1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
6 #include <linux/swap.h>
8 #include <linux/blkdev.h>
10 #include <linux/iocontext.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/export.h>
15 #include <linux/mempool.h>
16 #include <linux/workqueue.h>
17 #include <linux/cgroup.h>
18 #include <linux/blk-cgroup.h>
19 #include <linux/highmem.h>
21 #include <trace/events/block.h>
23 #include "blk-rq-qos.h"
26 * Test patch to inline a certain number of bi_io_vec's inside the bio
27 * itself, to shrink a bio data allocation from two mempool calls to one
29 #define BIO_INLINE_VECS 4
32 * if you change this list, also change bvec_alloc or things will
33 * break badly! cannot be bigger than what you can fit into an
36 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
37 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
38 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
43 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
44 * IO code that does not need private memory pools.
46 struct bio_set fs_bio_set;
47 EXPORT_SYMBOL(fs_bio_set);
50 * Our slab pool management
53 struct kmem_cache *slab;
54 unsigned int slab_ref;
55 unsigned int slab_size;
58 static DEFINE_MUTEX(bio_slab_lock);
59 static struct bio_slab *bio_slabs;
60 static unsigned int bio_slab_nr, bio_slab_max;
62 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
64 unsigned int sz = sizeof(struct bio) + extra_size;
65 struct kmem_cache *slab = NULL;
66 struct bio_slab *bslab, *new_bio_slabs;
67 unsigned int new_bio_slab_max;
68 unsigned int i, entry = -1;
70 mutex_lock(&bio_slab_lock);
73 while (i < bio_slab_nr) {
74 bslab = &bio_slabs[i];
76 if (!bslab->slab && entry == -1)
78 else if (bslab->slab_size == sz) {
89 if (bio_slab_nr == bio_slab_max && entry == -1) {
90 new_bio_slab_max = bio_slab_max << 1;
91 new_bio_slabs = krealloc(bio_slabs,
92 new_bio_slab_max * sizeof(struct bio_slab),
96 bio_slab_max = new_bio_slab_max;
97 bio_slabs = new_bio_slabs;
100 entry = bio_slab_nr++;
102 bslab = &bio_slabs[entry];
104 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
105 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
106 SLAB_HWCACHE_ALIGN, NULL);
112 bslab->slab_size = sz;
114 mutex_unlock(&bio_slab_lock);
118 static void bio_put_slab(struct bio_set *bs)
120 struct bio_slab *bslab = NULL;
123 mutex_lock(&bio_slab_lock);
125 for (i = 0; i < bio_slab_nr; i++) {
126 if (bs->bio_slab == bio_slabs[i].slab) {
127 bslab = &bio_slabs[i];
132 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
135 WARN_ON(!bslab->slab_ref);
137 if (--bslab->slab_ref)
140 kmem_cache_destroy(bslab->slab);
144 mutex_unlock(&bio_slab_lock);
147 unsigned int bvec_nr_vecs(unsigned short idx)
149 return bvec_slabs[--idx].nr_vecs;
152 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
158 BIO_BUG_ON(idx >= BVEC_POOL_NR);
160 if (idx == BVEC_POOL_MAX) {
161 mempool_free(bv, pool);
163 struct biovec_slab *bvs = bvec_slabs + idx;
165 kmem_cache_free(bvs->slab, bv);
169 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
175 * see comment near bvec_array define!
193 case 129 ... BIO_MAX_PAGES:
201 * idx now points to the pool we want to allocate from. only the
202 * 1-vec entry pool is mempool backed.
204 if (*idx == BVEC_POOL_MAX) {
206 bvl = mempool_alloc(pool, gfp_mask);
208 struct biovec_slab *bvs = bvec_slabs + *idx;
209 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
212 * Make this allocation restricted and don't dump info on
213 * allocation failures, since we'll fallback to the mempool
214 * in case of failure.
216 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
219 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
220 * is set, retry with the 1-entry mempool
222 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
223 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
224 *idx = BVEC_POOL_MAX;
233 void bio_uninit(struct bio *bio)
235 bio_disassociate_blkg(bio);
237 if (bio_integrity(bio))
238 bio_integrity_free(bio);
240 EXPORT_SYMBOL(bio_uninit);
242 static void bio_free(struct bio *bio)
244 struct bio_set *bs = bio->bi_pool;
250 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
253 * If we have front padding, adjust the bio pointer before freeing
258 mempool_free(p, &bs->bio_pool);
260 /* Bio was allocated by bio_kmalloc() */
266 * Users of this function have their own bio allocation. Subsequently,
267 * they must remember to pair any call to bio_init() with bio_uninit()
268 * when IO has completed, or when the bio is released.
270 void bio_init(struct bio *bio, struct bio_vec *table,
271 unsigned short max_vecs)
273 memset(bio, 0, sizeof(*bio));
274 atomic_set(&bio->__bi_remaining, 1);
275 atomic_set(&bio->__bi_cnt, 1);
277 bio->bi_io_vec = table;
278 bio->bi_max_vecs = max_vecs;
280 EXPORT_SYMBOL(bio_init);
283 * bio_reset - reinitialize a bio
287 * After calling bio_reset(), @bio will be in the same state as a freshly
288 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
289 * preserved are the ones that are initialized by bio_alloc_bioset(). See
290 * comment in struct bio.
292 void bio_reset(struct bio *bio)
294 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
298 memset(bio, 0, BIO_RESET_BYTES);
299 bio->bi_flags = flags;
300 atomic_set(&bio->__bi_remaining, 1);
302 EXPORT_SYMBOL(bio_reset);
304 static struct bio *__bio_chain_endio(struct bio *bio)
306 struct bio *parent = bio->bi_private;
308 if (bio->bi_status && !parent->bi_status)
309 parent->bi_status = bio->bi_status;
314 static void bio_chain_endio(struct bio *bio)
316 bio_endio(__bio_chain_endio(bio));
320 * bio_chain - chain bio completions
321 * @bio: the target bio
322 * @parent: the @bio's parent bio
324 * The caller won't have a bi_end_io called when @bio completes - instead,
325 * @parent's bi_end_io won't be called until both @parent and @bio have
326 * completed; the chained bio will also be freed when it completes.
328 * The caller must not set bi_private or bi_end_io in @bio.
