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/slab.h>
23 #include <linux/init.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/mempool.h>
27 #include <linux/workqueue.h>
28 #include <linux/blktrace_api.h>
29 #include <trace/block.h>
30 #include <scsi/sg.h> /* for struct sg_iovec */
32 DEFINE_TRACE(block_split);
34 static struct kmem_cache *bio_slab __read_mostly;
36 static mempool_t *bio_split_pool __read_mostly;
39 * if you change this list, also change bvec_alloc or things will
40 * break badly! cannot be bigger than what you can fit into an
44 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
45 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
46 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
51 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
52 * IO code that does not need private memory pools.
54 struct bio_set *fs_bio_set;
56 unsigned int bvec_nr_vecs(unsigned short idx)
58 return bvec_slabs[idx].nr_vecs;
61 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
66 * If 'bs' is given, lookup the pool and do the mempool alloc.
67 * If not, this is a bio_kmalloc() allocation and just do a
68 * kzalloc() for the exact number of vecs right away.
72 * see comment near bvec_array define!
90 case 129 ... BIO_MAX_PAGES:
98 * idx now points to the pool we want to allocate from
100 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
103 bvec_nr_vecs(*idx) * sizeof(struct bio_vec));
105 bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask);
110 void bio_free(struct bio *bio, struct bio_set *bio_set)
112 if (bio->bi_io_vec) {
113 const int pool_idx = BIO_POOL_IDX(bio);
115 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
117 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
120 if (bio_integrity(bio))
121 bio_integrity_free(bio, bio_set);
123 mempool_free(bio, bio_set->bio_pool);
127 * default destructor for a bio allocated with bio_alloc_bioset()
129 static void bio_fs_destructor(struct bio *bio)
131 bio_free(bio, fs_bio_set);
134 static void bio_kmalloc_destructor(struct bio *bio)
136 kfree(bio->bi_io_vec);
140 void bio_init(struct bio *bio)
142 memset(bio, 0, sizeof(*bio));
143 bio->bi_flags = 1 << BIO_UPTODATE;
144 bio->bi_comp_cpu = -1;
145 atomic_set(&bio->bi_cnt, 1);
149 * bio_alloc_bioset - allocate a bio for I/O
150 * @gfp_mask: the GFP_ mask given to the slab allocator
151 * @nr_iovecs: number of iovecs to pre-allocate
152 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
155 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
156 * If %__GFP_WAIT is set then we will block on the internal pool waiting
157 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
158 * fall back to just using @kmalloc to allocate the required memory.
160 * allocate bio and iovecs from the memory pools specified by the
161 * bio_set structure, or @kmalloc if none given.
163 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
168 bio = mempool_alloc(bs->bio_pool, gfp_mask);
170 bio = kmalloc(sizeof(*bio), gfp_mask);
173 struct bio_vec *bvl = NULL;
176 if (likely(nr_iovecs)) {
177 unsigned long uninitialized_var(idx);
179 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
180 if (unlikely(!bvl)) {
182 mempool_free(bio, bs->bio_pool);
188 bio->bi_flags |= idx << BIO_POOL_OFFSET;
189 bio->bi_max_vecs = bvec_nr_vecs(idx);
191 bio->bi_io_vec = bvl;
197 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
199 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
202 bio->bi_destructor = bio_fs_destructor;
208 * Like bio_alloc(), but doesn't use a mempool backing. This means that
209 * it CAN fail, but while bio_alloc() can only be used for allocations
210 * that have a short (finite) life span, bio_kmalloc() should be used
211 * for more permanent bio allocations (like allocating some bio's for
212 * initalization or setup purposes).
214 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
216 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL);
219 bio->bi_destructor = bio_kmalloc_destructor;
224 void zero_fill_bio(struct bio *bio)
230 bio_for_each_segment(bv, bio, i) {
231 char *data = bvec_kmap_irq(bv, &flags);
232 memset(data, 0, bv->bv_len);
233 flush_dcache_page(bv->bv_page);
234 bvec_kunmap_irq(data, &flags);
237 EXPORT_SYMBOL(zero_fill_bio);
240 * bio_put - release a reference to a bio
241 * @bio: bio to release reference to
244 * Put a reference to a &struct bio, either one you have gotten with
245 * bio_alloc or bio_get. The last put of a bio will free it.
