2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
32 #include <trace/events/block.h>
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
39 #define BIO_INLINE_VECS 4
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
46 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
47 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
48 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
56 struct bio_set fs_bio_set;
57 EXPORT_SYMBOL(fs_bio_set);
60 * Our slab pool management
63 struct kmem_cache *slab;
64 unsigned int slab_ref;
65 unsigned int slab_size;
68 static DEFINE_MUTEX(bio_slab_lock);
69 static struct bio_slab *bio_slabs;
70 static unsigned int bio_slab_nr, bio_slab_max;
72 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
74 unsigned int sz = sizeof(struct bio) + extra_size;
75 struct kmem_cache *slab = NULL;
76 struct bio_slab *bslab, *new_bio_slabs;
77 unsigned int new_bio_slab_max;
78 unsigned int i, entry = -1;
80 mutex_lock(&bio_slab_lock);
83 while (i < bio_slab_nr) {
84 bslab = &bio_slabs[i];
86 if (!bslab->slab && entry == -1)
88 else if (bslab->slab_size == sz) {
99 if (bio_slab_nr == bio_slab_max && entry == -1) {
100 new_bio_slab_max = bio_slab_max << 1;
101 new_bio_slabs = krealloc(bio_slabs,
102 new_bio_slab_max * sizeof(struct bio_slab),
106 bio_slab_max = new_bio_slab_max;
107 bio_slabs = new_bio_slabs;
110 entry = bio_slab_nr++;
112 bslab = &bio_slabs[entry];
114 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
115 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
116 SLAB_HWCACHE_ALIGN, NULL);
122 bslab->slab_size = sz;
124 mutex_unlock(&bio_slab_lock);
128 static void bio_put_slab(struct bio_set *bs)
130 struct bio_slab *bslab = NULL;
133 mutex_lock(&bio_slab_lock);
135 for (i = 0; i < bio_slab_nr; i++) {
136 if (bs->bio_slab == bio_slabs[i].slab) {
137 bslab = &bio_slabs[i];
142 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 WARN_ON(!bslab->slab_ref);
147 if (--bslab->slab_ref)
150 kmem_cache_destroy(bslab->slab);
154 mutex_unlock(&bio_slab_lock);
157 unsigned int bvec_nr_vecs(unsigned short idx)
159 return bvec_slabs[idx].nr_vecs;
162 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
168 BIO_BUG_ON(idx >= BVEC_POOL_NR);
170 if (idx == BVEC_POOL_MAX) {
171 mempool_free(bv, pool);
173 struct biovec_slab *bvs = bvec_slabs + idx;
175 kmem_cache_free(bvs->slab, bv);
179 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
185 * see comment near bvec_array define!
203 case 129 ... BIO_MAX_PAGES:
211 * idx now points to the pool we want to allocate from. only the
212 * 1-vec entry pool is mempool backed.
214 if (*idx == BVEC_POOL_MAX) {
216 bvl = mempool_alloc(pool, gfp_mask);
218 struct biovec_slab *bvs = bvec_slabs + *idx;
219 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
222 * Make this allocation restricted and don't dump info on
223 * allocation failures, since we'll fallback to the mempool
224 * in case of failure.
226 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
229 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
230 * is set, retry with the 1-entry mempool
232 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
233 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
234 *idx = BVEC_POOL_MAX;
243 void bio_uninit(struct bio *bio)
245 bio_disassociate_task(bio);
247 EXPORT_SYMBOL(bio_uninit);
249 static void bio_free(struct bio *bio)
251 struct bio_set *bs = bio->bi_pool;
257 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
260 * If we have front padding, adjust the bio pointer before freeing
265 mempool_free(p, &bs->bio_pool);
267 /* Bio was allocated by bio_kmalloc() */
273 * Users of this function have their own bio allocation. Subsequently,
274 * they must remember to pair any call to bio_init() with bio_uninit()
275 * when IO has completed, or when the bio is released.
277 void bio_init(struct bio *bio, struct bio_vec *table,
278 unsigned short max_vecs)
280 memset(bio, 0, sizeof(*bio));
281 atomic_set(&bio->__bi_remaining, 1);
282 atomic_set(&bio->__bi_cnt, 1);
284 bio->bi_io_vec = table;
285 bio->bi_max_vecs = max_vecs;
287 EXPORT_SYMBOL(bio_init);
290 * bio_reset - reinitialize a bio
294 * After calling bio_reset(), @bio will be in the same state as a freshly
295 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
296 * preserved are the ones that are initialized by bio_alloc_bioset(). See
297 * comment in struct bio.
299 void bio_reset(struct bio *bio)
301 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
305 memset(bio, 0, BIO_RESET_BYTES);
306 bio->bi_flags = flags;
307 atomic_set(&bio->__bi_remaining, 1);
309 EXPORT_SYMBOL(bio_reset);
311 static struct bio *__bio_chain_endio(struct bio *bio)
313 struct bio *parent = bio->bi_private;
315 if (!parent->bi_status)
316 parent->bi_status = bio->bi_status;
321 static void bio_chain_endio(struct bio *bio)
323 bio_endio(__bio_chain_endio(bio));
327 * bio_chain - chain bio completions
328 * @bio: the target bio
329 * @parent: the @bio's parent bio
331 * The caller won't have a bi_end_io called when @bio completes - instead,
332 * @parent's bi_end_io won't be called until both @parent and @bio have
333 * completed; the chained bio will also be freed when it completes.
335 * The caller must not set bi_private or bi_end_io in @bio.
