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[BVEC_POOL_NR] __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)
167 BIO_BUG_ON(idx >= BVEC_POOL_NR);
169 if (idx == BVEC_POOL_MAX) {
170 mempool_free(bv, pool);
172 struct biovec_slab *bvs = bvec_slabs + idx;
174 kmem_cache_free(bvs->slab, bv);
178 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
184 * see comment near bvec_array define!
202 case 129 ... BIO_MAX_PAGES:
210 * idx now points to the pool we want to allocate from. only the
211 * 1-vec entry pool is mempool backed.
213 if (*idx == BVEC_POOL_MAX) {
215 bvl = mempool_alloc(pool, gfp_mask);
217 struct biovec_slab *bvs = bvec_slabs + *idx;
218 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
221 * Make this allocation restricted and don't dump info on
222 * allocation failures, since we'll fallback to the mempool
223 * in case of failure.
225 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
228 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
229 * is set, retry with the 1-entry mempool
231 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
232 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
233 *idx = BVEC_POOL_MAX;
242 static void __bio_free(struct bio *bio)
244 bio_disassociate_task(bio);
246 if (bio_integrity(bio))
247 bio_integrity_free(bio);
250 static void bio_free(struct bio *bio)
252 struct bio_set *bs = bio->bi_pool;
258 bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
261 * If we have front padding, adjust the bio pointer before freeing
266 mempool_free(p, bs->bio_pool);
268 /* Bio was allocated by bio_kmalloc() */
273 void bio_init(struct bio *bio, struct bio_vec *table,
274 unsigned short max_vecs)
276 memset(bio, 0, sizeof(*bio));
277 atomic_set(&bio->__bi_remaining, 1);
278 atomic_set(&bio->__bi_cnt, 1);
280 bio->bi_io_vec = table;
281 bio->bi_max_vecs = max_vecs;
283 EXPORT_SYMBOL(bio_init);
286 * bio_reset - reinitialize a bio
290 * After calling bio_reset(), @bio will be in the same state as a freshly
291 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
292 * preserved are the ones that are initialized by bio_alloc_bioset(). See
293 * comment in struct bio.
295 void bio_reset(struct bio *bio)
297 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
301 memset(bio, 0, BIO_RESET_BYTES);
302 bio->bi_flags = flags;
303 atomic_set(&bio->__bi_remaining, 1);
305 EXPORT_SYMBOL(bio_reset);
307 static struct bio *__bio_chain_endio(struct bio *bio)
309 struct bio *parent = bio->bi_private;
311 if (!parent->bi_error)
312 parent->bi_error = bio->bi_error;
317 static void bio_chain_endio(struct bio *bio)
319 bio_endio(__bio_chain_endio(bio));
323 * bio_chain - chain bio completions
324 * @bio: the target bio
325 * @parent: the @bio's parent bio
327 * The caller won't have a bi_end_io called when @bio completes - instead,
328 * @parent's bi_end_io won't be called until both @parent and @bio have
329 * completed; the chained bio will also be freed when it completes.
331 * The caller must not set bi_private or bi_end_io in @bio.
333 void bio_chain(struct bio *bio, struct bio *parent)
335 BUG_ON(bio->bi_private || bio->bi_end_io);
337 bio->bi_private = parent;
338 bio->bi_end_io = bio_chain_endio;
339 bio_inc_remaining(parent);
341 EXPORT_SYMBOL(bio_chain);
343 static void bio_alloc_rescue(struct work_struct *work)
345 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
349 spin_lock(&bs->rescue_lock);
350 bio = bio_list_pop(&bs->rescue_list);
351 spin_unlock(&bs->rescue_lock);
356 generic_make_request(bio);
360 static void punt_bios_to_rescuer(struct bio_set *bs)
362 struct bio_list punt, nopunt;
366 * In order to guarantee forward progress we must punt only bios that
367 * were allocated from this bio_set; otherwise, if there was a bio on
368 * there for a stacking driver higher up in the stack, processing it
369 * could require allocating bios from this bio_set, and doing that from
370 * our own rescuer would be bad.
372 * Since bio lists are singly linked, pop them all instead of trying to
373 * remove from the middle of the list:
376 bio_list_init(&punt);
377 bio_list_init(&nopunt);
379 while ((bio = bio_list_pop(¤t->bio_list[0])))
380 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
381 current->bio_list[0] = nopunt;
383 bio_list_init(&nopunt);
384 while ((bio = bio_list_pop(¤t->bio_list[1])))
385 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
386 current->bio_list[1] = nopunt;
388 spin_lock(&bs->rescue_lock);
389 bio_list_merge(&bs->rescue_list, &punt);
390 spin_unlock(&bs->rescue_lock);
392 queue_work(bs->rescue_workqueue, &bs->rescue_work);
396 * bio_alloc_bioset - allocate a bio for I/O
397 * @gfp_mask: the GFP_ mask given to the slab allocator
398 * @nr_iovecs: number of iovecs to pre-allocate
399 * @bs: the bio_set to allocate from.
