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>
20 #include <linux/sched/sysctl.h>
22 #include <trace/events/block.h>
24 #include "blk-rq-qos.h"
27 * Test patch to inline a certain number of bi_io_vec's inside the bio
28 * itself, to shrink a bio data allocation from two mempool calls to one
30 #define BIO_INLINE_VECS 4
33 * if you change this list, also change bvec_alloc or things will
34 * break badly! cannot be bigger than what you can fit into an
37 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
38 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
39 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
44 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
45 * IO code that does not need private memory pools.
47 struct bio_set fs_bio_set;
48 EXPORT_SYMBOL(fs_bio_set);
51 * Our slab pool management
54 struct kmem_cache *slab;
55 unsigned int slab_ref;
56 unsigned int slab_size;
59 static DEFINE_MUTEX(bio_slab_lock);
60 static struct bio_slab *bio_slabs;
61 static unsigned int bio_slab_nr, bio_slab_max;
63 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
65 unsigned int sz = sizeof(struct bio) + extra_size;
66 struct kmem_cache *slab = NULL;
67 struct bio_slab *bslab, *new_bio_slabs;
68 unsigned int new_bio_slab_max;
69 unsigned int i, entry = -1;
71 mutex_lock(&bio_slab_lock);
74 while (i < bio_slab_nr) {
75 bslab = &bio_slabs[i];
77 if (!bslab->slab && entry == -1)
79 else if (bslab->slab_size == sz) {
90 if (bio_slab_nr == bio_slab_max && entry == -1) {
91 new_bio_slab_max = bio_slab_max << 1;
92 new_bio_slabs = krealloc(bio_slabs,
93 new_bio_slab_max * sizeof(struct bio_slab),
97 bio_slab_max = new_bio_slab_max;
98 bio_slabs = new_bio_slabs;
101 entry = bio_slab_nr++;
103 bslab = &bio_slabs[entry];
105 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
106 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
107 SLAB_HWCACHE_ALIGN, NULL);
113 bslab->slab_size = sz;
115 mutex_unlock(&bio_slab_lock);
119 static void bio_put_slab(struct bio_set *bs)
121 struct bio_slab *bslab = NULL;
124 mutex_lock(&bio_slab_lock);
126 for (i = 0; i < bio_slab_nr; i++) {
127 if (bs->bio_slab == bio_slabs[i].slab) {
128 bslab = &bio_slabs[i];
133 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
136 WARN_ON(!bslab->slab_ref);
138 if (--bslab->slab_ref)
141 kmem_cache_destroy(bslab->slab);
145 mutex_unlock(&bio_slab_lock);
148 unsigned int bvec_nr_vecs(unsigned short idx)
150 return bvec_slabs[--idx].nr_vecs;
153 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
159 BIO_BUG_ON(idx >= BVEC_POOL_NR);
161 if (idx == BVEC_POOL_MAX) {
162 mempool_free(bv, pool);
164 struct biovec_slab *bvs = bvec_slabs + idx;
166 kmem_cache_free(bvs->slab, bv);
170 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
176 * see comment near bvec_array define!
194 case 129 ... BIO_MAX_PAGES:
202 * idx now points to the pool we want to allocate from. only the
203 * 1-vec entry pool is mempool backed.
205 if (*idx == BVEC_POOL_MAX) {
207 bvl = mempool_alloc(pool, gfp_mask);
209 struct biovec_slab *bvs = bvec_slabs + *idx;
210 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
213 * Make this allocation restricted and don't dump info on
214 * allocation failures, since we'll fallback to the mempool
215 * in case of failure.
217 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
220 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
221 * is set, retry with the 1-entry mempool
223 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
224 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
225 *idx = BVEC_POOL_MAX;
234 void bio_uninit(struct bio *bio)
236 bio_disassociate_blkg(bio);
238 if (bio_integrity(bio))
239 bio_integrity_free(bio);
241 EXPORT_SYMBOL(bio_uninit);
243 static void bio_free(struct bio *bio)
245 struct bio_set *bs = bio->bi_pool;
251 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
254 * If we have front padding, adjust the bio pointer before freeing
259 mempool_free(p, &bs->bio_pool);
261 /* Bio was allocated by bio_kmalloc() */
267 * Users of this function have their own bio allocation. Subsequently,
268 * they must remember to pair any call to bio_init() with bio_uninit()
269 * when IO has completed, or when the bio is released.
271 void bio_init(struct bio *bio, struct bio_vec *table,
272 unsigned short max_vecs)
274 memset(bio, 0, sizeof(*bio));
275 atomic_set(&bio->__bi_remaining, 1);
276 atomic_set(&bio->__bi_cnt, 1);
278 bio->bi_io_vec = table;
279 bio->bi_max_vecs = max_vecs;
281 EXPORT_SYMBOL(bio_init);
284 * bio_reset - reinitialize a bio
288 * After calling bio_reset(), @bio will be in the same state as a freshly
289 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
290 * preserved are the ones that are initialized by bio_alloc_bioset(). See
291 * comment in struct bio.
293 void bio_reset(struct bio *bio)
295 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
299 memset(bio, 0, BIO_RESET_BYTES);
300 bio->bi_flags = flags;
301 atomic_set(&bio->__bi_remaining, 1);
303 EXPORT_SYMBOL(bio_reset);
305 static struct bio *__bio_chain_endio(struct bio *bio)
307 struct bio *parent = bio->bi_private;
309 if (!parent->bi_status)
310 parent->bi_status = bio->bi_status;
315 static void bio_chain_endio(struct bio *bio)
317 bio_endio(__bio_chain_endio(bio));
321 * bio_chain - chain bio completions
322 * @bio: the target bio
323 * @parent: the @bio's parent bio
325 * The caller won't have a bi_end_io called when @bio completes - instead,
326 * @parent's bi_end_io won't be called until both @parent and @bio have
327 * completed; the chained bio will also be freed when it completes.
