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>
21 #include <linux/blk-crypto.h>
22 #include <linux/xarray.h>
24 #include <trace/events/block.h>
26 #include "blk-rq-qos.h"
29 * Test patch to inline a certain number of bi_io_vec's inside the bio
30 * itself, to shrink a bio data allocation from two mempool calls to one
32 #define BIO_INLINE_VECS 4
35 * if you change this list, also change bvec_alloc or things will
36 * break badly! cannot be bigger than what you can fit into an
39 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
40 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
41 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
46 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
47 * IO code that does not need private memory pools.
49 struct bio_set fs_bio_set;
50 EXPORT_SYMBOL(fs_bio_set);
53 * Our slab pool management
56 struct kmem_cache *slab;
57 unsigned int slab_ref;
58 unsigned int slab_size;
61 static DEFINE_MUTEX(bio_slab_lock);
62 static DEFINE_XARRAY(bio_slabs);
64 static struct bio_slab *create_bio_slab(unsigned int size)
66 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
71 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
72 bslab->slab = kmem_cache_create(bslab->name, size,
73 ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL);
78 bslab->slab_size = size;
80 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
83 kmem_cache_destroy(bslab->slab);
90 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
92 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
95 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
97 unsigned int size = bs_bio_slab_size(bs);
98 struct bio_slab *bslab;
100 mutex_lock(&bio_slab_lock);
101 bslab = xa_load(&bio_slabs, size);
105 bslab = create_bio_slab(size);
106 mutex_unlock(&bio_slab_lock);
113 static void bio_put_slab(struct bio_set *bs)
115 struct bio_slab *bslab = NULL;
116 unsigned int slab_size = bs_bio_slab_size(bs);
118 mutex_lock(&bio_slab_lock);
120 bslab = xa_load(&bio_slabs, slab_size);
121 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
124 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
126 WARN_ON(!bslab->slab_ref);
128 if (--bslab->slab_ref)
131 xa_erase(&bio_slabs, slab_size);
133 kmem_cache_destroy(bslab->slab);
137 mutex_unlock(&bio_slab_lock);
140 unsigned int bvec_nr_vecs(unsigned short idx)
142 return bvec_slabs[--idx].nr_vecs;
145 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
151 BIO_BUG_ON(idx >= BVEC_POOL_NR);
153 if (idx == BVEC_POOL_MAX) {
154 mempool_free(bv, pool);
156 struct biovec_slab *bvs = bvec_slabs + idx;
158 kmem_cache_free(bvs->slab, bv);
162 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
168 * see comment near bvec_array define!
186 case 129 ... BIO_MAX_PAGES:
194 * idx now points to the pool we want to allocate from. only the
195 * 1-vec entry pool is mempool backed.
197 if (*idx == BVEC_POOL_MAX) {
199 bvl = mempool_alloc(pool, gfp_mask);
201 struct biovec_slab *bvs = bvec_slabs + *idx;
202 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
205 * Make this allocation restricted and don't dump info on
206 * allocation failures, since we'll fallback to the mempool
207 * in case of failure.
209 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
212 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
213 * is set, retry with the 1-entry mempool
215 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
216 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
217 *idx = BVEC_POOL_MAX;
226 void bio_uninit(struct bio *bio)
228 #ifdef CONFIG_BLK_CGROUP
230 blkg_put(bio->bi_blkg);
234 if (bio_integrity(bio))
235 bio_integrity_free(bio);
237 bio_crypt_free_ctx(bio);
239 EXPORT_SYMBOL(bio_uninit);
241 static void bio_free(struct bio *bio)
243 struct bio_set *bs = bio->bi_pool;
249 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
252 * If we have front padding, adjust the bio pointer before freeing
257 mempool_free(p, &bs->bio_pool);
259 /* Bio was allocated by bio_kmalloc() */
265 * Users of this function have their own bio allocation. Subsequently,
266 * they must remember to pair any call to bio_init() with bio_uninit()
267 * when IO has completed, or when the bio is released.
269 void bio_init(struct bio *bio, struct bio_vec *table,
270 unsigned short max_vecs)
272 memset(bio, 0, sizeof(*bio));
273 atomic_set(&bio->__bi_remaining, 1);
274 atomic_set(&bio->__bi_cnt, 1);
276 bio->bi_io_vec = table;
277 bio->bi_max_vecs = max_vecs;
279 EXPORT_SYMBOL(bio_init);
282 * bio_reset - reinitialize a bio
286 * After calling bio_reset(), @bio will be in the same state as a freshly
287 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
288 * preserved are the ones that are initialized by bio_alloc_bioset(). See
289 * comment in struct bio.
