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)
275 memset(bio, 0, sizeof(*bio));
276 atomic_set(&bio->__bi_remaining, 1);
277 atomic_set(&bio->__bi_cnt, 1);
279 EXPORT_SYMBOL(bio_init);
282 * bio_reset - reinitialize a bio
286 * After calling bio_reset(), @bio will be in the same state as a freshly
287 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
288 * preserved are the ones that are initialized by bio_alloc_bioset(). See
289 * comment in struct bio.
291 void bio_reset(struct bio *bio)
293 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
297 memset(bio, 0, BIO_RESET_BYTES);
298 bio->bi_flags = flags;
299 atomic_set(&bio->__bi_remaining, 1);
301 EXPORT_SYMBOL(bio_reset);
303 static struct bio *__bio_chain_endio(struct bio *bio)
305 struct bio *parent = bio->bi_private;
307 if (!parent->bi_error)
308 parent->bi_error = bio->bi_error;
313 static void bio_chain_endio(struct bio *bio)
315 bio_endio(__bio_chain_endio(bio));
319 * bio_chain - chain bio completions
320 * @bio: the target bio
321 * @parent: the @bio's parent bio
323 * The caller won't have a bi_end_io called when @bio completes - instead,
324 * @parent's bi_end_io won't be called until both @parent and @bio have
325 * completed; the chained bio will also be freed when it completes.
327 * The caller must not set bi_private or bi_end_io in @bio.
329 void bio_chain(struct bio *bio, struct bio *parent)
331 BUG_ON(bio->bi_private || bio->bi_end_io);
333 bio->bi_private = parent;
334 bio->bi_end_io = bio_chain_endio;
335 bio_inc_remaining(parent);
337 EXPORT_SYMBOL(bio_chain);
339 static void bio_alloc_rescue(struct work_struct *work)
341 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
345 spin_lock(&bs->rescue_lock);
346 bio = bio_list_pop(&bs->rescue_list);
347 spin_unlock(&bs->rescue_lock);
352 generic_make_request(bio);
356 static void punt_bios_to_rescuer(struct bio_set *bs)
358 struct bio_list punt, nopunt;
362 * In order to guarantee forward progress we must punt only bios that
363 * were allocated from this bio_set; otherwise, if there was a bio on
364 * there for a stacking driver higher up in the stack, processing it
365 * could require allocating bios from this bio_set, and doing that from
366 * our own rescuer would be bad.
368 * Since bio lists are singly linked, pop them all instead of trying to
369 * remove from the middle of the list:
372 bio_list_init(&punt);
373 bio_list_init(&nopunt);
375 while ((bio = bio_list_pop(current->bio_list)))
376 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
378 *current->bio_list = nopunt;
380 spin_lock(&bs->rescue_lock);
381 bio_list_merge(&bs->rescue_list, &punt);
382 spin_unlock(&bs->rescue_lock);
384 queue_work(bs->rescue_workqueue, &bs->rescue_work);
388 * bio_alloc_bioset - allocate a bio for I/O
389 * @gfp_mask: the GFP_ mask given to the slab allocator
390 * @nr_iovecs: number of iovecs to pre-allocate
391 * @bs: the bio_set to allocate from.
394 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
395 * backed by the @bs's mempool.
397 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
398 * always be able to allocate a bio. This is due to the mempool guarantees.
399 * To make this work, callers must never allocate more than 1 bio at a time
400 * from this pool. Callers that need to allocate more than 1 bio must always
401 * submit the previously allocated bio for IO before attempting to allocate
402 * a new one. Failure to do so can cause deadlocks under memory pressure.
404 * Note that when running under generic_make_request() (i.e. any block
405 * driver), bios are not submitted until after you return - see the code in
406 * generic_make_request() that converts recursion into iteration, to prevent
409 * This would normally mean allocating multiple bios under
410 * generic_make_request() would be susceptible to deadlocks, but we have
411 * deadlock avoidance code that resubmits any blocked bios from a rescuer
414 * However, we do not guarantee forward progress for allocations from other
415 * mempools. Doing multiple allocations from the same mempool under
416 * generic_make_request() should be avoided - instead, use bio_set's front_pad
417 * for per bio allocations.
420 * Pointer to new bio on success, NULL on failure.
422 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
424 gfp_t saved_gfp = gfp_mask;
426 unsigned inline_vecs;
427 struct bio_vec *bvl = NULL;
432 if (nr_iovecs > UIO_MAXIOV)
435 p = kmalloc(sizeof(struct bio) +
436 nr_iovecs * sizeof(struct bio_vec),
439 inline_vecs = nr_iovecs;
441 /* should not use nobvec bioset for nr_iovecs > 0 */
442 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
445 * generic_make_request() converts recursion to iteration; this
446 * means if we're running beneath it, any bios we allocate and
447 * submit will not be submitted (and thus freed) until after we
450 * This exposes us to a potential deadlock if we allocate
451 * multiple bios from the same bio_set() while running
452 * underneath generic_make_request(). If we were to allocate
453 * multiple bios (say a stacking block driver that was splitting
454 * bios), we would deadlock if we exhausted the mempool's
457 * We solve this, and guarantee forward progress, with a rescuer
458 * workqueue per bio_set. If we go to allocate and there are
459 * bios on current->bio_list, we first try the allocation
460 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
461 * bios we would be blocking to the rescuer workqueue before
462 * we retry with the original gfp_flags.
