2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
31 #include <linux/blk-cgroup.h>
33 #include <trace/events/block.h>
35 #include "blk-rq-qos.h"
38 * Test patch to inline a certain number of bi_io_vec's inside the bio
39 * itself, to shrink a bio data allocation from two mempool calls to one
41 #define BIO_INLINE_VECS 4
44 * if you change this list, also change bvec_alloc or things will
45 * break badly! cannot be bigger than what you can fit into an
48 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
49 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
50 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
55 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
56 * IO code that does not need private memory pools.
58 struct bio_set fs_bio_set;
59 EXPORT_SYMBOL(fs_bio_set);
62 * Our slab pool management
65 struct kmem_cache *slab;
66 unsigned int slab_ref;
67 unsigned int slab_size;
70 static DEFINE_MUTEX(bio_slab_lock);
71 static struct bio_slab *bio_slabs;
72 static unsigned int bio_slab_nr, bio_slab_max;
74 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
76 unsigned int sz = sizeof(struct bio) + extra_size;
77 struct kmem_cache *slab = NULL;
78 struct bio_slab *bslab, *new_bio_slabs;
79 unsigned int new_bio_slab_max;
80 unsigned int i, entry = -1;
82 mutex_lock(&bio_slab_lock);
85 while (i < bio_slab_nr) {
86 bslab = &bio_slabs[i];
88 if (!bslab->slab && entry == -1)
90 else if (bslab->slab_size == sz) {
101 if (bio_slab_nr == bio_slab_max && entry == -1) {
102 new_bio_slab_max = bio_slab_max << 1;
103 new_bio_slabs = krealloc(bio_slabs,
104 new_bio_slab_max * sizeof(struct bio_slab),
108 bio_slab_max = new_bio_slab_max;
109 bio_slabs = new_bio_slabs;
112 entry = bio_slab_nr++;
114 bslab = &bio_slabs[entry];
116 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
117 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
118 SLAB_HWCACHE_ALIGN, NULL);
124 bslab->slab_size = sz;
126 mutex_unlock(&bio_slab_lock);
130 static void bio_put_slab(struct bio_set *bs)
132 struct bio_slab *bslab = NULL;
135 mutex_lock(&bio_slab_lock);
137 for (i = 0; i < bio_slab_nr; i++) {
138 if (bs->bio_slab == bio_slabs[i].slab) {
139 bslab = &bio_slabs[i];
144 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
147 WARN_ON(!bslab->slab_ref);
149 if (--bslab->slab_ref)
152 kmem_cache_destroy(bslab->slab);
156 mutex_unlock(&bio_slab_lock);
159 unsigned int bvec_nr_vecs(unsigned short idx)
161 return bvec_slabs[--idx].nr_vecs;
164 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
170 BIO_BUG_ON(idx >= BVEC_POOL_NR);
172 if (idx == BVEC_POOL_MAX) {
173 mempool_free(bv, pool);
175 struct biovec_slab *bvs = bvec_slabs + idx;
177 kmem_cache_free(bvs->slab, bv);
181 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
187 * see comment near bvec_array define!
205 case 129 ... BIO_MAX_PAGES:
213 * idx now points to the pool we want to allocate from. only the
214 * 1-vec entry pool is mempool backed.
216 if (*idx == BVEC_POOL_MAX) {
218 bvl = mempool_alloc(pool, gfp_mask);
220 struct biovec_slab *bvs = bvec_slabs + *idx;
221 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
224 * Make this allocation restricted and don't dump info on
225 * allocation failures, since we'll fallback to the mempool
226 * in case of failure.
228 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
231 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
232 * is set, retry with the 1-entry mempool
234 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
235 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
236 *idx = BVEC_POOL_MAX;
245 void bio_uninit(struct bio *bio)
247 bio_disassociate_blkg(bio);
249 EXPORT_SYMBOL(bio_uninit);
251 static void bio_free(struct bio *bio)
253 struct bio_set *bs = bio->bi_pool;
259 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
262 * If we have front padding, adjust the bio pointer before freeing
267 mempool_free(p, &bs->bio_pool);
269 /* Bio was allocated by bio_kmalloc() */
275 * Users of this function have their own bio allocation. Subsequently,
276 * they must remember to pair any call to bio_init() with bio_uninit()
277 * when IO has completed, or when the bio is released.
279 void bio_init(struct bio *bio, struct bio_vec *table,
280 unsigned short max_vecs)
282 memset(bio, 0, sizeof(*bio));
283 atomic_set(&bio->__bi_remaining, 1);
284 atomic_set(&bio->__bi_cnt, 1);
286 bio->bi_io_vec = table;
287 bio->bi_max_vecs = max_vecs;
289 EXPORT_SYMBOL(bio_init);
292 * bio_reset - reinitialize a bio
296 * After calling bio_reset(), @bio will be in the same state as a freshly
297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
298 * preserved are the ones that are initialized by bio_alloc_bioset(). See
299 * comment in struct bio.
301 void bio_reset(struct bio *bio)
303 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
307 memset(bio, 0, BIO_RESET_BYTES);
308 bio->bi_flags = flags;
309 atomic_set(&bio->__bi_remaining, 1);
311 EXPORT_SYMBOL(bio_reset);
313 static struct bio *__bio_chain_endio(struct bio *bio)
315 struct bio *parent = bio->bi_private;
317 if (!parent->bi_status)
318 parent->bi_status = bio->bi_status;
323 static void bio_chain_endio(struct bio *bio)
325 bio_endio(__bio_chain_endio(bio));
329 * bio_chain - chain bio completions
330 * @bio: the target bio
331 * @parent: the @bio's parent bio
333 * The caller won't have a bi_end_io called when @bio completes - instead,
334 * @parent's bi_end_io won't be called until both @parent and @bio have
335 * completed; the chained bio will also be freed when it completes.
337 * The caller must not set bi_private or bi_end_io in @bio.