402 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
403 * backed by the @bs's mempool.
405 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
406 * always be able to allocate a bio. This is due to the mempool guarantees.
407 * To make this work, callers must never allocate more than 1 bio at a time
408 * from this pool. Callers that need to allocate more than 1 bio must always
409 * submit the previously allocated bio for IO before attempting to allocate
410 * a new one. Failure to do so can cause deadlocks under memory pressure.
412 * Note that when running under generic_make_request() (i.e. any block
413 * driver), bios are not submitted until after you return - see the code in
414 * generic_make_request() that converts recursion into iteration, to prevent
417 * This would normally mean allocating multiple bios under
418 * generic_make_request() would be susceptible to deadlocks, but we have
419 * deadlock avoidance code that resubmits any blocked bios from a rescuer
422 * However, we do not guarantee forward progress for allocations from other
423 * mempools. Doing multiple allocations from the same mempool under
424 * generic_make_request() should be avoided - instead, use bio_set's front_pad
425 * for per bio allocations.
428 * Pointer to new bio on success, NULL on failure.
430 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
432 gfp_t saved_gfp = gfp_mask;
434 unsigned inline_vecs;
435 struct bio_vec *bvl = NULL;
440 if (nr_iovecs > UIO_MAXIOV)
443 p = kmalloc(sizeof(struct bio) +
444 nr_iovecs * sizeof(struct bio_vec),
447 inline_vecs = nr_iovecs;
449 /* should not use nobvec bioset for nr_iovecs > 0 */
450 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
453 * generic_make_request() converts recursion to iteration; this
454 * means if we're running beneath it, any bios we allocate and
455 * submit will not be submitted (and thus freed) until after we
458 * This exposes us to a potential deadlock if we allocate
459 * multiple bios from the same bio_set() while running
460 * underneath generic_make_request(). If we were to allocate
461 * multiple bios (say a stacking block driver that was splitting
462 * bios), we would deadlock if we exhausted the mempool's
465 * We solve this, and guarantee forward progress, with a rescuer
466 * workqueue per bio_set. If we go to allocate and there are
467 * bios on current->bio_list, we first try the allocation
468 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
469 * bios we would be blocking to the rescuer workqueue before
470 * we retry with the original gfp_flags.
473 if (current->bio_list &&
474 (!bio_list_empty(¤t->bio_list[0]) ||
475 !bio_list_empty(¤t->bio_list[1])))
476 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
478 p = mempool_alloc(bs->bio_pool, gfp_mask);
479 if (!p && gfp_mask != saved_gfp) {
480 punt_bios_to_rescuer(bs);
481 gfp_mask = saved_gfp;
482 p = mempool_alloc(bs->bio_pool, gfp_mask);
485 front_pad = bs->front_pad;
486 inline_vecs = BIO_INLINE_VECS;
493 bio_init(bio, NULL, 0);
495 if (nr_iovecs > inline_vecs) {
496 unsigned long idx = 0;
498 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
499 if (!bvl && gfp_mask != saved_gfp) {
500 punt_bios_to_rescuer(bs);
501 gfp_mask = saved_gfp;
502 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
508 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
509 } else if (nr_iovecs) {
510 bvl = bio->bi_inline_vecs;
514 bio->bi_max_vecs = nr_iovecs;
515 bio->bi_io_vec = bvl;
519 mempool_free(p, bs->bio_pool);
522 EXPORT_SYMBOL(bio_alloc_bioset);
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_put - release a reference to a bio
541 * @bio: bio to release reference to
544 * Put a reference to a &struct bio, either one you have gotten with
545 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
547 void bio_put(struct bio *bio)
549 if (!bio_flagged(bio, BIO_REFFED))
552 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
557 if (atomic_dec_and_test(&bio->__bi_cnt))
561 EXPORT_SYMBOL(bio_put);
563 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
565 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
566 blk_recount_segments(q, bio);
568 return bio->bi_phys_segments;
570 EXPORT_SYMBOL(bio_phys_segments);
573 * __bio_clone_fast - clone a bio that shares the original bio's biovec
574 * @bio: destination bio
575 * @bio_src: bio to clone
577 * Clone a &bio. Caller will own the returned bio, but not
578 * the actual data it points to. Reference count of returned
581 * Caller must ensure that @bio_src is not freed before @bio.