247 void bio_put(struct bio *bio)
249 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
254 if (atomic_dec_and_test(&bio->bi_cnt)) {
256 bio->bi_destructor(bio);
260 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
262 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
263 blk_recount_segments(q, bio);
265 return bio->bi_phys_segments;
269 * __bio_clone - clone a bio
270 * @bio: destination bio
271 * @bio_src: bio to clone
273 * Clone a &bio. Caller will own the returned bio, but not
274 * the actual data it points to. Reference count of returned
277 void __bio_clone(struct bio *bio, struct bio *bio_src)
279 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
280 bio_src->bi_max_vecs * sizeof(struct bio_vec));
283 * most users will be overriding ->bi_bdev with a new target,
284 * so we don't set nor calculate new physical/hw segment counts here
286 bio->bi_sector = bio_src->bi_sector;
287 bio->bi_bdev = bio_src->bi_bdev;
288 bio->bi_flags |= 1 << BIO_CLONED;
289 bio->bi_rw = bio_src->bi_rw;
290 bio->bi_vcnt = bio_src->bi_vcnt;
291 bio->bi_size = bio_src->bi_size;
292 bio->bi_idx = bio_src->bi_idx;
296 * bio_clone - clone a bio
298 * @gfp_mask: allocation priority
300 * Like __bio_clone, only also allocates the returned bio
302 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
304 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
309 b->bi_destructor = bio_fs_destructor;
312 if (bio_integrity(bio)) {
315 ret = bio_integrity_clone(b, bio, fs_bio_set);
325 * bio_get_nr_vecs - return approx number of vecs
328 * Return the approximate number of pages we can send to this target.
329 * There's no guarantee that you will be able to fit this number of pages
330 * into a bio, it does not account for dynamic restrictions that vary
333 int bio_get_nr_vecs(struct block_device *bdev)
335 struct request_queue *q = bdev_get_queue(bdev);
338 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
339 if (nr_pages > q->max_phys_segments)
340 nr_pages = q->max_phys_segments;
341 if (nr_pages > q->max_hw_segments)
342 nr_pages = q->max_hw_segments;
347 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
348 *page, unsigned int len, unsigned int offset,
349 unsigned short max_sectors)
351 int retried_segments = 0;
352 struct bio_vec *bvec;
355 * cloned bio must not modify vec list
357 if (unlikely(bio_flagged(bio, BIO_CLONED)))
360 if (((bio->bi_size + len) >> 9) > max_sectors)
364 * For filesystems with a blocksize smaller than the pagesize
365 * we will often be called with the same page as last time and
366 * a consecutive offset. Optimize this special case.
368 if (bio->bi_vcnt > 0) {
369 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
371 if (page == prev->bv_page &&
372 offset == prev->bv_offset + prev->bv_len) {
375 if (q->merge_bvec_fn) {
376 struct bvec_merge_data bvm = {
377 .bi_bdev = bio->bi_bdev,
378 .bi_sector = bio->bi_sector,
379 .bi_size = bio->bi_size,
383 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
393 if (bio->bi_vcnt >= bio->bi_max_vecs)
397 * we might lose a segment or two here, but rather that than
398 * make this too complex.
401 while (bio->bi_phys_segments >= q->max_phys_segments
402 || bio->bi_phys_segments >= q->max_hw_segments) {
404 if (retried_segments)
407 retried_segments = 1;
408 blk_recount_segments(q, bio);
412 * setup the new entry, we might clear it again later if we
413 * cannot add the page
415 bvec = &bio->bi_io_vec[bio->bi_vcnt];
416 bvec->bv_page = page;
418 bvec->bv_offset = offset;
421 * if queue has other restrictions (eg varying max sector size
422 * depending on offset), it can specify a merge_bvec_fn in the
423 * queue to get further control
425 if (q->merge_bvec_fn) {
426 struct bvec_merge_data bvm = {
427 .bi_bdev = bio->bi_bdev,
428 .bi_sector = bio->bi_sector,
429 .bi_size = bio->bi_size,
434 * merge_bvec_fn() returns number of bytes it can accept
437 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
438 bvec->bv_page = NULL;
445 /* If we may be able to merge these biovecs, force a recount */
446 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
447 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
450 bio->bi_phys_segments++;
457 * bio_add_pc_page - attempt to add page to bio
458 * @q: the target queue
459 * @bio: destination bio
461 * @len: vec entry length
462 * @offset: vec entry offset
464 * Attempt to add a page to the bio_vec maplist. This can fail for a
465 * number of reasons, such as the bio being full or target block
466 * device limitations. The target block device must allow bio's
467 * smaller than PAGE_SIZE, so it is always possible to add a single
468 * page to an empty bio. This should only be used by REQ_PC bios.