465 if (current->bio_list && !bio_list_empty(current->bio_list))
466 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
468 p = mempool_alloc(bs->bio_pool, gfp_mask);
469 if (!p && gfp_mask != saved_gfp) {
470 punt_bios_to_rescuer(bs);
471 gfp_mask = saved_gfp;
472 p = mempool_alloc(bs->bio_pool, gfp_mask);
475 front_pad = bs->front_pad;
476 inline_vecs = BIO_INLINE_VECS;
485 if (nr_iovecs > inline_vecs) {
486 unsigned long idx = 0;
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 |= idx << BVEC_POOL_OFFSET;
499 } else if (nr_iovecs) {
500 bvl = bio->bi_inline_vecs;
504 bio->bi_max_vecs = nr_iovecs;
505 bio->bi_io_vec = bvl;
509 mempool_free(p, bs->bio_pool);
512 EXPORT_SYMBOL(bio_alloc_bioset);
514 void zero_fill_bio(struct bio *bio)
518 struct bvec_iter iter;
520 bio_for_each_segment(bv, bio, iter) {
521 char *data = bvec_kmap_irq(&bv, &flags);
522 memset(data, 0, bv.bv_len);
523 flush_dcache_page(bv.bv_page);
524 bvec_kunmap_irq(data, &flags);
527 EXPORT_SYMBOL(zero_fill_bio);
530 * bio_put - release a reference to a bio
531 * @bio: bio to release reference to
534 * Put a reference to a &struct bio, either one you have gotten with
535 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
537 void bio_put(struct bio *bio)
539 if (!bio_flagged(bio, BIO_REFFED))
542 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
547 if (atomic_dec_and_test(&bio->__bi_cnt))
551 EXPORT_SYMBOL(bio_put);
553 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
555 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
556 blk_recount_segments(q, bio);
558 return bio->bi_phys_segments;
560 EXPORT_SYMBOL(bio_phys_segments);
563 * __bio_clone_fast - clone a bio that shares the original bio's biovec
564 * @bio: destination bio
565 * @bio_src: bio to clone
567 * Clone a &bio. Caller will own the returned bio, but not
568 * the actual data it points to. Reference count of returned
571 * Caller must ensure that @bio_src is not freed before @bio.
573 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
575 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
578 * most users will be overriding ->bi_bdev with a new target,
579 * so we don't set nor calculate new physical/hw segment counts here
581 bio->bi_bdev = bio_src->bi_bdev;
582 bio_set_flag(bio, BIO_CLONED);
583 bio->bi_rw = bio_src->bi_rw;
584 bio->bi_iter = bio_src->bi_iter;
585 bio->bi_io_vec = bio_src->bi_io_vec;
587 EXPORT_SYMBOL(__bio_clone_fast);
590 * bio_clone_fast - clone a bio that shares the original bio's biovec
592 * @gfp_mask: allocation priority
593 * @bs: bio_set to allocate from
595 * Like __bio_clone_fast, only also allocates the returned bio
597 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
601 b = bio_alloc_bioset(gfp_mask, 0, bs);
605 __bio_clone_fast(b, bio);
607 if (bio_integrity(bio)) {
610 ret = bio_integrity_clone(b, bio, gfp_mask);
620 EXPORT_SYMBOL(bio_clone_fast);
623 * bio_clone_bioset - clone a bio
624 * @bio_src: bio to clone
625 * @gfp_mask: allocation priority
626 * @bs: bio_set to allocate from
628 * Clone bio. Caller will own the returned bio, but not the actual data it
629 * points to. Reference count of returned bio will be one.
631 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
634 struct bvec_iter iter;
639 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
640 * bio_src->bi_io_vec to bio->bi_io_vec.
642 * We can't do that anymore, because:
644 * - The point of cloning the biovec is to produce a bio with a biovec
645 * the caller can modify: bi_idx and bi_bvec_done should be 0.
647 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
648 * we tried to clone the whole thing bio_alloc_bioset() would fail.
649 * But the clone should succeed as long as the number of biovecs we
650 * actually need to allocate is fewer than BIO_MAX_PAGES.
652 * - Lastly, bi_vcnt should not be looked at or relied upon by code
653 * that does not own the bio - reason being drivers don't use it for
654 * iterating over the biovec anymore, so expecting it to be kept up
655 * to date (i.e. for clones that share the parent biovec) is just
656 * asking for trouble and would force extra work on
657 * __bio_clone_fast() anyways.