291 void bio_reset(struct bio *bio)
293 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
297 memset(bio, 0, BIO_RESET_BYTES);
298 bio->bi_flags = flags;
299 atomic_set(&bio->__bi_remaining, 1);
301 EXPORT_SYMBOL(bio_reset);
303 static struct bio *__bio_chain_endio(struct bio *bio)
305 struct bio *parent = bio->bi_private;
307 if (!parent->bi_status)
308 parent->bi_status = bio->bi_status;
313 static void bio_chain_endio(struct bio *bio)
315 bio_endio(__bio_chain_endio(bio));
319 * bio_chain - chain bio completions
320 * @bio: the target bio
321 * @parent: the parent bio of @bio
323 * The caller won't have a bi_end_io called when @bio completes - instead,
324 * @parent's bi_end_io won't be called until both @parent and @bio have
325 * completed; the chained bio will also be freed when it completes.
327 * The caller must not set bi_private or bi_end_io in @bio.
329 void bio_chain(struct bio *bio, struct bio *parent)
331 BUG_ON(bio->bi_private || bio->bi_end_io);
333 bio->bi_private = parent;
334 bio->bi_end_io = bio_chain_endio;
335 bio_inc_remaining(parent);
337 EXPORT_SYMBOL(bio_chain);
339 static void bio_alloc_rescue(struct work_struct *work)
341 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
345 spin_lock(&bs->rescue_lock);
346 bio = bio_list_pop(&bs->rescue_list);
347 spin_unlock(&bs->rescue_lock);
352 submit_bio_noacct(bio);
356 static void punt_bios_to_rescuer(struct bio_set *bs)
358 struct bio_list punt, nopunt;
361 if (WARN_ON_ONCE(!bs->rescue_workqueue))
364 * In order to guarantee forward progress we must punt only bios that
365 * were allocated from this bio_set; otherwise, if there was a bio on
366 * there for a stacking driver higher up in the stack, processing it
367 * could require allocating bios from this bio_set, and doing that from
368 * our own rescuer would be bad.
370 * Since bio lists are singly linked, pop them all instead of trying to
371 * remove from the middle of the list:
374 bio_list_init(&punt);
375 bio_list_init(&nopunt);
377 while ((bio = bio_list_pop(¤t->bio_list[0])))
378 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
379 current->bio_list[0] = nopunt;
381 bio_list_init(&nopunt);
382 while ((bio = bio_list_pop(¤t->bio_list[1])))
383 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
384 current->bio_list[1] = nopunt;
386 spin_lock(&bs->rescue_lock);
387 bio_list_merge(&bs->rescue_list, &punt);
388 spin_unlock(&bs->rescue_lock);
390 queue_work(bs->rescue_workqueue, &bs->rescue_work);
394 * bio_alloc_bioset - allocate a bio for I/O
395 * @gfp_mask: the GFP_* mask given to the slab allocator
396 * @nr_iovecs: number of iovecs to pre-allocate
397 * @bs: the bio_set to allocate from.
400 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
401 * backed by the @bs's mempool.
403 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
404 * always be able to allocate a bio. This is due to the mempool guarantees.
405 * To make this work, callers must never allocate more than 1 bio at a time
406 * from this pool. Callers that need to allocate more than 1 bio must always
407 * submit the previously allocated bio for IO before attempting to allocate
408 * a new one. Failure to do so can cause deadlocks under memory pressure.
410 * Note that when running under submit_bio_noacct() (i.e. any block
411 * driver), bios are not submitted until after you return - see the code in
412 * submit_bio_noacct() that converts recursion into iteration, to prevent
415 * This would normally mean allocating multiple bios under
416 * submit_bio_noacct() would be susceptible to deadlocks, but we have
417 * deadlock avoidance code that resubmits any blocked bios from a rescuer
420 * However, we do not guarantee forward progress for allocations from other
421 * mempools. Doing multiple allocations from the same mempool under
422 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
423 * for per bio allocations.
426 * Pointer to new bio on success, NULL on failure.
428 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
431 gfp_t saved_gfp = gfp_mask;
433 unsigned inline_vecs;
434 struct bio_vec *bvl = NULL;
439 if (nr_iovecs > UIO_MAXIOV)
442 p = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask);
444 inline_vecs = nr_iovecs;
446 /* should not use nobvec bioset for nr_iovecs > 0 */
447 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
451 * submit_bio_noacct() converts recursion to iteration; this
452 * means if we're running beneath it, any bios we allocate and
453 * submit will not be submitted (and thus freed) until after we
456 * This exposes us to a potential deadlock if we allocate
457 * multiple bios from the same bio_set() while running
458 * underneath submit_bio_noacct(). If we were to allocate
459 * multiple bios (say a stacking block driver that was splitting
460 * bios), we would deadlock if we exhausted the mempool's
463 * We solve this, and guarantee forward progress, with a rescuer
464 * workqueue per bio_set. If we go to allocate and there are
465 * bios on current->bio_list, we first try the allocation
466 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
467 * bios we would be blocking to the rescuer workqueue before
468 * we retry with the original gfp_flags.
