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,
67 * If 'bs' is given, lookup the pool and do the mempool alloc.
68 * If not, this is a bio_kmalloc() allocation and just do a
69 * kzalloc() for the exact number of vecs right away.
72 bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask);
75 * see comment near bvec_array define!
93 case 129 ... BIO_MAX_PAGES:
101 * idx now points to the pool we want to allocate from. only the
102 * 1-vec entry pool is mempool backed.
104 if (*idx == BIOVEC_MAX_IDX) {
106 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
108 struct biovec_slab *bvs = bvec_slabs + *idx;
109 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
112 * Make this allocation restricted and don't dump info on
113 * allocation failures, since we'll fallback to the mempool
114 * in case of failure.
116 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
119 * Try a slab allocation. If this fails and __GFP_WAIT
120 * is set, retry with the 1-entry mempool
122 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
123 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
124 *idx = BIOVEC_MAX_IDX;
130 memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec));
135 void bio_free(struct bio *bio, struct bio_set *bs)
137 if (bio->bi_io_vec) {
138 const int pool_idx = BIO_POOL_IDX(bio);
140 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
142 if (pool_idx == BIOVEC_MAX_IDX)
143 mempool_free(bio->bi_io_vec, bs->bvec_pool);
145 struct biovec_slab *bvs = bvec_slabs + pool_idx;
147 kmem_cache_free(bvs->slab, bio->bi_io_vec);
151 if (bio_integrity(bio))
152 bio_integrity_free(bio, bs);
154 mempool_free(bio, bs->bio_pool);
158 * default destructor for a bio allocated with bio_alloc_bioset()
160 static void bio_fs_destructor(struct bio *bio)
162 bio_free(bio, fs_bio_set);
165 static void bio_kmalloc_destructor(struct bio *bio)
167 kfree(bio->bi_io_vec);
171 void bio_init(struct bio *bio)
173 memset(bio, 0, sizeof(*bio));
174 bio->bi_flags = 1 << BIO_UPTODATE;
175 bio->bi_comp_cpu = -1;
176 atomic_set(&bio->bi_cnt, 1);
180 * bio_alloc_bioset - allocate a bio for I/O
181 * @gfp_mask: the GFP_ mask given to the slab allocator
182 * @nr_iovecs: number of iovecs to pre-allocate
183 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
186 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
187 * If %__GFP_WAIT is set then we will block on the internal pool waiting
188 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
189 * fall back to just using @kmalloc to allocate the required memory.
191 * allocate bio and iovecs from the memory pools specified by the
192 * bio_set structure, or @kmalloc if none given.
194 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
199 bio = mempool_alloc(bs->bio_pool, gfp_mask);
201 bio = kmalloc(sizeof(*bio), gfp_mask);
204 struct bio_vec *bvl = NULL;
207 if (likely(nr_iovecs)) {
208 unsigned long uninitialized_var(idx);
210 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
211 if (unlikely(!bvl)) {
213 mempool_free(bio, bs->bio_pool);
219 bio->bi_flags |= idx << BIO_POOL_OFFSET;
220 bio->bi_max_vecs = bvec_nr_vecs(idx);
222 bio->bi_io_vec = bvl;
228 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
230 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
233 bio->bi_destructor = bio_fs_destructor;
239 * Like bio_alloc(), but doesn't use a mempool backing. This means that
240 * it CAN fail, but while bio_alloc() can only be used for allocations
241 * that have a short (finite) life span, bio_kmalloc() should be used
242 * for more permanent bio allocations (like allocating some bio's for
243 * initalization or setup purposes).
245 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
247 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL);
250 bio->bi_destructor = bio_kmalloc_destructor;
255 void zero_fill_bio(struct bio *bio)
261 bio_for_each_segment(bv, bio, i) {
262 char *data = bvec_kmap_irq(bv, &flags);
263 memset(data, 0, bv->bv_len);
264 flush_dcache_page(bv->bv_page);
265 bvec_kunmap_irq(data, &flags);
268 EXPORT_SYMBOL(zero_fill_bio);
271 * bio_put - release a reference to a bio
272 * @bio: bio to release reference to
275 * Put a reference to a &struct bio, either one you have gotten with
276 * bio_alloc or bio_get. The last put of a bio will free it.
