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/iocontext.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/kernel.h>
26 #include <linux/export.h>
27 #include <linux/mempool.h>
28 #include <linux/workqueue.h>
29 #include <linux/cgroup.h>
30 #include <scsi/sg.h> /* for struct sg_iovec */
32 #include <trace/events/block.h>
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
38 #define BIO_INLINE_VECS 4
40 static mempool_t *bio_split_pool __read_mostly;
43 * if you change this list, also change bvec_alloc or things will
44 * break badly! cannot be bigger than what you can fit into an
47 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
48 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
49 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
54 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
55 * IO code that does not need private memory pools.
57 struct bio_set *fs_bio_set;
58 EXPORT_SYMBOL(fs_bio_set);
61 * Our slab pool management
64 struct kmem_cache *slab;
65 unsigned int slab_ref;
66 unsigned int slab_size;
69 static DEFINE_MUTEX(bio_slab_lock);
70 static struct bio_slab *bio_slabs;
71 static unsigned int bio_slab_nr, bio_slab_max;
73 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
75 unsigned int sz = sizeof(struct bio) + extra_size;
76 struct kmem_cache *slab = NULL;
77 struct bio_slab *bslab, *new_bio_slabs;
78 unsigned int new_bio_slab_max;
79 unsigned int i, entry = -1;
81 mutex_lock(&bio_slab_lock);
84 while (i < bio_slab_nr) {
85 bslab = &bio_slabs[i];
87 if (!bslab->slab && entry == -1)
89 else if (bslab->slab_size == sz) {
100 if (bio_slab_nr == bio_slab_max && entry == -1) {
101 new_bio_slab_max = bio_slab_max << 1;
102 new_bio_slabs = krealloc(bio_slabs,
103 new_bio_slab_max * sizeof(struct bio_slab),
107 bio_slab_max = new_bio_slab_max;
108 bio_slabs = new_bio_slabs;
111 entry = bio_slab_nr++;
113 bslab = &bio_slabs[entry];
115 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
116 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
120 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
123 bslab->slab_size = sz;
125 mutex_unlock(&bio_slab_lock);
129 static void bio_put_slab(struct bio_set *bs)
131 struct bio_slab *bslab = NULL;
134 mutex_lock(&bio_slab_lock);
136 for (i = 0; i < bio_slab_nr; i++) {
137 if (bs->bio_slab == bio_slabs[i].slab) {
138 bslab = &bio_slabs[i];
143 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
146 WARN_ON(!bslab->slab_ref);
148 if (--bslab->slab_ref)
151 kmem_cache_destroy(bslab->slab);
155 mutex_unlock(&bio_slab_lock);
158 unsigned int bvec_nr_vecs(unsigned short idx)
160 return bvec_slabs[idx].nr_vecs;
163 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
165 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
167 if (idx == BIOVEC_MAX_IDX)
168 mempool_free(bv, bs->bvec_pool);
170 struct biovec_slab *bvs = bvec_slabs + idx;
172 kmem_cache_free(bvs->slab, bv);
176 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
182 * see comment near bvec_array define!
200 case 129 ... BIO_MAX_PAGES:
208 * idx now points to the pool we want to allocate from. only the
209 * 1-vec entry pool is mempool backed.
211 if (*idx == BIOVEC_MAX_IDX) {
213 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
215 struct biovec_slab *bvs = bvec_slabs + *idx;
216 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
219 * Make this allocation restricted and don't dump info on
220 * allocation failures, since we'll fallback to the mempool
221 * in case of failure.
223 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
226 * Try a slab allocation. If this fails and __GFP_WAIT
227 * is set, retry with the 1-entry mempool
229 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
230 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
231 *idx = BIOVEC_MAX_IDX;
239 static void __bio_free(struct bio *bio)
241 bio_disassociate_task(bio);
243 if (bio_integrity(bio))
244 bio_integrity_free(bio);
247 static void bio_free(struct bio *bio)
249 struct bio_set *bs = bio->bi_pool;
255 if (bio_has_allocated_vec(bio))
256 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
259 * If we have front padding, adjust the bio pointer before freeing
264 mempool_free(p, bs->bio_pool);
266 /* Bio was allocated by bio_kmalloc() */
271 void bio_init(struct bio *bio)
273 memset(bio, 0, sizeof(*bio));
274 bio->bi_flags = 1 << BIO_UPTODATE;
275 atomic_set(&bio->bi_cnt, 1);
277 EXPORT_SYMBOL(bio_init);
280 * bio_reset - reinitialize a bio
284 * After calling bio_reset(), @bio will be in the same state as a freshly
285 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
286 * preserved are the ones that are initialized by bio_alloc_bioset(). See
287 * comment in struct bio.