471 if (current->bio_list &&
472 (!bio_list_empty(¤t->bio_list[0]) ||
473 !bio_list_empty(¤t->bio_list[1])) &&
474 bs->rescue_workqueue)
475 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
477 p = mempool_alloc(&bs->bio_pool, gfp_mask);
478 if (!p && gfp_mask != saved_gfp) {
479 punt_bios_to_rescuer(bs);
480 gfp_mask = saved_gfp;
481 p = mempool_alloc(&bs->bio_pool, gfp_mask);
484 front_pad = bs->front_pad;
485 inline_vecs = BIO_INLINE_VECS;
492 bio_init(bio, NULL, 0);
494 if (nr_iovecs > inline_vecs) {
495 unsigned long idx = 0;
497 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
498 if (!bvl && gfp_mask != saved_gfp) {
499 punt_bios_to_rescuer(bs);
500 gfp_mask = saved_gfp;
501 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
507 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
508 bio->bi_max_vecs = bvec_nr_vecs(idx);
509 } else if (nr_iovecs) {
510 bvl = bio->bi_inline_vecs;
511 bio->bi_max_vecs = inline_vecs;
515 bio->bi_io_vec = bvl;
519 mempool_free(p, &bs->bio_pool);
522 EXPORT_SYMBOL(bio_alloc_bioset);
524 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
528 struct bvec_iter iter;
530 __bio_for_each_segment(bv, bio, iter, start) {
531 char *data = bvec_kmap_irq(&bv, &flags);
532 memset(data, 0, bv.bv_len);
533 flush_dcache_page(bv.bv_page);
534 bvec_kunmap_irq(data, &flags);
537 EXPORT_SYMBOL(zero_fill_bio_iter);
540 * bio_truncate - truncate the bio to small size of @new_size
541 * @bio: the bio to be truncated
542 * @new_size: new size for truncating the bio
545 * Truncate the bio to new size of @new_size. If bio_op(bio) is
546 * REQ_OP_READ, zero the truncated part. This function should only
547 * be used for handling corner cases, such as bio eod.
549 void bio_truncate(struct bio *bio, unsigned new_size)
552 struct bvec_iter iter;
553 unsigned int done = 0;
554 bool truncated = false;
556 if (new_size >= bio->bi_iter.bi_size)
559 if (bio_op(bio) != REQ_OP_READ)
562 bio_for_each_segment(bv, bio, iter) {
563 if (done + bv.bv_len > new_size) {
567 offset = new_size - done;
570 zero_user(bv.bv_page, offset, bv.bv_len - offset);
578 * Don't touch bvec table here and make it really immutable, since
579 * fs bio user has to retrieve all pages via bio_for_each_segment_all
580 * in its .end_bio() callback.
582 * It is enough to truncate bio by updating .bi_size since we can make
583 * correct bvec with the updated .bi_size for drivers.
585 bio->bi_iter.bi_size = new_size;
589 * guard_bio_eod - truncate a BIO to fit the block device
590 * @bio: bio to truncate
592 * This allows us to do IO even on the odd last sectors of a device, even if the
593 * block size is some multiple of the physical sector size.
595 * We'll just truncate the bio to the size of the device, and clear the end of
596 * the buffer head manually. Truly out-of-range accesses will turn into actual
597 * I/O errors, this only handles the "we need to be able to do I/O at the final
600 void guard_bio_eod(struct bio *bio)
602 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
608 * If the *whole* IO is past the end of the device,
609 * let it through, and the IO layer will turn it into
612 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
615 maxsector -= bio->bi_iter.bi_sector;
616 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
619 bio_truncate(bio, maxsector << 9);
623 * bio_put - release a reference to a bio
624 * @bio: bio to release reference to
627 * Put a reference to a &struct bio, either one you have gotten with
628 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
630 void bio_put(struct bio *bio)
632 if (!bio_flagged(bio, BIO_REFFED))
635 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
640 if (atomic_dec_and_test(&bio->__bi_cnt))
644 EXPORT_SYMBOL(bio_put);
647 * __bio_clone_fast - clone a bio that shares the original bio's biovec
648 * @bio: destination bio
649 * @bio_src: bio to clone
651 * Clone a &bio. Caller will own the returned bio, but not
652 * the actual data it points to. Reference count of returned
655 * Caller must ensure that @bio_src is not freed before @bio.