330 void bio_chain(struct bio *bio, struct bio *parent)
332 BUG_ON(bio->bi_private || bio->bi_end_io);
334 bio->bi_private = parent;
335 bio->bi_end_io = bio_chain_endio;
336 bio_inc_remaining(parent);
338 EXPORT_SYMBOL(bio_chain);
340 static void bio_alloc_rescue(struct work_struct *work)
342 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
346 spin_lock(&bs->rescue_lock);
347 bio = bio_list_pop(&bs->rescue_list);
348 spin_unlock(&bs->rescue_lock);
353 generic_make_request(bio);
357 static void punt_bios_to_rescuer(struct bio_set *bs)
359 struct bio_list punt, nopunt;
362 if (WARN_ON_ONCE(!bs->rescue_workqueue))
365 * In order to guarantee forward progress we must punt only bios that
366 * were allocated from this bio_set; otherwise, if there was a bio on
367 * there for a stacking driver higher up in the stack, processing it
368 * could require allocating bios from this bio_set, and doing that from
369 * our own rescuer would be bad.
371 * Since bio lists are singly linked, pop them all instead of trying to
372 * remove from the middle of the list:
375 bio_list_init(&punt);
376 bio_list_init(&nopunt);
378 while ((bio = bio_list_pop(¤t->bio_list[0])))
379 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
380 current->bio_list[0] = nopunt;
382 bio_list_init(&nopunt);
383 while ((bio = bio_list_pop(¤t->bio_list[1])))
384 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
385 current->bio_list[1] = nopunt;
387 spin_lock(&bs->rescue_lock);
388 bio_list_merge(&bs->rescue_list, &punt);
389 spin_unlock(&bs->rescue_lock);
391 queue_work(bs->rescue_workqueue, &bs->rescue_work);
395 * bio_alloc_bioset - allocate a bio for I/O
396 * @gfp_mask: the GFP_* mask given to the slab allocator
397 * @nr_iovecs: number of iovecs to pre-allocate
398 * @bs: the bio_set to allocate from.
401 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
402 * backed by the @bs's mempool.
404 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
405 * always be able to allocate a bio. This is due to the mempool guarantees.
406 * To make this work, callers must never allocate more than 1 bio at a time
407 * from this pool. Callers that need to allocate more than 1 bio must always
408 * submit the previously allocated bio for IO before attempting to allocate
409 * a new one. Failure to do so can cause deadlocks under memory pressure.
411 * Note that when running under generic_make_request() (i.e. any block
412 * driver), bios are not submitted until after you return - see the code in
413 * generic_make_request() that converts recursion into iteration, to prevent
416 * This would normally mean allocating multiple bios under
417 * generic_make_request() would be susceptible to deadlocks, but we have
418 * deadlock avoidance code that resubmits any blocked bios from a rescuer
421 * However, we do not guarantee forward progress for allocations from other
422 * mempools. Doing multiple allocations from the same mempool under
423 * generic_make_request() should be avoided - instead, use bio_set's front_pad
424 * for per bio allocations.
427 * Pointer to new bio on success, NULL on failure.
429 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
432 gfp_t saved_gfp = gfp_mask;
434 unsigned inline_vecs;
435 struct bio_vec *bvl = NULL;
440 if (nr_iovecs > UIO_MAXIOV)
443 p = kmalloc(sizeof(struct bio) +
444 nr_iovecs * sizeof(struct bio_vec),
447 inline_vecs = nr_iovecs;
449 /* should not use nobvec bioset for nr_iovecs > 0 */
450 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
454 * generic_make_request() converts recursion to iteration; this
455 * means if we're running beneath it, any bios we allocate and
456 * submit will not be submitted (and thus freed) until after we
459 * This exposes us to a potential deadlock if we allocate
460 * multiple bios from the same bio_set() while running
461 * underneath generic_make_request(). If we were to allocate
462 * multiple bios (say a stacking block driver that was splitting
463 * bios), we would deadlock if we exhausted the mempool's
466 * We solve this, and guarantee forward progress, with a rescuer
467 * workqueue per bio_set. If we go to allocate and there are
468 * bios on current->bio_list, we first try the allocation
469 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
470 * bios we would be blocking to the rescuer workqueue before
471 * we retry with the original gfp_flags.
474 if (current->bio_list &&
475 (!bio_list_empty(¤t->bio_list[0]) ||
476 !bio_list_empty(¤t->bio_list[1])) &&
477 bs->rescue_workqueue)
478 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
480 p = mempool_alloc(&bs->bio_pool, gfp_mask);
481 if (!p && gfp_mask != saved_gfp) {
482 punt_bios_to_rescuer(bs);
483 gfp_mask = saved_gfp;
484 p = mempool_alloc(&bs->bio_pool, gfp_mask);
487 front_pad = bs->front_pad;
488 inline_vecs = BIO_INLINE_VECS;
495 bio_init(bio, NULL, 0);
497 if (nr_iovecs > inline_vecs) {
498 unsigned long idx = 0;
500 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
501 if (!bvl && gfp_mask != saved_gfp) {
502 punt_bios_to_rescuer(bs);
503 gfp_mask = saved_gfp;
504 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
510 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
511 } else if (nr_iovecs) {
512 bvl = bio->bi_inline_vecs;
516 bio->bi_max_vecs = nr_iovecs;
517 bio->bi_io_vec = bvl;
521 mempool_free(p, &bs->bio_pool);
524 EXPORT_SYMBOL(bio_alloc_bioset);
526 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
530 struct bvec_iter iter;
532 __bio_for_each_segment(bv, bio, iter, start) {
533 char *data = bvec_kmap_irq(&bv, &flags);
534 memset(data, 0, bv.bv_len);
535 flush_dcache_page(bv.bv_page);
536 bvec_kunmap_irq(data, &flags);
539 EXPORT_SYMBOL(zero_fill_bio_iter);
542 * bio_truncate - truncate the bio to small size of @new_size
543 * @bio: the bio to be truncated
544 * @new_size: new size for truncating the bio
547 * Truncate the bio to new size of @new_size. If bio_op(bio) is
548 * REQ_OP_READ, zero the truncated part. This function should only
549 * be used for handling corner cases, such as bio eod.