278 void bio_put(struct bio *bio)
280 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
285 if (atomic_dec_and_test(&bio->bi_cnt)) {
287 bio->bi_destructor(bio);
291 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
293 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
294 blk_recount_segments(q, bio);
296 return bio->bi_phys_segments;
300 * __bio_clone - clone a bio
301 * @bio: destination bio
302 * @bio_src: bio to clone
304 * Clone a &bio. Caller will own the returned bio, but not
305 * the actual data it points to. Reference count of returned
308 void __bio_clone(struct bio *bio, struct bio *bio_src)
310 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
311 bio_src->bi_max_vecs * sizeof(struct bio_vec));
314 * most users will be overriding ->bi_bdev with a new target,
315 * so we don't set nor calculate new physical/hw segment counts here
317 bio->bi_sector = bio_src->bi_sector;
318 bio->bi_bdev = bio_src->bi_bdev;
319 bio->bi_flags |= 1 << BIO_CLONED;
320 bio->bi_rw = bio_src->bi_rw;
321 bio->bi_vcnt = bio_src->bi_vcnt;
322 bio->bi_size = bio_src->bi_size;
323 bio->bi_idx = bio_src->bi_idx;
327 * bio_clone - clone a bio
329 * @gfp_mask: allocation priority
331 * Like __bio_clone, only also allocates the returned bio
333 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
335 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
340 b->bi_destructor = bio_fs_destructor;
343 if (bio_integrity(bio)) {
346 ret = bio_integrity_clone(b, bio, fs_bio_set);
356 * bio_get_nr_vecs - return approx number of vecs
359 * Return the approximate number of pages we can send to this target.
360 * There's no guarantee that you will be able to fit this number of pages
361 * into a bio, it does not account for dynamic restrictions that vary
364 int bio_get_nr_vecs(struct block_device *bdev)
366 struct request_queue *q = bdev_get_queue(bdev);
369 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
370 if (nr_pages > q->max_phys_segments)
371 nr_pages = q->max_phys_segments;
372 if (nr_pages > q->max_hw_segments)
373 nr_pages = q->max_hw_segments;
378 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
379 *page, unsigned int len, unsigned int offset,
380 unsigned short max_sectors)
382 int retried_segments = 0;
383 struct bio_vec *bvec;
386 * cloned bio must not modify vec list
388 if (unlikely(bio_flagged(bio, BIO_CLONED)))
391 if (((bio->bi_size + len) >> 9) > max_sectors)
395 * For filesystems with a blocksize smaller than the pagesize
396 * we will often be called with the same page as last time and
397 * a consecutive offset. Optimize this special case.
399 if (bio->bi_vcnt > 0) {
400 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
402 if (page == prev->bv_page &&
403 offset == prev->bv_offset + prev->bv_len) {
406 if (q->merge_bvec_fn) {
407 struct bvec_merge_data bvm = {
408 .bi_bdev = bio->bi_bdev,
409 .bi_sector = bio->bi_sector,
410 .bi_size = bio->bi_size,
414 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
424 if (bio->bi_vcnt >= bio->bi_max_vecs)
428 * we might lose a segment or two here, but rather that than
429 * make this too complex.