583 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
585 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
588 * most users will be overriding ->bi_bdev with a new target,
589 * so we don't set nor calculate new physical/hw segment counts here
591 bio->bi_bdev = bio_src->bi_bdev;
592 bio_set_flag(bio, BIO_CLONED);
593 bio->bi_opf = bio_src->bi_opf;
594 bio->bi_iter = bio_src->bi_iter;
595 bio->bi_io_vec = bio_src->bi_io_vec;
597 bio_clone_blkcg_association(bio, bio_src);
599 EXPORT_SYMBOL(__bio_clone_fast);
602 * bio_clone_fast - clone a bio that shares the original bio's biovec
604 * @gfp_mask: allocation priority
605 * @bs: bio_set to allocate from
607 * Like __bio_clone_fast, only also allocates the returned bio
609 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
613 b = bio_alloc_bioset(gfp_mask, 0, bs);
617 __bio_clone_fast(b, bio);
619 if (bio_integrity(bio)) {
622 ret = bio_integrity_clone(b, bio, gfp_mask);
632 EXPORT_SYMBOL(bio_clone_fast);
635 * bio_clone_bioset - clone a bio
636 * @bio_src: bio to clone
637 * @gfp_mask: allocation priority
638 * @bs: bio_set to allocate from
640 * Clone bio. Caller will own the returned bio, but not the actual data it
641 * points to. Reference count of returned bio will be one.
643 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
646 struct bvec_iter iter;
651 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
652 * bio_src->bi_io_vec to bio->bi_io_vec.
654 * We can't do that anymore, because:
656 * - The point of cloning the biovec is to produce a bio with a biovec
657 * the caller can modify: bi_idx and bi_bvec_done should be 0.
659 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
660 * we tried to clone the whole thing bio_alloc_bioset() would fail.
661 * But the clone should succeed as long as the number of biovecs we
662 * actually need to allocate is fewer than BIO_MAX_PAGES.
664 * - Lastly, bi_vcnt should not be looked at or relied upon by code
665 * that does not own the bio - reason being drivers don't use it for
666 * iterating over the biovec anymore, so expecting it to be kept up
667 * to date (i.e. for clones that share the parent biovec) is just
668 * asking for trouble and would force extra work on
669 * __bio_clone_fast() anyways.
672 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
675 bio->bi_bdev = bio_src->bi_bdev;
676 bio->bi_opf = bio_src->bi_opf;
677 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
678 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
680 switch (bio_op(bio)) {
682 case REQ_OP_SECURE_ERASE:
683 case REQ_OP_WRITE_ZEROES:
685 case REQ_OP_WRITE_SAME:
686 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
689 bio_for_each_segment(bv, bio_src, iter)
690 bio->bi_io_vec[bio->bi_vcnt++] = bv;
694 if (bio_integrity(bio_src)) {
697 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
704 bio_clone_blkcg_association(bio, bio_src);
708 EXPORT_SYMBOL(bio_clone_bioset);
711 * bio_add_pc_page - attempt to add page to bio
712 * @q: the target queue
713 * @bio: destination bio
715 * @len: vec entry length
716 * @offset: vec entry offset
718 * Attempt to add a page to the bio_vec maplist. This can fail for a
719 * number of reasons, such as the bio being full or target block device
720 * limitations. The target block device must allow bio's up to PAGE_SIZE,
721 * so it is always possible to add a single page to an empty bio.
723 * This should only be used by REQ_PC bios.
725 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
726 *page, unsigned int len, unsigned int offset)
728 int retried_segments = 0;
729 struct bio_vec *bvec;
732 * cloned bio must not modify vec list
734 if (unlikely(bio_flagged(bio, BIO_CLONED)))
737 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
741 * For filesystems with a blocksize smaller than the pagesize
742 * we will often be called with the same page as last time and
743 * a consecutive offset. Optimize this special case.
745 if (bio->bi_vcnt > 0) {
746 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
748 if (page == prev->bv_page &&
749 offset == prev->bv_offset + 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 (bvec_gap_to_prev(q, prev, offset))
763 if (bio->bi_vcnt >= bio->bi_max_vecs)
767 * setup the new entry, we might clear it again later if we
768 * cannot add the page
770 bvec = &bio->bi_io_vec[bio->bi_vcnt];
771 bvec->bv_page = page;
773 bvec->bv_offset = offset;
775 bio->bi_phys_segments++;
776 bio->bi_iter.bi_size += len;
779 * Perform a recount if the number of segments is greater
780 * than queue_max_segments(q).