470 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
471 unsigned int len, unsigned int offset)
473 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
477 * bio_add_page - attempt to add page to bio
478 * @bio: destination bio
480 * @len: vec entry length
481 * @offset: vec entry offset
483 * Attempt to add a page to the bio_vec maplist. This can fail for a
484 * number of reasons, such as the bio being full or target block
485 * device limitations. The target block device must allow bio's
486 * smaller than PAGE_SIZE, so it is always possible to add a single
487 * page to an empty bio.
489 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
492 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
493 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
496 struct bio_map_data {
497 struct bio_vec *iovecs;
498 struct sg_iovec *sgvecs;
503 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
504 struct sg_iovec *iov, int iov_count,
507 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
508 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
509 bmd->nr_sgvecs = iov_count;
510 bmd->is_our_pages = is_our_pages;
511 bio->bi_private = bmd;
514 static void bio_free_map_data(struct bio_map_data *bmd)
521 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
524 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
529 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
535 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
544 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
545 struct sg_iovec *iov, int iov_count, int uncopy,
549 struct bio_vec *bvec;
551 unsigned int iov_off = 0;
552 int read = bio_data_dir(bio) == READ;
554 __bio_for_each_segment(bvec, bio, i, 0) {
555 char *bv_addr = page_address(bvec->bv_page);
556 unsigned int bv_len = iovecs[i].bv_len;
558 while (bv_len && iov_idx < iov_count) {
562 bytes = min_t(unsigned int,
563 iov[iov_idx].iov_len - iov_off, bv_len);
564 iov_addr = iov[iov_idx].iov_base + iov_off;
567 if (!read && !uncopy)
568 ret = copy_from_user(bv_addr, iov_addr,
571 ret = copy_to_user(iov_addr, bv_addr,
583 if (iov[iov_idx].iov_len == iov_off) {
590 __free_page(bvec->bv_page);
597 * bio_uncopy_user - finish previously mapped bio
598 * @bio: bio being terminated
600 * Free pages allocated from bio_copy_user() and write back data
601 * to user space in case of a read.
603 int bio_uncopy_user(struct bio *bio)
605 struct bio_map_data *bmd = bio->bi_private;
608 if (!bio_flagged(bio, BIO_NULL_MAPPED))
609 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
610 bmd->nr_sgvecs, 1, bmd->is_our_pages);
611 bio_free_map_data(bmd);
617 * bio_copy_user_iov - copy user data to bio
618 * @q: destination block queue
619 * @map_data: pointer to the rq_map_data holding pages (if necessary)
621 * @iov_count: number of elements in the iovec
622 * @write_to_vm: bool indicating writing to pages or not
623 * @gfp_mask: memory allocation flags
625 * Prepares and returns a bio for indirect user io, bouncing data
626 * to/from kernel pages as necessary. Must be paired with
627 * call bio_uncopy_user() on io completion.