432 while (bio->bi_phys_segments >= q->max_phys_segments
433 || bio->bi_phys_segments >= q->max_hw_segments) {
435 if (retried_segments)
438 retried_segments = 1;
439 blk_recount_segments(q, bio);
443 * setup the new entry, we might clear it again later if we
444 * cannot add the page
446 bvec = &bio->bi_io_vec[bio->bi_vcnt];
447 bvec->bv_page = page;
449 bvec->bv_offset = offset;
452 * if queue has other restrictions (eg varying max sector size
453 * depending on offset), it can specify a merge_bvec_fn in the
454 * queue to get further control
456 if (q->merge_bvec_fn) {
457 struct bvec_merge_data bvm = {
458 .bi_bdev = bio->bi_bdev,
459 .bi_sector = bio->bi_sector,
460 .bi_size = bio->bi_size,
465 * merge_bvec_fn() returns number of bytes it can accept
468 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
469 bvec->bv_page = NULL;
476 /* If we may be able to merge these biovecs, force a recount */
477 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
478 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
481 bio->bi_phys_segments++;
488 * bio_add_pc_page - attempt to add page to bio
489 * @q: the target queue
490 * @bio: destination bio
492 * @len: vec entry length
493 * @offset: vec entry offset
495 * Attempt to add a page to the bio_vec maplist. This can fail for a
496 * number of reasons, such as the bio being full or target block
497 * device limitations. The target block device must allow bio's
498 * smaller than PAGE_SIZE, so it is always possible to add a single
499 * page to an empty bio. This should only be used by REQ_PC bios.
501 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
502 unsigned int len, unsigned int offset)
504 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
508 * bio_add_page - attempt to add page to bio
509 * @bio: destination bio
511 * @len: vec entry length
512 * @offset: vec entry offset
514 * Attempt to add a page to the bio_vec maplist. This can fail for a
515 * number of reasons, such as the bio being full or target block
516 * device limitations. The target block device must allow bio's
517 * smaller than PAGE_SIZE, so it is always possible to add a single
518 * page to an empty bio.
520 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
523 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
524 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
527 struct bio_map_data {
528 struct bio_vec *iovecs;
529 struct sg_iovec *sgvecs;
534 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
535 struct sg_iovec *iov, int iov_count,
538 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
539 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
540 bmd->nr_sgvecs = iov_count;
541 bmd->is_our_pages = is_our_pages;
542 bio->bi_private = bmd;
545 static void bio_free_map_data(struct bio_map_data *bmd)
552 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
555 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
560 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
566 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
575 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
576 struct sg_iovec *iov, int iov_count, int uncopy,
580 struct bio_vec *bvec;
582 unsigned int iov_off = 0;
583 int read = bio_data_dir(bio) == READ;
585 __bio_for_each_segment(bvec, bio, i, 0) {
586 char *bv_addr = page_address(bvec->bv_page);
587 unsigned int bv_len = iovecs[i].bv_len;
589 while (bv_len && iov_idx < iov_count) {
593 bytes = min_t(unsigned int,
594 iov[iov_idx].iov_len - iov_off, bv_len);
595 iov_addr = iov[iov_idx].iov_base + iov_off;
598 if (!read && !uncopy)
599 ret = copy_from_user(bv_addr, iov_addr,
602 ret = copy_to_user(iov_addr, bv_addr,
614 if (iov[iov_idx].iov_len == iov_off) {
621 __free_page(bvec->bv_page);
628 * bio_uncopy_user - finish previously mapped bio
629 * @bio: bio being terminated
631 * Free pages allocated from bio_copy_user() and write back data
632 * to user space in case of a read.
634 int bio_uncopy_user(struct bio *bio)
636 struct bio_map_data *bmd = bio->bi_private;
639 if (!bio_flagged(bio, BIO_NULL_MAPPED))
640 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
641 bmd->nr_sgvecs, 1, bmd->is_our_pages);
642 bio_free_map_data(bmd);
648 * bio_copy_user_iov - copy user data to bio
649 * @q: destination block queue
650 * @map_data: pointer to the rq_map_data holding pages (if necessary)
652 * @iov_count: number of elements in the iovec
653 * @write_to_vm: bool indicating writing to pages or not
654 * @gfp_mask: memory allocation flags
656 * Prepares and returns a bio for indirect user io, bouncing data
657 * to/from kernel pages as necessary. Must be paired with
658 * call bio_uncopy_user() on io completion.