329 * The caller must not set bi_private or bi_end_io in @bio.
331 void bio_chain(struct bio *bio, struct bio *parent)
333 BUG_ON(bio->bi_private || bio->bi_end_io);
335 bio->bi_private = parent;
336 bio->bi_end_io = bio_chain_endio;
337 bio_inc_remaining(parent);
339 EXPORT_SYMBOL(bio_chain);
341 static void bio_alloc_rescue(struct work_struct *work)
343 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
347 spin_lock(&bs->rescue_lock);
348 bio = bio_list_pop(&bs->rescue_list);
349 spin_unlock(&bs->rescue_lock);
354 generic_make_request(bio);
358 static void punt_bios_to_rescuer(struct bio_set *bs)
360 struct bio_list punt, nopunt;
363 if (WARN_ON_ONCE(!bs->rescue_workqueue))
366 * In order to guarantee forward progress we must punt only bios that
367 * were allocated from this bio_set; otherwise, if there was a bio on
368 * there for a stacking driver higher up in the stack, processing it
369 * could require allocating bios from this bio_set, and doing that from
370 * our own rescuer would be bad.
372 * Since bio lists are singly linked, pop them all instead of trying to
373 * remove from the middle of the list:
376 bio_list_init(&punt);
377 bio_list_init(&nopunt);
379 while ((bio = bio_list_pop(¤t->bio_list[0])))
380 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
381 current->bio_list[0] = nopunt;
383 bio_list_init(&nopunt);
384 while ((bio = bio_list_pop(¤t->bio_list[1])))
385 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
386 current->bio_list[1] = nopunt;
388 spin_lock(&bs->rescue_lock);
389 bio_list_merge(&bs->rescue_list, &punt);
390 spin_unlock(&bs->rescue_lock);
392 queue_work(bs->rescue_workqueue, &bs->rescue_work);
396 * bio_alloc_bioset - allocate a bio for I/O
397 * @gfp_mask: the GFP_* mask given to the slab allocator
398 * @nr_iovecs: number of iovecs to pre-allocate
399 * @bs: the bio_set to allocate from.
402 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
403 * backed by the @bs's mempool.
405 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
406 * always be able to allocate a bio. This is due to the mempool guarantees.
407 * To make this work, callers must never allocate more than 1 bio at a time
408 * from this pool. Callers that need to allocate more than 1 bio must always
409 * submit the previously allocated bio for IO before attempting to allocate
410 * a new one. Failure to do so can cause deadlocks under memory pressure.
412 * Note that when running under generic_make_request() (i.e. any block
413 * driver), bios are not submitted until after you return - see the code in
414 * generic_make_request() that converts recursion into iteration, to prevent
417 * This would normally mean allocating multiple bios under
418 * generic_make_request() would be susceptible to deadlocks, but we have
419 * deadlock avoidance code that resubmits any blocked bios from a rescuer
422 * However, we do not guarantee forward progress for allocations from other
423 * mempools. Doing multiple allocations from the same mempool under
424 * generic_make_request() should be avoided - instead, use bio_set's front_pad
425 * for per bio allocations.
428 * Pointer to new bio on success, NULL on failure.
430 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
433 gfp_t saved_gfp = gfp_mask;
435 unsigned inline_vecs;
436 struct bio_vec *bvl = NULL;
441 if (nr_iovecs > UIO_MAXIOV)
444 p = kmalloc(sizeof(struct bio) +
445 nr_iovecs * sizeof(struct bio_vec),
448 inline_vecs = nr_iovecs;
450 /* should not use nobvec bioset for nr_iovecs > 0 */
451 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
455 * generic_make_request() converts recursion to iteration; this
456 * means if we're running beneath it, any bios we allocate and
457 * submit will not be submitted (and thus freed) until after we
460 * This exposes us to a potential deadlock if we allocate
461 * multiple bios from the same bio_set() while running
462 * underneath generic_make_request(). If we were to allocate
463 * multiple bios (say a stacking block driver that was splitting
464 * bios), we would deadlock if we exhausted the mempool's
467 * We solve this, and guarantee forward progress, with a rescuer
468 * workqueue per bio_set. If we go to allocate and there are
469 * bios on current->bio_list, we first try the allocation
470 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
471 * bios we would be blocking to the rescuer workqueue before
472 * we retry with the original gfp_flags.