657 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
659 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
662 * most users will be overriding ->bi_bdev with a new target,
663 * so we don't set nor calculate new physical/hw segment counts here
665 bio->bi_bdev = bio_src->bi_bdev;
666 bio_set_flag(bio, BIO_CLONED);
667 if (bio_flagged(bio_src, BIO_THROTTLED))
668 bio_set_flag(bio, BIO_THROTTLED);
669 bio->bi_opf = bio_src->bi_opf;
670 bio->bi_ioprio = bio_src->bi_ioprio;
671 bio->bi_write_hint = bio_src->bi_write_hint;
672 bio->bi_iter = bio_src->bi_iter;
673 bio->bi_io_vec = bio_src->bi_io_vec;
675 bio_clone_blkg_association(bio, bio_src);
676 blkcg_bio_issue_init(bio);
678 EXPORT_SYMBOL(__bio_clone_fast);
681 * bio_clone_fast - clone a bio that shares the original bio's biovec
683 * @gfp_mask: allocation priority
684 * @bs: bio_set to allocate from
686 * Like __bio_clone_fast, only also allocates the returned bio
688 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
692 b = bio_alloc_bioset(gfp_mask, 0, bs);
696 __bio_clone_fast(b, bio);
698 if (bio_crypt_clone(b, bio, gfp_mask) < 0)
701 if (bio_integrity(bio) &&
702 bio_integrity_clone(b, bio, gfp_mask) < 0)
711 EXPORT_SYMBOL(bio_clone_fast);
713 const char *bio_devname(struct bio *bio, char *buf)
715 return bdevname(bio->bi_bdev, buf);
717 EXPORT_SYMBOL(bio_devname);
719 static inline bool page_is_mergeable(const struct bio_vec *bv,
720 struct page *page, unsigned int len, unsigned int off,
723 size_t bv_end = bv->bv_offset + bv->bv_len;
724 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
725 phys_addr_t page_addr = page_to_phys(page);
727 if (vec_end_addr + 1 != page_addr + off)
729 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
732 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
735 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
739 * Try to merge a page into a segment, while obeying the hardware segment
740 * size limit. This is not for normal read/write bios, but for passthrough
741 * or Zone Append operations that we can't split.
743 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
744 struct page *page, unsigned len,
745 unsigned offset, bool *same_page)
747 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
748 unsigned long mask = queue_segment_boundary(q);
749 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
750 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
752 if ((addr1 | mask) != (addr2 | mask))
754 if (bv->bv_len + len > queue_max_segment_size(q))
756 return __bio_try_merge_page(bio, page, len, offset, same_page);
760 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
761 * @q: the target queue
762 * @bio: destination bio
764 * @len: vec entry length
765 * @offset: vec entry offset
766 * @max_sectors: maximum number of sectors that can be added
767 * @same_page: return if the segment has been merged inside the same page
769 * Add a page to a bio while respecting the hardware max_sectors, max_segment
770 * and gap limitations.
772 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
773 struct page *page, unsigned int len, unsigned int offset,
774 unsigned int max_sectors, bool *same_page)
776 struct bio_vec *bvec;
778 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
781 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
784 if (bio->bi_vcnt > 0) {
785 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
789 * If the queue doesn't support SG gaps and adding this segment
790 * would create a gap, disallow it.
792 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
793 if (bvec_gap_to_prev(q, bvec, offset))
797 if (bio_full(bio, len))
800 if (bio->bi_vcnt >= queue_max_segments(q))
803 bvec = &bio->bi_io_vec[bio->bi_vcnt];
804 bvec->bv_page = page;
806 bvec->bv_offset = offset;
808 bio->bi_iter.bi_size += len;
813 * bio_add_pc_page - attempt to add page to passthrough bio
814 * @q: the target queue
815 * @bio: destination bio
817 * @len: vec entry length
818 * @offset: vec entry offset
820 * Attempt to add a page to the bio_vec maplist. This can fail for a
821 * number of reasons, such as the bio being full or target block device
822 * limitations. The target block device must allow bio's up to PAGE_SIZE,
823 * so it is always possible to add a single page to an empty bio.
825 * This should only be used by passthrough bios.
827 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
828 struct page *page, unsigned int len, unsigned int offset)
830 bool same_page = false;
831 return bio_add_hw_page(q, bio, page, len, offset,
832 queue_max_hw_sectors(q), &same_page);
834 EXPORT_SYMBOL(bio_add_pc_page);
837 * __bio_try_merge_page - try appending data to an existing bvec.
838 * @bio: destination bio
839 * @page: start page to add
840 * @len: length of the data to add
841 * @off: offset of the data relative to @page
842 * @same_page: return if the segment has been merged inside the same page
844 * Try to add the data at @page + @off to the last bvec of @bio. This is a
845 * useful optimisation for file systems with a block size smaller than the
848 * Warn if (@len, @off) crosses pages in case that @same_page is true.
850 * Return %true on success or %false on failure.