289 void bio_reset(struct bio *bio)
291 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
295 memset(bio, 0, BIO_RESET_BYTES);
296 bio->bi_flags = flags|(1 << BIO_UPTODATE);
298 EXPORT_SYMBOL(bio_reset);
301 * bio_alloc_bioset - allocate a bio for I/O
302 * @gfp_mask: the GFP_ mask given to the slab allocator
303 * @nr_iovecs: number of iovecs to pre-allocate
304 * @bs: the bio_set to allocate from.
307 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
308 * backed by the @bs's mempool.
310 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
311 * able to allocate a bio. This is due to the mempool guarantees. To make this
312 * work, callers must never allocate more than 1 bio at a time from this pool.
313 * Callers that need to allocate more than 1 bio must always submit the
314 * previously allocated bio for IO before attempting to allocate a new one.
315 * Failure to do so can cause deadlocks under memory pressure.
318 * Pointer to new bio on success, NULL on failure.
320 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
323 unsigned inline_vecs;
324 unsigned long idx = BIO_POOL_NONE;
325 struct bio_vec *bvl = NULL;
330 if (nr_iovecs > UIO_MAXIOV)
333 p = kmalloc(sizeof(struct bio) +
334 nr_iovecs * sizeof(struct bio_vec),
337 inline_vecs = nr_iovecs;
339 p = mempool_alloc(bs->bio_pool, gfp_mask);
340 front_pad = bs->front_pad;
341 inline_vecs = BIO_INLINE_VECS;
350 if (nr_iovecs > inline_vecs) {
351 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
354 } else if (nr_iovecs) {
355 bvl = bio->bi_inline_vecs;
359 bio->bi_flags |= idx << BIO_POOL_OFFSET;
360 bio->bi_max_vecs = nr_iovecs;
361 bio->bi_io_vec = bvl;
365 mempool_free(p, bs->bio_pool);
368 EXPORT_SYMBOL(bio_alloc_bioset);
370 void zero_fill_bio(struct bio *bio)
376 bio_for_each_segment(bv, bio, i) {
377 char *data = bvec_kmap_irq(bv, &flags);
378 memset(data, 0, bv->bv_len);
379 flush_dcache_page(bv->bv_page);
380 bvec_kunmap_irq(data, &flags);
383 EXPORT_SYMBOL(zero_fill_bio);
386 * bio_put - release a reference to a bio
387 * @bio: bio to release reference to
390 * Put a reference to a &struct bio, either one you have gotten with
391 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
393 void bio_put(struct bio *bio)
395 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
400 if (atomic_dec_and_test(&bio->bi_cnt))
403 EXPORT_SYMBOL(bio_put);
405 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
407 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
408 blk_recount_segments(q, bio);
410 return bio->bi_phys_segments;
412 EXPORT_SYMBOL(bio_phys_segments);
415 * __bio_clone - clone a bio
416 * @bio: destination bio
417 * @bio_src: bio to clone
419 * Clone a &bio. Caller will own the returned bio, but not
420 * the actual data it points to. Reference count of returned
423 void __bio_clone(struct bio *bio, struct bio *bio_src)
425 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
426 bio_src->bi_max_vecs * sizeof(struct bio_vec));
429 * most users will be overriding ->bi_bdev with a new target,
430 * so we don't set nor calculate new physical/hw segment counts here
432 bio->bi_sector = bio_src->bi_sector;
433 bio->bi_bdev = bio_src->bi_bdev;
434 bio->bi_flags |= 1 << BIO_CLONED;
435 bio->bi_rw = bio_src->bi_rw;
436 bio->bi_vcnt = bio_src->bi_vcnt;
437 bio->bi_size = bio_src->bi_size;
438 bio->bi_idx = bio_src->bi_idx;
440 EXPORT_SYMBOL(__bio_clone);
443 * bio_clone_bioset - clone a bio
445 * @gfp_mask: allocation priority
446 * @bs: bio_set to allocate from
448 * Like __bio_clone, only also allocates the returned bio
450 struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask,
455 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs);
461 if (bio_integrity(bio)) {
464 ret = bio_integrity_clone(b, bio, gfp_mask);
474 EXPORT_SYMBOL(bio_clone_bioset);
477 * bio_get_nr_vecs - return approx number of vecs
480 * Return the approximate number of pages we can send to this target.