475 if (current->bio_list &&
476 (!bio_list_empty(¤t->bio_list[0]) ||
477 !bio_list_empty(¤t->bio_list[1])) &&
478 bs->rescue_workqueue)
479 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
481 p = mempool_alloc(&bs->bio_pool, gfp_mask);
482 if (!p && gfp_mask != saved_gfp) {
483 punt_bios_to_rescuer(bs);
484 gfp_mask = saved_gfp;
485 p = mempool_alloc(&bs->bio_pool, gfp_mask);
488 front_pad = bs->front_pad;
489 inline_vecs = BIO_INLINE_VECS;
496 bio_init(bio, NULL, 0);
498 if (nr_iovecs > inline_vecs) {
499 unsigned long idx = 0;
501 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
502 if (!bvl && gfp_mask != saved_gfp) {
503 punt_bios_to_rescuer(bs);
504 gfp_mask = saved_gfp;
505 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
511 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
512 } else if (nr_iovecs) {
513 bvl = bio->bi_inline_vecs;
517 bio->bi_max_vecs = nr_iovecs;
518 bio->bi_io_vec = bvl;
522 mempool_free(p, &bs->bio_pool);
525 EXPORT_SYMBOL(bio_alloc_bioset);
527 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
531 struct bvec_iter iter;
533 __bio_for_each_segment(bv, bio, iter, start) {
534 char *data = bvec_kmap_irq(&bv, &flags);
535 memset(data, 0, bv.bv_len);
536 flush_dcache_page(bv.bv_page);
537 bvec_kunmap_irq(data, &flags);
540 EXPORT_SYMBOL(zero_fill_bio_iter);
543 * bio_truncate - truncate the bio to small size of @new_size
544 * @bio: the bio to be truncated
545 * @new_size: new size for truncating the bio
548 * Truncate the bio to new size of @new_size. If bio_op(bio) is
549 * REQ_OP_READ, zero the truncated part. This function should only
550 * be used for handling corner cases, such as bio eod.
552 void bio_truncate(struct bio *bio, unsigned new_size)
555 struct bvec_iter iter;
556 unsigned int done = 0;
557 bool truncated = false;
559 if (new_size >= bio->bi_iter.bi_size)
562 if (bio_op(bio) != REQ_OP_READ)
565 bio_for_each_segment(bv, bio, iter) {
566 if (done + bv.bv_len > new_size) {
570 offset = new_size - done;
573 zero_user(bv.bv_page, offset, bv.bv_len - 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 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) +
688 bv->bv_offset + bv->bv_len - 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);
697 if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page)
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)
814 bio->bi_iter.bi_size += len;
820 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
823 * __bio_add_page - add page(s) to a bio in a new segment
824 * @bio: destination bio
825 * @page: start page to add
826 * @len: length of the data to add, may cross pages
827 * @off: offset of the data relative to @page, may cross pages
829 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
830 * that @bio has space for another bvec.
832 void __bio_add_page(struct bio *bio, struct page *page,
833 unsigned int len, unsigned int off)
835 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
837 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
838 WARN_ON_ONCE(bio_full(bio, len));
844 bio->bi_iter.bi_size += len;
847 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
848 bio_set_flag(bio, BIO_WORKINGSET);
850 EXPORT_SYMBOL_GPL(__bio_add_page);
853 * bio_add_page - attempt to add page(s) to bio
854 * @bio: destination bio
855 * @page: start page to add
856 * @len: vec entry length, may cross pages
857 * @offset: vec entry offset relative to @page, may cross pages
859 * Attempt to add page(s) to the bio_vec maplist. This will only fail
860 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
862 int bio_add_page(struct bio *bio, struct page *page,
863 unsigned int len, unsigned int offset)
865 bool same_page = false;
867 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
868 if (bio_full(bio, len))
870 __bio_add_page(bio, page, len, offset);
874 EXPORT_SYMBOL(bio_add_page);
876 void bio_release_pages(struct bio *bio, bool mark_dirty)
878 struct bvec_iter_all iter_all;
879 struct bio_vec *bvec;
881 if (bio_flagged(bio, BIO_NO_PAGE_REF))
884 bio_for_each_segment_all(bvec, bio, iter_all) {
885 if (mark_dirty && !PageCompound(bvec->bv_page))
886 set_page_dirty_lock(bvec->bv_page);
887 put_page(bvec->bv_page);
891 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter)
893 const struct bio_vec *bv = iter->bvec;
897 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len))
900 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count);
901 size = bio_add_page(bio, bv->bv_page, len,
902 bv->bv_offset + iter->iov_offset);
903 if (unlikely(size != len))
905 iov_iter_advance(iter, size);
909 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
912 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
913 * @bio: bio to add pages to
914 * @iter: iov iterator describing the region to be mapped
916 * Pins pages from *iter and appends them to @bio's bvec array. The
917 * pages will have to be released using put_page() when done.
918 * For multi-segment *iter, this function only adds pages from the
919 * the next non-empty segment of the iov iterator.
921 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
923 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
924 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
925 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
926 struct page **pages = (struct page **)bv;
927 bool same_page = false;
933 * Move page array up in the allocated memory for the bio vecs as far as
934 * possible so that we can start filling biovecs from the beginning
935 * without overwriting the temporary page array.
937 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
938 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
940 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
941 if (unlikely(size <= 0))
942 return size ? size : -EFAULT;
944 for (left = size, i = 0; left > 0; left -= len, i++) {
945 struct page *page = pages[i];
947 len = min_t(size_t, PAGE_SIZE - offset, left);
949 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
953 if (WARN_ON_ONCE(bio_full(bio, len)))
955 __bio_add_page(bio, page, len, offset);
960 iov_iter_advance(iter, size);
965 * bio_iov_iter_get_pages - add user or kernel pages to a bio
966 * @bio: bio to add pages to
967 * @iter: iov iterator describing the region to be added
969 * This takes either an iterator pointing to user memory, or one pointing to
970 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
971 * map them into the kernel. On IO completion, the caller should put those
972 * pages. If we're adding kernel pages, and the caller told us it's safe to
973 * do so, we just have to add the pages to the bio directly. We don't grab an
974 * extra reference to those pages (the user should already have that), and we
975 * don't put the page on IO completion. The caller needs to check if the bio is
976 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
979 * The function tries, but does not guarantee, to pin as many pages as
980 * fit into the bio, or are requested in *iter, whatever is smaller. If
981 * MM encounters an error pinning the requested pages, it stops. Error
982 * is returned only if 0 pages could be pinned.