339 void bio_chain(struct bio *bio, struct bio *parent)
341 BUG_ON(bio->bi_private || bio->bi_end_io);
343 bio->bi_private = parent;
344 bio->bi_end_io = bio_chain_endio;
345 bio_inc_remaining(parent);
347 EXPORT_SYMBOL(bio_chain);
349 static void bio_alloc_rescue(struct work_struct *work)
351 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
355 spin_lock(&bs->rescue_lock);
356 bio = bio_list_pop(&bs->rescue_list);
357 spin_unlock(&bs->rescue_lock);
362 generic_make_request(bio);
366 static void punt_bios_to_rescuer(struct bio_set *bs)
368 struct bio_list punt, nopunt;
371 if (WARN_ON_ONCE(!bs->rescue_workqueue))
374 * In order to guarantee forward progress we must punt only bios that
375 * were allocated from this bio_set; otherwise, if there was a bio on
376 * there for a stacking driver higher up in the stack, processing it
377 * could require allocating bios from this bio_set, and doing that from
378 * our own rescuer would be bad.
380 * Since bio lists are singly linked, pop them all instead of trying to
381 * remove from the middle of the list:
384 bio_list_init(&punt);
385 bio_list_init(&nopunt);
387 while ((bio = bio_list_pop(¤t->bio_list[0])))
388 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
389 current->bio_list[0] = nopunt;
391 bio_list_init(&nopunt);
392 while ((bio = bio_list_pop(¤t->bio_list[1])))
393 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
394 current->bio_list[1] = nopunt;
396 spin_lock(&bs->rescue_lock);
397 bio_list_merge(&bs->rescue_list, &punt);
398 spin_unlock(&bs->rescue_lock);
400 queue_work(bs->rescue_workqueue, &bs->rescue_work);
404 * bio_alloc_bioset - allocate a bio for I/O
405 * @gfp_mask: the GFP_* mask given to the slab allocator
406 * @nr_iovecs: number of iovecs to pre-allocate
407 * @bs: the bio_set to allocate from.
410 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
411 * backed by the @bs's mempool.
413 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
414 * always be able to allocate a bio. This is due to the mempool guarantees.
415 * To make this work, callers must never allocate more than 1 bio at a time
416 * from this pool. Callers that need to allocate more than 1 bio must always
417 * submit the previously allocated bio for IO before attempting to allocate
418 * a new one. Failure to do so can cause deadlocks under memory pressure.
420 * Note that when running under generic_make_request() (i.e. any block
421 * driver), bios are not submitted until after you return - see the code in
422 * generic_make_request() that converts recursion into iteration, to prevent
425 * This would normally mean allocating multiple bios under
426 * generic_make_request() would be susceptible to deadlocks, but we have
427 * deadlock avoidance code that resubmits any blocked bios from a rescuer
430 * However, we do not guarantee forward progress for allocations from other
431 * mempools. Doing multiple allocations from the same mempool under
432 * generic_make_request() should be avoided - instead, use bio_set's front_pad
433 * for per bio allocations.
436 * Pointer to new bio on success, NULL on failure.
438 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
441 gfp_t saved_gfp = gfp_mask;
443 unsigned inline_vecs;
444 struct bio_vec *bvl = NULL;
449 if (nr_iovecs > UIO_MAXIOV)
452 p = kmalloc(sizeof(struct bio) +
453 nr_iovecs * sizeof(struct bio_vec),
456 inline_vecs = nr_iovecs;
458 /* should not use nobvec bioset for nr_iovecs > 0 */
459 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
463 * generic_make_request() converts recursion to iteration; this
464 * means if we're running beneath it, any bios we allocate and
465 * submit will not be submitted (and thus freed) until after we
468 * This exposes us to a potential deadlock if we allocate
469 * multiple bios from the same bio_set() while running
470 * underneath generic_make_request(). If we were to allocate
471 * multiple bios (say a stacking block driver that was splitting
472 * bios), we would deadlock if we exhausted the mempool's
475 * We solve this, and guarantee forward progress, with a rescuer
476 * workqueue per bio_set. If we go to allocate and there are
477 * bios on current->bio_list, we first try the allocation
478 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
479 * bios we would be blocking to the rescuer workqueue before
480 * we retry with the original gfp_flags.
483 if (current->bio_list &&
484 (!bio_list_empty(¤t->bio_list[0]) ||
485 !bio_list_empty(¤t->bio_list[1])) &&
486 bs->rescue_workqueue)
487 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
489 p = mempool_alloc(&bs->bio_pool, gfp_mask);
490 if (!p && gfp_mask != saved_gfp) {
491 punt_bios_to_rescuer(bs);
492 gfp_mask = saved_gfp;
493 p = mempool_alloc(&bs->bio_pool, gfp_mask);
496 front_pad = bs->front_pad;
497 inline_vecs = BIO_INLINE_VECS;
504 bio_init(bio, NULL, 0);
506 if (nr_iovecs > inline_vecs) {
507 unsigned long idx = 0;
509 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
510 if (!bvl && gfp_mask != saved_gfp) {
511 punt_bios_to_rescuer(bs);
512 gfp_mask = saved_gfp;
513 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
519 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
520 } else if (nr_iovecs) {
521 bvl = bio->bi_inline_vecs;
525 bio->bi_max_vecs = nr_iovecs;
526 bio->bi_io_vec = bvl;
530 mempool_free(p, &bs->bio_pool);
533 EXPORT_SYMBOL(bio_alloc_bioset);
535 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
539 struct bvec_iter iter;
541 __bio_for_each_segment(bv, bio, iter, start) {
542 char *data = bvec_kmap_irq(&bv, &flags);
543 memset(data, 0, bv.bv_len);
544 flush_dcache_page(bv.bv_page);
545 bvec_kunmap_irq(data, &flags);
548 EXPORT_SYMBOL(zero_fill_bio_iter);
551 * bio_put - release a reference to a bio
552 * @bio: bio to release reference to
555 * Put a reference to a &struct bio, either one you have gotten with
556 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
558 void bio_put(struct bio *bio)
560 if (!bio_flagged(bio, BIO_REFFED))
563 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
568 if (atomic_dec_and_test(&bio->__bi_cnt))
572 EXPORT_SYMBOL(bio_put);
574 int bio_phys_segments(struct request_queue *q, struct bio *bio)
576 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
577 blk_recount_segments(q, bio);
579 return bio->bi_phys_segments;
583 * __bio_clone_fast - clone a bio that shares the original bio's biovec
584 * @bio: destination bio
585 * @bio_src: bio to clone
587 * Clone a &bio. Caller will own the returned bio, but not
588 * the actual data it points to. Reference count of returned
591 * Caller must ensure that @bio_src is not freed before @bio.