660 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
663 bio->bi_bdev = bio_src->bi_bdev;
664 bio->bi_rw = bio_src->bi_rw;
665 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
666 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
668 if (bio_op(bio) == REQ_OP_DISCARD)
669 goto integrity_clone;
671 if (bio_op(bio) == REQ_OP_WRITE_SAME) {
672 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
673 goto integrity_clone;
676 bio_for_each_segment(bv, bio_src, iter)
677 bio->bi_io_vec[bio->bi_vcnt++] = bv;
680 if (bio_integrity(bio_src)) {
683 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
692 EXPORT_SYMBOL(bio_clone_bioset);
695 * bio_add_pc_page - attempt to add page to bio
696 * @q: the target queue
697 * @bio: destination bio
699 * @len: vec entry length
700 * @offset: vec entry offset
702 * Attempt to add a page to the bio_vec maplist. This can fail for a
703 * number of reasons, such as the bio being full or target block device
704 * limitations. The target block device must allow bio's up to PAGE_SIZE,
705 * so it is always possible to add a single page to an empty bio.
707 * This should only be used by REQ_PC bios.
709 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
710 *page, unsigned int len, unsigned int offset)
712 int retried_segments = 0;
713 struct bio_vec *bvec;
716 * cloned bio must not modify vec list
718 if (unlikely(bio_flagged(bio, BIO_CLONED)))
721 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
725 * For filesystems with a blocksize smaller than the pagesize
726 * we will often be called with the same page as last time and
727 * a consecutive offset. Optimize this special case.
729 if (bio->bi_vcnt > 0) {
730 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
732 if (page == prev->bv_page &&
733 offset == prev->bv_offset + prev->bv_len) {
735 bio->bi_iter.bi_size += len;
740 * If the queue doesn't support SG gaps and adding this
741 * offset would create a gap, disallow it.
743 if (bvec_gap_to_prev(q, prev, offset))
747 if (bio->bi_vcnt >= bio->bi_max_vecs)
751 * setup the new entry, we might clear it again later if we
752 * cannot add the page
754 bvec = &bio->bi_io_vec[bio->bi_vcnt];
755 bvec->bv_page = page;
757 bvec->bv_offset = offset;
759 bio->bi_phys_segments++;
760 bio->bi_iter.bi_size += len;
763 * Perform a recount if the number of segments is greater
764 * than queue_max_segments(q).
767 while (bio->bi_phys_segments > queue_max_segments(q)) {
769 if (retried_segments)
772 retried_segments = 1;
773 blk_recount_segments(q, bio);
776 /* If we may be able to merge these biovecs, force a recount */
777 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
778 bio_clear_flag(bio, BIO_SEG_VALID);
784 bvec->bv_page = NULL;
788 bio->bi_iter.bi_size -= len;
789 blk_recount_segments(q, bio);
792 EXPORT_SYMBOL(bio_add_pc_page);
795 * bio_add_page - attempt to add page to bio
796 * @bio: destination bio
798 * @len: vec entry length
799 * @offset: vec entry offset
801 * Attempt to add a page to the bio_vec maplist. This will only fail
802 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
804 int bio_add_page(struct bio *bio, struct page *page,
805 unsigned int len, unsigned int offset)
810 * cloned bio must not modify vec list
812 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
816 * For filesystems with a blocksize smaller than the pagesize
817 * we will often be called with the same page as last time and
818 * a consecutive offset. Optimize this special case.
820 if (bio->bi_vcnt > 0) {
821 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
823 if (page == bv->bv_page &&
824 offset == bv->bv_offset + bv->bv_len) {
830 if (bio->bi_vcnt >= bio->bi_max_vecs)
833 bv = &bio->bi_io_vec[bio->bi_vcnt];
836 bv->bv_offset = offset;
840 bio->bi_iter.bi_size += len;
843 EXPORT_SYMBOL(bio_add_page);
845 struct submit_bio_ret {
846 struct completion event;
850 static void submit_bio_wait_endio(struct bio *bio)
852 struct submit_bio_ret *ret = bio->bi_private;
854 ret->error = bio->bi_error;
855 complete(&ret->event);
859 * submit_bio_wait - submit a bio, and wait until it completes
860 * @bio: The &struct bio which describes the I/O
862 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
863 * bio_endio() on failure.
865 int submit_bio_wait(struct bio *bio)
867 struct submit_bio_ret ret;
869 init_completion(&ret.event);
870 bio->bi_private = &ret;
871 bio->bi_end_io = submit_bio_wait_endio;
872 bio->bi_rw |= REQ_SYNC;
874 wait_for_completion_io(&ret.event);
878 EXPORT_SYMBOL(submit_bio_wait);
881 * bio_advance - increment/complete a bio by some number of bytes
882 * @bio: bio to advance
883 * @bytes: number of bytes to complete
885 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
886 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
887 * be updated on the last bvec as well.