783 while (bio->bi_phys_segments > queue_max_segments(q)) {
785 if (retried_segments)
788 retried_segments = 1;
789 blk_recount_segments(q, bio);
792 /* If we may be able to merge these biovecs, force a recount */
793 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
794 bio_clear_flag(bio, BIO_SEG_VALID);
800 bvec->bv_page = NULL;
804 bio->bi_iter.bi_size -= len;
805 blk_recount_segments(q, bio);
808 EXPORT_SYMBOL(bio_add_pc_page);
811 * bio_add_page - attempt to add page to bio
812 * @bio: destination bio
814 * @len: vec entry length
815 * @offset: vec entry offset
817 * Attempt to add a page to the bio_vec maplist. This will only fail
818 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
820 int bio_add_page(struct bio *bio, struct page *page,
821 unsigned int len, unsigned int offset)
826 * cloned bio must not modify vec list
828 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
832 * For filesystems with a blocksize smaller than the pagesize
833 * we will often be called with the same page as last time and
834 * a consecutive offset. Optimize this special case.
836 if (bio->bi_vcnt > 0) {
837 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
839 if (page == bv->bv_page &&
840 offset == bv->bv_offset + bv->bv_len) {
846 if (bio->bi_vcnt >= bio->bi_max_vecs)
849 bv = &bio->bi_io_vec[bio->bi_vcnt];
852 bv->bv_offset = offset;
856 bio->bi_iter.bi_size += len;
859 EXPORT_SYMBOL(bio_add_page);
862 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
863 * @bio: bio to add pages to
864 * @iter: iov iterator describing the region to be mapped
866 * Pins as many pages from *iter and appends them to @bio's bvec array. The
867 * pages will have to be released using put_page() when done.
869 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
871 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
872 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
873 struct page **pages = (struct page **)bv;
877 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
878 if (unlikely(size <= 0))
879 return size ? size : -EFAULT;
880 nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE;
883 * Deep magic below: We need to walk the pinned pages backwards
884 * because we are abusing the space allocated for the bio_vecs
885 * for the page array. Because the bio_vecs are larger than the
886 * page pointers by definition this will always work. But it also
887 * means we can't use bio_add_page, so any changes to it's semantics
888 * need to be reflected here as well.
890 bio->bi_iter.bi_size += size;
891 bio->bi_vcnt += nr_pages;
893 diff = (nr_pages * PAGE_SIZE - offset) - size;
895 bv[nr_pages].bv_page = pages[nr_pages];
896 bv[nr_pages].bv_len = PAGE_SIZE;
897 bv[nr_pages].bv_offset = 0;
900 bv[0].bv_offset += offset;
901 bv[0].bv_len -= offset;
903 bv[bio->bi_vcnt - 1].bv_len -= diff;
905 iov_iter_advance(iter, size);
908 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
910 struct submit_bio_ret {
911 struct completion event;
915 static void submit_bio_wait_endio(struct bio *bio)
917 struct submit_bio_ret *ret = bio->bi_private;
919 ret->error = bio->bi_error;
920 complete(&ret->event);
924 * submit_bio_wait - submit a bio, and wait until it completes
925 * @bio: The &struct bio which describes the I/O
927 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
928 * bio_endio() on failure.
930 int submit_bio_wait(struct bio *bio)
932 struct submit_bio_ret ret;
934 init_completion(&ret.event);
935 bio->bi_private = &ret;
936 bio->bi_end_io = submit_bio_wait_endio;
937 bio->bi_opf |= REQ_SYNC;
939 wait_for_completion_io(&ret.event);
943 EXPORT_SYMBOL(submit_bio_wait);
946 * bio_advance - increment/complete a bio by some number of bytes
947 * @bio: bio to advance
948 * @bytes: number of bytes to complete
950 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
951 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
952 * be updated on the last bvec as well.
954 * @bio will then represent the remaining, uncompleted portion of the io.
956 void bio_advance(struct bio *bio, unsigned bytes)
958 if (bio_integrity(bio))
959 bio_integrity_advance(bio, bytes);
961 bio_advance_iter(bio, &bio->bi_iter, bytes);
963 EXPORT_SYMBOL(bio_advance);
966 * bio_alloc_pages - allocates a single page for each bvec in a bio
967 * @bio: bio to allocate pages for
968 * @gfp_mask: flags for allocation
970 * Allocates pages up to @bio->bi_vcnt.