593 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
595 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
598 * most users will be overriding ->bi_disk with a new target,
599 * so we don't set nor calculate new physical/hw segment counts here
601 bio->bi_disk = bio_src->bi_disk;
602 bio->bi_partno = bio_src->bi_partno;
603 bio_set_flag(bio, BIO_CLONED);
604 if (bio_flagged(bio_src, BIO_THROTTLED))
605 bio_set_flag(bio, BIO_THROTTLED);
606 bio->bi_opf = bio_src->bi_opf;
607 bio->bi_ioprio = bio_src->bi_ioprio;
608 bio->bi_write_hint = bio_src->bi_write_hint;
609 bio->bi_iter = bio_src->bi_iter;
610 bio->bi_io_vec = bio_src->bi_io_vec;
612 bio_clone_blkg_association(bio, bio_src);
613 blkcg_bio_issue_init(bio);
615 EXPORT_SYMBOL(__bio_clone_fast);
618 * bio_clone_fast - clone a bio that shares the original bio's biovec
620 * @gfp_mask: allocation priority
621 * @bs: bio_set to allocate from
623 * Like __bio_clone_fast, only also allocates the returned bio
625 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
629 b = bio_alloc_bioset(gfp_mask, 0, bs);
633 __bio_clone_fast(b, bio);
635 if (bio_integrity(bio)) {
638 ret = bio_integrity_clone(b, bio, gfp_mask);
648 EXPORT_SYMBOL(bio_clone_fast);
650 static inline bool page_is_mergeable(const struct bio_vec *bv,
651 struct page *page, unsigned int len, unsigned int off,
654 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) +
655 bv->bv_offset + bv->bv_len - 1;
656 phys_addr_t page_addr = page_to_phys(page);
658 if (vec_end_addr + 1 != page_addr + off)
660 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
663 if ((vec_end_addr & PAGE_MASK) != page_addr) {
666 if (pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page)
670 WARN_ON_ONCE(same_page && (len + off) > PAGE_SIZE);
676 * Check if the @page can be added to the current segment(@bv), and make
677 * sure to call it only if page_is_mergeable(@bv, @page) is true
679 static bool can_add_page_to_seg(struct request_queue *q,
680 struct bio_vec *bv, struct page *page, unsigned len,
683 unsigned long mask = queue_segment_boundary(q);
684 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
685 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
687 if ((addr1 | mask) != (addr2 | mask))
690 if (bv->bv_len + len > queue_max_segment_size(q))
697 * __bio_add_pc_page - attempt to add page to passthrough bio
698 * @q: the target queue
699 * @bio: destination bio
701 * @len: vec entry length
702 * @offset: vec entry offset
703 * @put_same_page: put the page if it is same with last added page
705 * Attempt to add a page to the bio_vec maplist. This can fail for a
706 * number of reasons, such as the bio being full or target block device
707 * limitations. The target block device must allow bio's up to PAGE_SIZE,
708 * so it is always possible to add a single page to an empty bio.
710 * This should only be used by passthrough bios.
712 static int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
713 struct page *page, unsigned int len, unsigned int offset,
716 struct bio_vec *bvec;
719 * cloned bio must not modify vec list
721 if (unlikely(bio_flagged(bio, BIO_CLONED)))
724 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
727 if (bio->bi_vcnt > 0) {
728 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
730 if (page == bvec->bv_page &&
731 offset == bvec->bv_offset + bvec->bv_len) {
736 bio->bi_iter.bi_size += len;
741 * If the queue doesn't support SG gaps and adding this
742 * offset would create a gap, disallow it.
744 if (bvec_gap_to_prev(q, bvec, offset))
747 if (page_is_mergeable(bvec, page, len, offset, false) &&
748 can_add_page_to_seg(q, bvec, page, len, offset))
755 if (bio->bi_phys_segments >= queue_max_segments(q))
758 bvec = &bio->bi_io_vec[bio->bi_vcnt];
759 bvec->bv_page = page;
761 bvec->bv_offset = offset;
763 bio->bi_iter.bi_size += len;
766 bio->bi_phys_segments = bio->bi_vcnt;
767 bio_set_flag(bio, BIO_SEG_VALID);
771 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
772 struct page *page, unsigned int len, unsigned int offset)
774 return __bio_add_pc_page(q, bio, page, len, offset, false);
776 EXPORT_SYMBOL(bio_add_pc_page);
779 * __bio_try_merge_page - try appending data to an existing bvec.
780 * @bio: destination bio
781 * @page: start page to add
782 * @len: length of the data to add
783 * @off: offset of the data relative to @page
784 * @same_page: if %true only merge if the new data is in the same physical
785 * page as the last segment of the bio.
787 * Try to add the data at @page + @off to the last bvec of @bio. This is a
788 * a useful optimisation for file systems with a block size smaller than the
791 * Warn if (@len, @off) crosses pages in case that @same_page is true.
793 * Return %true on success or %false on failure.
795 bool __bio_try_merge_page(struct bio *bio, struct page *page,
796 unsigned int len, unsigned int off, bool same_page)
798 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
801 if (bio->bi_vcnt > 0) {
802 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
804 if (page_is_mergeable(bv, page, len, off, same_page)) {
806 bio->bi_iter.bi_size += len;
812 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
815 * __bio_add_page - add page(s) to a bio in a new segment
816 * @bio: destination bio
817 * @page: start page to add
818 * @len: length of the data to add, may cross pages
819 * @off: offset of the data relative to @page, may cross pages
821 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
822 * that @bio has space for another bvec.