852 bool __bio_try_merge_page(struct bio *bio, struct page *page,
853 unsigned int len, unsigned int off, bool *same_page)
855 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
858 if (bio->bi_vcnt > 0) {
859 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
861 if (page_is_mergeable(bv, page, len, off, same_page)) {
862 if (bio->bi_iter.bi_size > UINT_MAX - len) {
867 bio->bi_iter.bi_size += len;
873 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
876 * __bio_add_page - add page(s) to a bio in a new segment
877 * @bio: destination bio
878 * @page: start page to add
879 * @len: length of the data to add, may cross pages
880 * @off: offset of the data relative to @page, may cross pages
882 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
883 * that @bio has space for another bvec.
885 void __bio_add_page(struct bio *bio, struct page *page,
886 unsigned int len, unsigned int off)
888 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
890 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
891 WARN_ON_ONCE(bio_full(bio, len));
897 bio->bi_iter.bi_size += len;
900 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
901 bio_set_flag(bio, BIO_WORKINGSET);
903 EXPORT_SYMBOL_GPL(__bio_add_page);
906 * bio_add_page - attempt to add page(s) to bio
907 * @bio: destination bio
908 * @page: start page to add
909 * @len: vec entry length, may cross pages
910 * @offset: vec entry offset relative to @page, may cross pages
912 * Attempt to add page(s) to the bio_vec maplist. This will only fail
913 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
915 int bio_add_page(struct bio *bio, struct page *page,
916 unsigned int len, unsigned int offset)
918 bool same_page = false;
920 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
921 if (bio_full(bio, len))
923 __bio_add_page(bio, page, len, offset);
927 EXPORT_SYMBOL(bio_add_page);
929 void bio_release_pages(struct bio *bio, bool mark_dirty)
931 struct bvec_iter_all iter_all;
932 struct bio_vec *bvec;
934 if (bio_flagged(bio, BIO_NO_PAGE_REF))
937 bio_for_each_segment_all(bvec, bio, iter_all) {
938 if (mark_dirty && !PageCompound(bvec->bv_page))
939 set_page_dirty_lock(bvec->bv_page);
940 put_page(bvec->bv_page);
943 EXPORT_SYMBOL_GPL(bio_release_pages);
945 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter)
947 const struct bio_vec *bv = iter->bvec;
951 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len))
954 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count);
955 size = bio_add_page(bio, bv->bv_page, len,
956 bv->bv_offset + iter->iov_offset);
957 if (unlikely(size != len))
959 iov_iter_advance(iter, size);
963 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
966 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
967 * @bio: bio to add pages to
968 * @iter: iov iterator describing the region to be mapped
970 * Pins pages from *iter and appends them to @bio's bvec array. The
971 * pages will have to be released using put_page() when done.
972 * For multi-segment *iter, this function only adds pages from the
973 * next non-empty segment of the iov iterator.
975 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
977 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
978 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
979 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
980 struct page **pages = (struct page **)bv;
981 bool same_page = false;
987 * Move page array up in the allocated memory for the bio vecs as far as
988 * possible so that we can start filling biovecs from the beginning
989 * without overwriting the temporary page array.
991 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
992 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
994 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
995 if (unlikely(size <= 0))
996 return size ? size : -EFAULT;
998 for (left = size, i = 0; left > 0; left -= len, i++) {
999 struct page *page = pages[i];
1001 len = min_t(size_t, PAGE_SIZE - offset, left);
1003 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1007 if (WARN_ON_ONCE(bio_full(bio, len)))
1009 __bio_add_page(bio, page, len, offset);
1014 iov_iter_advance(iter, size);
1018 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
1020 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1021 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1022 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1023 unsigned int max_append_sectors = queue_max_zone_append_sectors(q);
1024 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1025 struct page **pages = (struct page **)bv;
1031 if (WARN_ON_ONCE(!max_append_sectors))
1035 * Move page array up in the allocated memory for the bio vecs as far as
1036 * possible so that we can start filling biovecs from the beginning
1037 * without overwriting the temporary page array.
1039 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1040 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1042 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1043 if (unlikely(size <= 0))
1044 return size ? size : -EFAULT;
1046 for (left = size, i = 0; left > 0; left -= len, i++) {
1047 struct page *page = pages[i];
1048 bool same_page = false;
1050 len = min_t(size_t, PAGE_SIZE - offset, left);
1051 if (bio_add_hw_page(q, bio, page, len, offset,
1052 max_append_sectors, &same_page) != len) {
1061 iov_iter_advance(iter, size - left);
1066 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1067 * @bio: bio to add pages to
1068 * @iter: iov iterator describing the region to be added
1070 * This takes either an iterator pointing to user memory, or one pointing to
1071 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1072 * map them into the kernel. On IO completion, the caller should put those
1073 * pages. If we're adding kernel pages, and the caller told us it's safe to
1074 * do so, we just have to add the pages to the bio directly. We don't grab an
1075 * extra reference to those pages (the user should already have that), and we
1076 * don't put the page on IO completion. The caller needs to check if the bio is
1077 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
1080 * The function tries, but does not guarantee, to pin as many pages as
1081 * fit into the bio, or are requested in @iter, whatever is smaller. If
1082 * MM encounters an error pinning the requested pages, it stops. Error
1083 * is returned only if 0 pages could be pinned.