629 struct bio *bio_copy_user_iov(struct request_queue *q,
630 struct rq_map_data *map_data,
631 struct sg_iovec *iov, int iov_count,
632 int write_to_vm, gfp_t gfp_mask)
634 struct bio_map_data *bmd;
635 struct bio_vec *bvec;
640 unsigned int len = 0;
642 for (i = 0; i < iov_count; i++) {
647 uaddr = (unsigned long)iov[i].iov_base;
648 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
649 start = uaddr >> PAGE_SHIFT;
651 nr_pages += end - start;
652 len += iov[i].iov_len;
655 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
657 return ERR_PTR(-ENOMEM);
660 bio = bio_alloc(gfp_mask, nr_pages);
664 bio->bi_rw |= (!write_to_vm << BIO_RW);
672 bytes = 1U << (PAGE_SHIFT + map_data->page_order);
680 if (i == map_data->nr_entries) {
684 page = map_data->pages[i++];
686 page = alloc_page(q->bounce_gfp | gfp_mask);
692 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
705 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
710 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
714 bio_for_each_segment(bvec, bio, i)
715 __free_page(bvec->bv_page);
719 bio_free_map_data(bmd);
724 * bio_copy_user - copy user data to bio
725 * @q: destination block queue
726 * @map_data: pointer to the rq_map_data holding pages (if necessary)
727 * @uaddr: start of user address
728 * @len: length in bytes
729 * @write_to_vm: bool indicating writing to pages or not
730 * @gfp_mask: memory allocation flags
732 * Prepares and returns a bio for indirect user io, bouncing data
733 * to/from kernel pages as necessary. Must be paired with
734 * call bio_uncopy_user() on io completion.
736 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
737 unsigned long uaddr, unsigned int len,
738 int write_to_vm, gfp_t gfp_mask)
742 iov.iov_base = (void __user *)uaddr;
745 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
748 static struct bio *__bio_map_user_iov(struct request_queue *q,
749 struct block_device *bdev,
750 struct sg_iovec *iov, int iov_count,
751 int write_to_vm, gfp_t gfp_mask)
760 for (i = 0; i < iov_count; i++) {
761 unsigned long uaddr = (unsigned long)iov[i].iov_base;
762 unsigned long len = iov[i].iov_len;
763 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
764 unsigned long start = uaddr >> PAGE_SHIFT;
766 nr_pages += end - start;
768 * buffer must be aligned to at least hardsector size for now
770 if (uaddr & queue_dma_alignment(q))
771 return ERR_PTR(-EINVAL);
775 return ERR_PTR(-EINVAL);
777 bio = bio_alloc(gfp_mask, nr_pages);
779 return ERR_PTR(-ENOMEM);
782 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
786 for (i = 0; i < iov_count; i++) {
787 unsigned long uaddr = (unsigned long)iov[i].iov_base;
788 unsigned long len = iov[i].iov_len;
789 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
790 unsigned long start = uaddr >> PAGE_SHIFT;
791 const int local_nr_pages = end - start;
792 const int page_limit = cur_page + local_nr_pages;
794 ret = get_user_pages_fast(uaddr, local_nr_pages,
795 write_to_vm, &pages[cur_page]);
796 if (ret < local_nr_pages) {
801 offset = uaddr & ~PAGE_MASK;
802 for (j = cur_page; j < page_limit; j++) {
803 unsigned int bytes = PAGE_SIZE - offset;
814 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
824 * release the pages we didn't map into the bio, if any
826 while (j < page_limit)
827 page_cache_release(pages[j++]);
833 * set data direction, and check if mapped pages need bouncing
836 bio->bi_rw |= (1 << BIO_RW);
839 bio->bi_flags |= (1 << BIO_USER_MAPPED);
843 for (i = 0; i < nr_pages; i++) {
846 page_cache_release(pages[i]);
855 * bio_map_user - map user address into bio
856 * @q: the struct request_queue for the bio
857 * @bdev: destination block device
858 * @uaddr: start of user address
859 * @len: length in bytes
860 * @write_to_vm: bool indicating writing to pages or not
861 * @gfp_mask: memory allocation flags
863 * Map the user space address into a bio suitable for io to a block
864 * device. Returns an error pointer in case of error.
866 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
867 unsigned long uaddr, unsigned int len, int write_to_vm,
872 iov.iov_base = (void __user *)uaddr;
875 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
879 * bio_map_user_iov - map user sg_iovec table into bio
880 * @q: the struct request_queue for the bio
881 * @bdev: destination block device
883 * @iov_count: number of elements in the iovec
884 * @write_to_vm: bool indicating writing to pages or not
885 * @gfp_mask: memory allocation flags
887 * Map the user space address into a bio suitable for io to a block
888 * device. Returns an error pointer in case of error.
890 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
891 struct sg_iovec *iov, int iov_count,
892 int write_to_vm, gfp_t gfp_mask)
896 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
902 * subtle -- if __bio_map_user() ended up bouncing a bio,
903 * it would normally disappear when its bi_end_io is run.