824 void __bio_add_page(struct bio *bio, struct page *page,
825 unsigned int len, unsigned int off)
827 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
829 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
830 WARN_ON_ONCE(bio_full(bio));
836 bio->bi_iter.bi_size += len;
839 EXPORT_SYMBOL_GPL(__bio_add_page);
842 * bio_add_page - attempt to add page(s) to bio
843 * @bio: destination bio
844 * @page: start page to add
845 * @len: vec entry length, may cross pages
846 * @offset: vec entry offset relative to @page, may cross pages
848 * Attempt to add page(s) to the bio_vec maplist. This will only fail
849 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
851 int bio_add_page(struct bio *bio, struct page *page,
852 unsigned int len, unsigned int offset)
854 if (!__bio_try_merge_page(bio, page, len, offset, false)) {
857 __bio_add_page(bio, page, len, offset);
861 EXPORT_SYMBOL(bio_add_page);
863 static void bio_get_pages(struct bio *bio)
865 struct bvec_iter_all iter_all;
866 struct bio_vec *bvec;
868 bio_for_each_segment_all(bvec, bio, iter_all)
869 get_page(bvec->bv_page);
872 static void bio_release_pages(struct bio *bio)
874 struct bvec_iter_all iter_all;
875 struct bio_vec *bvec;
877 bio_for_each_segment_all(bvec, bio, iter_all)
878 put_page(bvec->bv_page);
881 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter)
883 const struct bio_vec *bv = iter->bvec;
887 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len))
890 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count);
891 size = bio_add_page(bio, bv->bv_page, len,
892 bv->bv_offset + iter->iov_offset);
893 if (unlikely(size != len))
895 iov_iter_advance(iter, size);
899 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
902 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
903 * @bio: bio to add pages to
904 * @iter: iov iterator describing the region to be mapped
906 * Pins pages from *iter and appends them to @bio's bvec array. The
907 * pages will have to be released using put_page() when done.
908 * For multi-segment *iter, this function only adds pages from the
909 * the next non-empty segment of the iov iterator.
911 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
913 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
914 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
915 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
916 struct page **pages = (struct page **)bv;
922 * Move page array up in the allocated memory for the bio vecs as far as
923 * possible so that we can start filling biovecs from the beginning
924 * without overwriting the temporary page array.
926 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
927 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
929 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
930 if (unlikely(size <= 0))
931 return size ? size : -EFAULT;
933 for (left = size, i = 0; left > 0; left -= len, i++) {
934 struct page *page = pages[i];
936 len = min_t(size_t, PAGE_SIZE - offset, left);
937 if (WARN_ON_ONCE(bio_add_page(bio, page, len, offset) != len))
942 iov_iter_advance(iter, size);
947 * bio_iov_iter_get_pages - add user or kernel pages to a bio
948 * @bio: bio to add pages to
949 * @iter: iov iterator describing the region to be added
951 * This takes either an iterator pointing to user memory, or one pointing to
952 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
953 * map them into the kernel. On IO completion, the caller should put those
954 * pages. If we're adding kernel pages, and the caller told us it's safe to
955 * do so, we just have to add the pages to the bio directly. We don't grab an
956 * extra reference to those pages (the user should already have that), and we
957 * don't put the page on IO completion. The caller needs to check if the bio is
958 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
961 * The function tries, but does not guarantee, to pin as many pages as
962 * fit into the bio, or are requested in *iter, whatever is smaller. If
963 * MM encounters an error pinning the requested pages, it stops. Error
964 * is returned only if 0 pages could be pinned.
966 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
968 const bool is_bvec = iov_iter_is_bvec(iter);
971 if (WARN_ON_ONCE(bio->bi_vcnt))
976 ret = __bio_iov_bvec_add_pages(bio, iter);
978 ret = __bio_iov_iter_get_pages(bio, iter);
979 } while (!ret && iov_iter_count(iter) && !bio_full(bio));
981 if (iov_iter_bvec_no_ref(iter))
982 bio_set_flag(bio, BIO_NO_PAGE_REF);
986 return bio->bi_vcnt ? 0 : ret;
989 static void submit_bio_wait_endio(struct bio *bio)
991 complete(bio->bi_private);
995 * submit_bio_wait - submit a bio, and wait until it completes
996 * @bio: The &struct bio which describes the I/O
998 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
999 * bio_endio() on failure.
1001 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1002 * result in bio reference to be consumed. The caller must drop the reference
1005 int submit_bio_wait(struct bio *bio)
1007 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
1009 bio->bi_private = &done;
1010 bio->bi_end_io = submit_bio_wait_endio;
1011 bio->bi_opf |= REQ_SYNC;
1013 wait_for_completion_io(&done);
1015 return blk_status_to_errno(bio->bi_status);
1017 EXPORT_SYMBOL(submit_bio_wait);
1020 * bio_advance - increment/complete a bio by some number of bytes
1021 * @bio: bio to advance
1022 * @bytes: number of bytes to complete
1024 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1025 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1026 * be updated on the last bvec as well.
1028 * @bio will then represent the remaining, uncompleted portion of the io.
1030 void bio_advance(struct bio *bio, unsigned bytes)
1032 if (bio_integrity(bio))
1033 bio_integrity_advance(bio, bytes);
1035 bio_advance_iter(bio, &bio->bi_iter, bytes);
1037 EXPORT_SYMBOL(bio_advance);
1039 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1040 struct bio *src, struct bvec_iter *src_iter)
1042 struct bio_vec src_bv, dst_bv;
1043 void *src_p, *dst_p;
1046 while (src_iter->bi_size && dst_iter->bi_size) {
1047 src_bv = bio_iter_iovec(src, *src_iter);
1048 dst_bv = bio_iter_iovec(dst, *dst_iter);
1050 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1052 src_p = kmap_atomic(src_bv.bv_page);
1053 dst_p = kmap_atomic(dst_bv.bv_page);
1055 memcpy(dst_p + dst_bv.bv_offset,
1056 src_p + src_bv.bv_offset,
1059 kunmap_atomic(dst_p);
1060 kunmap_atomic(src_p);
1062 flush_dcache_page(dst_bv.bv_page);
1064 bio_advance_iter(src, src_iter, bytes);
1065 bio_advance_iter(dst, dst_iter, bytes);
1068 EXPORT_SYMBOL(bio_copy_data_iter);
1071 * bio_copy_data - copy contents of data buffers from one bio to another
1073 * @dst: destination bio
1075 * Stops when it reaches the end of either @src or @dst - that is, copies
1076 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1078 void bio_copy_data(struct bio *dst, struct bio *src)
1080 struct bvec_iter src_iter = src->bi_iter;
1081 struct bvec_iter dst_iter = dst->bi_iter;
1083 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1085 EXPORT_SYMBOL(bio_copy_data);
1088 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1090 * @src: source bio list
1091 * @dst: destination bio list
1093 * Stops when it reaches the end of either the @src list or @dst list - that is,
1094 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1097 void bio_list_copy_data(struct bio *dst, struct bio *src)
1099 struct bvec_iter src_iter = src->bi_iter;
1100 struct bvec_iter dst_iter = dst->bi_iter;
1103 if (!src_iter.bi_size) {
1108 src_iter = src->bi_iter;
1111 if (!dst_iter.bi_size) {
1116 dst_iter = dst->bi_iter;
1119 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1122 EXPORT_SYMBOL(bio_list_copy_data);
1124 struct bio_map_data {
1126 struct iov_iter iter;
1130 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
1133 struct bio_map_data *bmd;
1134 if (data->nr_segs > UIO_MAXIOV)
1137 bmd = kmalloc(sizeof(struct bio_map_data) +
1138 sizeof(struct iovec) * data->nr_segs, gfp_mask);
1141 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1143 bmd->iter.iov = bmd->iov;
1148 * bio_copy_from_iter - copy all pages from iov_iter to bio
1149 * @bio: The &struct bio which describes the I/O as destination
1150 * @iter: iov_iter as source
1152 * Copy all pages from iov_iter to bio.