1085 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1087 const bool is_bvec = iov_iter_is_bvec(iter);
1090 if (WARN_ON_ONCE(bio->bi_vcnt))
1094 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1095 if (WARN_ON_ONCE(is_bvec))
1097 ret = __bio_iov_append_get_pages(bio, iter);
1100 ret = __bio_iov_bvec_add_pages(bio, iter);
1102 ret = __bio_iov_iter_get_pages(bio, iter);
1104 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1107 bio_set_flag(bio, BIO_NO_PAGE_REF);
1108 return bio->bi_vcnt ? 0 : ret;
1110 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1112 static void submit_bio_wait_endio(struct bio *bio)
1114 complete(bio->bi_private);
1118 * submit_bio_wait - submit a bio, and wait until it completes
1119 * @bio: The &struct bio which describes the I/O
1121 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1122 * bio_endio() on failure.
1124 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1125 * result in bio reference to be consumed. The caller must drop the reference
1128 int submit_bio_wait(struct bio *bio)
1130 DECLARE_COMPLETION_ONSTACK_MAP(done,
1131 bio->bi_bdev->bd_disk->lockdep_map);
1132 unsigned long hang_check;
1134 bio->bi_private = &done;
1135 bio->bi_end_io = submit_bio_wait_endio;
1136 bio->bi_opf |= REQ_SYNC;
1139 /* Prevent hang_check timer from firing at us during very long I/O */
1140 hang_check = sysctl_hung_task_timeout_secs;
1142 while (!wait_for_completion_io_timeout(&done,
1143 hang_check * (HZ/2)))
1146 wait_for_completion_io(&done);
1148 return blk_status_to_errno(bio->bi_status);
1150 EXPORT_SYMBOL(submit_bio_wait);
1153 * bio_advance - increment/complete a bio by some number of bytes
1154 * @bio: bio to advance
1155 * @bytes: number of bytes to complete
1157 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1158 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1159 * be updated on the last bvec as well.
1161 * @bio will then represent the remaining, uncompleted portion of the io.
1163 void bio_advance(struct bio *bio, unsigned bytes)
1165 if (bio_integrity(bio))
1166 bio_integrity_advance(bio, bytes);
1168 bio_crypt_advance(bio, bytes);
1169 bio_advance_iter(bio, &bio->bi_iter, bytes);
1171 EXPORT_SYMBOL(bio_advance);
1173 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1174 struct bio *src, struct bvec_iter *src_iter)
1176 struct bio_vec src_bv, dst_bv;
1177 void *src_p, *dst_p;
1180 while (src_iter->bi_size && dst_iter->bi_size) {
1181 src_bv = bio_iter_iovec(src, *src_iter);
1182 dst_bv = bio_iter_iovec(dst, *dst_iter);
1184 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1186 src_p = kmap_atomic(src_bv.bv_page);
1187 dst_p = kmap_atomic(dst_bv.bv_page);
1189 memcpy(dst_p + dst_bv.bv_offset,
1190 src_p + src_bv.bv_offset,
1193 kunmap_atomic(dst_p);
1194 kunmap_atomic(src_p);
1196 flush_dcache_page(dst_bv.bv_page);
1198 bio_advance_iter_single(src, src_iter, bytes);
1199 bio_advance_iter_single(dst, dst_iter, bytes);
1202 EXPORT_SYMBOL(bio_copy_data_iter);
1205 * bio_copy_data - copy contents of data buffers from one bio to another
1207 * @dst: destination bio
1209 * Stops when it reaches the end of either @src or @dst - that is, copies
1210 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1212 void bio_copy_data(struct bio *dst, struct bio *src)
1214 struct bvec_iter src_iter = src->bi_iter;
1215 struct bvec_iter dst_iter = dst->bi_iter;
1217 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1219 EXPORT_SYMBOL(bio_copy_data);
1222 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1224 * @src: source bio list
1225 * @dst: destination bio list
1227 * Stops when it reaches the end of either the @src list or @dst list - that is,
1228 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1231 void bio_list_copy_data(struct bio *dst, struct bio *src)
1233 struct bvec_iter src_iter = src->bi_iter;
1234 struct bvec_iter dst_iter = dst->bi_iter;
1237 if (!src_iter.bi_size) {
1242 src_iter = src->bi_iter;
1245 if (!dst_iter.bi_size) {
1250 dst_iter = dst->bi_iter;
1253 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1256 EXPORT_SYMBOL(bio_list_copy_data);
1258 void bio_free_pages(struct bio *bio)
1260 struct bio_vec *bvec;
1261 struct bvec_iter_all iter_all;
1263 bio_for_each_segment_all(bvec, bio, iter_all)
1264 __free_page(bvec->bv_page);
1266 EXPORT_SYMBOL(bio_free_pages);
1269 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1270 * for performing direct-IO in BIOs.