481 * There's no guarantee that you will be able to fit this number of pages
482 * into a bio, it does not account for dynamic restrictions that vary
485 int bio_get_nr_vecs(struct block_device *bdev)
487 struct request_queue *q = bdev_get_queue(bdev);
490 nr_pages = min_t(unsigned,
491 queue_max_segments(q),
492 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
494 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
497 EXPORT_SYMBOL(bio_get_nr_vecs);
499 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
500 *page, unsigned int len, unsigned int offset,
501 unsigned short max_sectors)
503 int retried_segments = 0;
504 struct bio_vec *bvec;
507 * cloned bio must not modify vec list
509 if (unlikely(bio_flagged(bio, BIO_CLONED)))
512 if (((bio->bi_size + len) >> 9) > max_sectors)
516 * For filesystems with a blocksize smaller than the pagesize
517 * we will often be called with the same page as last time and
518 * a consecutive offset. Optimize this special case.
520 if (bio->bi_vcnt > 0) {
521 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
523 if (page == prev->bv_page &&
524 offset == prev->bv_offset + prev->bv_len) {
525 unsigned int prev_bv_len = prev->bv_len;
528 if (q->merge_bvec_fn) {
529 struct bvec_merge_data bvm = {
530 /* prev_bvec is already charged in
531 bi_size, discharge it in order to
532 simulate merging updated prev_bvec
534 .bi_bdev = bio->bi_bdev,
535 .bi_sector = bio->bi_sector,
536 .bi_size = bio->bi_size - prev_bv_len,
540 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
550 if (bio->bi_vcnt >= bio->bi_max_vecs)
554 * we might lose a segment or two here, but rather that than
555 * make this too complex.
558 while (bio->bi_phys_segments >= queue_max_segments(q)) {
560 if (retried_segments)
563 retried_segments = 1;
564 blk_recount_segments(q, bio);
568 * setup the new entry, we might clear it again later if we
569 * cannot add the page
571 bvec = &bio->bi_io_vec[bio->bi_vcnt];
572 bvec->bv_page = page;
574 bvec->bv_offset = offset;
577 * if queue has other restrictions (eg varying max sector size
578 * depending on offset), it can specify a merge_bvec_fn in the
579 * queue to get further control
581 if (q->merge_bvec_fn) {
582 struct bvec_merge_data bvm = {
583 .bi_bdev = bio->bi_bdev,
584 .bi_sector = bio->bi_sector,
585 .bi_size = bio->bi_size,
590 * merge_bvec_fn() returns number of bytes it can accept
593 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
594 bvec->bv_page = NULL;
601 /* If we may be able to merge these biovecs, force a recount */
602 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
603 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
606 bio->bi_phys_segments++;
613 * bio_add_pc_page - attempt to add page to bio
614 * @q: the target queue
615 * @bio: destination bio
617 * @len: vec entry length
618 * @offset: vec entry offset
620 * Attempt to add a page to the bio_vec maplist. This can fail for a
621 * number of reasons, such as the bio being full or target block device
622 * limitations. The target block device must allow bio's up to PAGE_SIZE,
623 * so it is always possible to add a single page to an empty bio.
625 * This should only be used by REQ_PC bios.
627 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
628 unsigned int len, unsigned int offset)
630 return __bio_add_page(q, bio, page, len, offset,
631 queue_max_hw_sectors(q));
633 EXPORT_SYMBOL(bio_add_pc_page);
636 * bio_add_page - attempt to add page to bio
637 * @bio: destination bio
639 * @len: vec entry length
640 * @offset: vec entry offset
642 * Attempt to add a page to the bio_vec maplist. This can fail for a
643 * number of reasons, such as the bio being full or target block device
644 * limitations. The target block device must allow bio's up to PAGE_SIZE,
645 * so it is always possible to add a single page to an empty bio.