972 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
975 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
980 bio_for_each_segment_all(bv, bio, i) {
981 bv->bv_page = alloc_page(gfp_mask);
983 while (--bv >= bio->bi_io_vec)
984 __free_page(bv->bv_page);
991 EXPORT_SYMBOL(bio_alloc_pages);
994 * bio_copy_data - copy contents of data buffers from one chain of bios to
996 * @src: source bio list
997 * @dst: destination bio list
999 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
1000 * @src and @dst as linked lists of bios.
1002 * Stops when it reaches the end of either @src or @dst - that is, copies
1003 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1005 void bio_copy_data(struct bio *dst, struct bio *src)
1007 struct bvec_iter src_iter, dst_iter;
1008 struct bio_vec src_bv, dst_bv;
1009 void *src_p, *dst_p;
1012 src_iter = src->bi_iter;
1013 dst_iter = dst->bi_iter;
1016 if (!src_iter.bi_size) {
1021 src_iter = src->bi_iter;
1024 if (!dst_iter.bi_size) {
1029 dst_iter = dst->bi_iter;
1032 src_bv = bio_iter_iovec(src, src_iter);
1033 dst_bv = bio_iter_iovec(dst, dst_iter);
1035 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1037 src_p = kmap_atomic(src_bv.bv_page);
1038 dst_p = kmap_atomic(dst_bv.bv_page);
1040 memcpy(dst_p + dst_bv.bv_offset,
1041 src_p + src_bv.bv_offset,
1044 kunmap_atomic(dst_p);
1045 kunmap_atomic(src_p);
1047 bio_advance_iter(src, &src_iter, bytes);
1048 bio_advance_iter(dst, &dst_iter, bytes);
1051 EXPORT_SYMBOL(bio_copy_data);
1053 struct bio_map_data {
1055 struct iov_iter iter;
1059 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
1062 if (iov_count > UIO_MAXIOV)
1065 return kmalloc(sizeof(struct bio_map_data) +
1066 sizeof(struct iovec) * iov_count, gfp_mask);
1070 * bio_copy_from_iter - copy all pages from iov_iter to bio
1071 * @bio: The &struct bio which describes the I/O as destination
1072 * @iter: iov_iter as source
1074 * Copy all pages from iov_iter to bio.
1075 * Returns 0 on success, or error on failure.
1077 static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
1080 struct bio_vec *bvec;
1082 bio_for_each_segment_all(bvec, bio, i) {
1085 ret = copy_page_from_iter(bvec->bv_page,
1090 if (!iov_iter_count(&iter))
1093 if (ret < bvec->bv_len)
1101 * bio_copy_to_iter - copy all pages from bio to iov_iter
1102 * @bio: The &struct bio which describes the I/O as source
1103 * @iter: iov_iter as destination
1105 * Copy all pages from bio to iov_iter.
1106 * Returns 0 on success, or error on failure.
1108 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1111 struct bio_vec *bvec;
1113 bio_for_each_segment_all(bvec, bio, i) {
1116 ret = copy_page_to_iter(bvec->bv_page,
1121 if (!iov_iter_count(&iter))
1124 if (ret < bvec->bv_len)
1131 void bio_free_pages(struct bio *bio)
1133 struct bio_vec *bvec;
1136 bio_for_each_segment_all(bvec, bio, i)
1137 __free_page(bvec->bv_page);
1139 EXPORT_SYMBOL(bio_free_pages);
1142 * bio_uncopy_user - finish previously mapped bio
1143 * @bio: bio being terminated
1145 * Free pages allocated from bio_copy_user_iov() and write back data
1146 * to user space in case of a read.
1148 int bio_uncopy_user(struct bio *bio)
1150 struct bio_map_data *bmd = bio->bi_private;
1153 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1155 * if we're in a workqueue, the request is orphaned, so
1156 * don't copy into a random user address space, just free
1157 * and return -EINTR so user space doesn't expect any data.
1161 else if (bio_data_dir(bio) == READ)
1162 ret = bio_copy_to_iter(bio, bmd->iter);
1163 if (bmd->is_our_pages)
1164 bio_free_pages(bio);
1172 * bio_copy_user_iov - copy user data to bio
1173 * @q: destination block queue
1174 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1175 * @iter: iovec iterator
1176 * @gfp_mask: memory allocation flags
1178 * Prepares and returns a bio for indirect user io, bouncing data
1179 * to/from kernel pages as necessary. Must be paired with
1180 * call bio_uncopy_user() on io completion.
