio_vec);
2068 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2069 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2073 static int __init init_bio(void)
2077 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2079 panic("bio: can't allocate bios\n");
2081 bio_integrity_init();
2082 biovec_init_slabs();
2084 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2086 panic("bio: can't allocate bios\n");
2088 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2089 panic("bio: can't create integrity pool\n");
2093 subsys_initcall(init_bio);