647 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
650 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
651 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
653 EXPORT_SYMBOL(bio_add_page);
655 struct bio_map_data {
656 struct bio_vec *iovecs;
657 struct sg_iovec *sgvecs;
662 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
663 struct sg_iovec *iov, int iov_count,
666 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
667 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
668 bmd->nr_sgvecs = iov_count;
669 bmd->is_our_pages = is_our_pages;
670 bio->bi_private = bmd;
673 static void bio_free_map_data(struct bio_map_data *bmd)
680 static struct bio_map_data *bio_alloc_map_data(int nr_segs,
681 unsigned int iov_count,
684 struct bio_map_data *bmd;
686 if (iov_count > UIO_MAXIOV)
689 bmd = kmalloc(sizeof(*bmd), gfp_mask);
693 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
699 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
708 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
709 struct sg_iovec *iov, int iov_count,
710 int to_user, int from_user, int do_free_page)
713 struct bio_vec *bvec;
715 unsigned int iov_off = 0;
717 __bio_for_each_segment(bvec, bio, i, 0) {
718 char *bv_addr = page_address(bvec->bv_page);
719 unsigned int bv_len = iovecs[i].bv_len;
721 while (bv_len && iov_idx < iov_count) {
723 char __user *iov_addr;
725 bytes = min_t(unsigned int,
726 iov[iov_idx].iov_len - iov_off, bv_len);
727 iov_addr = iov[iov_idx].iov_base + iov_off;
731 ret = copy_to_user(iov_addr, bv_addr,
735 ret = copy_from_user(bv_addr, iov_addr,
747 if (iov[iov_idx].iov_len == iov_off) {
754 __free_page(bvec->bv_page);
761 * bio_uncopy_user - finish previously mapped bio
762 * @bio: bio being terminated
764 * Free pages allocated from bio_copy_user() and write back data
765 * to user space in case of a read.
767 int bio_uncopy_user(struct bio *bio)
769 struct bio_map_data *bmd = bio->bi_private;
772 if (!bio_flagged(bio, BIO_NULL_MAPPED))
773 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
774 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
775 0, bmd->is_our_pages);
776 bio_free_map_data(bmd);
780 EXPORT_SYMBOL(bio_uncopy_user);
783 * bio_copy_user_iov - copy user data to bio
784 * @q: destination block queue
785 * @map_data: pointer to the rq_map_data holding pages (if necessary)
787 * @iov_count: number of elements in the iovec
788 * @write_to_vm: bool indicating writing to pages or not
789 * @gfp_mask: memory allocation flags
791 * Prepares and returns a bio for indirect user io, bouncing data
792 * to/from kernel pages as necessary. Must be paired with
793 * call bio_uncopy_user() on io completion.
795 struct bio *bio_copy_user_iov(struct request_queue *q,
796 struct rq_map_data *map_data,
797 struct sg_iovec *iov, int iov_count,
798 int write_to_vm, gfp_t gfp_mask)
800 struct bio_map_data *bmd;
801 struct bio_vec *bvec;
806 unsigned int len = 0;
807 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
809 for (i = 0; i < iov_count; i++) {
814 uaddr = (unsigned long)iov[i].iov_base;
815 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
816 start = uaddr >> PAGE_SHIFT;
822 return ERR_PTR(-EINVAL);
824 nr_pages += end - start;
825 len += iov[i].iov_len;
831 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
833 return ERR_PTR(-ENOMEM);
836 bio = bio_kmalloc(gfp_mask, nr_pages);
841 bio->bi_rw |= REQ_WRITE;
846 nr_pages = 1 << map_data->page_order;
847 i = map_data->offset / PAGE_SIZE;
850 unsigned int bytes = PAGE_SIZE;
858 if (i == map_data->nr_entries * nr_pages) {
863 page = map_data->pages[i / nr_pages];
864 page += (i % nr_pages);
868 page = alloc_page(q->bounce_gfp | gfp_mask);
875 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
888 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
889 (map_data && map_data->from_user)) {
890 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
895 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
899 bio_for_each_segment(bvec, bio, i)
900 __free_page(bvec->bv_page);
904 bio_free_map_data(bmd);
909 * bio_copy_user - copy user data to bio
910 * @q: destination block queue
911 * @map_data: pointer to the rq_map_data holding pages (if necessary)
912 * @uaddr: start of user address
913 * @len: length in bytes
914 * @write_to_vm: bool indicating writing to pages or not
915 * @gfp_mask: memory allocation flags
917 * Prepares and returns a bio for indirect user io, bouncing data
918 * to/from kernel pages as necessary. Must be paired with
919 * call bio_uncopy_user() on io completion.
