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 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) {
101 new_bio_slabs = krealloc(bio_slabs,
102 bio_slab_max * sizeof(struct bio_slab),
106 bio_slabs = new_bio_slabs;
109 entry = bio_slab_nr++;
111 bslab = &bio_slabs[entry];
113 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
114 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
118 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
121 bslab->slab_size = sz;
123 mutex_unlock(&bio_slab_lock);
127 static void bio_put_slab(struct bio_set *bs)
129 struct bio_slab *bslab = NULL;
132 mutex_lock(&bio_slab_lock);
134 for (i = 0; i < bio_slab_nr; i++) {
135 if (bs->bio_slab == bio_slabs[i].slab) {
136 bslab = &bio_slabs[i];
141 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
144 WARN_ON(!bslab->slab_ref);
146 if (--bslab->slab_ref)
149 kmem_cache_destroy(bslab->slab);
153 mutex_unlock(&bio_slab_lock);
156 unsigned int bvec_nr_vecs(unsigned short idx)
158 return bvec_slabs[idx].nr_vecs;
161 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
163 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
165 if (idx == BIOVEC_MAX_IDX)
166 mempool_free(bv, bs->bvec_pool);
168 struct biovec_slab *bvs = bvec_slabs + idx;
170 kmem_cache_free(bvs->slab, bv);
174 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
180 * see comment near bvec_array define!
198 case 129 ... BIO_MAX_PAGES:
206 * idx now points to the pool we want to allocate from. only the
207 * 1-vec entry pool is mempool backed.
209 if (*idx == BIOVEC_MAX_IDX) {
211 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
213 struct biovec_slab *bvs = bvec_slabs + *idx;
214 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
217 * Make this allocation restricted and don't dump info on
218 * allocation failures, since we'll fallback to the mempool
219 * in case of failure.
221 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
224 * Try a slab allocation. If this fails and __GFP_WAIT
225 * is set, retry with the 1-entry mempool
227 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
228 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
229 *idx = BIOVEC_MAX_IDX;
237 static void __bio_free(struct bio *bio)
239 bio_disassociate_task(bio);
241 if (bio_integrity(bio))
242 bio_integrity_free(bio);
245 static void bio_free(struct bio *bio)
247 struct bio_set *bs = bio->bi_pool;
253 if (bio_has_allocated_vec(bio))
254 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
257 * If we have front padding, adjust the bio pointer before freeing
262 mempool_free(p, bs->bio_pool);
264 /* Bio was allocated by bio_kmalloc() */
269 void bio_init(struct bio *bio)
271 memset(bio, 0, sizeof(*bio));
272 bio->bi_flags = 1 << BIO_UPTODATE;
273 atomic_set(&bio->bi_cnt, 1);
275 EXPORT_SYMBOL(bio_init);
278 * bio_reset - reinitialize a bio
282 * After calling bio_reset(), @bio will be in the same state as a freshly
283 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
284 * preserved are the ones that are initialized by bio_alloc_bioset(). See
285 * comment in struct bio.
287 void bio_reset(struct bio *bio)
289 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
293 memset(bio, 0, BIO_RESET_BYTES);
294 bio->bi_flags = flags|(1 << BIO_UPTODATE);
296 EXPORT_SYMBOL(bio_reset);
299 * bio_alloc_bioset - allocate a bio for I/O
300 * @gfp_mask: the GFP_ mask given to the slab allocator
301 * @nr_iovecs: number of iovecs to pre-allocate
302 * @bs: the bio_set to allocate from.
305 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
306 * backed by the @bs's mempool.
308 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
309 * able to allocate a bio. This is due to the mempool guarantees. To make this
310 * work, callers must never allocate more than 1 bio at a time from this pool.
311 * Callers that need to allocate more than 1 bio must always submit the
312 * previously allocated bio for IO before attempting to allocate a new one.
313 * Failure to do so can cause deadlocks under memory pressure.
316 * Pointer to new bio on success, NULL on failure.