337 void bio_chain(struct bio *bio, struct bio *parent)
339 BUG_ON(bio->bi_private || bio->bi_end_io);
341 bio->bi_private = parent;
342 bio->bi_end_io = bio_chain_endio;
343 bio_inc_remaining(parent);
345 EXPORT_SYMBOL(bio_chain);
347 static void bio_alloc_rescue(struct work_struct *work)
349 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
353 spin_lock(&bs->rescue_lock);
354 bio = bio_list_pop(&bs->rescue_list);
355 spin_unlock(&bs->rescue_lock);
360 generic_make_request(bio);
364 static void punt_bios_to_rescuer(struct bio_set *bs)
366 struct bio_list punt, nopunt;
369 if (WARN_ON_ONCE(!bs->rescue_workqueue))
372 * In order to guarantee forward progress we must punt only bios that
373 * were allocated from this bio_set; otherwise, if there was a bio on
374 * there for a stacking driver higher up in the stack, processing it
375 * could require allocating bios from this bio_set, and doing that from
376 * our own rescuer would be bad.
378 * Since bio lists are singly linked, pop them all instead of trying to
379 * remove from the middle of the list:
382 bio_list_init(&punt);
383 bio_list_init(&nopunt);
385 while ((bio = bio_list_pop(¤t->bio_list[0])))
386 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
387 current->bio_list[0] = nopunt;
389 bio_list_init(&nopunt);
390 while ((bio = bio_list_pop(¤t->bio_list[1])))
391 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
392 current->bio_list[1] = nopunt;
394 spin_lock(&bs->rescue_lock);
395 bio_list_merge(&bs->rescue_list, &punt);
396 spin_unlock(&bs->rescue_lock);
398 queue_work(bs->rescue_workqueue, &bs->rescue_work);
402 * bio_alloc_bioset - allocate a bio for I/O
403 * @gfp_mask: the GFP_* mask given to the slab allocator
404 * @nr_iovecs: number of iovecs to pre-allocate
405 * @bs: the bio_set to allocate from.
408 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
409 * backed by the @bs's mempool.
411 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
412 * always be able to allocate a bio. This is due to the mempool guarantees.
413 * To make this work, callers must never allocate more than 1 bio at a time
414 * from this pool. Callers that need to allocate more than 1 bio must always
415 * submit the previously allocated bio for IO before attempting to allocate
416 * a new one. Failure to do so can cause deadlocks under memory pressure.
418 * Note that when running under generic_make_request() (i.e. any block
419 * driver), bios are not submitted until after you return - see the code in
420 * generic_make_request() that converts recursion into iteration, to prevent
423 * This would normally mean allocating multiple bios under
424 * generic_make_request() would be susceptible to deadlocks, but we have
425 * deadlock avoidance code that resubmits any blocked bios from a rescuer
428 * However, we do not guarantee forward progress for allocations from other
429 * mempools. Doing multiple allocations from the same mempool under
430 * generic_make_request() should be avoided - instead, use bio_set's front_pad
431 * for per bio allocations.
434 * Pointer to new bio on success, NULL on failure.
436 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
439 gfp_t saved_gfp = gfp_mask;
441 unsigned inline_vecs;
442 struct bio_vec *bvl = NULL;
447 if (nr_iovecs > UIO_MAXIOV)
450 p = kmalloc(sizeof(struct bio) +
451 nr_iovecs * sizeof(struct bio_vec),
454 inline_vecs = nr_iovecs;
456 /* should not use nobvec bioset for nr_iovecs > 0 */
457 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
461 * generic_make_request() converts recursion to iteration; this
462 * means if we're running beneath it, any bios we allocate and
463 * submit will not be submitted (and thus freed) until after we
466 * This exposes us to a potential deadlock if we allocate
467 * multiple bios from the same bio_set() while running
468 * underneath generic_make_request(). If we were to allocate
469 * multiple bios (say a stacking block driver that was splitting
470 * bios), we would deadlock if we exhausted the mempool's
473 * We solve this, and guarantee forward progress, with a rescuer
474 * workqueue per bio_set. If we go to allocate and there are
475 * bios on current->bio_list, we first try the allocation
476 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
477 * bios we would be blocking to the rescuer workqueue before
478 * we retry with the original gfp_flags.
481 if (current->bio_list &&
482 (!bio_list_empty(¤t->bio_list[0]) ||
483 !bio_list_empty(¤t->bio_list[1])) &&
484 bs->rescue_workqueue)
485 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
487 p = mempool_alloc(&bs->bio_pool, gfp_mask);
488 if (!p && gfp_mask != saved_gfp) {
489 punt_bios_to_rescuer(bs);
490 gfp_mask = saved_gfp;
491 p = mempool_alloc(&bs->bio_pool, gfp_mask);
494 front_pad = bs->front_pad;
495 inline_vecs = BIO_INLINE_VECS;
502 bio_init(bio, NULL, 0);
504 if (nr_iovecs > inline_vecs) {
505 unsigned long idx = 0;
507 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
508 if (!bvl && gfp_mask != saved_gfp) {
509 punt_bios_to_rescuer(bs);
510 gfp_mask = saved_gfp;
511 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
517 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
518 } else if (nr_iovecs) {
519 bvl = bio->bi_inline_vecs;
523 bio->bi_max_vecs = nr_iovecs;
524 bio->bi_io_vec = bvl;
528 mempool_free(p, &bs->bio_pool);
531 EXPORT_SYMBOL(bio_alloc_bioset);
533 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
537 struct bvec_iter iter;
539 __bio_for_each_segment(bv, bio, iter, start) {
540 char *data = bvec_kmap_irq(&bv, &flags);
541 memset(data, 0, bv.bv_len);
542 flush_dcache_page(bv.bv_page);
543 bvec_kunmap_irq(data, &flags);
546 EXPORT_SYMBOL(zero_fill_bio_iter);
549 * bio_put - release a reference to a bio
550 * @bio: bio to release reference to
553 * Put a reference to a &struct bio, either one you have gotten with
554 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
556 void bio_put(struct bio *bio)
558 if (!bio_flagged(bio, BIO_REFFED))
561 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
566 if (atomic_dec_and_test(&bio->__bi_cnt))
570 EXPORT_SYMBOL(bio_put);
572 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
574 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
575 blk_recount_segments(q, bio);
577 return bio->bi_phys_segments;
579 EXPORT_SYMBOL(bio_phys_segments);
582 * __bio_clone_fast - clone a bio that shares the original bio's biovec
583 * @bio: destination bio
584 * @bio_src: bio to clone
586 * Clone a &bio. Caller will own the returned bio, but not
587 * the actual data it points to. Reference count of returned
590 * Caller must ensure that @bio_src is not freed before @bio.