904 * however, we need it for the unmap, so grab an extra
912 static void __bio_unmap_user(struct bio *bio)
914 struct bio_vec *bvec;
918 * make sure we dirty pages we wrote to
920 __bio_for_each_segment(bvec, bio, i, 0) {
921 if (bio_data_dir(bio) == READ)
922 set_page_dirty_lock(bvec->bv_page);
924 page_cache_release(bvec->bv_page);
931 * bio_unmap_user - unmap a bio
932 * @bio: the bio being unmapped
934 * Unmap a bio previously mapped by bio_map_user(). Must be called with
937 * bio_unmap_user() may sleep.
939 void bio_unmap_user(struct bio *bio)
941 __bio_unmap_user(bio);
945 static void bio_map_kern_endio(struct bio *bio, int err)
951 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
952 unsigned int len, gfp_t gfp_mask)
954 unsigned long kaddr = (unsigned long)data;
955 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
956 unsigned long start = kaddr >> PAGE_SHIFT;
957 const int nr_pages = end - start;
961 bio = bio_alloc(gfp_mask, nr_pages);
963 return ERR_PTR(-ENOMEM);
965 offset = offset_in_page(kaddr);
966 for (i = 0; i < nr_pages; i++) {
967 unsigned int bytes = PAGE_SIZE - offset;
975 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
984 bio->bi_end_io = bio_map_kern_endio;
989 * bio_map_kern - map kernel address into bio
990 * @q: the struct request_queue for the bio
991 * @data: pointer to buffer to map
992 * @len: length in bytes
993 * @gfp_mask: allocation flags for bio allocation
995 * Map the kernel address into a bio suitable for io to a block
996 * device. Returns an error pointer in case of error.
998 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1003 bio = __bio_map_kern(q, data, len, gfp_mask);
1007 if (bio->bi_size == len)
1011 * Don't support partial mappings.
1014 return ERR_PTR(-EINVAL);
1017 static void bio_copy_kern_endio(struct bio *bio, int err)
1019 struct bio_vec *bvec;
1020 const int read = bio_data_dir(bio) == READ;
1021 struct bio_map_data *bmd = bio->bi_private;
1023 char *p = bmd->sgvecs[0].iov_base;
1025 __bio_for_each_segment(bvec, bio, i, 0) {
1026 char *addr = page_address(bvec->bv_page);
1027 int len = bmd->iovecs[i].bv_len;
1030 memcpy(p, addr, len);
1032 __free_page(bvec->bv_page);
1036 bio_free_map_data(bmd);
1041 * bio_copy_kern - copy kernel address into bio
1042 * @q: the struct request_queue for the bio
1043 * @data: pointer to buffer to copy
1044 * @len: length in bytes
1045 * @gfp_mask: allocation flags for bio and page allocation
1046 * @reading: data direction is READ
1048 * copy the kernel address into a bio suitable for io to a block
1049 * device. Returns an error pointer in case of error.
1051 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1052 gfp_t gfp_mask, int reading)
1055 struct bio_vec *bvec;
1058 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1065 bio_for_each_segment(bvec, bio, i) {
1066 char *addr = page_address(bvec->bv_page);
1068 memcpy(addr, p, bvec->bv_len);
1073 bio->bi_end_io = bio_copy_kern_endio;
1079 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1080 * for performing direct-IO in BIOs.
1082 * The problem is that we cannot run set_page_dirty() from interrupt context
1083 * because the required locks are not interrupt-safe. So what we can do is to
1084 * mark the pages dirty _before_ performing IO. And in interrupt context,
1085 * check that the pages are still dirty. If so, fine. If not, redirty them
1086 * in process context.
1088 * We special-case compound pages here: normally this means reads into hugetlb
1089 * pages. The logic in here doesn't really work right for compound pages
1090 * because the VM does not uniformly chase down the head page in all cases.
1091 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1092 * handle them at all. So we skip compound pages here at an early stage.
1094 * Note that this code is very hard to test under normal circumstances because
1095 * direct-io pins the pages with get_user_pages(). This makes
1096 * is_page_cache_freeable return false, and the VM will not clean the pages.