551 void bio_truncate(struct bio *bio, unsigned new_size)
554 struct bvec_iter iter;
555 unsigned int done = 0;
556 bool truncated = false;
558 if (new_size >= bio->bi_iter.bi_size)
561 if (bio_op(bio) != REQ_OP_READ)
564 bio_for_each_segment(bv, bio, iter) {
565 if (done + bv.bv_len > new_size) {
569 offset = new_size - done;
572 zero_user(bv.bv_page, bv.bv_offset + offset,
581 * Don't touch bvec table here and make it really immutable, since
582 * fs bio user has to retrieve all pages via bio_for_each_segment_all
583 * in its .end_bio() callback.
585 * It is enough to truncate bio by updating .bi_size since we can make
586 * correct bvec with the updated .bi_size for drivers.
588 bio->bi_iter.bi_size = new_size;
592 * bio_put - release a reference to a bio
593 * @bio: bio to release reference to
596 * Put a reference to a &struct bio, either one you have gotten with
597 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
599 void bio_put(struct bio *bio)
601 if (!bio_flagged(bio, BIO_REFFED))
604 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
609 if (atomic_dec_and_test(&bio->__bi_cnt))
613 EXPORT_SYMBOL(bio_put);
616 * __bio_clone_fast - clone a bio that shares the original bio's biovec
617 * @bio: destination bio
618 * @bio_src: bio to clone
620 * Clone a &bio. Caller will own the returned bio, but not
621 * the actual data it points to. Reference count of returned
624 * Caller must ensure that @bio_src is not freed before @bio.
626 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
628 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
631 * most users will be overriding ->bi_disk with a new target,
632 * so we don't set nor calculate new physical/hw segment counts here
634 bio->bi_disk = bio_src->bi_disk;
635 bio->bi_partno = bio_src->bi_partno;
636 bio_set_flag(bio, BIO_CLONED);
637 if (bio_flagged(bio_src, BIO_THROTTLED))
638 bio_set_flag(bio, BIO_THROTTLED);
639 bio->bi_opf = bio_src->bi_opf;
640 bio->bi_ioprio = bio_src->bi_ioprio;
641 bio->bi_write_hint = bio_src->bi_write_hint;
642 bio->bi_iter = bio_src->bi_iter;
643 bio->bi_io_vec = bio_src->bi_io_vec;
645 bio_clone_blkg_association(bio, bio_src);
646 blkcg_bio_issue_init(bio);
648 EXPORT_SYMBOL(__bio_clone_fast);
651 * bio_clone_fast - clone a bio that shares the original bio's biovec
653 * @gfp_mask: allocation priority
654 * @bs: bio_set to allocate from
656 * Like __bio_clone_fast, only also allocates the returned bio
658 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
662 b = bio_alloc_bioset(gfp_mask, 0, bs);
666 __bio_clone_fast(b, bio);
668 if (bio_integrity(bio)) {
671 ret = bio_integrity_clone(b, bio, gfp_mask);
681 EXPORT_SYMBOL(bio_clone_fast);
683 static inline bool page_is_mergeable(const struct bio_vec *bv,
684 struct page *page, unsigned int len, unsigned int off,
687 size_t bv_end = bv->bv_offset + bv->bv_len;
688 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
689 phys_addr_t page_addr = page_to_phys(page);
691 if (vec_end_addr + 1 != page_addr + off)
693 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
696 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
699 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
702 static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio,
703 struct page *page, unsigned len, unsigned offset,
706 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
707 unsigned long mask = queue_segment_boundary(q);
708 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
709 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
711 if ((addr1 | mask) != (addr2 | mask))
713 if (bv->bv_len + len > queue_max_segment_size(q))
715 return __bio_try_merge_page(bio, page, len, offset, same_page);
719 * __bio_add_pc_page - attempt to add page to passthrough bio
720 * @q: the target queue
721 * @bio: destination bio
723 * @len: vec entry length
724 * @offset: vec entry offset
725 * @same_page: return if the merge happen inside the same page
727 * Attempt to add a page to the bio_vec maplist. This can fail for a
728 * number of reasons, such as the bio being full or target block device
729 * limitations. The target block device must allow bio's up to PAGE_SIZE,
730 * so it is always possible to add a single page to an empty bio.
732 * This should only be used by passthrough bios.
734 static int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
735 struct page *page, unsigned int len, unsigned int offset,
738 struct bio_vec *bvec;
741 * cloned bio must not modify vec list
743 if (unlikely(bio_flagged(bio, BIO_CLONED)))
746 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
749 if (bio->bi_vcnt > 0) {
750 if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page))
754 * If the queue doesn't support SG gaps and adding this segment
755 * would create a gap, disallow it.
757 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
758 if (bvec_gap_to_prev(q, bvec, offset))
762 if (bio_full(bio, len))
765 if (bio->bi_vcnt >= queue_max_segments(q))
768 bvec = &bio->bi_io_vec[bio->bi_vcnt];
769 bvec->bv_page = page;
771 bvec->bv_offset = offset;
773 bio->bi_iter.bi_size += len;
777 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
778 struct page *page, unsigned int len, unsigned int offset)
780 bool same_page = false;
781 return __bio_add_pc_page(q, bio, page, len, offset, &same_page);
783 EXPORT_SYMBOL(bio_add_pc_page);
786 * __bio_try_merge_page - try appending data to an existing bvec.
787 * @bio: destination bio
788 * @page: start page to add
789 * @len: length of the data to add
790 * @off: offset of the data relative to @page
791 * @same_page: return if the segment has been merged inside the same page
793 * Try to add the data at @page + @off to the last bvec of @bio. This is a
794 * a useful optimisation for file systems with a block size smaller than the
797 * Warn if (@len, @off) crosses pages in case that @same_page is true.
799 * Return %true on success or %false on failure.
801 bool __bio_try_merge_page(struct bio *bio, struct page *page,
802 unsigned int len, unsigned int off, bool *same_page)
804 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
807 if (bio->bi_vcnt > 0) {
808 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
810 if (page_is_mergeable(bv, page, len, off, same_page)) {
811 if (bio->bi_iter.bi_size > UINT_MAX - len) {
816 bio->bi_iter.bi_size += len;
822 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
825 * __bio_add_page - add page(s) to a bio in a new segment
826 * @bio: destination bio
827 * @page: start page to add
828 * @len: length of the data to add, may cross pages
829 * @off: offset of the data relative to @page, may cross pages
831 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
832 * that @bio has space for another bvec.