592 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
594 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
597 * most users will be overriding ->bi_disk with a new target,
598 * so we don't set nor calculate new physical/hw segment counts here
600 bio->bi_disk = bio_src->bi_disk;
601 bio->bi_partno = bio_src->bi_partno;
602 bio_set_flag(bio, BIO_CLONED);
603 if (bio_flagged(bio_src, BIO_THROTTLED))
604 bio_set_flag(bio, BIO_THROTTLED);
605 bio->bi_opf = bio_src->bi_opf;
606 bio->bi_write_hint = bio_src->bi_write_hint;
607 bio->bi_iter = bio_src->bi_iter;
608 bio->bi_io_vec = bio_src->bi_io_vec;
610 bio_clone_blkcg_association(bio, bio_src);
612 EXPORT_SYMBOL(__bio_clone_fast);
615 * bio_clone_fast - clone a bio that shares the original bio's biovec
617 * @gfp_mask: allocation priority
618 * @bs: bio_set to allocate from
620 * Like __bio_clone_fast, only also allocates the returned bio
622 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
626 b = bio_alloc_bioset(gfp_mask, 0, bs);
630 __bio_clone_fast(b, bio);
632 if (bio_integrity(bio)) {
635 ret = bio_integrity_clone(b, bio, gfp_mask);
645 EXPORT_SYMBOL(bio_clone_fast);
648 * bio_clone_bioset - clone a bio
649 * @bio_src: bio to clone
650 * @gfp_mask: allocation priority
651 * @bs: bio_set to allocate from
653 * Clone bio. Caller will own the returned bio, but not the actual data it
654 * points to. Reference count of returned bio will be one.
656 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
659 struct bvec_iter iter;
664 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
665 * bio_src->bi_io_vec to bio->bi_io_vec.
667 * We can't do that anymore, because:
669 * - The point of cloning the biovec is to produce a bio with a biovec
670 * the caller can modify: bi_idx and bi_bvec_done should be 0.
672 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
673 * we tried to clone the whole thing bio_alloc_bioset() would fail.
674 * But the clone should succeed as long as the number of biovecs we
675 * actually need to allocate is fewer than BIO_MAX_PAGES.
677 * - Lastly, bi_vcnt should not be looked at or relied upon by code
678 * that does not own the bio - reason being drivers don't use it for
679 * iterating over the biovec anymore, so expecting it to be kept up
680 * to date (i.e. for clones that share the parent biovec) is just
681 * asking for trouble and would force extra work on
682 * __bio_clone_fast() anyways.
685 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
688 bio->bi_disk = bio_src->bi_disk;
689 bio->bi_opf = bio_src->bi_opf;
690 bio->bi_write_hint = bio_src->bi_write_hint;
691 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
692 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
694 switch (bio_op(bio)) {
696 case REQ_OP_SECURE_ERASE:
697 case REQ_OP_WRITE_ZEROES:
699 case REQ_OP_WRITE_SAME:
700 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
703 bio_for_each_segment(bv, bio_src, iter)
704 bio->bi_io_vec[bio->bi_vcnt++] = bv;
708 if (bio_integrity(bio_src)) {
711 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
718 bio_clone_blkcg_association(bio, bio_src);
722 EXPORT_SYMBOL(bio_clone_bioset);
725 * bio_add_pc_page - attempt to add page to bio
726 * @q: the target queue
727 * @bio: destination bio
729 * @len: vec entry length
730 * @offset: vec entry offset
732 * Attempt to add a page to the bio_vec maplist. This can fail for a
733 * number of reasons, such as the bio being full or target block device
734 * limitations. The target block device must allow bio's up to PAGE_SIZE,
735 * so it is always possible to add a single page to an empty bio.
737 * This should only be used by REQ_PC bios.
739 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
740 *page, unsigned int len, unsigned int offset)
742 int retried_segments = 0;
743 struct bio_vec *bvec;
746 * cloned bio must not modify vec list
748 if (unlikely(bio_flagged(bio, BIO_CLONED)))
751 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
755 * For filesystems with a blocksize smaller than the pagesize
756 * we will often be called with the same page as last time and
757 * a consecutive offset. Optimize this special case.
759 if (bio->bi_vcnt > 0) {
760 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
762 if (page == prev->bv_page &&
763 offset == prev->bv_offset + prev->bv_len) {
765 bio->bi_iter.bi_size += len;
770 * If the queue doesn't support SG gaps and adding this
771 * offset would create a gap, disallow it.