1153 * Returns 0 on success, or error on failure.
1155 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
1157 struct bio_vec *bvec;
1158 struct bvec_iter_all iter_all;
1160 bio_for_each_segment_all(bvec, bio, iter_all) {
1163 ret = copy_page_from_iter(bvec->bv_page,
1168 if (!iov_iter_count(iter))
1171 if (ret < bvec->bv_len)
1179 * bio_copy_to_iter - copy all pages from bio to iov_iter
1180 * @bio: The &struct bio which describes the I/O as source
1181 * @iter: iov_iter as destination
1183 * Copy all pages from bio to iov_iter.
1184 * Returns 0 on success, or error on failure.
1186 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1188 struct bio_vec *bvec;
1189 struct bvec_iter_all iter_all;
1191 bio_for_each_segment_all(bvec, bio, iter_all) {
1194 ret = copy_page_to_iter(bvec->bv_page,
1199 if (!iov_iter_count(&iter))
1202 if (ret < bvec->bv_len)
1209 void bio_free_pages(struct bio *bio)
1211 struct bio_vec *bvec;
1212 struct bvec_iter_all iter_all;
1214 bio_for_each_segment_all(bvec, bio, iter_all)
1215 __free_page(bvec->bv_page);
1217 EXPORT_SYMBOL(bio_free_pages);
1220 * bio_uncopy_user - finish previously mapped bio
1221 * @bio: bio being terminated
1223 * Free pages allocated from bio_copy_user_iov() and write back data
1224 * to user space in case of a read.
1226 int bio_uncopy_user(struct bio *bio)
1228 struct bio_map_data *bmd = bio->bi_private;
1231 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1233 * if we're in a workqueue, the request is orphaned, so
1234 * don't copy into a random user address space, just free
1235 * and return -EINTR so user space doesn't expect any data.
1239 else if (bio_data_dir(bio) == READ)
1240 ret = bio_copy_to_iter(bio, bmd->iter);
1241 if (bmd->is_our_pages)
1242 bio_free_pages(bio);
1250 * bio_copy_user_iov - copy user data to bio
1251 * @q: destination block queue
1252 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1253 * @iter: iovec iterator
1254 * @gfp_mask: memory allocation flags
1256 * Prepares and returns a bio for indirect user io, bouncing data
1257 * to/from kernel pages as necessary. Must be paired with
1258 * call bio_uncopy_user() on io completion.
1260 struct bio *bio_copy_user_iov(struct request_queue *q,
1261 struct rq_map_data *map_data,
1262 struct iov_iter *iter,
1265 struct bio_map_data *bmd;
1270 unsigned int len = iter->count;
1271 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1273 bmd = bio_alloc_map_data(iter, gfp_mask);
1275 return ERR_PTR(-ENOMEM);
1278 * We need to do a deep copy of the iov_iter including the iovecs.
1279 * The caller provided iov might point to an on-stack or otherwise
1282 bmd->is_our_pages = map_data ? 0 : 1;
1284 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1285 if (nr_pages > BIO_MAX_PAGES)
1286 nr_pages = BIO_MAX_PAGES;
1289 bio = bio_kmalloc(gfp_mask, nr_pages);
1296 nr_pages = 1 << map_data->page_order;
1297 i = map_data->offset / PAGE_SIZE;
1300 unsigned int bytes = PAGE_SIZE;
1308 if (i == map_data->nr_entries * nr_pages) {
1313 page = map_data->pages[i / nr_pages];
1314 page += (i % nr_pages);
1318 page = alloc_page(q->bounce_gfp | gfp_mask);
1325 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) {
1339 map_data->offset += bio->bi_iter.bi_size;
1344 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) ||
1345 (map_data && map_data->from_user)) {
1346 ret = bio_copy_from_iter(bio, iter);
1350 if (bmd->is_our_pages)
1352 iov_iter_advance(iter, bio->bi_iter.bi_size);
1355 bio->bi_private = bmd;
1356 if (map_data && map_data->null_mapped)
1357 bio_set_flag(bio, BIO_NULL_MAPPED);
1361 bio_free_pages(bio);
1365 return ERR_PTR(ret);
1369 * bio_map_user_iov - map user iovec into bio
1370 * @q: the struct request_queue for the bio
1371 * @iter: iovec iterator
1372 * @gfp_mask: memory allocation flags
1374 * Map the user space address into a bio suitable for io to a block
1375 * device. Returns an error pointer in case of error.