984 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
986 const bool is_bvec = iov_iter_is_bvec(iter);
989 if (WARN_ON_ONCE(bio->bi_vcnt))
994 ret = __bio_iov_bvec_add_pages(bio, iter);
996 ret = __bio_iov_iter_get_pages(bio, iter);
997 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1000 bio_set_flag(bio, BIO_NO_PAGE_REF);
1001 return bio->bi_vcnt ? 0 : ret;
1004 static void submit_bio_wait_endio(struct bio *bio)
1006 complete(bio->bi_private);
1010 * submit_bio_wait - submit a bio, and wait until it completes
1011 * @bio: The &struct bio which describes the I/O
1013 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1014 * bio_endio() on failure.
1016 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1017 * result in bio reference to be consumed. The caller must drop the reference
1020 int submit_bio_wait(struct bio *bio)
1022 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
1023 unsigned long hang_check;
1025 bio->bi_private = &done;
1026 bio->bi_end_io = submit_bio_wait_endio;
1027 bio->bi_opf |= REQ_SYNC;
1030 /* Prevent hang_check timer from firing at us during very long I/O */
1031 hang_check = sysctl_hung_task_timeout_secs;
1033 while (!wait_for_completion_io_timeout(&done,
1034 hang_check * (HZ/2)))
1037 wait_for_completion_io(&done);
1039 return blk_status_to_errno(bio->bi_status);
1041 EXPORT_SYMBOL(submit_bio_wait);
1044 * bio_advance - increment/complete a bio by some number of bytes
1045 * @bio: bio to advance
1046 * @bytes: number of bytes to complete
1048 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1049 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1050 * be updated on the last bvec as well.
1052 * @bio will then represent the remaining, uncompleted portion of the io.
1054 void bio_advance(struct bio *bio, unsigned bytes)
1056 if (bio_integrity(bio))
1057 bio_integrity_advance(bio, bytes);
1059 bio_advance_iter(bio, &bio->bi_iter, bytes);
1061 EXPORT_SYMBOL(bio_advance);
1063 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1064 struct bio *src, struct bvec_iter *src_iter)
1066 struct bio_vec src_bv, dst_bv;
1067 void *src_p, *dst_p;
1070 while (src_iter->bi_size && dst_iter->bi_size) {
1071 src_bv = bio_iter_iovec(src, *src_iter);
1072 dst_bv = bio_iter_iovec(dst, *dst_iter);
1074 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1076 src_p = kmap_atomic(src_bv.bv_page);
1077 dst_p = kmap_atomic(dst_bv.bv_page);
1079 memcpy(dst_p + dst_bv.bv_offset,
1080 src_p + src_bv.bv_offset,
1083 kunmap_atomic(dst_p);
1084 kunmap_atomic(src_p);
1086 flush_dcache_page(dst_bv.bv_page);
1088 bio_advance_iter(src, src_iter, bytes);
1089 bio_advance_iter(dst, dst_iter, bytes);
1092 EXPORT_SYMBOL(bio_copy_data_iter);
1095 * bio_copy_data - copy contents of data buffers from one bio to another
1097 * @dst: destination bio
1099 * Stops when it reaches the end of either @src or @dst - that is, copies
1100 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1102 void bio_copy_data(struct bio *dst, struct bio *src)
1104 struct bvec_iter src_iter = src->bi_iter;
1105 struct bvec_iter dst_iter = dst->bi_iter;
1107 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1109 EXPORT_SYMBOL(bio_copy_data);
1112 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1114 * @src: source bio list
1115 * @dst: destination bio list
1117 * Stops when it reaches the end of either the @src list or @dst list - that is,
1118 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1121 void bio_list_copy_data(struct bio *dst, struct bio *src)
1123 struct bvec_iter src_iter = src->bi_iter;
1124 struct bvec_iter dst_iter = dst->bi_iter;
1127 if (!src_iter.bi_size) {
1132 src_iter = src->bi_iter;
1135 if (!dst_iter.bi_size) {
1140 dst_iter = dst->bi_iter;
1143 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1146 EXPORT_SYMBOL(bio_list_copy_data);
1148 struct bio_map_data {
1150 struct iov_iter iter;
1154 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
1157 struct bio_map_data *bmd;
1158 if (data->nr_segs > UIO_MAXIOV)
1161 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask);
1164 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1166 bmd->iter.iov = bmd->iov;
1171 * bio_copy_from_iter - copy all pages from iov_iter to bio
1172 * @bio: The &struct bio which describes the I/O as destination
1173 * @iter: iov_iter as source
1175 * Copy all pages from iov_iter to bio.
1176 * Returns 0 on success, or error on failure.
1178 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
1180 struct bio_vec *bvec;
1181 struct bvec_iter_all iter_all;
1183 bio_for_each_segment_all(bvec, bio, iter_all) {
1186 ret = copy_page_from_iter(bvec->bv_page,
1191 if (!iov_iter_count(iter))
1194 if (ret < bvec->bv_len)
1202 * bio_copy_to_iter - copy all pages from bio to iov_iter
1203 * @bio: The &struct bio which describes the I/O as source
1204 * @iter: iov_iter as destination
1206 * Copy all pages from bio to iov_iter.
1207 * Returns 0 on success, or error on failure.