834 void __bio_add_page(struct bio *bio, struct page *page,
835 unsigned int len, unsigned int off)
837 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
839 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
840 WARN_ON_ONCE(bio_full(bio, len));
846 bio->bi_iter.bi_size += len;
849 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
850 bio_set_flag(bio, BIO_WORKINGSET);
852 EXPORT_SYMBOL_GPL(__bio_add_page);
855 * bio_add_page - attempt to add page(s) to bio
856 * @bio: destination bio
857 * @page: start page to add
858 * @len: vec entry length, may cross pages
859 * @offset: vec entry offset relative to @page, may cross pages
861 * Attempt to add page(s) to the bio_vec maplist. This will only fail
862 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
864 int bio_add_page(struct bio *bio, struct page *page,
865 unsigned int len, unsigned int offset)
867 bool same_page = false;
869 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
870 if (bio_full(bio, len))
872 __bio_add_page(bio, page, len, offset);
876 EXPORT_SYMBOL(bio_add_page);
878 void bio_release_pages(struct bio *bio, bool mark_dirty)
880 struct bvec_iter_all iter_all;
881 struct bio_vec *bvec;
883 if (bio_flagged(bio, BIO_NO_PAGE_REF))
886 bio_for_each_segment_all(bvec, bio, iter_all) {
887 if (mark_dirty && !PageCompound(bvec->bv_page))
888 set_page_dirty_lock(bvec->bv_page);
889 put_page(bvec->bv_page);
893 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter)
895 const struct bio_vec *bv = iter->bvec;
899 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len))
902 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count);
903 size = bio_add_page(bio, bv->bv_page, len,
904 bv->bv_offset + iter->iov_offset);
905 if (unlikely(size != len))
907 iov_iter_advance(iter, size);
911 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
914 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
915 * @bio: bio to add pages to
916 * @iter: iov iterator describing the region to be mapped
918 * Pins pages from *iter and appends them to @bio's bvec array. The
919 * pages will have to be released using put_page() when done.
920 * For multi-segment *iter, this function only adds pages from the
921 * the next non-empty segment of the iov iterator.
923 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
925 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
926 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
927 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
928 struct page **pages = (struct page **)bv;
929 bool same_page = false;
935 * Move page array up in the allocated memory for the bio vecs as far as
936 * possible so that we can start filling biovecs from the beginning
937 * without overwriting the temporary page array.
939 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
940 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
942 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
943 if (unlikely(size <= 0))
944 return size ? size : -EFAULT;
946 for (left = size, i = 0; left > 0; left -= len, i++) {
947 struct page *page = pages[i];
949 len = min_t(size_t, PAGE_SIZE - offset, left);
951 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
955 if (WARN_ON_ONCE(bio_full(bio, len)))
957 __bio_add_page(bio, page, len, offset);
962 iov_iter_advance(iter, size);
967 * bio_iov_iter_get_pages - add user or kernel pages to a bio
968 * @bio: bio to add pages to
969 * @iter: iov iterator describing the region to be added
971 * This takes either an iterator pointing to user memory, or one pointing to
972 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
973 * map them into the kernel. On IO completion, the caller should put those
974 * pages. If we're adding kernel pages, and the caller told us it's safe to
975 * do so, we just have to add the pages to the bio directly. We don't grab an
976 * extra reference to those pages (the user should already have that), and we
977 * don't put the page on IO completion. The caller needs to check if the bio is
978 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
981 * The function tries, but does not guarantee, to pin as many pages as
982 * fit into the bio, or are requested in *iter, whatever is smaller. If
983 * MM encounters an error pinning the requested pages, it stops. Error
984 * is returned only if 0 pages could be pinned.
986 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
988 const bool is_bvec = iov_iter_is_bvec(iter);
991 if (WARN_ON_ONCE(bio->bi_vcnt))
996 ret = __bio_iov_bvec_add_pages(bio, iter);
998 ret = __bio_iov_iter_get_pages(bio, iter);
999 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1002 bio_set_flag(bio, BIO_NO_PAGE_REF);
1003 return bio->bi_vcnt ? 0 : ret;
1006 static void submit_bio_wait_endio(struct bio *bio)
1008 complete(bio->bi_private);
1012 * submit_bio_wait - submit a bio, and wait until it completes
1013 * @bio: The &struct bio which describes the I/O
1015 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1016 * bio_endio() on failure.
1018 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1019 * result in bio reference to be consumed. The caller must drop the reference
1022 int submit_bio_wait(struct bio *bio)
1024 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
1026 bio->bi_private = &done;
1027 bio->bi_end_io = submit_bio_wait_endio;
1028 bio->bi_opf |= REQ_SYNC;
1030 wait_for_completion_io(&done);
1032 return blk_status_to_errno(bio->bi_status);
1034 EXPORT_SYMBOL(submit_bio_wait);
1037 * bio_advance - increment/complete a bio by some number of bytes
1038 * @bio: bio to advance
1039 * @bytes: number of bytes to complete
1041 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1042 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1043 * be updated on the last bvec as well.
1045 * @bio will then represent the remaining, uncompleted portion of the io.