1097 * But other code (eg, pdflush) could clean the pages if they are mapped
1100 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1101 * deferred bio dirtying paths.
1105 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1107 void bio_set_pages_dirty(struct bio *bio)
1109 struct bio_vec *bvec = bio->bi_io_vec;
1112 for (i = 0; i < bio->bi_vcnt; i++) {
1113 struct page *page = bvec[i].bv_page;
1115 if (page && !PageCompound(page))
1116 set_page_dirty_lock(page);
1120 static void bio_release_pages(struct bio *bio)
1122 struct bio_vec *bvec = bio->bi_io_vec;
1125 for (i = 0; i < bio->bi_vcnt; i++) {
1126 struct page *page = bvec[i].bv_page;
1134 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1135 * If they are, then fine. If, however, some pages are clean then they must
1136 * have been written out during the direct-IO read. So we take another ref on
1137 * the BIO and the offending pages and re-dirty the pages in process context.
1139 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1140 * here on. It will run one page_cache_release() against each page and will
1141 * run one bio_put() against the BIO.
1144 static void bio_dirty_fn(struct work_struct *work);
1146 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1147 static DEFINE_SPINLOCK(bio_dirty_lock);
1148 static struct bio *bio_dirty_list;
1151 * This runs in process context
1153 static void bio_dirty_fn(struct work_struct *work)
1155 unsigned long flags;
1158 spin_lock_irqsave(&bio_dirty_lock, flags);
1159 bio = bio_dirty_list;
1160 bio_dirty_list = NULL;
1161 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1164 struct bio *next = bio->bi_private;
1166 bio_set_pages_dirty(bio);
1167 bio_release_pages(bio);
1173 void bio_check_pages_dirty(struct bio *bio)
1175 struct bio_vec *bvec = bio->bi_io_vec;
1176 int nr_clean_pages = 0;
1179 for (i = 0; i < bio->bi_vcnt; i++) {
1180 struct page *page = bvec[i].bv_page;
1182 if (PageDirty(page) || PageCompound(page)) {
1183 page_cache_release(page);
1184 bvec[i].bv_page = NULL;
1190 if (nr_clean_pages) {
1191 unsigned long flags;
1193 spin_lock_irqsave(&bio_dirty_lock, flags);
1194 bio->bi_private = bio_dirty_list;
1195 bio_dirty_list = bio;
1196 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1197 schedule_work(&bio_dirty_work);
1204 * bio_endio - end I/O on a bio
1206 * @error: error, if any
1209 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1210 * preferred way to end I/O on a bio, it takes care of clearing
1211 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1212 * established -Exxxx (-EIO, for instance) error values in case
1213 * something went wrong. Noone should call bi_end_io() directly on a
1214 * bio unless they own it and thus know that it has an end_io
1217 void bio_endio(struct bio *bio, int error)
1220 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1221 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1225 bio->bi_end_io(bio, error);
1228 void bio_pair_release(struct bio_pair *bp)
1230 if (atomic_dec_and_test(&bp->cnt)) {
1231 struct bio *master = bp->bio1.bi_private;
1233 bio_endio(master, bp->error);
1234 mempool_free(bp, bp->bio2.bi_private);
1238 static void bio_pair_end_1(struct bio *bi, int err)
1240 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1245 bio_pair_release(bp);
1248 static void bio_pair_end_2(struct bio *bi, int err)
1250 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1255 bio_pair_release(bp);
1259 * split a bio - only worry about a bio with a single page
1262 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1264 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1269 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1270 bi->bi_sector + first_sectors);
1272 BUG_ON(bi->bi_vcnt != 1);
1273 BUG_ON(bi->bi_idx != 0);
1274 atomic_set(&bp->cnt, 3);
1278 bp->bio2.bi_sector += first_sectors;
1279 bp->bio2.bi_size -= first_sectors << 9;
1280 bp->bio1.bi_size = first_sectors << 9;
1282 bp->bv1 = bi->bi_io_vec[0];
1283 bp->bv2 = bi->bi_io_vec[0];
1284 bp->bv2.bv_offset += first_sectors << 9;
1285 bp->bv2.bv_len -= first_sectors << 9;
1286 bp->bv1.bv_len = first_sectors << 9;
1288 bp->bio1.bi_io_vec = &bp->bv1;
1289 bp->bio2.bi_io_vec = &bp->bv2;
1291 bp->bio1.bi_max_vecs = 1;
1292 bp->bio2.bi_max_vecs = 1;
1294 bp->bio1.bi_end_io = bio_pair_end_1;
1295 bp->bio2.bi_end_io = bio_pair_end_2;
1297 bp->bio1.bi_private = bi;
1298 bp->bio2.bi_private = bio_split_pool;
1300 if (bio_integrity(bi))
1301 bio_integrity_split(bi, bp, first_sectors);
1307 * bio_sector_offset - Find hardware sector offset in bio
1308 * @bio: bio to inspect
1309 * @index: bio_vec index
1310 * @offset: offset in bv_page
1312 * Return the number of hardware sectors between beginning of bio
1313 * and an end point indicated by a bio_vec index and an offset
1314 * within that vector's page.