660 struct bio *bio_copy_user_iov(struct request_queue *q,
661 struct rq_map_data *map_data,
662 struct sg_iovec *iov, int iov_count,
663 int write_to_vm, gfp_t gfp_mask)
665 struct bio_map_data *bmd;
666 struct bio_vec *bvec;
671 unsigned int len = 0;
673 for (i = 0; i < iov_count; i++) {
678 uaddr = (unsigned long)iov[i].iov_base;
679 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
680 start = uaddr >> PAGE_SHIFT;
682 nr_pages += end - start;
683 len += iov[i].iov_len;
686 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
688 return ERR_PTR(-ENOMEM);
691 bio = bio_alloc(gfp_mask, nr_pages);
695 bio->bi_rw |= (!write_to_vm << BIO_RW);
703 bytes = 1U << (PAGE_SHIFT + map_data->page_order);
711 if (i == map_data->nr_entries) {
715 page = map_data->pages[i++];
717 page = alloc_page(q->bounce_gfp | gfp_mask);
723 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
736 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
741 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
745 bio_for_each_segment(bvec, bio, i)
746 __free_page(bvec->bv_page);
750 bio_free_map_data(bmd);
755 * bio_copy_user - copy user data to bio
756 * @q: destination block queue
757 * @map_data: pointer to the rq_map_data holding pages (if necessary)
758 * @uaddr: start of user address
759 * @len: length in bytes
760 * @write_to_vm: bool indicating writing to pages or not
761 * @gfp_mask: memory allocation flags
763 * Prepares and returns a bio for indirect user io, bouncing data
764 * to/from kernel pages as necessary. Must be paired with
765 * call bio_uncopy_user() on io completion.
767 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
768 unsigned long uaddr, unsigned int len,
769 int write_to_vm, gfp_t gfp_mask)
773 iov.iov_base = (void __user *)uaddr;
776 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
779 static struct bio *__bio_map_user_iov(struct request_queue *q,
780 struct block_device *bdev,
781 struct sg_iovec *iov, int iov_count,
782 int write_to_vm, gfp_t gfp_mask)
791 for (i = 0; i < iov_count; i++) {
792 unsigned long uaddr = (unsigned long)iov[i].iov_base;
793 unsigned long len = iov[i].iov_len;
794 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
795 unsigned long start = uaddr >> PAGE_SHIFT;
797 nr_pages += end - start;
799 * buffer must be aligned to at least hardsector size for now
801 if (uaddr & queue_dma_alignment(q))
802 return ERR_PTR(-EINVAL);
806 return ERR_PTR(-EINVAL);
808 bio = bio_alloc(gfp_mask, nr_pages);
810 return ERR_PTR(-ENOMEM);
813 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
817 for (i = 0; i < iov_count; i++) {
818 unsigned long uaddr = (unsigned long)iov[i].iov_base;
819 unsigned long len = iov[i].iov_len;
820 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
821 unsigned long start = uaddr >> PAGE_SHIFT;
822 const int local_nr_pages = end - start;
823 const int page_limit = cur_page + local_nr_pages;
825 ret = get_user_pages_fast(uaddr, local_nr_pages,
826 write_to_vm, &pages[cur_page]);
827 if (ret < local_nr_pages) {
832 offset = uaddr & ~PAGE_MASK;
833 for (j = cur_page; j < page_limit; j++) {
834 unsigned int bytes = PAGE_SIZE - offset;
845 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
855 * release the pages we didn't map into the bio, if any
857 while (j < page_limit)
858 page_cache_release(pages[j++]);
864 * set data direction, and check if mapped pages need bouncing
867 bio->bi_rw |= (1 << BIO_RW);
870 bio->bi_flags |= (1 << BIO_USER_MAPPED);
874 for (i = 0; i < nr_pages; i++) {
877 page_cache_release(pages[i]);
886 * bio_map_user - map user address into bio
887 * @q: the struct request_queue for the bio
888 * @bdev: destination block device
889 * @uaddr: start of user address
890 * @len: length in bytes
891 * @write_to_vm: bool indicating writing to pages or not
892 * @gfp_mask: memory allocation flags
894 * Map the user space address into a bio suitable for io to a block
895 * device. Returns an error pointer in case of error.