1047 void bio_advance(struct bio *bio, unsigned bytes)
1049 if (bio_integrity(bio))
1050 bio_integrity_advance(bio, bytes);
1052 bio_advance_iter(bio, &bio->bi_iter, bytes);
1054 EXPORT_SYMBOL(bio_advance);
1056 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1057 struct bio *src, struct bvec_iter *src_iter)
1059 struct bio_vec src_bv, dst_bv;
1060 void *src_p, *dst_p;
1063 while (src_iter->bi_size && dst_iter->bi_size) {
1064 src_bv = bio_iter_iovec(src, *src_iter);
1065 dst_bv = bio_iter_iovec(dst, *dst_iter);
1067 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1069 src_p = kmap_atomic(src_bv.bv_page);
1070 dst_p = kmap_atomic(dst_bv.bv_page);
1072 memcpy(dst_p + dst_bv.bv_offset,
1073 src_p + src_bv.bv_offset,
1076 kunmap_atomic(dst_p);
1077 kunmap_atomic(src_p);
1079 flush_dcache_page(dst_bv.bv_page);
1081 bio_advance_iter(src, src_iter, bytes);
1082 bio_advance_iter(dst, dst_iter, bytes);
1085 EXPORT_SYMBOL(bio_copy_data_iter);
1088 * bio_copy_data - copy contents of data buffers from one bio to another
1090 * @dst: destination bio
1092 * Stops when it reaches the end of either @src or @dst - that is, copies
1093 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1095 void bio_copy_data(struct bio *dst, struct bio *src)
1097 struct bvec_iter src_iter = src->bi_iter;
1098 struct bvec_iter dst_iter = dst->bi_iter;
1100 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1102 EXPORT_SYMBOL(bio_copy_data);
1105 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1107 * @src: source bio list
1108 * @dst: destination bio list
1110 * Stops when it reaches the end of either the @src list or @dst list - that is,
1111 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1114 void bio_list_copy_data(struct bio *dst, struct bio *src)
1116 struct bvec_iter src_iter = src->bi_iter;
1117 struct bvec_iter dst_iter = dst->bi_iter;
1120 if (!src_iter.bi_size) {
1125 src_iter = src->bi_iter;
1128 if (!dst_iter.bi_size) {
1133 dst_iter = dst->bi_iter;
1136 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1139 EXPORT_SYMBOL(bio_list_copy_data);
1141 struct bio_map_data {
1143 struct iov_iter iter;
1147 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
1150 struct bio_map_data *bmd;
1151 if (data->nr_segs > UIO_MAXIOV)
1154 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask);
1157 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1159 bmd->iter.iov = bmd->iov;
1164 * bio_copy_from_iter - copy all pages from iov_iter to bio
1165 * @bio: The &struct bio which describes the I/O as destination
1166 * @iter: iov_iter as source
1168 * Copy all pages from iov_iter to bio.
1169 * Returns 0 on success, or error on failure.
1171 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
1173 struct bio_vec *bvec;
1174 struct bvec_iter_all iter_all;
1176 bio_for_each_segment_all(bvec, bio, iter_all) {
1179 ret = copy_page_from_iter(bvec->bv_page,
1184 if (!iov_iter_count(iter))
1187 if (ret < bvec->bv_len)
1195 * bio_copy_to_iter - copy all pages from bio to iov_iter
1196 * @bio: The &struct bio which describes the I/O as source
1197 * @iter: iov_iter as destination
1199 * Copy all pages from bio to iov_iter.
1200 * Returns 0 on success, or error on failure.
1202 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1204 struct bio_vec *bvec;
1205 struct bvec_iter_all iter_all;
1207 bio_for_each_segment_all(bvec, bio, iter_all) {
1210 ret = copy_page_to_iter(bvec->bv_page,
1215 if (!iov_iter_count(&iter))
1218 if (ret < bvec->bv_len)
1225 void bio_free_pages(struct bio *bio)
1227 struct bio_vec *bvec;
1228 struct bvec_iter_all iter_all;
1230 bio_for_each_segment_all(bvec, bio, iter_all)
1231 __free_page(bvec->bv_page);
1233 EXPORT_SYMBOL(bio_free_pages);
1236 * bio_uncopy_user - finish previously mapped bio
1237 * @bio: bio being terminated
1239 * Free pages allocated from bio_copy_user_iov() and write back data
1240 * to user space in case of a read.
1242 int bio_uncopy_user(struct bio *bio)
1244 struct bio_map_data *bmd = bio->bi_private;
1247 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1249 * if we're in a workqueue, the request is orphaned, so
1250 * don't copy into a random user address space, just free
1251 * and return -EINTR so user space doesn't expect any data.
1255 else if (bio_data_dir(bio) == READ)
1256 ret = bio_copy_to_iter(bio, bmd->iter);
1257 if (bmd->is_our_pages)
1258 bio_free_pages(bio);
1266 * bio_copy_user_iov - copy user data to bio
1267 * @q: destination block queue
1268 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1269 * @iter: iovec iterator
1270 * @gfp_mask: memory allocation flags
1272 * Prepares and returns a bio for indirect user io, bouncing data
1273 * to/from kernel pages as necessary. Must be paired with
1274 * call bio_uncopy_user() on io completion.
1276 struct bio *bio_copy_user_iov(struct request_queue *q,
1277 struct rq_map_data *map_data,
1278 struct iov_iter *iter,
1281 struct bio_map_data *bmd;
1286 unsigned int len = iter->count;
1287 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1289 bmd = bio_alloc_map_data(iter, gfp_mask);
1291 return ERR_PTR(-ENOMEM);
1294 * We need to do a deep copy of the iov_iter including the iovecs.
1295 * The caller provided iov might point to an on-stack or otherwise
1298 bmd->is_our_pages = map_data ? 0 : 1;
1300 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1301 if (nr_pages > BIO_MAX_PAGES)
1302 nr_pages = BIO_MAX_PAGES;
1305 bio = bio_kmalloc(gfp_mask, nr_pages);
1312 nr_pages = 1 << map_data->page_order;
1313 i = map_data->offset / PAGE_SIZE;
1316 unsigned int bytes = PAGE_SIZE;
1324 if (i == map_data->nr_entries * nr_pages) {
1329 page = map_data->pages[i / nr_pages];
1330 page += (i % nr_pages);
1334 page = alloc_page(q->bounce_gfp | gfp_mask);
1341 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) {
1355 map_data->offset += bio->bi_iter.bi_size;
1360 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) ||
1361 (map_data && map_data->from_user)) {
1362 ret = bio_copy_from_iter(bio, iter);
1366 if (bmd->is_our_pages)
1368 iov_iter_advance(iter, bio->bi_iter.bi_size);
1371 bio->bi_private = bmd;
1372 if (map_data && map_data->null_mapped)
1373 bio_set_flag(bio, BIO_NULL_MAPPED);
1377 bio_free_pages(bio);
1381 return ERR_PTR(ret);
1385 * bio_map_user_iov - map user iovec into bio
1386 * @q: the struct request_queue for the bio
1387 * @iter: iovec iterator
1388 * @gfp_mask: memory allocation flags
1390 * Map the user space address into a bio suitable for io to a block
1391 * device. Returns an error pointer in case of error.