921 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
922 unsigned long uaddr, unsigned int len,
923 int write_to_vm, gfp_t gfp_mask)
927 iov.iov_base = (void __user *)uaddr;
930 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
932 EXPORT_SYMBOL(bio_copy_user);
934 static struct bio *__bio_map_user_iov(struct request_queue *q,
935 struct block_device *bdev,
936 struct sg_iovec *iov, int iov_count,
937 int write_to_vm, gfp_t gfp_mask)
946 for (i = 0; i < iov_count; i++) {
947 unsigned long uaddr = (unsigned long)iov[i].iov_base;
948 unsigned long len = iov[i].iov_len;
949 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
950 unsigned long start = uaddr >> PAGE_SHIFT;
956 return ERR_PTR(-EINVAL);
958 nr_pages += end - start;
960 * buffer must be aligned to at least hardsector size for now
962 if (uaddr & queue_dma_alignment(q))
963 return ERR_PTR(-EINVAL);
967 return ERR_PTR(-EINVAL);
969 bio = bio_kmalloc(gfp_mask, nr_pages);
971 return ERR_PTR(-ENOMEM);
974 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
978 for (i = 0; i < iov_count; i++) {
979 unsigned long uaddr = (unsigned long)iov[i].iov_base;
980 unsigned long len = iov[i].iov_len;
981 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
982 unsigned long start = uaddr >> PAGE_SHIFT;
983 const int local_nr_pages = end - start;
984 const int page_limit = cur_page + local_nr_pages;
986 ret = get_user_pages_fast(uaddr, local_nr_pages,
987 write_to_vm, &pages[cur_page]);
988 if (ret < local_nr_pages) {
993 offset = uaddr & ~PAGE_MASK;
994 for (j = cur_page; j < page_limit; j++) {
995 unsigned int bytes = PAGE_SIZE - offset;
1006 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1016 * release the pages we didn't map into the bio, if any
1018 while (j < page_limit)
1019 page_cache_release(pages[j++]);
1025 * set data direction, and check if mapped pages need bouncing
1028 bio->bi_rw |= REQ_WRITE;
1030 bio->bi_bdev = bdev;
1031 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1035 for (i = 0; i < nr_pages; i++) {
1038 page_cache_release(pages[i]);
1043 return ERR_PTR(ret);
1047 * bio_map_user - map user address into bio
1048 * @q: the struct request_queue for the bio
1049 * @bdev: destination block device
1050 * @uaddr: start of user address
1051 * @len: length in bytes
1052 * @write_to_vm: bool indicating writing to pages or not
1053 * @gfp_mask: memory allocation flags
1055 * Map the user space address into a bio suitable for io to a block
1056 * device. Returns an error pointer in case of error.
1058 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1059 unsigned long uaddr, unsigned int len, int write_to_vm,
1062 struct sg_iovec iov;
1064 iov.iov_base = (void __user *)uaddr;
1067 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1069 EXPORT_SYMBOL(bio_map_user);
1072 * bio_map_user_iov - map user sg_iovec table into bio
1073 * @q: the struct request_queue for the bio
1074 * @bdev: destination block device
1076 * @iov_count: number of elements in the iovec
1077 * @write_to_vm: bool indicating writing to pages or not
1078 * @gfp_mask: memory allocation flags
1080 * Map the user space address into a bio suitable for io to a block
1081 * device. Returns an error pointer in case of error.
1083 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1084 struct sg_iovec *iov, int iov_count,
1085 int write_to_vm, gfp_t gfp_mask)
1089 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1095 * subtle -- if __bio_map_user() ended up bouncing a bio,
1096 * it would normally disappear when its bi_end_io is run.
1097 * however, we need it for the unmap, so grab an extra
1105 static void __bio_unmap_user(struct bio *bio)
1107 struct bio_vec *bvec;
1111 * make sure we dirty pages we wrote to
1113 __bio_for_each_segment(bvec, bio, i, 0) {
1114 if (bio_data_dir(bio) == READ)
1115 set_page_dirty_lock(bvec->bv_page);
1117 page_cache_release(bvec->bv_page);
1124 * bio_unmap_user - unmap a bio
1125 * @bio: the bio being unmapped
1127 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1128 * a process context.
1130 * bio_unmap_user() may sleep.