773 if (bvec_gap_to_prev(q, prev, offset))
777 if (bio->bi_vcnt >= bio->bi_max_vecs)
781 * setup the new entry, we might clear it again later if we
782 * cannot add the page
784 bvec = &bio->bi_io_vec[bio->bi_vcnt];
785 bvec->bv_page = page;
787 bvec->bv_offset = offset;
789 bio->bi_phys_segments++;
790 bio->bi_iter.bi_size += len;
793 * Perform a recount if the number of segments is greater
794 * than queue_max_segments(q).
797 while (bio->bi_phys_segments > queue_max_segments(q)) {
799 if (retried_segments)
802 retried_segments = 1;
803 blk_recount_segments(q, bio);
806 /* If we may be able to merge these biovecs, force a recount */
807 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
808 bio_clear_flag(bio, BIO_SEG_VALID);
814 bvec->bv_page = NULL;
818 bio->bi_iter.bi_size -= len;
819 blk_recount_segments(q, bio);
822 EXPORT_SYMBOL(bio_add_pc_page);
825 * bio_add_page - attempt to add page to bio
826 * @bio: destination bio
828 * @len: vec entry length
829 * @offset: vec entry offset
831 * Attempt to add a page to the bio_vec maplist. This will only fail
832 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
834 int bio_add_page(struct bio *bio, struct page *page,
835 unsigned int len, unsigned int offset)
840 * cloned bio must not modify vec list
842 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
846 * For filesystems with a blocksize smaller than the pagesize
847 * we will often be called with the same page as last time and
848 * a consecutive offset. Optimize this special case.
850 if (bio->bi_vcnt > 0) {
851 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
853 if (page == bv->bv_page &&
854 offset == bv->bv_offset + bv->bv_len) {
860 if (bio->bi_vcnt >= bio->bi_max_vecs)
863 bv = &bio->bi_io_vec[bio->bi_vcnt];
866 bv->bv_offset = offset;
870 bio->bi_iter.bi_size += len;
873 EXPORT_SYMBOL(bio_add_page);
876 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
877 * @bio: bio to add pages to
878 * @iter: iov iterator describing the region to be mapped
880 * Pins as many pages from *iter and appends them to @bio's bvec array. The
881 * pages will have to be released using put_page() when done.
883 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
885 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
886 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
887 struct page **pages = (struct page **)bv;
891 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
892 if (unlikely(size <= 0))
893 return size ? size : -EFAULT;
894 nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE;
897 * Deep magic below: We need to walk the pinned pages backwards
898 * because we are abusing the space allocated for the bio_vecs
899 * for the page array. Because the bio_vecs are larger than the
900 * page pointers by definition this will always work. But it also
901 * means we can't use bio_add_page, so any changes to it's semantics
902 * need to be reflected here as well.
904 bio->bi_iter.bi_size += size;
905 bio->bi_vcnt += nr_pages;
907 diff = (nr_pages * PAGE_SIZE - offset) - size;
909 bv[nr_pages].bv_page = pages[nr_pages];
910 bv[nr_pages].bv_len = PAGE_SIZE;
911 bv[nr_pages].bv_offset = 0;
914 bv[0].bv_offset += offset;
915 bv[0].bv_len -= offset;
917 bv[bio->bi_vcnt - 1].bv_len -= diff;
919 iov_iter_advance(iter, size);
922 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
924 static void submit_bio_wait_endio(struct bio *bio)
926 complete(bio->bi_private);
930 * submit_bio_wait - submit a bio, and wait until it completes
931 * @bio: The &struct bio which describes the I/O
933 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
934 * bio_endio() on failure.
936 * WARNING: Unlike to how submit_bio() is usually used, this function does not
937 * result in bio reference to be consumed. The caller must drop the reference
940 int submit_bio_wait(struct bio *bio)
942 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
944 bio->bi_private = &done;
945 bio->bi_end_io = submit_bio_wait_endio;
946 bio->bi_opf |= REQ_SYNC;
948 wait_for_completion_io(&done);
950 return blk_status_to_errno(bio->bi_status);
952 EXPORT_SYMBOL(submit_bio_wait);
955 * bio_advance - increment/complete a bio by some number of bytes
956 * @bio: bio to advance
957 * @bytes: number of bytes to complete
959 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
960 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
961 * be updated on the last bvec as well.
963 * @bio will then represent the remaining, uncompleted portion of the io.
965 void bio_advance(struct bio *bio, unsigned bytes)
967 if (bio_integrity(bio))
968 bio_integrity_advance(bio, bytes);
970 bio_advance_iter(bio, &bio->bi_iter, bytes);
972 EXPORT_SYMBOL(bio_advance);
974 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
975 struct bio *src, struct bvec_iter *src_iter)
977 struct bio_vec src_bv, dst_bv;
981 while (src_iter->bi_size && dst_iter->bi_size) {
982 src_bv = bio_iter_iovec(src, *src_iter);
983 dst_bv = bio_iter_iovec(dst, *dst_iter);
985 bytes = min(src_bv.bv_len, dst_bv.bv_len);
987 src_p = kmap_atomic(src_bv.bv_page);
988 dst_p = kmap_atomic(dst_bv.bv_page);
990 memcpy(dst_p + dst_bv.bv_offset,
991 src_p + src_bv.bv_offset,
994 kunmap_atomic(dst_p);
995 kunmap_atomic(src_p);
997 flush_dcache_page(dst_bv.bv_page);
999 bio_advance_iter(src, src_iter, bytes);
1000 bio_advance_iter(dst, dst_iter, bytes);
1003 EXPORT_SYMBOL(bio_copy_data_iter);
1006 * bio_copy_data - copy contents of data buffers from one bio to another
1008 * @dst: destination bio
1010 * Stops when it reaches the end of either @src or @dst - that is, copies
1011 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1013 void bio_copy_data(struct bio *dst, struct bio *src)
1015 struct bvec_iter src_iter = src->bi_iter;
1016 struct bvec_iter dst_iter = dst->bi_iter;
1018 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1020 EXPORT_SYMBOL(bio_copy_data);
1023 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1025 * @src: source bio list
1026 * @dst: destination bio list
1028 * Stops when it reaches the end of either the @src list or @dst list - that is,
1029 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1032 void bio_list_copy_data(struct bio *dst, struct bio *src)
1034 struct bvec_iter src_iter = src->bi_iter;
1035 struct bvec_iter dst_iter = dst->bi_iter;
1038 if (!src_iter.bi_size) {
1043 src_iter = src->bi_iter;
1046 if (!dst_iter.bi_size) {
1051 dst_iter = dst->bi_iter;
1054 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1057 EXPORT_SYMBOL(bio_list_copy_data);
1059 struct bio_map_data {
1061 struct iov_iter iter;
1065 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
1068 struct bio_map_data *bmd;
1069 if (data->nr_segs > UIO_MAXIOV)
1072 bmd = kmalloc(sizeof(struct bio_map_data) +
1073 sizeof(struct iovec) * data->nr_segs, gfp_mask);
1076 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1078 bmd->iter.iov = bmd->iov;
1083 * bio_copy_from_iter - copy all pages from iov_iter to bio
1084 * @bio: The &struct bio which describes the I/O as destination
1085 * @iter: iov_iter as source
1087 * Copy all pages from iov_iter to bio.