1393 struct bio *bio_map_user_iov(struct request_queue *q,
1394 struct iov_iter *iter,
1401 if (!iov_iter_count(iter))
1402 return ERR_PTR(-EINVAL);
1404 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
1406 return ERR_PTR(-ENOMEM);
1408 while (iov_iter_count(iter)) {
1409 struct page **pages;
1411 size_t offs, added = 0;
1414 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
1415 if (unlikely(bytes <= 0)) {
1416 ret = bytes ? bytes : -EFAULT;
1420 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
1422 if (unlikely(offs & queue_dma_alignment(q))) {
1426 for (j = 0; j < npages; j++) {
1427 struct page *page = pages[j];
1428 unsigned int n = PAGE_SIZE - offs;
1429 bool same_page = false;
1434 if (!__bio_add_pc_page(q, bio, page, n, offs,
1445 iov_iter_advance(iter, added);
1448 * release the pages we didn't map into the bio, if any
1451 put_page(pages[j++]);
1453 /* couldn't stuff something into bio? */
1458 bio_set_flag(bio, BIO_USER_MAPPED);
1461 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1462 * it would normally disappear when its bi_end_io is run.
1463 * however, we need it for the unmap, so grab an extra
1470 bio_release_pages(bio, false);
1472 return ERR_PTR(ret);
1476 * bio_unmap_user - unmap a bio
1477 * @bio: the bio being unmapped
1479 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1482 * bio_unmap_user() may sleep.
1484 void bio_unmap_user(struct bio *bio)
1486 bio_release_pages(bio, bio_data_dir(bio) == READ);
1491 static void bio_invalidate_vmalloc_pages(struct bio *bio)
1493 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
1494 if (bio->bi_private && !op_is_write(bio_op(bio))) {
1495 unsigned long i, len = 0;
1497 for (i = 0; i < bio->bi_vcnt; i++)
1498 len += bio->bi_io_vec[i].bv_len;
1499 invalidate_kernel_vmap_range(bio->bi_private, len);
1504 static void bio_map_kern_endio(struct bio *bio)
1506 bio_invalidate_vmalloc_pages(bio);
1511 * bio_map_kern - map kernel address into bio
1512 * @q: the struct request_queue for the bio
1513 * @data: pointer to buffer to map
1514 * @len: length in bytes
1515 * @gfp_mask: allocation flags for bio allocation
1517 * Map the kernel address into a bio suitable for io to a block
1518 * device. Returns an error pointer in case of error.
1520 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1523 unsigned long kaddr = (unsigned long)data;
1524 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1525 unsigned long start = kaddr >> PAGE_SHIFT;
1526 const int nr_pages = end - start;
1527 bool is_vmalloc = is_vmalloc_addr(data);
1532 bio = bio_kmalloc(gfp_mask, nr_pages);
1534 return ERR_PTR(-ENOMEM);
1537 flush_kernel_vmap_range(data, len);
1538 bio->bi_private = data;
1541 offset = offset_in_page(kaddr);
1542 for (i = 0; i < nr_pages; i++) {
1543 unsigned int bytes = PAGE_SIZE - offset;
1552 page = virt_to_page(data);
1554 page = vmalloc_to_page(data);
1555 if (bio_add_pc_page(q, bio, page, bytes,
1557 /* we don't support partial mappings */
1559 return ERR_PTR(-EINVAL);
1567 bio->bi_end_io = bio_map_kern_endio;
1571 static void bio_copy_kern_endio(struct bio *bio)
1573 bio_free_pages(bio);
1577 static void bio_copy_kern_endio_read(struct bio *bio)
1579 char *p = bio->bi_private;
1580 struct bio_vec *bvec;
1581 struct bvec_iter_all iter_all;
1583 bio_for_each_segment_all(bvec, bio, iter_all) {
1584 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1588 bio_copy_kern_endio(bio);
1592 * bio_copy_kern - copy kernel address into bio
1593 * @q: the struct request_queue for the bio
1594 * @data: pointer to buffer to copy
1595 * @len: length in bytes
1596 * @gfp_mask: allocation flags for bio and page allocation
1597 * @reading: data direction is READ
1599 * copy the kernel address into a bio suitable for io to a block
1600 * device. Returns an error pointer in case of error.
1602 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1603 gfp_t gfp_mask, int reading)
1605 unsigned long kaddr = (unsigned long)data;
1606 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1607 unsigned long start = kaddr >> PAGE_SHIFT;
1616 return ERR_PTR(-EINVAL);
1618 nr_pages = end - start;
1619 bio = bio_kmalloc(gfp_mask, nr_pages);
1621 return ERR_PTR(-ENOMEM);
1625 unsigned int bytes = PAGE_SIZE;
1630 page = alloc_page(q->bounce_gfp | __GFP_ZERO | gfp_mask);
1635 memcpy(page_address(page), p, bytes);
1637 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1645 bio->bi_end_io = bio_copy_kern_endio_read;
1646 bio->bi_private = data;
1648 bio->bi_end_io = bio_copy_kern_endio;
1654 bio_free_pages(bio);
1656 return ERR_PTR(-ENOMEM);
1660 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1661 * for performing direct-IO in BIOs.
1663 * The problem is that we cannot run set_page_dirty() from interrupt context
1664 * because the required locks are not interrupt-safe. So what we can do is to
1665 * mark the pages dirty _before_ performing IO. And in interrupt context,
1666 * check that the pages are still dirty. If so, fine. If not, redirty them
1667 * in process context.
1669 * We special-case compound pages here: normally this means reads into hugetlb
1670 * pages. The logic in here doesn't really work right for compound pages
1671 * because the VM does not uniformly chase down the head page in all cases.
1672 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1673 * handle them at all. So we skip compound pages here at an early stage.
1675 * Note that this code is very hard to test under normal circumstances because
1676 * direct-io pins the pages with get_user_pages(). This makes
1677 * is_page_cache_freeable return false, and the VM will not clean the pages.
1678 * But other code (eg, flusher threads) could clean the pages if they are mapped
1681 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1682 * deferred bio dirtying paths.