1182 struct bio *bio_copy_user_iov(struct request_queue *q,
1183 struct rq_map_data *map_data,
1184 const struct iov_iter *iter,
1187 struct bio_map_data *bmd;
1192 unsigned int len = iter->count;
1193 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1195 for (i = 0; i < iter->nr_segs; i++) {
1196 unsigned long uaddr;
1198 unsigned long start;
1200 uaddr = (unsigned long) iter->iov[i].iov_base;
1201 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1203 start = uaddr >> PAGE_SHIFT;
1209 return ERR_PTR(-EINVAL);
1211 nr_pages += end - start;
1217 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
1219 return ERR_PTR(-ENOMEM);
1222 * We need to do a deep copy of the iov_iter including the iovecs.
1223 * The caller provided iov might point to an on-stack or otherwise
1226 bmd->is_our_pages = map_data ? 0 : 1;
1227 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1228 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1229 iter->nr_segs, iter->count);
1232 bio = bio_kmalloc(gfp_mask, nr_pages);
1239 nr_pages = 1 << map_data->page_order;
1240 i = map_data->offset / PAGE_SIZE;
1243 unsigned int bytes = PAGE_SIZE;
1251 if (i == map_data->nr_entries * nr_pages) {
1256 page = map_data->pages[i / nr_pages];
1257 page += (i % nr_pages);
1261 page = alloc_page(q->bounce_gfp | gfp_mask);
1268 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1281 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1282 (map_data && map_data->from_user)) {
1283 ret = bio_copy_from_iter(bio, *iter);
1288 bio->bi_private = bmd;
1292 bio_free_pages(bio);
1296 return ERR_PTR(ret);
1300 * bio_map_user_iov - map user iovec into bio
1301 * @q: the struct request_queue for the bio
1302 * @iter: iovec iterator
1303 * @gfp_mask: memory allocation flags
1305 * Map the user space address into a bio suitable for io to a block
1306 * device. Returns an error pointer in case of error.
1308 struct bio *bio_map_user_iov(struct request_queue *q,
1309 const struct iov_iter *iter,
1314 struct page **pages;
1321 iov_for_each(iov, i, *iter) {
1322 unsigned long uaddr = (unsigned long) iov.iov_base;
1323 unsigned long len = iov.iov_len;
1324 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1325 unsigned long start = uaddr >> PAGE_SHIFT;
1331 return ERR_PTR(-EINVAL);
1333 nr_pages += end - start;
1335 * buffer must be aligned to at least logical block size for now
1337 if (uaddr & queue_dma_alignment(q))
1338 return ERR_PTR(-EINVAL);
1342 return ERR_PTR(-EINVAL);
1344 bio = bio_kmalloc(gfp_mask, nr_pages);
1346 return ERR_PTR(-ENOMEM);
1349 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1353 iov_for_each(iov, i, *iter) {
1354 unsigned long uaddr = (unsigned long) iov.iov_base;
1355 unsigned long len = iov.iov_len;
1356 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1357 unsigned long start = uaddr >> PAGE_SHIFT;
1358 const int local_nr_pages = end - start;
1359 const int page_limit = cur_page + local_nr_pages;
1361 ret = get_user_pages_fast(uaddr, local_nr_pages,
1362 (iter->type & WRITE) != WRITE,
1364 if (ret < local_nr_pages) {
1369 offset = offset_in_page(uaddr);
1370 for (j = cur_page; j < page_limit; j++) {
1371 unsigned int bytes = PAGE_SIZE - offset;
1382 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1392 * release the pages we didn't map into the bio, if any
1394 while (j < page_limit)
1395 put_page(pages[j++]);
1400 bio_set_flag(bio, BIO_USER_MAPPED);
1403 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1404 * it would normally disappear when its bi_end_io is run.
1405 * however, we need it for the unmap, so grab an extra
1412 for (j = 0; j < nr_pages; j++) {
1420 return ERR_PTR(ret);
1423 static void __bio_unmap_user(struct bio *bio)
1425 struct bio_vec *bvec;
1429 * make sure we dirty pages we wrote to
1431 bio_for_each_segment_all(bvec, bio, i) {
1432 if (bio_data_dir(bio) == READ)
1433 set_page_dirty_lock(bvec->bv_page);
1435 put_page(bvec->bv_page);
1442 * bio_unmap_user - unmap a bio
1443 * @bio: the bio being unmapped
1445 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1448 * bio_unmap_user() may sleep.