897 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
898 unsigned long uaddr, unsigned int len, int write_to_vm,
903 iov.iov_base = (void __user *)uaddr;
906 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
910 * bio_map_user_iov - map user sg_iovec table into bio
911 * @q: the struct request_queue for the bio
912 * @bdev: destination block device
914 * @iov_count: number of elements in the iovec
915 * @write_to_vm: bool indicating writing to pages or not
916 * @gfp_mask: memory allocation flags
918 * Map the user space address into a bio suitable for io to a block
919 * device. Returns an error pointer in case of error.
921 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
922 struct sg_iovec *iov, int iov_count,
923 int write_to_vm, gfp_t gfp_mask)
927 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
933 * subtle -- if __bio_map_user() ended up bouncing a bio,
934 * it would normally disappear when its bi_end_io is run.
935 * however, we need it for the unmap, so grab an extra
943 static void __bio_unmap_user(struct bio *bio)
945 struct bio_vec *bvec;
949 * make sure we dirty pages we wrote to
951 __bio_for_each_segment(bvec, bio, i, 0) {
952 if (bio_data_dir(bio) == READ)
953 set_page_dirty_lock(bvec->bv_page);
955 page_cache_release(bvec->bv_page);
962 * bio_unmap_user - unmap a bio
963 * @bio: the bio being unmapped
965 * Unmap a bio previously mapped by bio_map_user(). Must be called with
968 * bio_unmap_user() may sleep.
970 void bio_unmap_user(struct bio *bio)
972 __bio_unmap_user(bio);
976 static void bio_map_kern_endio(struct bio *bio, int err)
982 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
983 unsigned int len, gfp_t gfp_mask)
985 unsigned long kaddr = (unsigned long)data;
986 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
987 unsigned long start = kaddr >> PAGE_SHIFT;
988 const int nr_pages = end - start;
992 bio = bio_alloc(gfp_mask, nr_pages);
994 return ERR_PTR(-ENOMEM);
996 offset = offset_in_page(kaddr);
997 for (i = 0; i < nr_pages; i++) {
998 unsigned int bytes = PAGE_SIZE - offset;
1006 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1015 bio->bi_end_io = bio_map_kern_endio;
1020 * bio_map_kern - map kernel address into bio
1021 * @q: the struct request_queue for the bio
1022 * @data: pointer to buffer to map
1023 * @len: length in bytes
1024 * @gfp_mask: allocation flags for bio allocation
1026 * Map the kernel address into a bio suitable for io to a block
1027 * device. Returns an error pointer in case of error.
1029 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1034 bio = __bio_map_kern(q, data, len, gfp_mask);
1038 if (bio->bi_size == len)
1042 * Don't support partial mappings.
1045 return ERR_PTR(-EINVAL);
1048 static void bio_copy_kern_endio(struct bio *bio, int err)
1050 struct bio_vec *bvec;
1051 const int read = bio_data_dir(bio) == READ;
1052 struct bio_map_data *bmd = bio->bi_private;
1054 char *p = bmd->sgvecs[0].iov_base;
1056 __bio_for_each_segment(bvec, bio, i, 0) {
1057 char *addr = page_address(bvec->bv_page);
1058 int len = bmd->iovecs[i].bv_len;
1061 memcpy(p, addr, len);
1063 __free_page(bvec->bv_page);
1067 bio_free_map_data(bmd);
1072 * bio_copy_kern - copy kernel address into bio
1073 * @q: the struct request_queue for the bio
1074 * @data: pointer to buffer to copy
1075 * @len: length in bytes
1076 * @gfp_mask: allocation flags for bio and page allocation
1077 * @reading: data direction is READ
1079 * copy the kernel address into a bio suitable for io to a block
1080 * device. Returns an error pointer in case of error.