1316 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1317 unsigned int offset)
1319 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1326 if (index >= bio->bi_idx)
1327 index = bio->bi_vcnt - 1;
1329 __bio_for_each_segment(bv, bio, i, 0) {
1331 if (offset > bv->bv_offset)
1332 sectors += (offset - bv->bv_offset) / sector_sz;
1336 sectors += bv->bv_len / sector_sz;
1341 EXPORT_SYMBOL(bio_sector_offset);
1344 * create memory pools for biovec's in a bio_set.
1345 * use the global biovec slabs created for general use.
1347 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1351 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1352 struct biovec_slab *bp = bvec_slabs + i;
1353 mempool_t **bvp = bs->bvec_pools + i;
1355 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
1362 static void biovec_free_pools(struct bio_set *bs)
1366 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1367 mempool_t *bvp = bs->bvec_pools[i];
1370 mempool_destroy(bvp);
1375 void bioset_free(struct bio_set *bs)
1378 mempool_destroy(bs->bio_pool);
1380 bioset_integrity_free(bs);
1381 biovec_free_pools(bs);
1386 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
1388 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1393 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
1397 if (bioset_integrity_create(bs, bio_pool_size))
1400 if (!biovec_create_pools(bs, bvec_pool_size))
1408 static void __init biovec_init_slabs(void)
1412 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1414 struct biovec_slab *bvs = bvec_slabs + i;
1416 size = bvs->nr_vecs * sizeof(struct bio_vec);
1417 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1418 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1422 static int __init init_bio(void)
1424 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
1426 bio_integrity_init_slab();
1427 biovec_init_slabs();
1429 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
1431 panic("bio: can't allocate bios\n");
1433 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1434 sizeof(struct bio_pair));
1435 if (!bio_split_pool)
1436 panic("bio: can't create split pool\n");
1441 subsys_initcall(init_bio);
1443 EXPORT_SYMBOL(bio_alloc);
1444 EXPORT_SYMBOL(bio_kmalloc);
1445 EXPORT_SYMBOL(bio_put);
1446 EXPORT_SYMBOL(bio_free);
1447 EXPORT_SYMBOL(bio_endio);
1448 EXPORT_SYMBOL(bio_init);
1449 EXPORT_SYMBOL(__bio_clone);
1450 EXPORT_SYMBOL(bio_clone);
1451 EXPORT_SYMBOL(bio_phys_segments);
1452 EXPORT_SYMBOL(bio_add_page);
1453 EXPORT_SYMBOL(bio_add_pc_page);
1454 EXPORT_SYMBOL(bio_get_nr_vecs);
1455 EXPORT_SYMBOL(bio_map_user);
1456 EXPORT_SYMBOL(bio_unmap_user);
1457 EXPORT_SYMBOL(bio_map_kern);
1458 EXPORT_SYMBOL(bio_copy_kern);
1459 EXPORT_SYMBOL(bio_pair_release);
1460 EXPORT_SYMBOL(bio_split);
1461 EXPORT_SYMBOL(bio_copy_user);
1462 EXPORT_SYMBOL(bio_uncopy_user);
1463 EXPORT_SYMBOL(bioset_create);
1464 EXPORT_SYMBOL(bioset_free);
1465 EXPORT_SYMBOL(bio_alloc_bioset);