1450 void bio_unmap_user(struct bio *bio)
1452 __bio_unmap_user(bio);
1456 static void bio_map_kern_endio(struct bio *bio)
1462 * bio_map_kern - map kernel address into bio
1463 * @q: the struct request_queue for the bio
1464 * @data: pointer to buffer to map
1465 * @len: length in bytes
1466 * @gfp_mask: allocation flags for bio allocation
1468 * Map the kernel address into a bio suitable for io to a block
1469 * device. Returns an error pointer in case of error.
1471 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1474 unsigned long kaddr = (unsigned long)data;
1475 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1476 unsigned long start = kaddr >> PAGE_SHIFT;
1477 const int nr_pages = end - start;
1481 bio = bio_kmalloc(gfp_mask, nr_pages);
1483 return ERR_PTR(-ENOMEM);
1485 offset = offset_in_page(kaddr);
1486 for (i = 0; i < nr_pages; i++) {
1487 unsigned int bytes = PAGE_SIZE - offset;
1495 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1497 /* we don't support partial mappings */
1499 return ERR_PTR(-EINVAL);
1507 bio->bi_end_io = bio_map_kern_endio;
1510 EXPORT_SYMBOL(bio_map_kern);
1512 static void bio_copy_kern_endio(struct bio *bio)
1514 bio_free_pages(bio);
1518 static void bio_copy_kern_endio_read(struct bio *bio)
1520 char *p = bio->bi_private;
1521 struct bio_vec *bvec;
1524 bio_for_each_segment_all(bvec, bio, i) {
1525 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1529 bio_copy_kern_endio(bio);
1533 * bio_copy_kern - copy kernel address into bio
1534 * @q: the struct request_queue for the bio
1535 * @data: pointer to buffer to copy
1536 * @len: length in bytes
1537 * @gfp_mask: allocation flags for bio and page allocation
1538 * @reading: data direction is READ
1540 * copy the kernel address into a bio suitable for io to a block
1541 * device. Returns an error pointer in case of error.
1543 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1544 gfp_t gfp_mask, int reading)
1546 unsigned long kaddr = (unsigned long)data;
1547 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1548 unsigned long start = kaddr >> PAGE_SHIFT;
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_end_io = bio_copy_kern_endio_read;
1587 bio->bi_private = data;
1589 bio->bi_end_io = bio_copy_kern_endio;
1595 bio_free_pages(bio);
1597 return ERR_PTR(-ENOMEM);
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 put_page() against each page and will run one
1663 * 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)) {
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);
1765 static inline bool bio_remaining_done(struct bio *bio)
1768 * If we're not chaining, then ->__bi_remaining is always 1 and
1769 * we always end io on the first invocation.
1771 if (!bio_flagged(bio, BIO_CHAIN))
1774 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1776 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1777 bio_clear_flag(bio, BIO_CHAIN);
1785 * bio_endio - end I/O on a bio
1789 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1790 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1791 * bio unless they own it and thus know that it has an end_io function.
1793 void bio_endio(struct bio *bio)
1796 if (!bio_remaining_done(bio))
1800 * Need to have a real endio function for chained bios, otherwise
1801 * various corner cases will break (like stacking block devices that
1802 * save/restore bi_end_io) - however, we want to avoid unbounded
1803 * recursion and blowing the stack. Tail call optimization would
1804 * handle this, but compiling with frame pointers also disables
1805 * gcc's sibling call optimization.
1807 if (bio->bi_end_io == bio_chain_endio) {
1808 bio = __bio_chain_endio(bio);
1813 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 * Unless this is a discard request the newly allocated bio will point
1828 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1829 * @bio is not freed before the split.
1831 struct bio *bio_split(struct bio *bio, int sectors,
1832 gfp_t gfp, struct bio_set *bs)
1834 struct bio *split = NULL;
1836 BUG_ON(sectors <= 0);
1837 BUG_ON(sectors >= bio_sectors(bio));
1839 split = bio_clone_fast(bio, gfp, bs);
1843 split->bi_iter.bi_size = sectors << 9;
1845 if (bio_integrity(split))
1846 bio_integrity_trim(split, 0, sectors);
1848 bio_advance(bio, split->bi_iter.bi_size);
1852 EXPORT_SYMBOL(bio_split);
1855 * bio_trim - trim a bio
1857 * @offset: number of sectors to trim from the front of @bio
1858 * @size: size we want to trim @bio to, in sectors
1860 void bio_trim(struct bio *bio, int offset, int size)
1862 /* 'bio' is a cloned bio which we need to trim to match
1863 * the given offset and size.
1867 if (offset == 0 && size == bio->bi_iter.bi_size)
1870 bio_clear_flag(bio, BIO_SEG_VALID);
1872 bio_advance(bio, offset << 9);
1874 bio->bi_iter.bi_size = size;
1876 EXPORT_SYMBOL_GPL(bio_trim);
1879 * create memory pools for biovec's in a bio_set.