889 * @bio will then represent the remaining, uncompleted portion of the io.
891 void bio_advance(struct bio *bio, unsigned bytes)
893 if (bio_integrity(bio))
894 bio_integrity_advance(bio, bytes);
896 bio_advance_iter(bio, &bio->bi_iter, bytes);
898 EXPORT_SYMBOL(bio_advance);
901 * bio_alloc_pages - allocates a single page for each bvec in a bio
902 * @bio: bio to allocate pages for
903 * @gfp_mask: flags for allocation
905 * Allocates pages up to @bio->bi_vcnt.
907 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
910 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
915 bio_for_each_segment_all(bv, bio, i) {
916 bv->bv_page = alloc_page(gfp_mask);
918 while (--bv >= bio->bi_io_vec)
919 __free_page(bv->bv_page);
926 EXPORT_SYMBOL(bio_alloc_pages);
929 * bio_copy_data - copy contents of data buffers from one chain of bios to
931 * @src: source bio list
932 * @dst: destination bio list
934 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
935 * @src and @dst as linked lists of bios.
937 * Stops when it reaches the end of either @src or @dst - that is, copies
938 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
940 void bio_copy_data(struct bio *dst, struct bio *src)
942 struct bvec_iter src_iter, dst_iter;
943 struct bio_vec src_bv, dst_bv;
947 src_iter = src->bi_iter;
948 dst_iter = dst->bi_iter;
951 if (!src_iter.bi_size) {
956 src_iter = src->bi_iter;
959 if (!dst_iter.bi_size) {
964 dst_iter = dst->bi_iter;
967 src_bv = bio_iter_iovec(src, src_iter);
968 dst_bv = bio_iter_iovec(dst, dst_iter);
970 bytes = min(src_bv.bv_len, dst_bv.bv_len);
972 src_p = kmap_atomic(src_bv.bv_page);
973 dst_p = kmap_atomic(dst_bv.bv_page);
975 memcpy(dst_p + dst_bv.bv_offset,
976 src_p + src_bv.bv_offset,
979 kunmap_atomic(dst_p);
980 kunmap_atomic(src_p);
982 bio_advance_iter(src, &src_iter, bytes);
983 bio_advance_iter(dst, &dst_iter, bytes);
986 EXPORT_SYMBOL(bio_copy_data);
988 struct bio_map_data {
990 struct iov_iter iter;
994 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
997 if (iov_count > UIO_MAXIOV)
1000 return kmalloc(sizeof(struct bio_map_data) +
1001 sizeof(struct iovec) * iov_count, gfp_mask);
1005 * bio_copy_from_iter - copy all pages from iov_iter to bio
1006 * @bio: The &struct bio which describes the I/O as destination
1007 * @iter: iov_iter as source
1009 * Copy all pages from iov_iter to bio.
1010 * Returns 0 on success, or error on failure.
1012 static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
1015 struct bio_vec *bvec;
1017 bio_for_each_segment_all(bvec, bio, i) {
1020 ret = copy_page_from_iter(bvec->bv_page,
1025 if (!iov_iter_count(&iter))
1028 if (ret < bvec->bv_len)
1036 * bio_copy_to_iter - copy all pages from bio to iov_iter
1037 * @bio: The &struct bio which describes the I/O as source
1038 * @iter: iov_iter as destination
1040 * Copy all pages from bio to iov_iter.
1041 * Returns 0 on success, or error on failure.
1043 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1046 struct bio_vec *bvec;
1048 bio_for_each_segment_all(bvec, bio, i) {
1051 ret = copy_page_to_iter(bvec->bv_page,
1056 if (!iov_iter_count(&iter))
1059 if (ret < bvec->bv_len)
1066 static void bio_free_pages(struct bio *bio)
1068 struct bio_vec *bvec;
1071 bio_for_each_segment_all(bvec, bio, i)
1072 __free_page(bvec->bv_page);
1076 * bio_uncopy_user - finish previously mapped bio
1077 * @bio: bio being terminated
1079 * Free pages allocated from bio_copy_user_iov() and write back data
1080 * to user space in case of a read.
1082 int bio_uncopy_user(struct bio *bio)
1084 struct bio_map_data *bmd = bio->bi_private;
1087 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1089 * if we're in a workqueue, the request is orphaned, so
1090 * don't copy into a random user address space, just free
1091 * and return -EINTR so user space doesn't expect any data.
1095 else if (bio_data_dir(bio) == READ)
1096 ret = bio_copy_to_iter(bio, bmd->iter);
1097 if (bmd->is_our_pages)
1098 bio_free_pages(bio);
1106 * bio_copy_user_iov - copy user data to bio
1107 * @q: destination block queue
1108 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1109 * @iter: iovec iterator
1110 * @gfp_mask: memory allocation flags
1112 * Prepares and returns a bio for indirect user io, bouncing data
1113 * to/from kernel pages as necessary. Must be paired with
1114 * call bio_uncopy_user() on io completion.