1377 struct bio *bio_map_user_iov(struct request_queue *q,
1378 struct iov_iter *iter,
1384 struct bio_vec *bvec;
1385 struct bvec_iter_all iter_all;
1387 if (!iov_iter_count(iter))
1388 return ERR_PTR(-EINVAL);
1390 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
1392 return ERR_PTR(-ENOMEM);
1394 while (iov_iter_count(iter)) {
1395 struct page **pages;
1397 size_t offs, added = 0;
1400 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
1401 if (unlikely(bytes <= 0)) {
1402 ret = bytes ? bytes : -EFAULT;
1406 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
1408 if (unlikely(offs & queue_dma_alignment(q))) {
1412 for (j = 0; j < npages; j++) {
1413 struct page *page = pages[j];
1414 unsigned int n = PAGE_SIZE - offs;
1419 if (!__bio_add_pc_page(q, bio, page, n, offs,
1427 iov_iter_advance(iter, added);
1430 * release the pages we didn't map into the bio, if any
1433 put_page(pages[j++]);
1435 /* couldn't stuff something into bio? */
1440 bio_set_flag(bio, BIO_USER_MAPPED);
1443 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1444 * it would normally disappear when its bi_end_io is run.
1445 * however, we need it for the unmap, so grab an extra
1452 bio_for_each_segment_all(bvec, bio, iter_all) {
1453 put_page(bvec->bv_page);
1456 return ERR_PTR(ret);
1459 static void __bio_unmap_user(struct bio *bio)
1461 struct bio_vec *bvec;
1462 struct bvec_iter_all iter_all;
1465 * make sure we dirty pages we wrote to
1467 bio_for_each_segment_all(bvec, bio, iter_all) {
1468 if (bio_data_dir(bio) == READ)
1469 set_page_dirty_lock(bvec->bv_page);
1471 put_page(bvec->bv_page);
1478 * bio_unmap_user - unmap a bio
1479 * @bio: the bio being unmapped
1481 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1484 * bio_unmap_user() may sleep.
1486 void bio_unmap_user(struct bio *bio)
1488 __bio_unmap_user(bio);
1492 static void bio_map_kern_endio(struct bio *bio)
1498 * bio_map_kern - map kernel address into bio
1499 * @q: the struct request_queue for the bio
1500 * @data: pointer to buffer to map
1501 * @len: length in bytes
1502 * @gfp_mask: allocation flags for bio allocation
1504 * Map the kernel address into a bio suitable for io to a block
1505 * device. Returns an error pointer in case of error.
1507 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1510 unsigned long kaddr = (unsigned long)data;
1511 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1512 unsigned long start = kaddr >> PAGE_SHIFT;
1513 const int nr_pages = end - start;
1517 bio = bio_kmalloc(gfp_mask, nr_pages);
1519 return ERR_PTR(-ENOMEM);
1521 offset = offset_in_page(kaddr);
1522 for (i = 0; i < nr_pages; i++) {
1523 unsigned int bytes = PAGE_SIZE - offset;
1531 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1533 /* we don't support partial mappings */
1535 return ERR_PTR(-EINVAL);
1543 bio->bi_end_io = bio_map_kern_endio;
1546 EXPORT_SYMBOL(bio_map_kern);
1548 static void bio_copy_kern_endio(struct bio *bio)
1550 bio_free_pages(bio);
1554 static void bio_copy_kern_endio_read(struct bio *bio)
1556 char *p = bio->bi_private;
1557 struct bio_vec *bvec;
1558 struct bvec_iter_all iter_all;
1560 bio_for_each_segment_all(bvec, bio, iter_all) {
1561 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1565 bio_copy_kern_endio(bio);
1569 * bio_copy_kern - copy kernel address into bio
1570 * @q: the struct request_queue for the bio
1571 * @data: pointer to buffer to copy
1572 * @len: length in bytes
1573 * @gfp_mask: allocation flags for bio and page allocation
1574 * @reading: data direction is READ
1576 * copy the kernel address into a bio suitable for io to a block
1577 * device. Returns an error pointer in case of error.
1579 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1580 gfp_t gfp_mask, int reading)
1582 unsigned long kaddr = (unsigned long)data;
1583 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1584 unsigned long start = kaddr >> PAGE_SHIFT;
1593 return ERR_PTR(-EINVAL);
1595 nr_pages = end - start;
1596 bio = bio_kmalloc(gfp_mask, nr_pages);
1598 return ERR_PTR(-ENOMEM);
1602 unsigned int bytes = PAGE_SIZE;
1607 page = alloc_page(q->bounce_gfp | gfp_mask);
1612 memcpy(page_address(page), p, bytes);
1614 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1622 bio->bi_end_io = bio_copy_kern_endio_read;
1623 bio->bi_private = data;
1625 bio->bi_end_io = bio_copy_kern_endio;
1631 bio_free_pages(bio);
1633 return ERR_PTR(-ENOMEM);
1637 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1638 * for performing direct-IO in BIOs.
1640 * The problem is that we cannot run set_page_dirty() from interrupt context
1641 * because the required locks are not interrupt-safe. So what we can do is to
1642 * mark the pages dirty _before_ performing IO. And in interrupt context,
1643 * check that the pages are still dirty. If so, fine. If not, redirty them
1644 * in process context.
1646 * We special-case compound pages here: normally this means reads into hugetlb
1647 * pages. The logic in here doesn't really work right for compound pages
1648 * because the VM does not uniformly chase down the head page in all cases.
1649 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1650 * handle them at all. So we skip compound pages here at an early stage.
1652 * Note that this code is very hard to test under normal circumstances because
1653 * direct-io pins the pages with get_user_pages(). This makes
1654 * is_page_cache_freeable return false, and the VM will not clean the pages.
1655 * But other code (eg, flusher threads) could clean the pages if they are mapped
1658 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1659 * deferred bio dirtying paths.
1663 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1665 void bio_set_pages_dirty(struct bio *bio)
1667 struct bio_vec *bvec;
1668 struct bvec_iter_all iter_all;
1670 bio_for_each_segment_all(bvec, bio, iter_all) {
1671 if (!PageCompound(bvec->bv_page))
1672 set_page_dirty_lock(bvec->bv_page);
1677 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1678 * If they are, then fine. If, however, some pages are clean then they must
1679 * have been written out during the direct-IO read. So we take another ref on
1680 * the BIO and re-dirty the pages in process context.
1682 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1683 * here on. It will run one put_page() against each page and will run one
1684 * bio_put() against the BIO.