1880 * use the global biovec slabs created for general use.
1882 mempool_t *biovec_create_pool(int pool_entries)
1884 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1886 return mempool_create_slab_pool(pool_entries, bp->slab);
1889 void bioset_free(struct bio_set *bs)
1891 if (bs->rescue_workqueue)
1892 destroy_workqueue(bs->rescue_workqueue);
1895 mempool_destroy(bs->bio_pool);
1898 mempool_destroy(bs->bvec_pool);
1900 bioset_integrity_free(bs);
1905 EXPORT_SYMBOL(bioset_free);
1907 static struct bio_set *__bioset_create(unsigned int pool_size,
1908 unsigned int front_pad,
1909 bool create_bvec_pool)
1911 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1914 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1918 bs->front_pad = front_pad;
1920 spin_lock_init(&bs->rescue_lock);
1921 bio_list_init(&bs->rescue_list);
1922 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1924 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1925 if (!bs->bio_slab) {
1930 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1934 if (create_bvec_pool) {
1935 bs->bvec_pool = biovec_create_pool(pool_size);
1940 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1941 if (!bs->rescue_workqueue)
1951 * bioset_create - Create a bio_set
1952 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1953 * @front_pad: Number of bytes to allocate in front of the returned bio
1956 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1957 * to ask for a number of bytes to be allocated in front of the bio.
1958 * Front pad allocation is useful for embedding the bio inside
1959 * another structure, to avoid allocating extra data to go with the bio.
1960 * Note that the bio must be embedded at the END of that structure always,
1961 * or things will break badly.
1963 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1965 return __bioset_create(pool_size, front_pad, true);
1967 EXPORT_SYMBOL(bioset_create);
1970 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1971 * @pool_size: Number of bio to cache in the mempool
1972 * @front_pad: Number of bytes to allocate in front of the returned bio
1975 * Same functionality as bioset_create() except that mempool is not
1976 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1978 struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1980 return __bioset_create(pool_size, front_pad, false);
1982 EXPORT_SYMBOL(bioset_create_nobvec);
1984 #ifdef CONFIG_BLK_CGROUP
1987 * bio_associate_blkcg - associate a bio with the specified blkcg
1989 * @blkcg_css: css of the blkcg to associate
1991 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1992 * treat @bio as if it were issued by a task which belongs to the blkcg.
1994 * This function takes an extra reference of @blkcg_css which will be put
1995 * when @bio is released. The caller must own @bio and is responsible for
1996 * synchronizing calls to this function.
1998 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
2000 if (unlikely(bio->bi_css))
2003 bio->bi_css = blkcg_css;
2006 EXPORT_SYMBOL_GPL(bio_associate_blkcg);
2009 * bio_associate_current - associate a bio with %current
2012 * Associate @bio with %current if it hasn't been associated yet. Block
2013 * layer will treat @bio as if it were issued by %current no matter which
2014 * task actually issues it.
2016 * This function takes an extra reference of @task's io_context and blkcg
2017 * which will be put when @bio is released. The caller must own @bio,
2018 * ensure %current->io_context exists, and is responsible for synchronizing
2019 * calls to this function.
2021 int bio_associate_current(struct bio *bio)
2023 struct io_context *ioc;
2028 ioc = current->io_context;
2032 get_io_context_active(ioc);
2034 bio->bi_css = task_get_css(current, io_cgrp_id);
2037 EXPORT_SYMBOL_GPL(bio_associate_current);
2040 * bio_disassociate_task - undo bio_associate_current()
2043 void bio_disassociate_task(struct bio *bio)
2046 put_io_context(bio->bi_ioc);
2050 css_put(bio->bi_css);
2056 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2057 * @dst: destination bio
2060 void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2063 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2066 #endif /* CONFIG_BLK_CGROUP */
2068 static void __init biovec_init_slabs(void)
2072 for (i = 0; i < BVEC_POOL_NR; i++) {
2074 struct biovec_slab *bvs = bvec_slabs + i;
2076 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2081 size = bvs->nr_vecs * sizeof(struct bio_vec);
2082 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2083 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2087 static int __init init_bio(void)
2091 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2093 panic("bio: can't allocate bios\n");
2095 bio_integrity_init();
2096 biovec_init_slabs();
2098 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2100 panic("bio: can't allocate bios\n");
2102 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2103 panic("bio: can't create integrity pool\n");
2107 subsys_initcall(init_bio);