318 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
321 unsigned inline_vecs;
322 unsigned long idx = BIO_POOL_NONE;
323 struct bio_vec *bvl = NULL;
328 if (nr_iovecs > UIO_MAXIOV)
331 p = kmalloc(sizeof(struct bio) +
332 nr_iovecs * sizeof(struct bio_vec),
335 inline_vecs = nr_iovecs;
337 p = mempool_alloc(bs->bio_pool, gfp_mask);
338 front_pad = bs->front_pad;
339 inline_vecs = BIO_INLINE_VECS;
348 if (nr_iovecs > inline_vecs) {
349 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
352 } else if (nr_iovecs) {
353 bvl = bio->bi_inline_vecs;
357 bio->bi_flags |= idx << BIO_POOL_OFFSET;
358 bio->bi_max_vecs = nr_iovecs;
359 bio->bi_io_vec = bvl;
363 mempool_free(p, bs->bio_pool);
366 EXPORT_SYMBOL(bio_alloc_bioset);
368 void zero_fill_bio(struct bio *bio)
374 bio_for_each_segment(bv, bio, i) {
375 char *data = bvec_kmap_irq(bv, &flags);
376 memset(data, 0, bv->bv_len);
377 flush_dcache_page(bv->bv_page);
378 bvec_kunmap_irq(data, &flags);
381 EXPORT_SYMBOL(zero_fill_bio);
384 * bio_put - release a reference to a bio
385 * @bio: bio to release reference to
388 * Put a reference to a &struct bio, either one you have gotten with
389 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
391 void bio_put(struct bio *bio)
393 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
398 if (atomic_dec_and_test(&bio->bi_cnt))
401 EXPORT_SYMBOL(bio_put);
403 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
405 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
406 blk_recount_segments(q, bio);
408 return bio->bi_phys_segments;
410 EXPORT_SYMBOL(bio_phys_segments);
413 * __bio_clone - clone a bio
414 * @bio: destination bio
415 * @bio_src: bio to clone
417 * Clone a &bio. Caller will own the returned bio, but not
418 * the actual data it points to. Reference count of returned
421 void __bio_clone(struct bio *bio, struct bio *bio_src)
423 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
424 bio_src->bi_max_vecs * sizeof(struct bio_vec));
427 * most users will be overriding ->bi_bdev with a new target,
428 * so we don't set nor calculate new physical/hw segment counts here
430 bio->bi_sector = bio_src->bi_sector;
431 bio->bi_bdev = bio_src->bi_bdev;
432 bio->bi_flags |= 1 << BIO_CLONED;
433 bio->bi_rw = bio_src->bi_rw;
434 bio->bi_vcnt = bio_src->bi_vcnt;
435 bio->bi_size = bio_src->bi_size;
436 bio->bi_idx = bio_src->bi_idx;
438 EXPORT_SYMBOL(__bio_clone);
441 * bio_clone_bioset - clone a bio
443 * @gfp_mask: allocation priority
444 * @bs: bio_set to allocate from
446 * Like __bio_clone, only also allocates the returned bio
448 struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask,
453 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs);
459 if (bio_integrity(bio)) {
462 ret = bio_integrity_clone(b, bio, gfp_mask);
472 EXPORT_SYMBOL(bio_clone_bioset);
475 * bio_get_nr_vecs - return approx number of vecs
478 * Return the approximate number of pages we can send to this target.
479 * There's no guarantee that you will be able to fit this number of pages
480 * into a bio, it does not account for dynamic restrictions that vary
483 int bio_get_nr_vecs(struct block_device *bdev)
485 struct request_queue *q = bdev_get_queue(bdev);
488 nr_pages = min_t(unsigned,
489 queue_max_segments(q),
490 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
492 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
495 EXPORT_SYMBOL(bio_get_nr_vecs);
497 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
498 *page, unsigned int len, unsigned int offset,
499 unsigned short max_sectors)
501 int retried_segments = 0;
502 struct bio_vec *bvec;
505 * cloned bio must not modify vec list
507 if (unlikely(bio_flagged(bio, BIO_CLONED)))
510 if (((bio->bi_size + len) >> 9) > max_sectors)
514 * For filesystems with a blocksize smaller than the pagesize
515 * we will often be called with the same page as last time and
516 * a consecutive offset. Optimize this special case.