1082 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1083 gfp_t gfp_mask, int reading)
1086 struct bio_vec *bvec;
1089 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1096 bio_for_each_segment(bvec, bio, i) {
1097 char *addr = page_address(bvec->bv_page);
1099 memcpy(addr, p, bvec->bv_len);
1104 bio->bi_end_io = bio_copy_kern_endio;
1110 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1111 * for performing direct-IO in BIOs.
1113 * The problem is that we cannot run set_page_dirty() from interrupt context
1114 * because the required locks are not interrupt-safe. So what we can do is to
1115 * mark the pages dirty _before_ performing IO. And in interrupt context,
1116 * check that the pages are still dirty. If so, fine. If not, redirty them
1117 * in process context.
1119 * We special-case compound pages here: normally this means reads into hugetlb
1120 * pages. The logic in here doesn't really work right for compound pages
1121 * because the VM does not uniformly chase down the head page in all cases.
1122 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1123 * handle them at all. So we skip compound pages here at an early stage.
1125 * Note that this code is very hard to test under normal circumstances because
1126 * direct-io pins the pages with get_user_pages(). This makes
1127 * is_page_cache_freeable return false, and the VM will not clean the pages.
1128 * But other code (eg, pdflush) could clean the pages if they are mapped
1131 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1132 * deferred bio dirtying paths.
1136 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1138 void bio_set_pages_dirty(struct bio *bio)
1140 struct bio_vec *bvec = bio->bi_io_vec;
1143 for (i = 0; i < bio->bi_vcnt; i++) {
1144 struct page *page = bvec[i].bv_page;
1146 if (page && !PageCompound(page))
1147 set_page_dirty_lock(page);
1151 static void bio_release_pages(struct bio *bio)
1153 struct bio_vec *bvec = bio->bi_io_vec;
1156 for (i = 0; i < bio->bi_vcnt; i++) {
1157 struct page *page = bvec[i].bv_page;
1165 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1166 * If they are, then fine. If, however, some pages are clean then they must
1167 * have been written out during the direct-IO read. So we take another ref on
1168 * the BIO and the offending pages and re-dirty the pages in process context.
1170 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1171 * here on. It will run one page_cache_release() against each page and will
1172 * run one bio_put() against the BIO.
1175 static void bio_dirty_fn(struct work_struct *work);
1177 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1178 static DEFINE_SPINLOCK(bio_dirty_lock);
1179 static struct bio *bio_dirty_list;
1182 * This runs in process context
1184 static void bio_dirty_fn(struct work_struct *work)
1186 unsigned long flags;
1189 spin_lock_irqsave(&bio_dirty_lock, flags);
1190 bio = bio_dirty_list;
1191 bio_dirty_list = NULL;
1192 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1195 struct bio *next = bio->bi_private;
1197 bio_set_pages_dirty(bio);
1198 bio_release_pages(bio);
1204 void bio_check_pages_dirty(struct bio *bio)
1206 struct bio_vec *bvec = bio->bi_io_vec;
1207 int nr_clean_pages = 0;
1210 for (i = 0; i < bio->bi_vcnt; i++) {
1211 struct page *page = bvec[i].bv_page;
1213 if (PageDirty(page) || PageCompound(page)) {
1214 page_cache_release(page);
1215 bvec[i].bv_page = NULL;
1221 if (nr_clean_pages) {
1222 unsigned long flags;
1224 spin_lock_irqsave(&bio_dirty_lock, flags);
1225 bio->bi_private = bio_dirty_list;
1226 bio_dirty_list = bio;
1227 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1228 schedule_work(&bio_dirty_work);
1235 * bio_endio - end I/O on a bio
1237 * @error: error, if any
1240 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1241 * preferred way to end I/O on a bio, it takes care of clearing
1242 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1243 * established -Exxxx (-EIO, for instance) error values in case
1244 * something went wrong. Noone should call bi_end_io() directly on a
1245 * bio unless they own it and thus know that it has an end_io