1209 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1211 struct bio_vec *bvec;
1212 struct bvec_iter_all iter_all;
1214 bio_for_each_segment_all(bvec, bio, iter_all) {
1217 ret = copy_page_to_iter(bvec->bv_page,
1222 if (!iov_iter_count(&iter))
1225 if (ret < bvec->bv_len)
1232 void bio_free_pages(struct bio *bio)
1234 struct bio_vec *bvec;
1235 struct bvec_iter_all iter_all;
1237 bio_for_each_segment_all(bvec, bio, iter_all)
1238 __free_page(bvec->bv_page);
1240 EXPORT_SYMBOL(bio_free_pages);
1243 * bio_uncopy_user - finish previously mapped bio
1244 * @bio: bio being terminated
1246 * Free pages allocated from bio_copy_user_iov() and write back data
1247 * to user space in case of a read.
1249 int bio_uncopy_user(struct bio *bio)
1251 struct bio_map_data *bmd = bio->bi_private;
1254 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1256 * if we're in a workqueue, the request is orphaned, so
1257 * don't copy into a random user address space, just free
1258 * and return -EINTR so user space doesn't expect any data.
1262 else if (bio_data_dir(bio) == READ)
1263 ret = bio_copy_to_iter(bio, bmd->iter);
1264 if (bmd->is_our_pages)
1265 bio_free_pages(bio);
1273 * bio_copy_user_iov - copy user data to bio
1274 * @q: destination block queue
1275 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1276 * @iter: iovec iterator
1277 * @gfp_mask: memory allocation flags
1279 * Prepares and returns a bio for indirect user io, bouncing data
1280 * to/from kernel pages as necessary. Must be paired with
1281 * call bio_uncopy_user() on io completion.
1283 struct bio *bio_copy_user_iov(struct request_queue *q,
1284 struct rq_map_data *map_data,
1285 struct iov_iter *iter,
1288 struct bio_map_data *bmd;
1293 unsigned int len = iter->count;
1294 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1296 bmd = bio_alloc_map_data(iter, gfp_mask);
1298 return ERR_PTR(-ENOMEM);
1301 * We need to do a deep copy of the iov_iter including the iovecs.
1302 * The caller provided iov might point to an on-stack or otherwise
1305 bmd->is_our_pages = map_data ? 0 : 1;
1307 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1308 if (nr_pages > BIO_MAX_PAGES)
1309 nr_pages = BIO_MAX_PAGES;
1312 bio = bio_kmalloc(gfp_mask, nr_pages);
1319 nr_pages = 1 << map_data->page_order;
1320 i = map_data->offset / PAGE_SIZE;
1323 unsigned int bytes = PAGE_SIZE;
1331 if (i == map_data->nr_entries * nr_pages) {
1336 page = map_data->pages[i / nr_pages];
1337 page += (i % nr_pages);
1341 page = alloc_page(q->bounce_gfp | gfp_mask);
1348 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) {
1362 map_data->offset += bio->bi_iter.bi_size;
1367 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) ||
1368 (map_data && map_data->from_user)) {
1369 ret = bio_copy_from_iter(bio, iter);
1373 if (bmd->is_our_pages)
1375 iov_iter_advance(iter, bio->bi_iter.bi_size);
1378 bio->bi_private = bmd;
1379 if (map_data && map_data->null_mapped)
1380 bio_set_flag(bio, BIO_NULL_MAPPED);
1384 bio_free_pages(bio);
1388 return ERR_PTR(ret);
1392 * bio_map_user_iov - map user iovec into bio
1393 * @q: the struct request_queue for the bio
1394 * @iter: iovec iterator
1395 * @gfp_mask: memory allocation flags
1397 * Map the user space address into a bio suitable for io to a block
1398 * device. Returns an error pointer in case of error.
1400 struct bio *bio_map_user_iov(struct request_queue *q,
1401 struct iov_iter *iter,
1408 if (!iov_iter_count(iter))
1409 return ERR_PTR(-EINVAL);
1411 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
1413 return ERR_PTR(-ENOMEM);
1415 while (iov_iter_count(iter)) {
1416 struct page **pages;
1418 size_t offs, added = 0;
1421 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
1422 if (unlikely(bytes <= 0)) {
1423 ret = bytes ? bytes : -EFAULT;
1427 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
1429 if (unlikely(offs & queue_dma_alignment(q))) {
1433 for (j = 0; j < npages; j++) {
1434 struct page *page = pages[j];
1435 unsigned int n = PAGE_SIZE - offs;
1436 bool same_page = false;
1441 if (!__bio_add_pc_page(q, bio, page, n, offs,
1452 iov_iter_advance(iter, added);
1455 * release the pages we didn't map into the bio, if any
1458 put_page(pages[j++]);
1460 /* couldn't stuff something into bio? */
1465 bio_set_flag(bio, BIO_USER_MAPPED);
1468 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1469 * it would normally disappear when its bi_end_io is run.
1470 * however, we need it for the unmap, so grab an extra
1477 bio_release_pages(bio, false);
1479 return ERR_PTR(ret);
1483 * bio_unmap_user - unmap a bio
1484 * @bio: the bio being unmapped
1486 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1489 * bio_unmap_user() may sleep.
1491 void bio_unmap_user(struct bio *bio)
1493 bio_release_pages(bio, bio_data_dir(bio) == READ);
1498 static void bio_invalidate_vmalloc_pages(struct bio *bio)
1500 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
1501 if (bio->bi_private && !op_is_write(bio_op(bio))) {
1502 unsigned long i, len = 0;
1504 for (i = 0; i < bio->bi_vcnt; i++)
1505 len += bio->bi_io_vec[i].bv_len;
1506 invalidate_kernel_vmap_range(bio->bi_private, len);
1511 static void bio_map_kern_endio(struct bio *bio)
1513 bio_invalidate_vmalloc_pages(bio);
1518 * bio_map_kern - map kernel address into bio
1519 * @q: the struct request_queue for the bio
1520 * @data: pointer to buffer to map
1521 * @len: length in bytes
1522 * @gfp_mask: allocation flags for bio allocation
1524 * Map the kernel address into a bio suitable for io to a block
1525 * device. Returns an error pointer in case of error.