1116 struct bio *bio_copy_user_iov(struct request_queue *q,
1117 struct rq_map_data *map_data,
1118 const struct iov_iter *iter,
1121 struct bio_map_data *bmd;
1126 unsigned int len = iter->count;
1127 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1129 for (i = 0; i < iter->nr_segs; i++) {
1130 unsigned long uaddr;
1132 unsigned long start;
1134 uaddr = (unsigned long) iter->iov[i].iov_base;
1135 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1137 start = uaddr >> PAGE_SHIFT;
1143 return ERR_PTR(-EINVAL);
1145 nr_pages += end - start;
1151 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
1153 return ERR_PTR(-ENOMEM);
1156 * We need to do a deep copy of the iov_iter including the iovecs.
1157 * The caller provided iov might point to an on-stack or otherwise
1160 bmd->is_our_pages = map_data ? 0 : 1;
1161 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1162 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1163 iter->nr_segs, iter->count);
1166 bio = bio_kmalloc(gfp_mask, nr_pages);
1170 if (iter->type & WRITE)
1171 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1176 nr_pages = 1 << map_data->page_order;
1177 i = map_data->offset / PAGE_SIZE;
1180 unsigned int bytes = PAGE_SIZE;
1188 if (i == map_data->nr_entries * nr_pages) {
1193 page = map_data->pages[i / nr_pages];
1194 page += (i % nr_pages);
1198 page = alloc_page(q->bounce_gfp | gfp_mask);
1205 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1218 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1219 (map_data && map_data->from_user)) {
1220 ret = bio_copy_from_iter(bio, *iter);
1225 bio->bi_private = bmd;
1229 bio_free_pages(bio);
1233 return ERR_PTR(ret);
1237 * bio_map_user_iov - map user iovec into bio
1238 * @q: the struct request_queue for the bio
1239 * @iter: iovec iterator
1240 * @gfp_mask: memory allocation flags
1242 * Map the user space address into a bio suitable for io to a block
1243 * device. Returns an error pointer in case of error.
1245 struct bio *bio_map_user_iov(struct request_queue *q,
1246 const struct iov_iter *iter,
1251 struct page **pages;
1258 iov_for_each(iov, i, *iter) {
1259 unsigned long uaddr = (unsigned long) iov.iov_base;
1260 unsigned long len = iov.iov_len;
1261 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1262 unsigned long start = uaddr >> PAGE_SHIFT;
1268 return ERR_PTR(-EINVAL);
1270 nr_pages += end - start;
1272 * buffer must be aligned to at least hardsector size for now
1274 if (uaddr & queue_dma_alignment(q))
1275 return ERR_PTR(-EINVAL);
1279 return ERR_PTR(-EINVAL);
1281 bio = bio_kmalloc(gfp_mask, nr_pages);
1283 return ERR_PTR(-ENOMEM);
1286 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1290 iov_for_each(iov, i, *iter) {
1291 unsigned long uaddr = (unsigned long) iov.iov_base;
1292 unsigned long len = iov.iov_len;
1293 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1294 unsigned long start = uaddr >> PAGE_SHIFT;
1295 const int local_nr_pages = end - start;
1296 const int page_limit = cur_page + local_nr_pages;
1298 ret = get_user_pages_fast(uaddr, local_nr_pages,
1299 (iter->type & WRITE) != WRITE,
1301 if (ret < local_nr_pages) {
1306 offset = offset_in_page(uaddr);
1307 for (j = cur_page; j < page_limit; j++) {
1308 unsigned int bytes = PAGE_SIZE - offset;
1319 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1329 * release the pages we didn't map into the bio, if any
1331 while (j < page_limit)
1332 put_page(pages[j++]);
1338 * set data direction, and check if mapped pages need bouncing
1340 if (iter->type & WRITE)
1341 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1343 bio_set_flag(bio, BIO_USER_MAPPED);
1346 * subtle -- if __bio_map_user() ended up bouncing a bio,
1347 * it would normally disappear when its bi_end_io is run.
1348 * however, we need it for the unmap, so grab an extra
1355 for (j = 0; j < nr_pages; j++) {
1363 return ERR_PTR(ret);
1366 static void __bio_unmap_user(struct bio *bio)
1368 struct bio_vec *bvec;
1372 * make sure we dirty pages we wrote to
1374 bio_for_each_segment_all(bvec, bio, i) {
1375 if (bio_data_dir(bio) == READ)
1376 set_page_dirty_lock(bvec->bv_page);
1378 put_page(bvec->bv_page);
1385 * bio_unmap_user - unmap a bio
1386 * @bio: the bio being unmapped
1388 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1389 * a process context.
1391 * bio_unmap_user() may sleep.