1687 static void bio_dirty_fn(struct work_struct *work);
1689 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1690 static DEFINE_SPINLOCK(bio_dirty_lock);
1691 static struct bio *bio_dirty_list;
1694 * This runs in process context
1696 static void bio_dirty_fn(struct work_struct *work)
1698 struct bio *bio, *next;
1700 spin_lock_irq(&bio_dirty_lock);
1701 next = bio_dirty_list;
1702 bio_dirty_list = NULL;
1703 spin_unlock_irq(&bio_dirty_lock);
1705 while ((bio = next) != NULL) {
1706 next = bio->bi_private;
1708 bio_set_pages_dirty(bio);
1709 if (!bio_flagged(bio, BIO_NO_PAGE_REF))
1710 bio_release_pages(bio);
1715 void bio_check_pages_dirty(struct bio *bio)
1717 struct bio_vec *bvec;
1718 unsigned long flags;
1719 struct bvec_iter_all iter_all;
1721 bio_for_each_segment_all(bvec, bio, iter_all) {
1722 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1726 if (!bio_flagged(bio, BIO_NO_PAGE_REF))
1727 bio_release_pages(bio);
1731 spin_lock_irqsave(&bio_dirty_lock, flags);
1732 bio->bi_private = bio_dirty_list;
1733 bio_dirty_list = bio;
1734 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1735 schedule_work(&bio_dirty_work);
1738 void update_io_ticks(struct hd_struct *part, unsigned long now)
1740 unsigned long stamp;
1742 stamp = READ_ONCE(part->stamp);
1743 if (unlikely(stamp != now)) {
1744 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) {
1745 __part_stat_add(part, io_ticks, 1);
1749 part = &part_to_disk(part)->part0;
1754 void generic_start_io_acct(struct request_queue *q, int op,
1755 unsigned long sectors, struct hd_struct *part)
1757 const int sgrp = op_stat_group(op);
1761 update_io_ticks(part, jiffies);
1762 part_stat_inc(part, ios[sgrp]);
1763 part_stat_add(part, sectors[sgrp], sectors);
1764 part_inc_in_flight(q, part, op_is_write(op));
1768 EXPORT_SYMBOL(generic_start_io_acct);
1770 void generic_end_io_acct(struct request_queue *q, int req_op,
1771 struct hd_struct *part, unsigned long start_time)
1773 unsigned long now = jiffies;
1774 unsigned long duration = now - start_time;
1775 const int sgrp = op_stat_group(req_op);
1779 update_io_ticks(part, now);
1780 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1781 part_stat_add(part, time_in_queue, duration);
1782 part_dec_in_flight(q, part, op_is_write(req_op));
1786 EXPORT_SYMBOL(generic_end_io_acct);
1788 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1789 void bio_flush_dcache_pages(struct bio *bi)
1791 struct bio_vec bvec;
1792 struct bvec_iter iter;
1794 bio_for_each_segment(bvec, bi, iter)
1795 flush_dcache_page(bvec.bv_page);
1797 EXPORT_SYMBOL(bio_flush_dcache_pages);
1800 static inline bool bio_remaining_done(struct bio *bio)
1803 * If we're not chaining, then ->__bi_remaining is always 1 and
1804 * we always end io on the first invocation.
1806 if (!bio_flagged(bio, BIO_CHAIN))
1809 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1811 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1812 bio_clear_flag(bio, BIO_CHAIN);
1820 * bio_endio - end I/O on a bio
1824 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1825 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1826 * bio unless they own it and thus know that it has an end_io function.
1828 * bio_endio() can be called several times on a bio that has been chained
1829 * using bio_chain(). The ->bi_end_io() function will only be called the
1830 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1831 * generated if BIO_TRACE_COMPLETION is set.
1833 void bio_endio(struct bio *bio)
1836 if (!bio_remaining_done(bio))
1838 if (!bio_integrity_endio(bio))
1842 rq_qos_done_bio(bio->bi_disk->queue, bio);
1845 * Need to have a real endio function for chained bios, otherwise
1846 * various corner cases will break (like stacking block devices that
1847 * save/restore bi_end_io) - however, we want to avoid unbounded
1848 * recursion and blowing the stack. Tail call optimization would
1849 * handle this, but compiling with frame pointers also disables
1850 * gcc's sibling call optimization.
1852 if (bio->bi_end_io == bio_chain_endio) {
1853 bio = __bio_chain_endio(bio);
1857 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1858 trace_block_bio_complete(bio->bi_disk->queue, bio,
1859 blk_status_to_errno(bio->bi_status));
1860 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1863 blk_throtl_bio_endio(bio);
1864 /* release cgroup info */
1867 bio->bi_end_io(bio);
1869 EXPORT_SYMBOL(bio_endio);
1872 * bio_split - split a bio
1873 * @bio: bio to split
1874 * @sectors: number of sectors to split from the front of @bio
1876 * @bs: bio set to allocate from
1878 * Allocates and returns a new bio which represents @sectors from the start of
1879 * @bio, and updates @bio to represent the remaining sectors.
1881 * Unless this is a discard request the newly allocated bio will point
1882 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1883 * @bio is not freed before the split.
1885 struct bio *bio_split(struct bio *bio, int sectors,
1886 gfp_t gfp, struct bio_set *bs)
1890 BUG_ON(sectors <= 0);
1891 BUG_ON(sectors >= bio_sectors(bio));
1893 split = bio_clone_fast(bio, gfp, bs);
1897 split->bi_iter.bi_size = sectors << 9;
1899 if (bio_integrity(split))
1900 bio_integrity_trim(split);
1902 bio_advance(bio, split->bi_iter.bi_size);
1904 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1905 bio_set_flag(split, BIO_TRACE_COMPLETION);
1909 EXPORT_SYMBOL(bio_split);
1912 * bio_trim - trim a bio
1914 * @offset: number of sectors to trim from the front of @bio
1915 * @size: size we want to trim @bio to, in sectors
1917 void bio_trim(struct bio *bio, int offset, int size)
1919 /* 'bio' is a cloned bio which we need to trim to match
1920 * the given offset and size.
1924 if (offset == 0 && size == bio->bi_iter.bi_size)
1927 bio_clear_flag(bio, BIO_SEG_VALID);
1929 bio_advance(bio, offset << 9);
1931 bio->bi_iter.bi_size = size;
1933 if (bio_integrity(bio))
1934 bio_integrity_trim(bio);
1937 EXPORT_SYMBOL_GPL(bio_trim);
1940 * create memory pools for biovec's in a bio_set.