518 if (bio->bi_vcnt > 0) {
519 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
521 if (page == prev->bv_page &&
522 offset == prev->bv_offset + prev->bv_len) {
523 unsigned int prev_bv_len = prev->bv_len;
526 if (q->merge_bvec_fn) {
527 struct bvec_merge_data bvm = {
528 /* prev_bvec is already charged in
529 bi_size, discharge it in order to
530 simulate merging updated prev_bvec
532 .bi_bdev = bio->bi_bdev,
533 .bi_sector = bio->bi_sector,
534 .bi_size = bio->bi_size - prev_bv_len,
538 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
548 if (bio->bi_vcnt >= bio->bi_max_vecs)
552 * we might lose a segment or two here, but rather that than
553 * make this too complex.
556 while (bio->bi_phys_segments >= queue_max_segments(q)) {
558 if (retried_segments)
561 retried_segments = 1;
562 blk_recount_segments(q, bio);
566 * setup the new entry, we might clear it again later if we
567 * cannot add the page
569 bvec = &bio->bi_io_vec[bio->bi_vcnt];
570 bvec->bv_page = page;
572 bvec->bv_offset = offset;
575 * if queue has other restrictions (eg varying max sector size
576 * depending on offset), it can specify a merge_bvec_fn in the
577 * queue to get further control
579 if (q->merge_bvec_fn) {
580 struct bvec_merge_data bvm = {
581 .bi_bdev = bio->bi_bdev,
582 .bi_sector = bio->bi_sector,
583 .bi_size = bio->bi_size,
588 * merge_bvec_fn() returns number of bytes it can accept
591 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
592 bvec->bv_page = NULL;
599 /* If we may be able to merge these biovecs, force a recount */
600 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
601 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
604 bio->bi_phys_segments++;
611 * bio_add_pc_page - attempt to add page to bio
612 * @q: the target queue
613 * @bio: destination bio
615 * @len: vec entry length
616 * @offset: vec entry offset
618 * Attempt to add a page to the bio_vec maplist. This can fail for a
619 * number of reasons, such as the bio being full or target block device
620 * limitations. The target block device must allow bio's up to PAGE_SIZE,
621 * so it is always possible to add a single page to an empty bio.
623 * This should only be used by REQ_PC bios.
625 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
626 unsigned int len, unsigned int offset)
628 return __bio_add_page(q, bio, page, len, offset,
629 queue_max_hw_sectors(q));
631 EXPORT_SYMBOL(bio_add_pc_page);
634 * bio_add_page - attempt to add page to bio
635 * @bio: destination bio
637 * @len: vec entry length
638 * @offset: vec entry offset
640 * Attempt to add a page to the bio_vec maplist. This can fail for a
641 * number of reasons, such as the bio being full or target block device
642 * limitations. The target block device must allow bio's up to PAGE_SIZE,
643 * so it is always possible to add a single page to an empty bio.
645 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
648 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
649 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
651 EXPORT_SYMBOL(bio_add_page);
653 struct bio_map_data {
654 struct bio_vec *iovecs;
655 struct sg_iovec *sgvecs;
660 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
661 struct sg_iovec *iov, int iov_count,
664 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
665 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
666 bmd->nr_sgvecs = iov_count;
667 bmd->is_our_pages = is_our_pages;
668 bio->bi_private = bmd;
671 static void bio_free_map_data(struct bio_map_data *bmd)
678 static struct bio_map_data *bio_alloc_map_data(int nr_segs,
679 unsigned int iov_count,
682 struct bio_map_data *bmd;
684 if (iov_count > UIO_MAXIOV)
687 bmd = kmalloc(sizeof(*bmd), gfp_mask);
691 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
697 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
706 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
707 struct sg_iovec *iov, int iov_count,
708 int to_user, int from_user, int do_free_page)
711 struct bio_vec *bvec;
713 unsigned int iov_off = 0;
715 __bio_for_each_segment(bvec, bio, i, 0) {
716 char *bv_addr = page_address(bvec->bv_page);
717 unsigned int bv_len = iovecs[i].bv_len;
719 while (bv_len && iov_idx < iov_count) {
721 char __user *iov_addr;
723 bytes = min_t(unsigned int,
724 iov[iov_idx].iov_len - iov_off, bv_len);
725 iov_addr = iov[iov_idx].iov_base + iov_off;
729 ret = copy_to_user(iov_addr, bv_addr,
733 ret = copy_from_user(bv_addr, iov_addr,
745 if (iov[iov_idx].iov_len == iov_off) {
752 __free_page(bvec->bv_page);
759 * bio_uncopy_user - finish previously mapped bio
760 * @bio: bio being terminated
762 * Free pages allocated from bio_copy_user() and write back data
763 * to user space in case of a read.