1527 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1530 unsigned long kaddr = (unsigned long)data;
1531 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1532 unsigned long start = kaddr >> PAGE_SHIFT;
1533 const int nr_pages = end - start;
1534 bool is_vmalloc = is_vmalloc_addr(data);
1539 bio = bio_kmalloc(gfp_mask, nr_pages);
1541 return ERR_PTR(-ENOMEM);
1544 flush_kernel_vmap_range(data, len);
1545 bio->bi_private = data;
1548 offset = offset_in_page(kaddr);
1549 for (i = 0; i < nr_pages; i++) {
1550 unsigned int bytes = PAGE_SIZE - offset;
1559 page = virt_to_page(data);
1561 page = vmalloc_to_page(data);
1562 if (bio_add_pc_page(q, bio, page, bytes,
1564 /* we don't support partial mappings */
1566 return ERR_PTR(-EINVAL);
1574 bio->bi_end_io = bio_map_kern_endio;
1578 static void bio_copy_kern_endio(struct bio *bio)
1580 bio_free_pages(bio);
1584 static void bio_copy_kern_endio_read(struct bio *bio)
1586 char *p = bio->bi_private;
1587 struct bio_vec *bvec;
1588 struct bvec_iter_all iter_all;
1590 bio_for_each_segment_all(bvec, bio, iter_all) {
1591 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1595 bio_copy_kern_endio(bio);
1599 * bio_copy_kern - copy kernel address into bio
1600 * @q: the struct request_queue for the bio
1601 * @data: pointer to buffer to copy
1602 * @len: length in bytes
1603 * @gfp_mask: allocation flags for bio and page allocation
1604 * @reading: data direction is READ
1606 * copy the kernel address into a bio suitable for io to a block
1607 * device. Returns an error pointer in case of error.
1609 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1610 gfp_t gfp_mask, int reading)
1612 unsigned long kaddr = (unsigned long)data;
1613 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1614 unsigned long start = kaddr >> PAGE_SHIFT;
1623 return ERR_PTR(-EINVAL);
1625 nr_pages = end - start;
1626 bio = bio_kmalloc(gfp_mask, nr_pages);
1628 return ERR_PTR(-ENOMEM);
1632 unsigned int bytes = PAGE_SIZE;
1637 page = alloc_page(q->bounce_gfp | gfp_mask);
1642 memcpy(page_address(page), p, bytes);
1644 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1652 bio->bi_end_io = bio_copy_kern_endio_read;
1653 bio->bi_private = data;
1655 bio->bi_end_io = bio_copy_kern_endio;
1661 bio_free_pages(bio);
1663 return ERR_PTR(-ENOMEM);
1667 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1668 * for performing direct-IO in BIOs.
1670 * The problem is that we cannot run set_page_dirty() from interrupt context
1671 * because the required locks are not interrupt-safe. So what we can do is to
1672 * mark the pages dirty _before_ performing IO. And in interrupt context,
1673 * check that the pages are still dirty. If so, fine. If not, redirty them
1674 * in process context.
1676 * We special-case compound pages here: normally this means reads into hugetlb
1677 * pages. The logic in here doesn't really work right for compound pages
1678 * because the VM does not uniformly chase down the head page in all cases.
1679 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1680 * handle them at all. So we skip compound pages here at an early stage.
1682 * Note that this code is very hard to test under normal circumstances because
1683 * direct-io pins the pages with get_user_pages(). This makes
1684 * is_page_cache_freeable return false, and the VM will not clean the pages.
1685 * But other code (eg, flusher threads) could clean the pages if they are mapped
1688 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1689 * deferred bio dirtying paths.
1693 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1695 void bio_set_pages_dirty(struct bio *bio)
1697 struct bio_vec *bvec;
1698 struct bvec_iter_all iter_all;
1700 bio_for_each_segment_all(bvec, bio, iter_all) {
1701 if (!PageCompound(bvec->bv_page))
1702 set_page_dirty_lock(bvec->bv_page);
1707 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1708 * If they are, then fine. If, however, some pages are clean then they must
1709 * have been written out during the direct-IO read. So we take another ref on
1710 * the BIO and re-dirty the pages in process context.
1712 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1713 * here on. It will run one put_page() against each page and will run one
1714 * bio_put() against the BIO.