1941 * use the global biovec slabs created for general use.
1943 int biovec_init_pool(mempool_t *pool, int pool_entries)
1945 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1947 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1951 * bioset_exit - exit a bioset initialized with bioset_init()
1953 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1956 void bioset_exit(struct bio_set *bs)
1958 if (bs->rescue_workqueue)
1959 destroy_workqueue(bs->rescue_workqueue);
1960 bs->rescue_workqueue = NULL;
1962 mempool_exit(&bs->bio_pool);
1963 mempool_exit(&bs->bvec_pool);
1965 bioset_integrity_free(bs);
1968 bs->bio_slab = NULL;
1970 EXPORT_SYMBOL(bioset_exit);
1973 * bioset_init - Initialize a bio_set
1974 * @bs: pool to initialize
1975 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1976 * @front_pad: Number of bytes to allocate in front of the returned bio
1977 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1978 * and %BIOSET_NEED_RESCUER
1981 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1982 * to ask for a number of bytes to be allocated in front of the bio.
1983 * Front pad allocation is useful for embedding the bio inside
1984 * another structure, to avoid allocating extra data to go with the bio.
1985 * Note that the bio must be embedded at the END of that structure always,
1986 * or things will break badly.
1987 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1988 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1989 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1990 * dispatch queued requests when the mempool runs out of space.
1993 int bioset_init(struct bio_set *bs,
1994 unsigned int pool_size,
1995 unsigned int front_pad,
1998 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
2000 bs->front_pad = front_pad;
2002 spin_lock_init(&bs->rescue_lock);
2003 bio_list_init(&bs->rescue_list);
2004 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
2006 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
2010 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
2013 if ((flags & BIOSET_NEED_BVECS) &&
2014 biovec_init_pool(&bs->bvec_pool, pool_size))
2017 if (!(flags & BIOSET_NEED_RESCUER))
2020 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
2021 if (!bs->rescue_workqueue)
2029 EXPORT_SYMBOL(bioset_init);
2032 * Initialize and setup a new bio_set, based on the settings from
2035 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
2040 if (src->bvec_pool.min_nr)
2041 flags |= BIOSET_NEED_BVECS;
2042 if (src->rescue_workqueue)
2043 flags |= BIOSET_NEED_RESCUER;
2045 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
2047 EXPORT_SYMBOL(bioset_init_from_src);
2049 #ifdef CONFIG_BLK_CGROUP
2052 * bio_disassociate_blkg - puts back the blkg reference if associated
2055 * Helper to disassociate the blkg from @bio if a blkg is associated.
2057 void bio_disassociate_blkg(struct bio *bio)
2060 blkg_put(bio->bi_blkg);
2061 bio->bi_blkg = NULL;
2064 EXPORT_SYMBOL_GPL(bio_disassociate_blkg);
2067 * __bio_associate_blkg - associate a bio with the a blkg
2069 * @blkg: the blkg to associate
2071 * This tries to associate @bio with the specified @blkg. Association failure
2072 * is handled by walking up the blkg tree. Therefore, the blkg associated can
2073 * be anything between @blkg and the root_blkg. This situation only happens
2074 * when a cgroup is dying and then the remaining bios will spill to the closest
2077 * A reference will be taken on the @blkg and will be released when @bio is
2080 static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg)
2082 bio_disassociate_blkg(bio);
2084 bio->bi_blkg = blkg_tryget_closest(blkg);
2088 * bio_associate_blkg_from_css - associate a bio with a specified css
2092 * Associate @bio with the blkg found by combining the css's blkg and the
2093 * request_queue of the @bio. This falls back to the queue's root_blkg if
2094 * the association fails with the css.
2096 void bio_associate_blkg_from_css(struct bio *bio,
2097 struct cgroup_subsys_state *css)
2099 struct request_queue *q = bio->bi_disk->queue;
2100 struct blkcg_gq *blkg;
2104 if (!css || !css->parent)
2105 blkg = q->root_blkg;
2107 blkg = blkg_lookup_create(css_to_blkcg(css), q);
2109 __bio_associate_blkg(bio, blkg);
2113 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css);
2117 * bio_associate_blkg_from_page - associate a bio with the page's blkg
2119 * @page: the page to lookup the blkcg from
2121 * Associate @bio with the blkg from @page's owning memcg and the respective
2122 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's
2125 void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
2127 struct cgroup_subsys_state *css;
2129 if (!page->mem_cgroup)
2134 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
2135 bio_associate_blkg_from_css(bio, css);
2139 #endif /* CONFIG_MEMCG */
2142 * bio_associate_blkg - associate a bio with a blkg
2145 * Associate @bio with the blkg found from the bio's css and request_queue.
2146 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is
2147 * already associated, the css is reused and association redone as the
2148 * request_queue may have changed.
2150 void bio_associate_blkg(struct bio *bio)
2152 struct cgroup_subsys_state *css;
2157 css = &bio_blkcg(bio)->css;
2161 bio_associate_blkg_from_css(bio, css);
2165 EXPORT_SYMBOL_GPL(bio_associate_blkg);
2168 * bio_clone_blkg_association - clone blkg association from src to dst bio
2169 * @dst: destination bio
2172 void bio_clone_blkg_association(struct bio *dst, struct bio *src)
2177 __bio_associate_blkg(dst, src->bi_blkg);
2181 EXPORT_SYMBOL_GPL(bio_clone_blkg_association);
2182 #endif /* CONFIG_BLK_CGROUP */
2184 static void __init biovec_init_slabs(void)
2188 for (i = 0; i < BVEC_POOL_NR; i++) {
2190 struct biovec_slab *bvs = bvec_slabs + i;
2192 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2197 size = bvs->nr_vecs * sizeof(struct bio_vec);
2198 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2199 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2203 static int __init init_bio(void)
2207 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab),
2210 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET);
2213 panic("bio: can't allocate bios\n");
2215 bio_integrity_init();
2216 biovec_init_slabs();
2218 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
2219 panic("bio: can't allocate bios\n");
2221 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
2222 panic("bio: can't create integrity pool\n");
2226 subsys_initcall(init_bio);