765 int bio_uncopy_user(struct bio *bio)
767 struct bio_map_data *bmd = bio->bi_private;
770 if (!bio_flagged(bio, BIO_NULL_MAPPED))
771 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
772 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
773 0, bmd->is_our_pages);
774 bio_free_map_data(bmd);
778 EXPORT_SYMBOL(bio_uncopy_user);
781 * bio_copy_user_iov - copy user data to bio
782 * @q: destination block queue
783 * @map_data: pointer to the rq_map_data holding pages (if necessary)
785 * @iov_count: number of elements in the iovec
786 * @write_to_vm: bool indicating writing to pages or not
787 * @gfp_mask: memory allocation flags
789 * Prepares and returns a bio for indirect user io, bouncing data
790 * to/from kernel pages as necessary. Must be paired with
791 * call bio_uncopy_user() on io completion.
793 struct bio *bio_copy_user_iov(struct request_queue *q,
794 struct rq_map_data *map_data,
795 struct sg_iovec *iov, int iov_count,
796 int write_to_vm, gfp_t gfp_mask)
798 struct bio_map_data *bmd;
799 struct bio_vec *bvec;
804 unsigned int len = 0;
805 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
807 for (i = 0; i < iov_count; i++) {
812 uaddr = (unsigned long)iov[i].iov_base;
813 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
814 start = uaddr >> PAGE_SHIFT;
820 return ERR_PTR(-EINVAL);
822 nr_pages += end - start;
823 len += iov[i].iov_len;
829 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
831 return ERR_PTR(-ENOMEM);
834 bio = bio_kmalloc(gfp_mask, nr_pages);
839 bio->bi_rw |= REQ_WRITE;
844 nr_pages = 1 << map_data->page_order;
845 i = map_data->offset / PAGE_SIZE;
848 unsigned int bytes = PAGE_SIZE;
856 if (i == map_data->nr_entries * nr_pages) {
861 page = map_data->pages[i / nr_pages];
862 page += (i % nr_pages);
866 page = alloc_page(q->bounce_gfp | gfp_mask);
873 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
886 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
887 (map_data && map_data->from_user)) {
888 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
893 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
897 bio_for_each_segment(bvec, bio, i)
898 __free_page(bvec->bv_page);
902 bio_free_map_data(bmd);
907 * bio_copy_user - copy user data to bio
908 * @q: destination block queue
909 * @map_data: pointer to the rq_map_data holding pages (if necessary)
910 * @uaddr: start of user address
911 * @len: length in bytes
912 * @write_to_vm: bool indicating writing to pages or not
913 * @gfp_mask: memory allocation flags
915 * Prepares and returns a bio for indirect user io, bouncing data
916 * to/from kernel pages as necessary. Must be paired with
917 * call bio_uncopy_user() on io completion.