1393 void bio_unmap_user(struct bio *bio)
1395 __bio_unmap_user(bio);
1399 static void bio_map_kern_endio(struct bio *bio)
1405 * bio_map_kern - map kernel address into bio
1406 * @q: the struct request_queue for the bio
1407 * @data: pointer to buffer to map
1408 * @len: length in bytes
1409 * @gfp_mask: allocation flags for bio allocation
1411 * Map the kernel address into a bio suitable for io to a block
1412 * device. Returns an error pointer in case of error.
1414 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1417 unsigned long kaddr = (unsigned long)data;
1418 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1419 unsigned long start = kaddr >> PAGE_SHIFT;
1420 const int nr_pages = end - start;
1424 bio = bio_kmalloc(gfp_mask, nr_pages);
1426 return ERR_PTR(-ENOMEM);
1428 offset = offset_in_page(kaddr);
1429 for (i = 0; i < nr_pages; i++) {
1430 unsigned int bytes = PAGE_SIZE - offset;
1438 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1440 /* we don't support partial mappings */
1442 return ERR_PTR(-EINVAL);
1450 bio->bi_end_io = bio_map_kern_endio;
1453 EXPORT_SYMBOL(bio_map_kern);
1455 static void bio_copy_kern_endio(struct bio *bio)
1457 bio_free_pages(bio);
1461 static void bio_copy_kern_endio_read(struct bio *bio)
1463 char *p = bio->bi_private;
1464 struct bio_vec *bvec;
1467 bio_for_each_segment_all(bvec, bio, i) {
1468 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1472 bio_copy_kern_endio(bio);
1476 * bio_copy_kern - copy kernel address into bio
1477 * @q: the struct request_queue for the bio
1478 * @data: pointer to buffer to copy
1479 * @len: length in bytes
1480 * @gfp_mask: allocation flags for bio and page allocation
1481 * @reading: data direction is READ
1483 * copy the kernel address into a bio suitable for io to a block
1484 * device. Returns an error pointer in case of error.
1486 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1487 gfp_t gfp_mask, int reading)
1489 unsigned long kaddr = (unsigned long)data;
1490 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1491 unsigned long start = kaddr >> PAGE_SHIFT;
1500 return ERR_PTR(-EINVAL);
1502 nr_pages = end - start;
1503 bio = bio_kmalloc(gfp_mask, nr_pages);
1505 return ERR_PTR(-ENOMEM);
1509 unsigned int bytes = PAGE_SIZE;
1514 page = alloc_page(q->bounce_gfp | gfp_mask);
1519 memcpy(page_address(page), p, bytes);
1521 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1529 bio->bi_end_io = bio_copy_kern_endio_read;
1530 bio->bi_private = data;
1532 bio->bi_end_io = bio_copy_kern_endio;
1533 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1539 bio_free_pages(bio);
1541 return ERR_PTR(-ENOMEM);
1545 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1546 * for performing direct-IO in BIOs.
1548 * The problem is that we cannot run set_page_dirty() from interrupt context
1549 * because the required locks are not interrupt-safe. So what we can do is to
1550 * mark the pages dirty _before_ performing IO. And in interrupt context,
1551 * check that the pages are still dirty. If so, fine. If not, redirty them
1552 * in process context.
1554 * We special-case compound pages here: normally this means reads into hugetlb
1555 * pages. The logic in here doesn't really work right for compound pages
1556 * because the VM does not uniformly chase down the head page in all cases.
1557 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1558 * handle them at all. So we skip compound pages here at an early stage.
1560 * Note that this code is very hard to test under normal circumstances because
1561 * direct-io pins the pages with get_user_pages(). This makes
1562 * is_page_cache_freeable return false, and the VM will not clean the pages.
1563 * But other code (eg, flusher threads) could clean the pages if they are mapped
1566 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1567 * deferred bio dirtying paths.
1571 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1573 void bio_set_pages_dirty(struct bio *bio)
1575 struct bio_vec *bvec;
1578 bio_for_each_segment_all(bvec, bio, i) {
1579 struct page *page = bvec->bv_page;
1581 if (page && !PageCompound(page))
1582 set_page_dirty_lock(page);
1586 static void bio_release_pages(struct bio *bio)
1588 struct bio_vec *bvec;
1591 bio_for_each_segment_all(bvec, bio, i) {
1592 struct page *page = bvec->bv_page;
1600 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1601 * If they are, then fine. If, however, some pages are clean then they must
1602 * have been written out during the direct-IO read. So we take another ref on
1603 * the BIO and the offending pages and re-dirty the pages in process context.
1605 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1606 * here on. It will run one put_page() against each page and will run one
1607 * bio_put() against the BIO.