1088 * Returns 0 on success, or error on failure.
1090 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
1093 struct bio_vec *bvec;
1095 bio_for_each_segment_all(bvec, bio, i) {
1098 ret = copy_page_from_iter(bvec->bv_page,
1103 if (!iov_iter_count(iter))
1106 if (ret < bvec->bv_len)
1114 * bio_copy_to_iter - copy all pages from bio to iov_iter
1115 * @bio: The &struct bio which describes the I/O as source
1116 * @iter: iov_iter as destination
1118 * Copy all pages from bio to iov_iter.
1119 * Returns 0 on success, or error on failure.
1121 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1124 struct bio_vec *bvec;
1126 bio_for_each_segment_all(bvec, bio, i) {
1129 ret = copy_page_to_iter(bvec->bv_page,
1134 if (!iov_iter_count(&iter))
1137 if (ret < bvec->bv_len)
1144 void bio_free_pages(struct bio *bio)
1146 struct bio_vec *bvec;
1149 bio_for_each_segment_all(bvec, bio, i)
1150 __free_page(bvec->bv_page);
1152 EXPORT_SYMBOL(bio_free_pages);
1155 * bio_uncopy_user - finish previously mapped bio
1156 * @bio: bio being terminated
1158 * Free pages allocated from bio_copy_user_iov() and write back data
1159 * to user space in case of a read.
1161 int bio_uncopy_user(struct bio *bio)
1163 struct bio_map_data *bmd = bio->bi_private;
1166 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1168 * if we're in a workqueue, the request is orphaned, so
1169 * don't copy into a random user address space, just free
1170 * and return -EINTR so user space doesn't expect any data.
1174 else if (bio_data_dir(bio) == READ)
1175 ret = bio_copy_to_iter(bio, bmd->iter);
1176 if (bmd->is_our_pages)
1177 bio_free_pages(bio);
1185 * bio_copy_user_iov - copy user data to bio
1186 * @q: destination block queue
1187 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1188 * @iter: iovec iterator
1189 * @gfp_mask: memory allocation flags
1191 * Prepares and returns a bio for indirect user io, bouncing data
1192 * to/from kernel pages as necessary. Must be paired with
1193 * call bio_uncopy_user() on io completion.
1195 struct bio *bio_copy_user_iov(struct request_queue *q,
1196 struct rq_map_data *map_data,
1197 struct iov_iter *iter,
1200 struct bio_map_data *bmd;
1205 unsigned int len = iter->count;
1206 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1208 bmd = bio_alloc_map_data(iter, gfp_mask);
1210 return ERR_PTR(-ENOMEM);
1213 * We need to do a deep copy of the iov_iter including the iovecs.
1214 * The caller provided iov might point to an on-stack or otherwise
1217 bmd->is_our_pages = map_data ? 0 : 1;
1219 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1220 if (nr_pages > BIO_MAX_PAGES)
1221 nr_pages = BIO_MAX_PAGES;
1224 bio = bio_kmalloc(gfp_mask, nr_pages);
1231 nr_pages = 1 << map_data->page_order;
1232 i = map_data->offset / PAGE_SIZE;
1235 unsigned int bytes = PAGE_SIZE;
1243 if (i == map_data->nr_entries * nr_pages) {
1248 page = map_data->pages[i / nr_pages];
1249 page += (i % nr_pages);
1253 page = alloc_page(q->bounce_gfp | gfp_mask);
1260 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1271 map_data->offset += bio->bi_iter.bi_size;
1276 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1277 (map_data && map_data->from_user)) {
1278 ret = bio_copy_from_iter(bio, iter);
1282 iov_iter_advance(iter, bio->bi_iter.bi_size);
1285 bio->bi_private = bmd;
1286 if (map_data && map_data->null_mapped)
1287 bio_set_flag(bio, BIO_NULL_MAPPED);
1291 bio_free_pages(bio);
1295 return ERR_PTR(ret);
1299 * bio_map_user_iov - map user iovec into bio
1300 * @q: the struct request_queue for the bio
1301 * @iter: iovec iterator
1302 * @gfp_mask: memory allocation flags
1304 * Map the user space address into a bio suitable for io to a block
1305 * device. Returns an error pointer in case of error.