1717 static void bio_dirty_fn(struct work_struct *work);
1719 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1720 static DEFINE_SPINLOCK(bio_dirty_lock);
1721 static struct bio *bio_dirty_list;
1724 * This runs in process context
1726 static void bio_dirty_fn(struct work_struct *work)
1728 struct bio *bio, *next;
1730 spin_lock_irq(&bio_dirty_lock);
1731 next = bio_dirty_list;
1732 bio_dirty_list = NULL;
1733 spin_unlock_irq(&bio_dirty_lock);
1735 while ((bio = next) != NULL) {
1736 next = bio->bi_private;
1738 bio_release_pages(bio, true);
1743 void bio_check_pages_dirty(struct bio *bio)
1745 struct bio_vec *bvec;
1746 unsigned long flags;
1747 struct bvec_iter_all iter_all;
1749 bio_for_each_segment_all(bvec, bio, iter_all) {
1750 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1754 bio_release_pages(bio, false);
1758 spin_lock_irqsave(&bio_dirty_lock, flags);
1759 bio->bi_private = bio_dirty_list;
1760 bio_dirty_list = bio;
1761 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1762 schedule_work(&bio_dirty_work);
1765 void update_io_ticks(struct hd_struct *part, unsigned long now)
1767 unsigned long stamp;
1769 stamp = READ_ONCE(part->stamp);
1770 if (unlikely(stamp != now)) {
1771 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) {
1772 __part_stat_add(part, io_ticks, 1);
1776 part = &part_to_disk(part)->part0;
1781 void generic_start_io_acct(struct request_queue *q, int op,
1782 unsigned long sectors, struct hd_struct *part)
1784 const int sgrp = op_stat_group(op);
1788 update_io_ticks(part, jiffies);
1789 part_stat_inc(part, ios[sgrp]);
1790 part_stat_add(part, sectors[sgrp], sectors);
1791 part_inc_in_flight(q, part, op_is_write(op));
1795 EXPORT_SYMBOL(generic_start_io_acct);
1797 void generic_end_io_acct(struct request_queue *q, int req_op,
1798 struct hd_struct *part, unsigned long start_time)
1800 unsigned long now = jiffies;
1801 unsigned long duration = now - start_time;
1802 const int sgrp = op_stat_group(req_op);
1806 update_io_ticks(part, now);
1807 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1808 part_stat_add(part, time_in_queue, duration);
1809 part_dec_in_flight(q, part, op_is_write(req_op));
1813 EXPORT_SYMBOL(generic_end_io_acct);
1815 static inline bool bio_remaining_done(struct bio *bio)
1818 * If we're not chaining, then ->__bi_remaining is always 1 and
1819 * we always end io on the first invocation.
1821 if (!bio_flagged(bio, BIO_CHAIN))
1824 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1826 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1827 bio_clear_flag(bio, BIO_CHAIN);
1835 * bio_endio - end I/O on a bio
1839 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1840 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1841 * bio unless they own it and thus know that it has an end_io function.
1843 * bio_endio() can be called several times on a bio that has been chained
1844 * using bio_chain(). The ->bi_end_io() function will only be called the
1845 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1846 * generated if BIO_TRACE_COMPLETION is set.
1848 void bio_endio(struct bio *bio)
1851 if (!bio_remaining_done(bio))
1853 if (!bio_integrity_endio(bio))
1857 rq_qos_done_bio(bio->bi_disk->queue, bio);
1860 * Need to have a real endio function for chained bios, otherwise
1861 * various corner cases will break (like stacking block devices that
1862 * save/restore bi_end_io) - however, we want to avoid unbounded
1863 * recursion and blowing the stack. Tail call optimization would
1864 * handle this, but compiling with frame pointers also disables
1865 * gcc's sibling call optimization.
1867 if (bio->bi_end_io == bio_chain_endio) {
1868 bio = __bio_chain_endio(bio);
1872 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1873 trace_block_bio_complete(bio->bi_disk->queue, bio,
1874 blk_status_to_errno(bio->bi_status));
1875 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1878 blk_throtl_bio_endio(bio);
1879 /* release cgroup info */
1882 bio->bi_end_io(bio);
1884 EXPORT_SYMBOL(bio_endio);
1887 * bio_split - split a bio
1888 * @bio: bio to split
1889 * @sectors: number of sectors to split from the front of @bio
1891 * @bs: bio set to allocate from
1893 * Allocates and returns a new bio which represents @sectors from the start of
1894 * @bio, and updates @bio to represent the remaining sectors.
1896 * Unless this is a discard request the newly allocated bio will point
1897 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1898 * neither @bio nor @bs are freed before the split bio.
1900 struct bio *bio_split(struct bio *bio, int sectors,
1901 gfp_t gfp, struct bio_set *bs)
1905 BUG_ON(sectors <= 0);
1906 BUG_ON(sectors >= bio_sectors(bio));
1908 split = bio_clone_fast(bio, gfp, bs);
1912 split->bi_iter.bi_size = sectors << 9;
1914 if (bio_integrity(split))
1915 bio_integrity_trim(split);
1917 bio_advance(bio, split->bi_iter.bi_size);
1919 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1920 bio_set_flag(split, BIO_TRACE_COMPLETION);
1924 EXPORT_SYMBOL(bio_split);
1927 * bio_trim - trim a bio
1929 * @offset: number of sectors to trim from the front of @bio
1930 * @size: size we want to trim @bio to, in sectors
1932 void bio_trim(struct bio *bio, int offset, int size)
1934 /* 'bio' is a cloned bio which we need to trim to match
1935 * the given offset and size.
1939 if (offset == 0 && size == bio->bi_iter.bi_size)
1942 bio_advance(bio, offset << 9);
1943 bio->bi_iter.bi_size = size;
1945 if (bio_integrity(bio))
1946 bio_integrity_trim(bio);
1949 EXPORT_SYMBOL_GPL(bio_trim);
1952 * create memory pools for biovec's in a bio_set.
1953 * use the global biovec slabs created for general use.
1955 int biovec_init_pool(mempool_t *pool, int pool_entries)
1957 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1959 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1963 * bioset_exit - exit a bioset initialized with bioset_init()
1965 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1968 void bioset_exit(struct bio_set *bs)
1970 if (bs->rescue_workqueue)
1971 destroy_workqueue(bs->rescue_workqueue);
1972 bs->rescue_workqueue = NULL;
1974 mempool_exit(&bs->bio_pool);
1975 mempool_exit(&bs->bvec_pool);
1977 bioset_integrity_free(bs);
1980 bs->bio_slab = NULL;
1982 EXPORT_SYMBOL(bioset_exit);
1985 * bioset_init - Initialize a bio_set
1986 * @bs: pool to initialize
1987 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1988 * @front_pad: Number of bytes to allocate in front of the returned bio
1989 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1990 * and %BIOSET_NEED_RESCUER
1993 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1994 * to ask for a number of bytes to be allocated in front of the bio.