919 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
920 unsigned long uaddr, unsigned int len,
921 int write_to_vm, gfp_t gfp_mask)
925 iov.iov_base = (void __user *)uaddr;
928 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
930 EXPORT_SYMBOL(bio_copy_user);
932 static struct bio *__bio_map_user_iov(struct request_queue *q,
933 struct block_device *bdev,
934 struct sg_iovec *iov, int iov_count,
935 int write_to_vm, gfp_t gfp_mask)
944 for (i = 0; i < iov_count; i++) {
945 unsigned long uaddr = (unsigned long)iov[i].iov_base;
946 unsigned long len = iov[i].iov_len;
947 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
948 unsigned long start = uaddr >> PAGE_SHIFT;
954 return ERR_PTR(-EINVAL);
956 nr_pages += end - start;
958 * buffer must be aligned to at least hardsector size for now
960 if (uaddr & queue_dma_alignment(q))
961 return ERR_PTR(-EINVAL);
965 return ERR_PTR(-EINVAL);
967 bio = bio_kmalloc(gfp_mask, nr_pages);
969 return ERR_PTR(-ENOMEM);
972 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
976 for (i = 0; i < iov_count; i++) {
977 unsigned long uaddr = (unsigned long)iov[i].iov_base;
978 unsigned long len = iov[i].iov_len;
979 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
980 unsigned long start = uaddr >> PAGE_SHIFT;
981 const int local_nr_pages = end - start;
982 const int page_limit = cur_page + local_nr_pages;
984 ret = get_user_pages_fast(uaddr, local_nr_pages,
985 write_to_vm, &pages[cur_page]);
986 if (ret < local_nr_pages) {
991 offset = uaddr & ~PAGE_MASK;
992 for (j = cur_page; j < page_limit; j++) {
993 unsigned int bytes = PAGE_SIZE - offset;
1004 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1014 * release the pages we didn't map into the bio, if any
1016 while (j < page_limit)
1017 page_cache_release(pages[j++]);
1023 * set data direction, and check if mapped pages need bouncing
1026 bio->bi_rw |= REQ_WRITE;
1028 bio->bi_bdev = bdev;
1029 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1033 for (i = 0; i < nr_pages; i++) {
1036 page_cache_release(pages[i]);
1041 return ERR_PTR(ret);
1045 * bio_map_user - map user address into bio
1046 * @q: the struct request_queue for the bio
1047 * @bdev: destination block device
1048 * @uaddr: start of user address
1049 * @len: length in bytes
1050 * @write_to_vm: bool indicating writing to pages or not
1051 * @gfp_mask: memory allocation flags
1053 * Map the user space address into a bio suitable for io to a block
1054 * device. Returns an error pointer in case of error.
1056 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1057 unsigned long uaddr, unsigned int len, int write_to_vm,
1060 struct sg_iovec iov;
1062 iov.iov_base = (void __user *)uaddr;
1065 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1067 EXPORT_SYMBOL(bio_map_user);
1070 * bio_map_user_iov - map user sg_iovec table into bio
1071 * @q: the struct request_queue for the bio
1072 * @bdev: destination block device
1074 * @iov_count: number of elements in the iovec
1075 * @write_to_vm: bool indicating writing to pages or not
1076 * @gfp_mask: memory allocation flags
1078 * Map the user space address into a bio suitable for io to a block
1079 * device. Returns an error pointer in case of error.
1081 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1082 struct sg_iovec *iov, int iov_count,
1083 int write_to_vm, gfp_t gfp_mask)
1087 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1093 * subtle -- if __bio_map_user() ended up bouncing a bio,
1094 * it would normally disappear when its bi_end_io is run.
1095 * however, we need it for the unmap, so grab an extra
1103 static void __bio_unmap_user(struct bio *bio)
1105 struct bio_vec *bvec;
1109 * make sure we dirty pages we wrote to
1111 __bio_for_each_segment(bvec, bio, i, 0) {
1112 if (bio_data_dir(bio) == READ)
1113 set_page_dirty_lock(bvec->bv_page);
1115 page_cache_release(bvec->bv_page);
1122 * bio_unmap_user - unmap a bio
1123 * @bio: the bio being unmapped
1125 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1126 * a process context.
1128 * bio_unmap_user() may sleep.
1130 void bio_unmap_user(struct bio *bio)
1132 __bio_unmap_user(bio);
1135 EXPORT_SYMBOL(bio_unmap_user);
1137 static void bio_map_kern_endio(struct bio *bio, int err)
1142 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1143 unsigned int len, gfp_t gfp_mask)
1145 unsigned long kaddr = (unsigned long)data;
1146 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1147 unsigned long start = kaddr >> PAGE_SHIFT;
1148 const int nr_pages = end - start;
1152 bio = bio_kmalloc(gfp_mask, nr_pages);
1154 return ERR_PTR(-ENOMEM);
1156 offset = offset_in_page(kaddr);
1157 for (i = 0; i < nr_pages; i++) {
1158 unsigned int bytes = PAGE_SIZE - offset;
1166 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1175 bio->bi_end_io = bio_map_kern_endio;
1180 * bio_map_kern - map kernel address into bio
1181 * @q: the struct request_queue for the bio
1182 * @data: pointer to buffer to map
1183 * @len: length in bytes
1184 * @gfp_mask: allocation flags for bio allocation
1186 * Map the kernel address into a bio suitable for io to a block
1187 * device. Returns an error pointer in case of error.