1272 * The problem is that we cannot run set_page_dirty() from interrupt context
1273 * because the required locks are not interrupt-safe. So what we can do is to
1274 * mark the pages dirty _before_ performing IO. And in interrupt context,
1275 * check that the pages are still dirty. If so, fine. If not, redirty them
1276 * in process context.
1278 * We special-case compound pages here: normally this means reads into hugetlb
1279 * pages. The logic in here doesn't really work right for compound pages
1280 * because the VM does not uniformly chase down the head page in all cases.
1281 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1282 * handle them at all. So we skip compound pages here at an early stage.
1284 * Note that this code is very hard to test under normal circumstances because
1285 * direct-io pins the pages with get_user_pages(). This makes
1286 * is_page_cache_freeable return false, and the VM will not clean the pages.
1287 * But other code (eg, flusher threads) could clean the pages if they are mapped
1290 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1291 * deferred bio dirtying paths.
1295 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1297 void bio_set_pages_dirty(struct bio *bio)
1299 struct bio_vec *bvec;
1300 struct bvec_iter_all iter_all;
1302 bio_for_each_segment_all(bvec, bio, iter_all) {
1303 if (!PageCompound(bvec->bv_page))
1304 set_page_dirty_lock(bvec->bv_page);
1309 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1310 * If they are, then fine. If, however, some pages are clean then they must
1311 * have been written out during the direct-IO read. So we take another ref on
1312 * the BIO and re-dirty the pages in process context.
1314 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1315 * here on. It will run one put_page() against each page and will run one
1316 * bio_put() against the BIO.
1319 static void bio_dirty_fn(struct work_struct *work);
1321 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1322 static DEFINE_SPINLOCK(bio_dirty_lock);
1323 static struct bio *bio_dirty_list;
1326 * This runs in process context
1328 static void bio_dirty_fn(struct work_struct *work)
1330 struct bio *bio, *next;
1332 spin_lock_irq(&bio_dirty_lock);
1333 next = bio_dirty_list;
1334 bio_dirty_list = NULL;
1335 spin_unlock_irq(&bio_dirty_lock);
1337 while ((bio = next) != NULL) {
1338 next = bio->bi_private;
1340 bio_release_pages(bio, true);
1345 void bio_check_pages_dirty(struct bio *bio)
1347 struct bio_vec *bvec;
1348 unsigned long flags;
1349 struct bvec_iter_all iter_all;
1351 bio_for_each_segment_all(bvec, bio, iter_all) {
1352 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1356 bio_release_pages(bio, false);
1360 spin_lock_irqsave(&bio_dirty_lock, flags);
1361 bio->bi_private = bio_dirty_list;
1362 bio_dirty_list = bio;
1363 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1364 schedule_work(&bio_dirty_work);
1367 static inline bool bio_remaining_done(struct bio *bio)
1370 * If we're not chaining, then ->__bi_remaining is always 1 and
1371 * we always end io on the first invocation.
1373 if (!bio_flagged(bio, BIO_CHAIN))
1376 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1378 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1379 bio_clear_flag(bio, BIO_CHAIN);
1387 * bio_endio - end I/O on a bio
1391 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1392 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1393 * bio unless they own it and thus know that it has an end_io function.
1395 * bio_endio() can be called several times on a bio that has been chained
1396 * using bio_chain(). The ->bi_end_io() function will only be called the
1397 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1398 * generated if BIO_TRACE_COMPLETION is set.
1400 void bio_endio(struct bio *bio)
1403 if (!bio_remaining_done(bio))
1405 if (!bio_integrity_endio(bio))
1409 rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio);
1412 * Need to have a real endio function for chained bios, otherwise
1413 * various corner cases will break (like stacking block devices that
1414 * save/restore bi_end_io) - however, we want to avoid unbounded
1415 * recursion and blowing the stack. Tail call optimization would
1416 * handle this, but compiling with frame pointers also disables
1417 * gcc's sibling call optimization.
1419 if (bio->bi_end_io == bio_chain_endio) {
1420 bio = __bio_chain_endio(bio);
1424 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1425 trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio);
1426 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1429 blk_throtl_bio_endio(bio);
1430 /* release cgroup info */
1433 bio->bi_end_io(bio);
1435 EXPORT_SYMBOL(bio_endio);
1438 * bio_split - split a bio
1439 * @bio: bio to split
1440 * @sectors: number of sectors to split from the front of @bio
1442 * @bs: bio set to allocate from
1444 * Allocates and returns a new bio which represents @sectors from the start of
1445 * @bio, and updates @bio to represent the remaining sectors.