1995 * Front pad allocation is useful for embedding the bio inside
1996 * another structure, to avoid allocating extra data to go with the bio.
1997 * Note that the bio must be embedded at the END of that structure always,
1998 * or things will break badly.
1999 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
2000 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
2001 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
2002 * dispatch queued requests when the mempool runs out of space.
2005 int bioset_init(struct bio_set *bs,
2006 unsigned int pool_size,
2007 unsigned int front_pad,
2010 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
2012 bs->front_pad = front_pad;
2014 spin_lock_init(&bs->rescue_lock);
2015 bio_list_init(&bs->rescue_list);
2016 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
2018 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
2022 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
2025 if ((flags & BIOSET_NEED_BVECS) &&
2026 biovec_init_pool(&bs->bvec_pool, pool_size))
2029 if (!(flags & BIOSET_NEED_RESCUER))
2032 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
2033 if (!bs->rescue_workqueue)
2041 EXPORT_SYMBOL(bioset_init);
2044 * Initialize and setup a new bio_set, based on the settings from
2047 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
2052 if (src->bvec_pool.min_nr)
2053 flags |= BIOSET_NEED_BVECS;
2054 if (src->rescue_workqueue)
2055 flags |= BIOSET_NEED_RESCUER;
2057 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
2059 EXPORT_SYMBOL(bioset_init_from_src);
2061 #ifdef CONFIG_BLK_CGROUP
2064 * bio_disassociate_blkg - puts back the blkg reference if associated
2067 * Helper to disassociate the blkg from @bio if a blkg is associated.
2069 void bio_disassociate_blkg(struct bio *bio)
2072 blkg_put(bio->bi_blkg);
2073 bio->bi_blkg = NULL;
2076 EXPORT_SYMBOL_GPL(bio_disassociate_blkg);
2079 * __bio_associate_blkg - associate a bio with the a blkg
2081 * @blkg: the blkg to associate
2083 * This tries to associate @bio with the specified @blkg. Association failure
2084 * is handled by walking up the blkg tree. Therefore, the blkg associated can
2085 * be anything between @blkg and the root_blkg. This situation only happens
2086 * when a cgroup is dying and then the remaining bios will spill to the closest
2089 * A reference will be taken on the @blkg and will be released when @bio is
2092 static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg)
2094 bio_disassociate_blkg(bio);
2096 bio->bi_blkg = blkg_tryget_closest(blkg);
2100 * bio_associate_blkg_from_css - associate a bio with a specified css
2104 * Associate @bio with the blkg found by combining the css's blkg and the
2105 * request_queue of the @bio. This falls back to the queue's root_blkg if
2106 * the association fails with the css.
2108 void bio_associate_blkg_from_css(struct bio *bio,
2109 struct cgroup_subsys_state *css)
2111 struct request_queue *q = bio->bi_disk->queue;
2112 struct blkcg_gq *blkg;
2116 if (!css || !css->parent)
2117 blkg = q->root_blkg;
2119 blkg = blkg_lookup_create(css_to_blkcg(css), q);
2121 __bio_associate_blkg(bio, blkg);
2125 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css);
2129 * bio_associate_blkg_from_page - associate a bio with the page's blkg
2131 * @page: the page to lookup the blkcg from
2133 * Associate @bio with the blkg from @page's owning memcg and the respective
2134 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's
2137 void bio_associate_blkg_from_page(struct bio *bio, struct page *page)
2139 struct cgroup_subsys_state *css;
2141 if (!page->mem_cgroup)
2146 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
2147 bio_associate_blkg_from_css(bio, css);
2151 #endif /* CONFIG_MEMCG */
2154 * bio_associate_blkg - associate a bio with a blkg
2157 * Associate @bio with the blkg found from the bio's css and request_queue.
2158 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is
2159 * already associated, the css is reused and association redone as the
2160 * request_queue may have changed.
2162 void bio_associate_blkg(struct bio *bio)
2164 struct cgroup_subsys_state *css;
2169 css = &bio_blkcg(bio)->css;
2173 bio_associate_blkg_from_css(bio, css);
2177 EXPORT_SYMBOL_GPL(bio_associate_blkg);
2180 * bio_clone_blkg_association - clone blkg association from src to dst bio
2181 * @dst: destination bio
2184 void bio_clone_blkg_association(struct bio *dst, struct bio *src)
2189 __bio_associate_blkg(dst, src->bi_blkg);
2193 EXPORT_SYMBOL_GPL(bio_clone_blkg_association);
2194 #endif /* CONFIG_BLK_CGROUP */
2196 static void __init biovec_init_slabs(void)
2200 for (i = 0; i < BVEC_POOL_NR; i++) {
2202 struct biovec_slab *bvs = bvec_slabs + i;
2204 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2209 size = bvs->nr_vecs * sizeof(struct bio_vec);
2210 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2211 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2215 static int __init init_bio(void)
2219 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab),
2222 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET);
2225 panic("bio: can't allocate bios\n");
2227 bio_integrity_init();
2228 biovec_init_slabs();
2230 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
2231 panic("bio: can't allocate bios\n");
2233 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
2234 panic("bio: can't create integrity pool\n");
2238 subsys_initcall(init_bio);