1189 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1194 bio = __bio_map_kern(q, data, len, gfp_mask);
1198 if (bio->bi_size == len)
1202 * Don't support partial mappings.
1205 return ERR_PTR(-EINVAL);
1207 EXPORT_SYMBOL(bio_map_kern);
1209 static void bio_copy_kern_endio(struct bio *bio, int err)
1211 struct bio_vec *bvec;
1212 const int read = bio_data_dir(bio) == READ;
1213 struct bio_map_data *bmd = bio->bi_private;
1215 char *p = bmd->sgvecs[0].iov_base;
1217 __bio_for_each_segment(bvec, bio, i, 0) {
1218 char *addr = page_address(bvec->bv_page);
1219 int len = bmd->iovecs[i].bv_len;
1222 memcpy(p, addr, len);
1224 __free_page(bvec->bv_page);
1228 bio_free_map_data(bmd);
1233 * bio_copy_kern - copy kernel address into bio
1234 * @q: the struct request_queue for the bio
1235 * @data: pointer to buffer to copy
1236 * @len: length in bytes
1237 * @gfp_mask: allocation flags for bio and page allocation
1238 * @reading: data direction is READ
1240 * copy the kernel address into a bio suitable for io to a block
1241 * device. Returns an error pointer in case of error.
1243 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1244 gfp_t gfp_mask, int reading)
1247 struct bio_vec *bvec;
1250 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1257 bio_for_each_segment(bvec, bio, i) {
1258 char *addr = page_address(bvec->bv_page);
1260 memcpy(addr, p, bvec->bv_len);
1265 bio->bi_end_io = bio_copy_kern_endio;
1269 EXPORT_SYMBOL(bio_copy_kern);
1272 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1273 * for performing direct-IO in BIOs.
1275 * The problem is that we cannot run set_page_dirty() from interrupt context
1276 * because the required locks are not interrupt-safe. So what we can do is to
1277 * mark the pages dirty _before_ performing IO. And in interrupt context,
1278 * check that the pages are still dirty. If so, fine. If not, redirty them
1279 * in process context.
1281 * We special-case compound pages here: normally this means reads into hugetlb
1282 * pages. The logic in here doesn't really work right for compound pages
1283 * because the VM does not uniformly chase down the head page in all cases.
1284 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1285 * handle them at all. So we skip compound pages here at an early stage.
1287 * Note that this code is very hard to test under normal circumstances because
1288 * direct-io pins the pages with get_user_pages(). This makes
1289 * is_page_cache_freeable return false, and the VM will not clean the pages.
1290 * But other code (eg, flusher threads) could clean the pages if they are mapped
1293 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1294 * deferred bio dirtying paths.
1298 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1300 void bio_set_pages_dirty(struct bio *bio)
1302 struct bio_vec *bvec = bio->bi_io_vec;
1305 for (i = 0; i < bio->bi_vcnt; i++) {
1306 struct page *page = bvec[i].bv_page;
1308 if (page && !PageCompound(page))
1309 set_page_dirty_lock(page);
1313 static void bio_release_pages(struct bio *bio)
1315 struct bio_vec *bvec = bio->bi_io_vec;
1318 for (i = 0; i < bio->bi_vcnt; i++) {
1319 struct page *page = bvec[i].bv_page;
1327 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1328 * If they are, then fine. If, however, some pages are clean then they must
1329 * have been written out during the direct-IO read. So we take another ref on
1330 * the BIO and the offending pages and re-dirty the pages in process context.
1332 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1333 * here on. It will run one page_cache_release() against each page and will
1334 * run one bio_put() against the BIO.