1610 static void bio_dirty_fn(struct work_struct *work);
1612 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1613 static DEFINE_SPINLOCK(bio_dirty_lock);
1614 static struct bio *bio_dirty_list;
1617 * This runs in process context
1619 static void bio_dirty_fn(struct work_struct *work)
1621 unsigned long flags;
1624 spin_lock_irqsave(&bio_dirty_lock, flags);
1625 bio = bio_dirty_list;
1626 bio_dirty_list = NULL;
1627 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1630 struct bio *next = bio->bi_private;
1632 bio_set_pages_dirty(bio);
1633 bio_release_pages(bio);
1639 void bio_check_pages_dirty(struct bio *bio)
1641 struct bio_vec *bvec;
1642 int nr_clean_pages = 0;
1645 bio_for_each_segment_all(bvec, bio, i) {
1646 struct page *page = bvec->bv_page;
1648 if (PageDirty(page) || PageCompound(page)) {
1650 bvec->bv_page = NULL;
1656 if (nr_clean_pages) {
1657 unsigned long flags;
1659 spin_lock_irqsave(&bio_dirty_lock, flags);
1660 bio->bi_private = bio_dirty_list;
1661 bio_dirty_list = bio;
1662 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1663 schedule_work(&bio_dirty_work);
1669 void generic_start_io_acct(int rw, unsigned long sectors,
1670 struct hd_struct *part)
1672 int cpu = part_stat_lock();
1674 part_round_stats(cpu, part);
1675 part_stat_inc(cpu, part, ios[rw]);
1676 part_stat_add(cpu, part, sectors[rw], sectors);
1677 part_inc_in_flight(part, rw);
1681 EXPORT_SYMBOL(generic_start_io_acct);
1683 void generic_end_io_acct(int rw, struct hd_struct *part,
1684 unsigned long start_time)
1686 unsigned long duration = jiffies - start_time;
1687 int cpu = part_stat_lock();
1689 part_stat_add(cpu, part, ticks[rw], duration);
1690 part_round_stats(cpu, part);
1691 part_dec_in_flight(part, rw);
1695 EXPORT_SYMBOL(generic_end_io_acct);
1697 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1698 void bio_flush_dcache_pages(struct bio *bi)
1700 struct bio_vec bvec;
1701 struct bvec_iter iter;
1703 bio_for_each_segment(bvec, bi, iter)
1704 flush_dcache_page(bvec.bv_page);
1706 EXPORT_SYMBOL(bio_flush_dcache_pages);
1709 static inline bool bio_remaining_done(struct bio *bio)
1712 * If we're not chaining, then ->__bi_remaining is always 1 and
1713 * we always end io on the first invocation.
1715 if (!bio_flagged(bio, BIO_CHAIN))
1718 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1720 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1721 bio_clear_flag(bio, BIO_CHAIN);
1729 * bio_endio - end I/O on a bio
1733 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1734 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1735 * bio unless they own it and thus know that it has an end_io function.
1737 void bio_endio(struct bio *bio)
1740 if (!bio_remaining_done(bio))
1744 * Need to have a real endio function for chained bios, otherwise
1745 * various corner cases will break (like stacking block devices that
1746 * save/restore bi_end_io) - however, we want to avoid unbounded
1747 * recursion and blowing the stack. Tail call optimization would
1748 * handle this, but compiling with frame pointers also disables
1749 * gcc's sibling call optimization.
1751 if (bio->bi_end_io == bio_chain_endio) {
1752 bio = __bio_chain_endio(bio);
1757 bio->bi_end_io(bio);
1759 EXPORT_SYMBOL(bio_endio);
1762 * bio_split - split a bio
1763 * @bio: bio to split
1764 * @sectors: number of sectors to split from the front of @bio
1766 * @bs: bio set to allocate from
1768 * Allocates and returns a new bio which represents @sectors from the start of
1769 * @bio, and updates @bio to represent the remaining sectors.
1771 * Unless this is a discard request the newly allocated bio will point
1772 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1773 * @bio is not freed before the split.
1775 struct bio *bio_split(struct bio *bio, int sectors,
1776 gfp_t gfp, struct bio_set *bs)
1778 struct bio *split = NULL;
1780 BUG_ON(sectors <= 0);
1781 BUG_ON(sectors >= bio_sectors(bio));
1784 * Discards need a mutable bio_vec to accommodate the payload
1785 * required by the DSM TRIM and UNMAP commands.
1787 if (bio_op(bio) == REQ_OP_DISCARD)
1788 split = bio_clone_bioset(bio, gfp, bs);
1790 split = bio_clone_fast(bio, gfp, bs);
1795 split->bi_iter.bi_size = sectors << 9;
1797 if (bio_integrity(split))
1798 bio_integrity_trim(split, 0, sectors);
1800 bio_advance(bio, split->bi_iter.bi_size);
1804 EXPORT_SYMBOL(bio_split);
1807 * bio_trim - trim a bio
1809 * @offset: number of sectors to trim from the front of @bio
1810 * @size: size we want to trim @bio to, in sectors
1812 void bio_trim(struct bio *bio, int offset, int size)
1814 /* 'bio' is a cloned bio which we need to trim to match
1815 * the given offset and size.