1132 void bio_unmap_user(struct bio *bio)
1134 __bio_unmap_user(bio);
1137 EXPORT_SYMBOL(bio_unmap_user);
1139 static void bio_map_kern_endio(struct bio *bio, int err)
1144 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1145 unsigned int len, gfp_t gfp_mask)
1147 unsigned long kaddr = (unsigned long)data;
1148 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1149 unsigned long start = kaddr >> PAGE_SHIFT;
1150 const int nr_pages = end - start;
1154 bio = bio_kmalloc(gfp_mask, nr_pages);
1156 return ERR_PTR(-ENOMEM);
1158 offset = offset_in_page(kaddr);
1159 for (i = 0; i < nr_pages; i++) {
1160 unsigned int bytes = PAGE_SIZE - offset;
1168 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1177 bio->bi_end_io = bio_map_kern_endio;
1182 * bio_map_kern - map kernel address into bio
1183 * @q: the struct request_queue for the bio
1184 * @data: pointer to buffer to map
1185 * @len: length in bytes
1186 * @gfp_mask: allocation flags for bio allocation
1188 * Map the kernel address into a bio suitable for io to a block
1189 * device. Returns an error pointer in case of error.
1191 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1196 bio = __bio_map_kern(q, data, len, gfp_mask);
1200 if (bio->bi_size == len)
1204 * Don't support partial mappings.
1207 return ERR_PTR(-EINVAL);
1209 EXPORT_SYMBOL(bio_map_kern);
1211 static void bio_copy_kern_endio(struct bio *bio, int err)
1213 struct bio_vec *bvec;
1214 const int read = bio_data_dir(bio) == READ;
1215 struct bio_map_data *bmd = bio->bi_private;
1217 char *p = bmd->sgvecs[0].iov_base;
1219 __bio_for_each_segment(bvec, bio, i, 0) {
1220 char *addr = page_address(bvec->bv_page);
1221 int len = bmd->iovecs[i].bv_len;
1224 memcpy(p, addr, len);
1226 __free_page(bvec->bv_page);
1230 bio_free_map_data(bmd);
1235 * bio_copy_kern - copy kernel address into bio
1236 * @q: the struct request_queue for the bio
1237 * @data: pointer to buffer to copy
1238 * @len: length in bytes
1239 * @gfp_mask: allocation flags for bio and page allocation
1240 * @reading: data direction is READ
1242 * copy the kernel address into a bio suitable for io to a block
1243 * device. Returns an error pointer in case of error.
1245 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1246 gfp_t gfp_mask, int reading)
1249 struct bio_vec *bvec;
1252 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1259 bio_for_each_segment(bvec, bio, i) {
1260 char *addr = page_address(bvec->bv_page);
1262 memcpy(addr, p, bvec->bv_len);
1267 bio->bi_end_io = bio_copy_kern_endio;
1271 EXPORT_SYMBOL(bio_copy_kern);
1274 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1275 * for performing direct-IO in BIOs.
1277 * The problem is that we cannot run set_page_dirty() from interrupt context
1278 * because the required locks are not interrupt-safe. So what we can do is to
1279 * mark the pages dirty _before_ performing IO. And in interrupt context,
1280 * check that the pages are still dirty. If so, fine. If not, redirty them
1281 * in process context.
1283 * We special-case compound pages here: normally this means reads into hugetlb
1284 * pages. The logic in here doesn't really work right for compound pages
1285 * because the VM does not uniformly chase down the head page in all cases.
1286 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1287 * handle them at all. So we skip compound pages here at an early stage.
1289 * Note that this code is very hard to test under normal circumstances because
1290 * direct-io pins the pages with get_user_pages(). This makes
1291 * is_page_cache_freeable return false, and the VM will not clean the pages.
1292 * But other code (eg, flusher threads) could clean the pages if they are mapped
1295 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1296 * deferred bio dirtying paths.
1300 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1302 void bio_set_pages_dirty(struct bio *bio)
1304 struct bio_vec *bvec = bio->bi_io_vec;
1307 for (i = 0; i < bio->bi_vcnt; i++) {
1308 struct page *page = bvec[i].bv_page;
1310 if (page && !PageCompound(page))
1311 set_page_dirty_lock(page);
1315 static void bio_release_pages(struct bio *bio)
1317 struct bio_vec *bvec = bio->bi_io_vec;
1320 for (i = 0; i < bio->bi_vcnt; i++) {
1321 struct page *page = bvec[i].bv_page;
1329 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1330 * If they are, then fine. If, however, some pages are clean then they must
1331 * have been written out during the direct-IO read. So we take another ref on
1332 * the BIO and the offending pages and re-dirty the pages in process context.