1337 static void bio_dirty_fn(struct work_struct *work);
1339 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1340 static DEFINE_SPINLOCK(bio_dirty_lock);
1341 static struct bio *bio_dirty_list;
1344 * This runs in process context
1346 static void bio_dirty_fn(struct work_struct *work)
1348 unsigned long flags;
1351 spin_lock_irqsave(&bio_dirty_lock, flags);
1352 bio = bio_dirty_list;
1353 bio_dirty_list = NULL;
1354 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1357 struct bio *next = bio->bi_private;
1359 bio_set_pages_dirty(bio);
1360 bio_release_pages(bio);
1366 void bio_check_pages_dirty(struct bio *bio)
1368 struct bio_vec *bvec = bio->bi_io_vec;
1369 int nr_clean_pages = 0;
1372 for (i = 0; i < bio->bi_vcnt; i++) {
1373 struct page *page = bvec[i].bv_page;
1375 if (PageDirty(page) || PageCompound(page)) {
1376 page_cache_release(page);
1377 bvec[i].bv_page = NULL;
1383 if (nr_clean_pages) {
1384 unsigned long flags;
1386 spin_lock_irqsave(&bio_dirty_lock, flags);
1387 bio->bi_private = bio_dirty_list;
1388 bio_dirty_list = bio;
1389 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1390 schedule_work(&bio_dirty_work);
1396 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1397 void bio_flush_dcache_pages(struct bio *bi)
1400 struct bio_vec *bvec;
1402 bio_for_each_segment(bvec, bi, i)
1403 flush_dcache_page(bvec->bv_page);
1405 EXPORT_SYMBOL(bio_flush_dcache_pages);
1409 * bio_endio - end I/O on a bio
1411 * @error: error, if any
1414 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1415 * preferred way to end I/O on a bio, it takes care of clearing
1416 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1417 * established -Exxxx (-EIO, for instance) error values in case
1418 * something went wrong. No one should call bi_end_io() directly on a
1419 * bio unless they own it and thus know that it has an end_io
1422 void bio_endio(struct bio *bio, int error)
1425 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1426 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1430 bio->bi_end_io(bio, error);
1432 EXPORT_SYMBOL(bio_endio);
1434 void bio_pair_release(struct bio_pair *bp)
1436 if (atomic_dec_and_test(&bp->cnt)) {
1437 struct bio *master = bp->bio1.bi_private;
1439 bio_endio(master, bp->error);
1440 mempool_free(bp, bp->bio2.bi_private);
1443 EXPORT_SYMBOL(bio_pair_release);
1445 static void bio_pair_end_1(struct bio *bi, int err)
1447 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1452 bio_pair_release(bp);
1455 static void bio_pair_end_2(struct bio *bi, int err)
1457 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1462 bio_pair_release(bp);
1466 * split a bio - only worry about a bio with a single page in its iovec
1468 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1470 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1475 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1476 bi->bi_sector + first_sectors);
1478 BUG_ON(bi->bi_vcnt != 1 && bi->bi_vcnt != 0);
1479 BUG_ON(bi->bi_idx != 0);
1480 atomic_set(&bp->cnt, 3);
1484 bp->bio2.bi_sector += first_sectors;
1485 bp->bio2.bi_size -= first_sectors << 9;
1486 bp->bio1.bi_size = first_sectors << 9;
1488 if (bi->bi_vcnt != 0) {
1489 bp->bv1 = bi->bi_io_vec[0];
1490 bp->bv2 = bi->bi_io_vec[0];
1492 if (bio_is_rw(bi)) {
1493 bp->bv2.bv_offset += first_sectors << 9;
1494 bp->bv2.bv_len -= first_sectors << 9;
1495 bp->bv1.bv_len = first_sectors << 9;
1498 bp->bio1.bi_io_vec = &bp->bv1;
1499 bp->bio2.bi_io_vec = &bp->bv2;
1501 bp->bio1.bi_max_vecs = 1;
1502 bp->bio2.bi_max_vecs = 1;
1505 bp->bio1.bi_end_io = bio_pair_end_1;
1506 bp->bio2.bi_end_io = bio_pair_end_2;
1508 bp->bio1.bi_private = bi;
1509 bp->bio2.bi_private = bio_split_pool;
1511 if (bio_integrity(bi))
1512 bio_integrity_split(bi, bp, first_sectors);
1516 EXPORT_SYMBOL(bio_split);
1519 * bio_sector_offset - Find hardware sector offset in bio
1520 * @bio: bio to inspect
1521 * @index: bio_vec index
1522 * @offset: offset in bv_page
1524 * Return the number of hardware sectors between beginning of bio
1525 * and an end point indicated by a bio_vec index and an offset
1526 * within that vector's page.