1307 struct bio *bio_map_user_iov(struct request_queue *q,
1308 struct iov_iter *iter,
1314 struct bio_vec *bvec;
1316 if (!iov_iter_count(iter))
1317 return ERR_PTR(-EINVAL);
1319 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
1321 return ERR_PTR(-ENOMEM);
1323 while (iov_iter_count(iter)) {
1324 struct page **pages;
1326 size_t offs, added = 0;
1329 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
1330 if (unlikely(bytes <= 0)) {
1331 ret = bytes ? bytes : -EFAULT;
1335 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
1337 if (unlikely(offs & queue_dma_alignment(q))) {
1341 for (j = 0; j < npages; j++) {
1342 struct page *page = pages[j];
1343 unsigned int n = PAGE_SIZE - offs;
1344 unsigned short prev_bi_vcnt = bio->bi_vcnt;
1349 if (!bio_add_pc_page(q, bio, page, n, offs))
1353 * check if vector was merged with previous
1354 * drop page reference if needed
1356 if (bio->bi_vcnt == prev_bi_vcnt)
1363 iov_iter_advance(iter, added);
1366 * release the pages we didn't map into the bio, if any
1369 put_page(pages[j++]);
1371 /* couldn't stuff something into bio? */
1376 bio_set_flag(bio, BIO_USER_MAPPED);
1379 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1380 * it would normally disappear when its bi_end_io is run.
1381 * however, we need it for the unmap, so grab an extra
1388 bio_for_each_segment_all(bvec, bio, j) {
1389 put_page(bvec->bv_page);
1392 return ERR_PTR(ret);
1395 static void __bio_unmap_user(struct bio *bio)
1397 struct bio_vec *bvec;
1401 * make sure we dirty pages we wrote to
1403 bio_for_each_segment_all(bvec, bio, i) {
1404 if (bio_data_dir(bio) == READ)
1405 set_page_dirty_lock(bvec->bv_page);
1407 put_page(bvec->bv_page);
1414 * bio_unmap_user - unmap a bio
1415 * @bio: the bio being unmapped
1417 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1420 * bio_unmap_user() may sleep.
1422 void bio_unmap_user(struct bio *bio)
1424 __bio_unmap_user(bio);
1428 static void bio_map_kern_endio(struct bio *bio)
1434 * bio_map_kern - map kernel address into bio
1435 * @q: the struct request_queue for the bio
1436 * @data: pointer to buffer to map
1437 * @len: length in bytes
1438 * @gfp_mask: allocation flags for bio allocation
1440 * Map the kernel address into a bio suitable for io to a block
1441 * device. Returns an error pointer in case of error.
1443 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1446 unsigned long kaddr = (unsigned long)data;
1447 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1448 unsigned long start = kaddr >> PAGE_SHIFT;
1449 const int nr_pages = end - start;
1453 bio = bio_kmalloc(gfp_mask, nr_pages);
1455 return ERR_PTR(-ENOMEM);
1457 offset = offset_in_page(kaddr);
1458 for (i = 0; i < nr_pages; i++) {
1459 unsigned int bytes = PAGE_SIZE - offset;
1467 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1469 /* we don't support partial mappings */
1471 return ERR_PTR(-EINVAL);
1479 bio->bi_end_io = bio_map_kern_endio;
1482 EXPORT_SYMBOL(bio_map_kern);
1484 static void bio_copy_kern_endio(struct bio *bio)
1486 bio_free_pages(bio);
1490 static void bio_copy_kern_endio_read(struct bio *bio)
1492 char *p = bio->bi_private;
1493 struct bio_vec *bvec;
1496 bio_for_each_segment_all(bvec, bio, i) {
1497 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1501 bio_copy_kern_endio(bio);
1505 * bio_copy_kern - copy kernel address into bio
1506 * @q: the struct request_queue for the bio
1507 * @data: pointer to buffer to copy
1508 * @len: length in bytes
1509 * @gfp_mask: allocation flags for bio and page allocation
1510 * @reading: data direction is READ
1512 * copy the kernel address into a bio suitable for io to a block
1513 * device. Returns an error pointer in case of error.
1515 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1516 gfp_t gfp_mask, int reading)
1518 unsigned long kaddr = (unsigned long)data;
1519 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1520 unsigned long start = kaddr >> PAGE_SHIFT;
1529 return ERR_PTR(-EINVAL);
1531 nr_pages = end - start;
1532 bio = bio_kmalloc(gfp_mask, nr_pages);
1534 return ERR_PTR(-ENOMEM);
1538 unsigned int bytes = PAGE_SIZE;
1543 page = alloc_page(q->bounce_gfp | gfp_mask);
1548 memcpy(page_address(page), p, bytes);
1550 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1558 bio->bi_end_io = bio_copy_kern_endio_read;
1559 bio->bi_private = data;
1561 bio->bi_end_io = bio_copy_kern_endio;
1567 bio_free_pages(bio);
1569 return ERR_PTR(-ENOMEM);
1573 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1574 * for performing direct-IO in BIOs.
1576 * The problem is that we cannot run set_page_dirty() from interrupt context
1577 * because the required locks are not interrupt-safe. So what we can do is to
1578 * mark the pages dirty _before_ performing IO. And in interrupt context,
1579 * check that the pages are still dirty. If so, fine. If not, redirty them
1580 * in process context.
1582 * We special-case compound pages here: normally this means reads into hugetlb
1583 * pages. The logic in here doesn't really work right for compound pages
1584 * because the VM does not uniformly chase down the head page in all cases.
1585 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1586 * handle them at all. So we skip compound pages here at an early stage.