1334 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1335 * here on. It will run one page_cache_release() against each page and will
1336 * run one bio_put() against the BIO.
1339 static void bio_dirty_fn(struct work_struct *work);
1341 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1342 static DEFINE_SPINLOCK(bio_dirty_lock);
1343 static struct bio *bio_dirty_list;
1346 * This runs in process context
1348 static void bio_dirty_fn(struct work_struct *work)
1350 unsigned long flags;
1353 spin_lock_irqsave(&bio_dirty_lock, flags);
1354 bio = bio_dirty_list;
1355 bio_dirty_list = NULL;
1356 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1359 struct bio *next = bio->bi_private;
1361 bio_set_pages_dirty(bio);
1362 bio_release_pages(bio);
1368 void bio_check_pages_dirty(struct bio *bio)
1370 struct bio_vec *bvec = bio->bi_io_vec;
1371 int nr_clean_pages = 0;
1374 for (i = 0; i < bio->bi_vcnt; i++) {
1375 struct page *page = bvec[i].bv_page;
1377 if (PageDirty(page) || PageCompound(page)) {
1378 page_cache_release(page);
1379 bvec[i].bv_page = NULL;
1385 if (nr_clean_pages) {
1386 unsigned long flags;
1388 spin_lock_irqsave(&bio_dirty_lock, flags);
1389 bio->bi_private = bio_dirty_list;
1390 bio_dirty_list = bio;
1391 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1392 schedule_work(&bio_dirty_work);
1398 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1399 void bio_flush_dcache_pages(struct bio *bi)
1402 struct bio_vec *bvec;
1404 bio_for_each_segment(bvec, bi, i)
1405 flush_dcache_page(bvec->bv_page);
1407 EXPORT_SYMBOL(bio_flush_dcache_pages);
1411 * bio_endio - end I/O on a bio
1413 * @error: error, if any
1416 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1417 * preferred way to end I/O on a bio, it takes care of clearing
1418 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1419 * established -Exxxx (-EIO, for instance) error values in case
1420 * something went wrong. No one should call bi_end_io() directly on a
1421 * bio unless they own it and thus know that it has an end_io
1424 void bio_endio(struct bio *bio, int error)
1427 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1428 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1431 trace_block_bio_complete(bio, error);
1434 bio->bi_end_io(bio, error);
1436 EXPORT_SYMBOL(bio_endio);
1438 void bio_pair_release(struct bio_pair *bp)
1440 if (atomic_dec_and_test(&bp->cnt)) {
1441 struct bio *master = bp->bio1.bi_private;
1443 bio_endio(master, bp->error);
1444 mempool_free(bp, bp->bio2.bi_private);
1447 EXPORT_SYMBOL(bio_pair_release);
1449 static void bio_pair_end_1(struct bio *bi, int err)
1451 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1456 bio_pair_release(bp);
1459 static void bio_pair_end_2(struct bio *bi, int err)
1461 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1466 bio_pair_release(bp);
1470 * split a bio - only worry about a bio with a single page in its iovec
1472 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1474 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1479 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1480 bi->bi_sector + first_sectors);
1482 BUG_ON(bi->bi_vcnt != 1 && bi->bi_vcnt != 0);
1483 BUG_ON(bi->bi_idx != 0);
1484 atomic_set(&bp->cnt, 3);
1488 bp->bio2.bi_sector += first_sectors;
1489 bp->bio2.bi_size -= first_sectors << 9;
1490 bp->bio1.bi_size = first_sectors << 9;
1492 if (bi->bi_vcnt != 0) {
1493 bp->bv1 = bi->bi_io_vec[0];
1494 bp->bv2 = bi->bi_io_vec[0];
1496 if (bio_is_rw(bi)) {
1497 bp->bv2.bv_offset += first_sectors << 9;
1498 bp->bv2.bv_len -= first_sectors << 9;
1499 bp->bv1.bv_len = first_sectors << 9;
1502 bp->bio1.bi_io_vec = &bp->bv1;
1503 bp->bio2.bi_io_vec = &bp->bv2;
1505 bp->bio1.bi_max_vecs = 1;
1506 bp->bio2.bi_max_vecs = 1;
1509 bp->bio1.bi_end_io = bio_pair_end_1;
1510 bp->bio2.bi_end_io = bio_pair_end_2;
1512 bp->bio1.bi_private = bi;
1513 bp->bio2.bi_private = bio_split_pool;
1515 if (bio_integrity(bi))
1516 bio_integrity_split(bi, bp, first_sectors);
1520 EXPORT_SYMBOL(bio_split);
1523 * bio_sector_offset - Find hardware sector offset in bio
1524 * @bio: bio to inspect
1525 * @index: bio_vec index
1526 * @offset: offset in bv_page
1528 * Return the number of hardware sectors between beginning of bio
1529 * and an end point indicated by a bio_vec index and an offset
1530 * within that vector's page.