1686 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1688 void bio_set_pages_dirty(struct bio *bio)
1690 struct bio_vec *bvec;
1691 struct bvec_iter_all iter_all;
1693 bio_for_each_segment_all(bvec, bio, iter_all) {
1694 if (!PageCompound(bvec->bv_page))
1695 set_page_dirty_lock(bvec->bv_page);
1700 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1701 * If they are, then fine. If, however, some pages are clean then they must
1702 * have been written out during the direct-IO read. So we take another ref on
1703 * the BIO and re-dirty the pages in process context.
1705 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1706 * here on. It will run one put_page() against each page and will run one
1707 * bio_put() against the BIO.
1710 static void bio_dirty_fn(struct work_struct *work);
1712 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1713 static DEFINE_SPINLOCK(bio_dirty_lock);
1714 static struct bio *bio_dirty_list;
1717 * This runs in process context
1719 static void bio_dirty_fn(struct work_struct *work)
1721 struct bio *bio, *next;
1723 spin_lock_irq(&bio_dirty_lock);
1724 next = bio_dirty_list;
1725 bio_dirty_list = NULL;
1726 spin_unlock_irq(&bio_dirty_lock);
1728 while ((bio = next) != NULL) {
1729 next = bio->bi_private;
1731 bio_release_pages(bio, true);
1736 void bio_check_pages_dirty(struct bio *bio)
1738 struct bio_vec *bvec;
1739 unsigned long flags;
1740 struct bvec_iter_all iter_all;
1742 bio_for_each_segment_all(bvec, bio, iter_all) {
1743 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1747 bio_release_pages(bio, false);
1751 spin_lock_irqsave(&bio_dirty_lock, flags);
1752 bio->bi_private = bio_dirty_list;
1753 bio_dirty_list = bio;
1754 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1755 schedule_work(&bio_dirty_work);
1758 void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
1760 unsigned long stamp;
1762 stamp = READ_ONCE(part->stamp);
1763 if (unlikely(stamp != now)) {
1764 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) {
1765 __part_stat_add(part, io_ticks, end ? now - stamp : 1);
1769 part = &part_to_disk(part)->part0;
1774 void generic_start_io_acct(struct request_queue *q, int op,
1775 unsigned long sectors, struct hd_struct *part)
1777 const int sgrp = op_stat_group(op);
1781 update_io_ticks(part, jiffies, false);
1782 part_stat_inc(part, ios[sgrp]);
1783 part_stat_add(part, sectors[sgrp], sectors);
1784 part_inc_in_flight(q, part, op_is_write(op));
1788 EXPORT_SYMBOL(generic_start_io_acct);
1790 void generic_end_io_acct(struct request_queue *q, int req_op,
1791 struct hd_struct *part, unsigned long start_time)
1793 unsigned long now = jiffies;
1794 unsigned long duration = now - start_time;
1795 const int sgrp = op_stat_group(req_op);
1799 update_io_ticks(part, now, true);
1800 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1801 part_stat_add(part, time_in_queue, duration);
1802 part_dec_in_flight(q, part, op_is_write(req_op));
1806 EXPORT_SYMBOL(generic_end_io_acct);
1808 static inline bool bio_remaining_done(struct bio *bio)
1811 * If we're not chaining, then ->__bi_remaining is always 1 and
1812 * we always end io on the first invocation.
1814 if (!bio_flagged(bio, BIO_CHAIN))
1817 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1819 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1820 bio_clear_flag(bio, BIO_CHAIN);
1828 * bio_endio - end I/O on a bio
1832 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1833 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1834 * bio unless they own it and thus know that it has an end_io function.
1836 * bio_endio() can be called several times on a bio that has been chained
1837 * using bio_chain(). The ->bi_end_io() function will only be called the
1838 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1839 * generated if BIO_TRACE_COMPLETION is set.
1841 void bio_endio(struct bio *bio)
1844 if (!bio_remaining_done(bio))
1846 if (!bio_integrity_endio(bio))
1850 rq_qos_done_bio(bio->bi_disk->queue, bio);
1853 * Need to have a real endio function for chained bios, otherwise
1854 * various corner cases will break (like stacking block devices that
1855 * save/restore bi_end_io) - however, we want to avoid unbounded
1856 * recursion and blowing the stack. Tail call optimization would
1857 * handle this, but compiling with frame pointers also disables
1858 * gcc's sibling call optimization.
1860 if (bio->bi_end_io == bio_chain_endio) {
1861 bio = __bio_chain_endio(bio);
1865 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1866 trace_block_bio_complete(bio->bi_disk->queue, bio,
1867 blk_status_to_errno(bio->bi_status));
1868 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1871 blk_throtl_bio_endio(bio);
1872 /* release cgroup info */
1875 bio->bi_end_io(bio);
1877 EXPORT_SYMBOL(bio_endio);
1880 * bio_split - split a bio
1881 * @bio: bio to split
1882 * @sectors: number of sectors to split from the front of @bio
1884 * @bs: bio set to allocate from
1886 * Allocates and returns a new bio which represents @sectors from the start of
1887 * @bio, and updates @bio to represent the remaining sectors.
1889 * Unless this is a discard request the newly allocated bio will point
1890 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1891 * neither @bio nor @bs are freed before the split bio.
1893 struct bio *bio_split(struct bio *bio, int sectors,
1894 gfp_t gfp, struct bio_set *bs)
1898 BUG_ON(sectors <= 0);
1899 BUG_ON(sectors >= bio_sectors(bio));
1901 split = bio_clone_fast(bio, gfp, bs);
1905 split->bi_iter.bi_size = sectors << 9;
1907 if (bio_integrity(split))
1908 bio_integrity_trim(split);
1910 bio_advance(bio, split->bi_iter.bi_size);
1912 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1913 bio_set_flag(split, BIO_TRACE_COMPLETION);
1917 EXPORT_SYMBOL(bio_split);
1920 * bio_trim - trim a bio
1922 * @offset: number of sectors to trim from the front of @bio
1923 * @size: size we want to trim @bio to, in sectors
1925 void bio_trim(struct bio *bio, int offset, int size)
1927 /* 'bio' is a cloned bio which we need to trim to match
1928 * the given offset and size.
1932 if (offset == 0 && size == bio->bi_iter.bi_size)
1935 bio_advance(bio, offset << 9);
1936 bio->bi_iter.bi_size = size;
1938 if (bio_integrity(bio))
1939 bio_integrity_trim(bio);
1942 EXPORT_SYMBOL_GPL(bio_trim);
1945 * create memory pools for biovec's in a bio_set.