1819 if (offset == 0 && size == bio->bi_iter.bi_size)
1822 bio_clear_flag(bio, BIO_SEG_VALID);
1824 bio_advance(bio, offset << 9);
1826 bio->bi_iter.bi_size = size;
1828 EXPORT_SYMBOL_GPL(bio_trim);
1831 * create memory pools for biovec's in a bio_set.
1832 * use the global biovec slabs created for general use.
1834 mempool_t *biovec_create_pool(int pool_entries)
1836 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1838 return mempool_create_slab_pool(pool_entries, bp->slab);
1841 void bioset_free(struct bio_set *bs)
1843 if (bs->rescue_workqueue)
1844 destroy_workqueue(bs->rescue_workqueue);
1847 mempool_destroy(bs->bio_pool);
1850 mempool_destroy(bs->bvec_pool);
1852 bioset_integrity_free(bs);
1857 EXPORT_SYMBOL(bioset_free);
1859 static struct bio_set *__bioset_create(unsigned int pool_size,
1860 unsigned int front_pad,
1861 bool create_bvec_pool)
1863 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1866 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1870 bs->front_pad = front_pad;
1872 spin_lock_init(&bs->rescue_lock);
1873 bio_list_init(&bs->rescue_list);
1874 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1876 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1877 if (!bs->bio_slab) {
1882 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1886 if (create_bvec_pool) {
1887 bs->bvec_pool = biovec_create_pool(pool_size);
1892 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1893 if (!bs->rescue_workqueue)
1903 * bioset_create - Create a bio_set
1904 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1905 * @front_pad: Number of bytes to allocate in front of the returned bio
1908 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1909 * to ask for a number of bytes to be allocated in front of the bio.
1910 * Front pad allocation is useful for embedding the bio inside
1911 * another structure, to avoid allocating extra data to go with the bio.
1912 * Note that the bio must be embedded at the END of that structure always,
1913 * or things will break badly.
1915 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1917 return __bioset_create(pool_size, front_pad, true);
1919 EXPORT_SYMBOL(bioset_create);
1922 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1923 * @pool_size: Number of bio to cache in the mempool
1924 * @front_pad: Number of bytes to allocate in front of the returned bio
1927 * Same functionality as bioset_create() except that mempool is not
1928 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1930 struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1932 return __bioset_create(pool_size, front_pad, false);
1934 EXPORT_SYMBOL(bioset_create_nobvec);
1936 #ifdef CONFIG_BLK_CGROUP
1939 * bio_associate_blkcg - associate a bio with the specified blkcg
1941 * @blkcg_css: css of the blkcg to associate
1943 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1944 * treat @bio as if it were issued by a task which belongs to the blkcg.
1946 * This function takes an extra reference of @blkcg_css which will be put
1947 * when @bio is released. The caller must own @bio and is responsible for
1948 * synchronizing calls to this function.
1950 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1952 if (unlikely(bio->bi_css))
1955 bio->bi_css = blkcg_css;
1958 EXPORT_SYMBOL_GPL(bio_associate_blkcg);
1961 * bio_associate_current - associate a bio with %current
1964 * Associate @bio with %current if it hasn't been associated yet. Block
1965 * layer will treat @bio as if it were issued by %current no matter which
1966 * task actually issues it.
1968 * This function takes an extra reference of @task's io_context and blkcg
1969 * which will be put when @bio is released. The caller must own @bio,
1970 * ensure %current->io_context exists, and is responsible for synchronizing
1971 * calls to this function.
1973 int bio_associate_current(struct bio *bio)
1975 struct io_context *ioc;
1980 ioc = current->io_context;
1984 get_io_context_active(ioc);
1986 bio->bi_css = task_get_css(current, io_cgrp_id);
1989 EXPORT_SYMBOL_GPL(bio_associate_current);
1992 * bio_disassociate_task - undo bio_associate_current()
1995 void bio_disassociate_task(struct bio *bio)
1998 put_io_context(bio->bi_ioc);
2002 css_put(bio->bi_css);
2007 #endif /* CONFIG_BLK_CGROUP */
2009 static void __init biovec_init_slabs(void)
2013 for (i = 0; i < BVEC_POOL_NR; i++) {
2015 struct biovec_slab *bvs = bvec_slabs + i;
2017 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2022 size = bvs->nr_vecs * sizeof(struct bio_vec);
2023 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2024 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2028 static int __init init_bio(void)
2032 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2034 panic("bio: can't allocate bios\n");
2036 bio_integrity_init();
2037 biovec_init_slabs();
2039 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2041 panic("bio: can't allocate bios\n");
2043 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2044 panic("bio: can't create integrity pool\n");
2048 subsys_initcall(init_bio);