1588 * Note that this code is very hard to test under normal circumstances because
1589 * direct-io pins the pages with get_user_pages(). This makes
1590 * is_page_cache_freeable return false, and the VM will not clean the pages.
1591 * But other code (eg, flusher threads) could clean the pages if they are mapped
1594 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1595 * deferred bio dirtying paths.
1599 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1601 void bio_set_pages_dirty(struct bio *bio)
1603 struct bio_vec *bvec;
1606 bio_for_each_segment_all(bvec, bio, i) {
1607 struct page *page = bvec->bv_page;
1609 if (page && !PageCompound(page))
1610 set_page_dirty_lock(page);
1613 EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1615 static void bio_release_pages(struct bio *bio)
1617 struct bio_vec *bvec;
1620 bio_for_each_segment_all(bvec, bio, i) {
1621 struct page *page = bvec->bv_page;
1629 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1630 * If they are, then fine. If, however, some pages are clean then they must
1631 * have been written out during the direct-IO read. So we take another ref on
1632 * the BIO and the offending pages and re-dirty the pages in process context.
1634 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1635 * here on. It will run one put_page() against each page and will run one
1636 * bio_put() against the BIO.
1639 static void bio_dirty_fn(struct work_struct *work);
1641 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1642 static DEFINE_SPINLOCK(bio_dirty_lock);
1643 static struct bio *bio_dirty_list;
1646 * This runs in process context
1648 static void bio_dirty_fn(struct work_struct *work)
1650 unsigned long flags;
1653 spin_lock_irqsave(&bio_dirty_lock, flags);
1654 bio = bio_dirty_list;
1655 bio_dirty_list = NULL;
1656 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1659 struct bio *next = bio->bi_private;
1661 bio_set_pages_dirty(bio);
1662 bio_release_pages(bio);
1668 void bio_check_pages_dirty(struct bio *bio)
1670 struct bio_vec *bvec;
1671 int nr_clean_pages = 0;
1674 bio_for_each_segment_all(bvec, bio, i) {
1675 struct page *page = bvec->bv_page;
1677 if (PageDirty(page) || PageCompound(page)) {
1679 bvec->bv_page = NULL;
1685 if (nr_clean_pages) {
1686 unsigned long flags;
1688 spin_lock_irqsave(&bio_dirty_lock, flags);
1689 bio->bi_private = bio_dirty_list;
1690 bio_dirty_list = bio;
1691 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1692 schedule_work(&bio_dirty_work);
1697 EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1699 void generic_start_io_acct(struct request_queue *q, int rw,
1700 unsigned long sectors, struct hd_struct *part)
1702 int cpu = part_stat_lock();
1704 part_round_stats(q, cpu, part);
1705 part_stat_inc(cpu, part, ios[rw]);
1706 part_stat_add(cpu, part, sectors[rw], sectors);
1707 part_inc_in_flight(q, part, rw);
1711 EXPORT_SYMBOL(generic_start_io_acct);
1713 void generic_end_io_acct(struct request_queue *q, int rw,
1714 struct hd_struct *part, unsigned long start_time)
1716 unsigned long duration = jiffies - start_time;
1717 int cpu = part_stat_lock();
1719 part_stat_add(cpu, part, ticks[rw], duration);
1720 part_round_stats(q, cpu, part);
1721 part_dec_in_flight(q, part, rw);
1725 EXPORT_SYMBOL(generic_end_io_acct);
1727 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1728 void bio_flush_dcache_pages(struct bio *bi)
1730 struct bio_vec bvec;
1731 struct bvec_iter iter;
1733 bio_for_each_segment(bvec, bi, iter)
1734 flush_dcache_page(bvec.bv_page);
1736 EXPORT_SYMBOL(bio_flush_dcache_pages);
1739 static inline bool bio_remaining_done(struct bio *bio)
1742 * If we're not chaining, then ->__bi_remaining is always 1 and
1743 * we always end io on the first invocation.
1745 if (!bio_flagged(bio, BIO_CHAIN))
1748 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1750 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1751 bio_clear_flag(bio, BIO_CHAIN);
1759 * bio_endio - end I/O on a bio
1763 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1764 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1765 * bio unless they own it and thus know that it has an end_io function.
1767 * bio_endio() can be called several times on a bio that has been chained
1768 * using bio_chain(). The ->bi_end_io() function will only be called the
1769 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1770 * generated if BIO_TRACE_COMPLETION is set.
1772 void bio_endio(struct bio *bio)
1775 if (!bio_remaining_done(bio))
1777 if (!bio_integrity_endio(bio))
1780 if (WARN_ONCE(bio->bi_next, "driver left bi_next not NULL"))
1781 bio->bi_next = NULL;
1784 * Need to have a real endio function for chained bios, otherwise
1785 * various corner cases will break (like stacking block devices that
1786 * save/restore bi_end_io) - however, we want to avoid unbounded
1787 * recursion and blowing the stack. Tail call optimization would
1788 * handle this, but compiling with frame pointers also disables
1789 * gcc's sibling call optimization.
1791 if (bio->bi_end_io == bio_chain_endio) {
1792 bio = __bio_chain_endio(bio);
1796 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1797 trace_block_bio_complete(bio->bi_disk->queue, bio,
1798 blk_status_to_errno(bio->bi_status));
1799 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1802 blk_throtl_bio_endio(bio);
1803 /* release cgroup info */
1806 bio->bi_end_io(bio);
1808 EXPORT_SYMBOL(bio_endio);
1811 * bio_split - split a bio
1812 * @bio: bio to split
1813 * @sectors: number of sectors to split from the front of @bio
1815 * @bs: bio set to allocate from
1817 * Allocates and returns a new bio which represents @sectors from the start of
1818 * @bio, and updates @bio to represent the remaining sectors.