1946 * use the global biovec slabs created for general use.
1948 int biovec_init_pool(mempool_t *pool, int pool_entries)
1950 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1952 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1956 * bioset_exit - exit a bioset initialized with bioset_init()
1958 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1961 void bioset_exit(struct bio_set *bs)
1963 if (bs->rescue_workqueue)
1964 destroy_workqueue(bs->rescue_workqueue);
1965 bs->rescue_workqueue = NULL;
1967 mempool_exit(&bs->bio_pool);
1968 mempool_exit(&bs->bvec_pool);
1970 bioset_integrity_free(bs);
1973 bs->bio_slab = NULL;
1975 EXPORT_SYMBOL(bioset_exit);
1978 * bioset_init - Initialize a bio_set
1979 * @bs: pool to initialize
1980 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1981 * @front_pad: Number of bytes to allocate in front of the returned bio
1982 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1983 * and %BIOSET_NEED_RESCUER
1986 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1987 * to ask for a number of bytes to be allocated in front of the bio.
1988 * Front pad allocation is useful for embedding the bio inside
1989 * another structure, to avoid allocating extra data to go with the bio.
1990 * Note that the bio must be embedded at the END of that structure always,
1991 * or things will break badly.
1992 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1993 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1994 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1995 * dispatch queued requests when the mempool runs out of space.
1998 int bioset_init(struct bio_set *bs,
1999 unsigned int pool_size,
2000 unsigned int front_pad,
2003 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
2005 bs->front_pad = front_pad;
2007 spin_lock_init(&bs->rescue_lock);
2008 bio_list_init(&bs->rescue_list);
2009 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
2011 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
2015 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
2018 if ((flags & BIOSET_NEED_BVECS) &&
2019 biovec_init_pool(&bs->bvec_pool, pool_size))
2022 if (!(flags & BIOSET_NEED_RESCUER))
2025 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
2026 if (!bs->rescue_workqueue)
2034 EXPORT_SYMBOL(bioset_init);
2037 * Initialize and setup a new bio_set, based on the settings from
2040 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
2045 if (src->bvec_pool.min_nr)
2046 flags |= BIOSET_NEED_BVECS;
2047 if (src->rescue_workqueue)
2048 flags |= BIOSET_NEED_RESCUER;
2050 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
2052 EXPORT_SYMBOL(bioset_init_from_src);
2054 #ifdef CONFIG_BLK_CGROUP
2057 * bio_disassociate_blkg - puts back the blkg reference if associated
2060 * Helper to disassociate the blkg from @bio if a blkg is associated.
2062 void bio_disassociate_blkg(struct bio *bio)
2065 blkg_put(bio->bi_blkg);
2066 bio->bi_blkg = NULL;
2069 EXPORT_SYMBOL_GPL(bio_disassociate_blkg);
2072 * __bio_associate_blkg - associate a bio with the a blkg
2074 * @blkg: the blkg to associate
2076 * This tries to associate @bio with the specified @blkg. Association failure
2077 * is handled by walking up the blkg tree. Therefore, the blkg associated can
2078 * be anything between @blkg and the root_blkg. This situation only happens
2079 * when a cgroup is dying and then the remaining bios will spill to the closest
2082 * A reference will be taken on the @blkg and will be released when @bio is
2085 static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg)
2087 bio_disassociate_blkg(bio);
2089 bio->bi_blkg = blkg_tryget_closest(blkg);
2093 * bio_associate_blkg_from_css - associate a bio with a specified css
2097 * Associate @bio with the blkg found by combining the css's blkg and the
2098 * request_queue of the @bio. This falls back to the queue's root_blkg if
2099 * the association fails with the css.
2101 void bio_associate_blkg_from_css(struct bio *bio,
2102 struct cgroup_subsys_state *css)
2104 struct request_queue *q = bio->bi_disk->queue;
2105 struct blkcg_gq *blkg;
2109 if (!css || !css->parent)
2110 blkg = q->root_blkg;
2112 blkg = blkg_lookup_create(css_to_blkcg(css), q);
2114 __bio_associate_blkg(bio, blkg);
2118 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css);
2122 * bio_associate_blkg_from_page - associate a bio with the page's blkg
2124 * @page: the page to lookup the blkcg from
2126 * Associate @bio with the blkg from @page's owning memcg and the respective
2127 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's
2130 void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
2132 struct cgroup_subsys_state *css;
2134 if (!page->mem_cgroup)
2139 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
2140 bio_associate_blkg_from_css(bio, css);
2144 #endif /* CONFIG_MEMCG */
2147 * bio_associate_blkg - associate a bio with a blkg
2150 * Associate @bio with the blkg found from the bio's css and request_queue.
2151 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is
2152 * already associated, the css is reused and association redone as the
2153 * request_queue may have changed.
2155 void bio_associate_blkg(struct bio *bio)
2157 struct cgroup_subsys_state *css;
2162 css = &bio_blkcg(bio)->css;
2166 bio_associate_blkg_from_css(bio, css);
2170 EXPORT_SYMBOL_GPL(bio_associate_blkg);
2173 * bio_clone_blkg_association - clone blkg association from src to dst bio
2174 * @dst: destination bio
2177 void bio_clone_blkg_association(struct bio *dst, struct bio *src)
2182 bio_associate_blkg_from_css(dst, &bio_blkcg(src)->css);
2186 EXPORT_SYMBOL_GPL(bio_clone_blkg_association);
2187 #endif /* CONFIG_BLK_CGROUP */
2189 static void __init biovec_init_slabs(void)
2193 for (i = 0; i < BVEC_POOL_NR; i++) {
2195 struct biovec_slab *bvs = bvec_slabs + i;
2197 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2202 size = bvs->nr_vecs * sizeof(struct bio_vec);
2203 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2204 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2208 static int __init init_bio(void)
2212 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab),
2215 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET);
2218 panic("bio: can't allocate bios\n");
2220 bio_integrity_init();
2221 biovec_init_slabs();
2223 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
2224 panic("bio: can't allocate bios\n");
2226 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
2227 panic("bio: can't create integrity pool\n");
2231 subsys_initcall(init_bio);