1528 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1529 unsigned int offset)
1531 unsigned int sector_sz;
1536 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1539 if (index >= bio->bi_idx)
1540 index = bio->bi_vcnt - 1;
1542 __bio_for_each_segment(bv, bio, i, 0) {
1544 if (offset > bv->bv_offset)
1545 sectors += (offset - bv->bv_offset) / sector_sz;
1549 sectors += bv->bv_len / sector_sz;
1554 EXPORT_SYMBOL(bio_sector_offset);
1557 * create memory pools for biovec's in a bio_set.
1558 * use the global biovec slabs created for general use.
1560 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1562 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1564 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1571 static void biovec_free_pools(struct bio_set *bs)
1573 mempool_destroy(bs->bvec_pool);
1576 void bioset_free(struct bio_set *bs)
1579 mempool_destroy(bs->bio_pool);
1581 bioset_integrity_free(bs);
1582 biovec_free_pools(bs);
1587 EXPORT_SYMBOL(bioset_free);
1590 * bioset_create - Create a bio_set
1591 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1592 * @front_pad: Number of bytes to allocate in front of the returned bio
1595 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1596 * to ask for a number of bytes to be allocated in front of the bio.
1597 * Front pad allocation is useful for embedding the bio inside
1598 * another structure, to avoid allocating extra data to go with the bio.
1599 * Note that the bio must be embedded at the END of that structure always,
1600 * or things will break badly.
1602 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1604 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1607 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1611 bs->front_pad = front_pad;
1613 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1614 if (!bs->bio_slab) {
1619 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1623 if (!biovec_create_pools(bs, pool_size))
1630 EXPORT_SYMBOL(bioset_create);
1632 #ifdef CONFIG_BLK_CGROUP
1634 * bio_associate_current - associate a bio with %current
1637 * Associate @bio with %current if it hasn't been associated yet. Block
1638 * layer will treat @bio as if it were issued by %current no matter which
1639 * task actually issues it.
1641 * This function takes an extra reference of @task's io_context and blkcg
1642 * which will be put when @bio is released. The caller must own @bio,
1643 * ensure %current->io_context exists, and is responsible for synchronizing
1644 * calls to this function.
1646 int bio_associate_current(struct bio *bio)
1648 struct io_context *ioc;
1649 struct cgroup_subsys_state *css;
1654 ioc = current->io_context;
1658 /* acquire active ref on @ioc and associate */
1659 get_io_context_active(ioc);
1662 /* associate blkcg if exists */
1664 css = task_subsys_state(current, blkio_subsys_id);
1665 if (css && css_tryget(css))
1673 * bio_disassociate_task - undo bio_associate_current()
1676 void bio_disassociate_task(struct bio *bio)
1679 put_io_context(bio->bi_ioc);
1683 css_put(bio->bi_css);
1688 #endif /* CONFIG_BLK_CGROUP */
1690 static void __init biovec_init_slabs(void)
1694 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1696 struct biovec_slab *bvs = bvec_slabs + i;
1698 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1703 size = bvs->nr_vecs * sizeof(struct bio_vec);
1704 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1705 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1709 static int __init init_bio(void)
1713 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1715 panic("bio: can't allocate bios\n");
1717 bio_integrity_init();
1718 biovec_init_slabs();
1720 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1722 panic("bio: can't allocate bios\n");
1724 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1725 panic("bio: can't create integrity pool\n");
1727 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1728 sizeof(struct bio_pair));
1729 if (!bio_split_pool)
1730 panic("bio: can't create split pool\n");
1734 subsys_initcall(init_bio);