1820 * Unless this is a discard request the newly allocated bio will point
1821 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1822 * @bio is not freed before the split.
1824 struct bio *bio_split(struct bio *bio, int sectors,
1825 gfp_t gfp, struct bio_set *bs)
1829 BUG_ON(sectors <= 0);
1830 BUG_ON(sectors >= bio_sectors(bio));
1832 split = bio_clone_fast(bio, gfp, bs);
1836 split->bi_iter.bi_size = sectors << 9;
1838 if (bio_integrity(split))
1839 bio_integrity_trim(split);
1841 bio_advance(bio, split->bi_iter.bi_size);
1843 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1844 bio_set_flag(split, BIO_TRACE_COMPLETION);
1848 EXPORT_SYMBOL(bio_split);
1851 * bio_trim - trim a bio
1853 * @offset: number of sectors to trim from the front of @bio
1854 * @size: size we want to trim @bio to, in sectors
1856 void bio_trim(struct bio *bio, int offset, int size)
1858 /* 'bio' is a cloned bio which we need to trim to match
1859 * the given offset and size.
1863 if (offset == 0 && size == bio->bi_iter.bi_size)
1866 bio_clear_flag(bio, BIO_SEG_VALID);
1868 bio_advance(bio, offset << 9);
1870 bio->bi_iter.bi_size = size;
1872 if (bio_integrity(bio))
1873 bio_integrity_trim(bio);
1876 EXPORT_SYMBOL_GPL(bio_trim);
1879 * create memory pools for biovec's in a bio_set.
1880 * use the global biovec slabs created for general use.
1882 int biovec_init_pool(mempool_t *pool, int pool_entries)
1884 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1886 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1890 * bioset_exit - exit a bioset initialized with bioset_init()
1892 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1895 void bioset_exit(struct bio_set *bs)
1897 if (bs->rescue_workqueue)
1898 destroy_workqueue(bs->rescue_workqueue);
1899 bs->rescue_workqueue = NULL;
1901 mempool_exit(&bs->bio_pool);
1902 mempool_exit(&bs->bvec_pool);
1904 bioset_integrity_free(bs);
1907 bs->bio_slab = NULL;
1909 EXPORT_SYMBOL(bioset_exit);
1912 * bioset_init - Initialize a bio_set
1913 * @bs: pool to initialize
1914 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1915 * @front_pad: Number of bytes to allocate in front of the returned bio
1916 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1917 * and %BIOSET_NEED_RESCUER
1920 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1921 * to ask for a number of bytes to be allocated in front of the bio.
1922 * Front pad allocation is useful for embedding the bio inside
1923 * another structure, to avoid allocating extra data to go with the bio.
1924 * Note that the bio must be embedded at the END of that structure always,
1925 * or things will break badly.
1926 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1927 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1928 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1929 * dispatch queued requests when the mempool runs out of space.
1932 int bioset_init(struct bio_set *bs,
1933 unsigned int pool_size,
1934 unsigned int front_pad,
1937 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1939 bs->front_pad = front_pad;
1941 spin_lock_init(&bs->rescue_lock);
1942 bio_list_init(&bs->rescue_list);
1943 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1945 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1949 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1952 if ((flags & BIOSET_NEED_BVECS) &&
1953 biovec_init_pool(&bs->bvec_pool, pool_size))
1956 if (!(flags & BIOSET_NEED_RESCUER))
1959 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1960 if (!bs->rescue_workqueue)
1968 EXPORT_SYMBOL(bioset_init);
1971 * Initialize and setup a new bio_set, based on the settings from
1974 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1979 if (src->bvec_pool.min_nr)
1980 flags |= BIOSET_NEED_BVECS;
1981 if (src->rescue_workqueue)
1982 flags |= BIOSET_NEED_RESCUER;
1984 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1986 EXPORT_SYMBOL(bioset_init_from_src);
1988 #ifdef CONFIG_BLK_CGROUP
1991 * bio_associate_blkcg - associate a bio with the specified blkcg
1993 * @blkcg_css: css of the blkcg to associate
1995 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1996 * treat @bio as if it were issued by a task which belongs to the blkcg.
1998 * This function takes an extra reference of @blkcg_css which will be put
1999 * when @bio is released. The caller must own @bio and is responsible for
2000 * synchronizing calls to this function.
2002 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
2004 if (unlikely(bio->bi_css))
2007 bio->bi_css = blkcg_css;
2010 EXPORT_SYMBOL_GPL(bio_associate_blkcg);
2013 * bio_disassociate_task - undo bio_associate_current()
2016 void bio_disassociate_task(struct bio *bio)
2019 put_io_context(bio->bi_ioc);
2023 css_put(bio->bi_css);
2029 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2030 * @dst: destination bio
2033 void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2036 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2038 EXPORT_SYMBOL_GPL(bio_clone_blkcg_association);
2039 #endif /* CONFIG_BLK_CGROUP */
2041 static void __init biovec_init_slabs(void)
2045 for (i = 0; i < BVEC_POOL_NR; i++) {
2047 struct biovec_slab *bvs = bvec_slabs + i;
2049 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2054 size = bvs->nr_vecs * sizeof(struct bio_vec);
2055 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2056 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2060 static int __init init_bio(void)
2064 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2066 panic("bio: can't allocate bios\n");
2068 bio_integrity_init();
2069 biovec_init_slabs();
2071 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
2072 panic("bio: can't allocate bios\n");
2074 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
2075 panic("bio: can't create integrity pool\n");
2079 subsys_initcall(init_bio);