1447 * Unless this is a discard request the newly allocated bio will point
1448 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1449 * neither @bio nor @bs are freed before the split bio.
1451 struct bio *bio_split(struct bio *bio, int sectors,
1452 gfp_t gfp, struct bio_set *bs)
1456 BUG_ON(sectors <= 0);
1457 BUG_ON(sectors >= bio_sectors(bio));
1459 /* Zone append commands cannot be split */
1460 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1463 split = bio_clone_fast(bio, gfp, bs);
1467 split->bi_iter.bi_size = sectors << 9;
1469 if (bio_integrity(split))
1470 bio_integrity_trim(split);
1472 bio_advance(bio, split->bi_iter.bi_size);
1474 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1475 bio_set_flag(split, BIO_TRACE_COMPLETION);
1479 EXPORT_SYMBOL(bio_split);
1482 * bio_trim - trim a bio
1484 * @offset: number of sectors to trim from the front of @bio
1485 * @size: size we want to trim @bio to, in sectors
1487 void bio_trim(struct bio *bio, int offset, int size)
1489 /* 'bio' is a cloned bio which we need to trim to match
1490 * the given offset and size.
1494 if (offset == 0 && size == bio->bi_iter.bi_size)
1497 bio_advance(bio, offset << 9);
1498 bio->bi_iter.bi_size = size;
1500 if (bio_integrity(bio))
1501 bio_integrity_trim(bio);
1504 EXPORT_SYMBOL_GPL(bio_trim);
1507 * create memory pools for biovec's in a bio_set.
1508 * use the global biovec slabs created for general use.
1510 int biovec_init_pool(mempool_t *pool, int pool_entries)
1512 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1514 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1518 * bioset_exit - exit a bioset initialized with bioset_init()
1520 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1523 void bioset_exit(struct bio_set *bs)
1525 if (bs->rescue_workqueue)
1526 destroy_workqueue(bs->rescue_workqueue);
1527 bs->rescue_workqueue = NULL;
1529 mempool_exit(&bs->bio_pool);
1530 mempool_exit(&bs->bvec_pool);
1532 bioset_integrity_free(bs);
1535 bs->bio_slab = NULL;
1537 EXPORT_SYMBOL(bioset_exit);
1540 * bioset_init - Initialize a bio_set
1541 * @bs: pool to initialize
1542 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1543 * @front_pad: Number of bytes to allocate in front of the returned bio
1544 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1545 * and %BIOSET_NEED_RESCUER
1548 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1549 * to ask for a number of bytes to be allocated in front of the bio.
1550 * Front pad allocation is useful for embedding the bio inside
1551 * another structure, to avoid allocating extra data to go with the bio.
1552 * Note that the bio must be embedded at the END of that structure always,
1553 * or things will break badly.
1554 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1555 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1556 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1557 * dispatch queued requests when the mempool runs out of space.
1560 int bioset_init(struct bio_set *bs,
1561 unsigned int pool_size,
1562 unsigned int front_pad,
1565 bs->front_pad = front_pad;
1566 if (flags & BIOSET_NEED_BVECS)
1567 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1571 spin_lock_init(&bs->rescue_lock);
1572 bio_list_init(&bs->rescue_list);
1573 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1575 bs->bio_slab = bio_find_or_create_slab(bs);
1579 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1582 if ((flags & BIOSET_NEED_BVECS) &&
1583 biovec_init_pool(&bs->bvec_pool, pool_size))
1586 if (!(flags & BIOSET_NEED_RESCUER))
1589 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1590 if (!bs->rescue_workqueue)
1598 EXPORT_SYMBOL(bioset_init);
1601 * Initialize and setup a new bio_set, based on the settings from
1604 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1609 if (src->bvec_pool.min_nr)
1610 flags |= BIOSET_NEED_BVECS;
1611 if (src->rescue_workqueue)
1612 flags |= BIOSET_NEED_RESCUER;
1614 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1616 EXPORT_SYMBOL(bioset_init_from_src);
1618 static void __init biovec_init_slabs(void)
1622 for (i = 0; i < BVEC_POOL_NR; i++) {
1624 struct biovec_slab *bvs = bvec_slabs + i;
1626 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1631 size = bvs->nr_vecs * sizeof(struct bio_vec);
1632 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1633 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1637 static int __init init_bio(void)
1639 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET);
1641 bio_integrity_init();
1642 biovec_init_slabs();
1644 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
1645 panic("bio: can't allocate bios\n");
1647 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1648 panic("bio: can't create integrity pool\n");
1652 subsys_initcall(init_bio);