1532 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1533 unsigned int offset)
1535 unsigned int sector_sz;
1540 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1543 if (index >= bio->bi_idx)
1544 index = bio->bi_vcnt - 1;
1546 __bio_for_each_segment(bv, bio, i, 0) {
1548 if (offset > bv->bv_offset)
1549 sectors += (offset - bv->bv_offset) / sector_sz;
1553 sectors += bv->bv_len / sector_sz;
1558 EXPORT_SYMBOL(bio_sector_offset);
1561 * create memory pools for biovec's in a bio_set.
1562 * use the global biovec slabs created for general use.
1564 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1566 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1568 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1575 static void biovec_free_pools(struct bio_set *bs)
1577 mempool_destroy(bs->bvec_pool);
1580 void bioset_free(struct bio_set *bs)
1583 mempool_destroy(bs->bio_pool);
1585 bioset_integrity_free(bs);
1586 biovec_free_pools(bs);
1591 EXPORT_SYMBOL(bioset_free);
1594 * bioset_create - Create a bio_set
1595 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1596 * @front_pad: Number of bytes to allocate in front of the returned bio
1599 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1600 * to ask for a number of bytes to be allocated in front of the bio.
1601 * Front pad allocation is useful for embedding the bio inside
1602 * another structure, to avoid allocating extra data to go with the bio.
1603 * Note that the bio must be embedded at the END of that structure always,
1604 * or things will break badly.
1606 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1608 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1611 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1615 bs->front_pad = front_pad;
1617 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1618 if (!bs->bio_slab) {
1623 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1627 if (!biovec_create_pools(bs, pool_size))
1634 EXPORT_SYMBOL(bioset_create);
1636 #ifdef CONFIG_BLK_CGROUP
1638 * bio_associate_current - associate a bio with %current
1641 * Associate @bio with %current if it hasn't been associated yet. Block
1642 * layer will treat @bio as if it were issued by %current no matter which
1643 * task actually issues it.
1645 * This function takes an extra reference of @task's io_context and blkcg
1646 * which will be put when @bio is released. The caller must own @bio,
1647 * ensure %current->io_context exists, and is responsible for synchronizing
1648 * calls to this function.
1650 int bio_associate_current(struct bio *bio)
1652 struct io_context *ioc;
1653 struct cgroup_subsys_state *css;
1658 ioc = current->io_context;
1662 /* acquire active ref on @ioc and associate */
1663 get_io_context_active(ioc);
1666 /* associate blkcg if exists */
1668 css = task_subsys_state(current, blkio_subsys_id);
1669 if (css && css_tryget(css))
1677 * bio_disassociate_task - undo bio_associate_current()
1680 void bio_disassociate_task(struct bio *bio)
1683 put_io_context(bio->bi_ioc);
1687 css_put(bio->bi_css);
1692 #endif /* CONFIG_BLK_CGROUP */
1694 static void __init biovec_init_slabs(void)
1698 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1700 struct biovec_slab *bvs = bvec_slabs + i;
1702 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1707 size = bvs->nr_vecs * sizeof(struct bio_vec);
1708 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1709 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1713 static int __init init_bio(void)
1717 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1719 panic("bio: can't allocate bios\n");
1721 bio_integrity_init();
1722 biovec_init_slabs();
1724 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1726 panic("bio: can't allocate bios\n");
1728 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1729 panic("bio: can't create integrity pool\n");
1731 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1732 sizeof(struct bio_pair));
1733 if (!bio_split_pool)
1734 panic("bio: can't create split pool\n");
1738 subsys_initcall(init_bio);