1248 void bio_endio(struct bio *bio, int error)
1251 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1252 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1256 bio->bi_end_io(bio, error);
1259 void bio_pair_release(struct bio_pair *bp)
1261 if (atomic_dec_and_test(&bp->cnt)) {
1262 struct bio *master = bp->bio1.bi_private;
1264 bio_endio(master, bp->error);
1265 mempool_free(bp, bp->bio2.bi_private);
1269 static void bio_pair_end_1(struct bio *bi, int err)
1271 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1276 bio_pair_release(bp);
1279 static void bio_pair_end_2(struct bio *bi, int err)
1281 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1286 bio_pair_release(bp);
1290 * split a bio - only worry about a bio with a single page
1293 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1295 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1300 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1301 bi->bi_sector + first_sectors);
1303 BUG_ON(bi->bi_vcnt != 1);
1304 BUG_ON(bi->bi_idx != 0);
1305 atomic_set(&bp->cnt, 3);
1309 bp->bio2.bi_sector += first_sectors;
1310 bp->bio2.bi_size -= first_sectors << 9;
1311 bp->bio1.bi_size = first_sectors << 9;
1313 bp->bv1 = bi->bi_io_vec[0];
1314 bp->bv2 = bi->bi_io_vec[0];
1315 bp->bv2.bv_offset += first_sectors << 9;
1316 bp->bv2.bv_len -= first_sectors << 9;
1317 bp->bv1.bv_len = first_sectors << 9;
1319 bp->bio1.bi_io_vec = &bp->bv1;
1320 bp->bio2.bi_io_vec = &bp->bv2;
1322 bp->bio1.bi_max_vecs = 1;
1323 bp->bio2.bi_max_vecs = 1;
1325 bp->bio1.bi_end_io = bio_pair_end_1;
1326 bp->bio2.bi_end_io = bio_pair_end_2;
1328 bp->bio1.bi_private = bi;
1329 bp->bio2.bi_private = bio_split_pool;
1331 if (bio_integrity(bi))
1332 bio_integrity_split(bi, bp, first_sectors);
1338 * bio_sector_offset - Find hardware sector offset in bio
1339 * @bio: bio to inspect
1340 * @index: bio_vec index
1341 * @offset: offset in bv_page
1343 * Return the number of hardware sectors between beginning of bio
1344 * and an end point indicated by a bio_vec index and an offset
1345 * within that vector's page.
1347 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1348 unsigned int offset)
1350 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1357 if (index >= bio->bi_idx)
1358 index = bio->bi_vcnt - 1;
1360 __bio_for_each_segment(bv, bio, i, 0) {
1362 if (offset > bv->bv_offset)
1363 sectors += (offset - bv->bv_offset) / sector_sz;
1367 sectors += bv->bv_len / sector_sz;
1372 EXPORT_SYMBOL(bio_sector_offset);
1375 * create memory pools for biovec's in a bio_set.
1376 * use the global biovec slabs created for general use.
1378 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1380 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1382 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1389 static void biovec_free_pools(struct bio_set *bs)
1391 mempool_destroy(bs->bvec_pool);
1394 void bioset_free(struct bio_set *bs)
1397 mempool_destroy(bs->bio_pool);
1399 bioset_integrity_free(bs);
1400 biovec_free_pools(bs);
1405 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
1409 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1413 bs->bio_slab = bio_slab;
1415 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bs->bio_slab);
1419 if (bioset_integrity_create(bs, bio_pool_size))
1422 if (!biovec_create_pools(bs, bvec_pool_size))
1430 static void __init biovec_init_slabs(void)
1434 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1436 struct biovec_slab *bvs = bvec_slabs + i;
1438 size = bvs->nr_vecs * sizeof(struct bio_vec);
1439 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1440 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1444 static int __init init_bio(void)
1446 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
1448 bio_integrity_init_slab();
1449 biovec_init_slabs();
1451 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
1453 panic("bio: can't allocate bios\n");
1455 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1456 sizeof(struct bio_pair));
1457 if (!bio_split_pool)
1458 panic("bio: can't create split pool\n");
1463 subsys_initcall(init_bio);
1465 EXPORT_SYMBOL(bio_alloc);
1466 EXPORT_SYMBOL(bio_kmalloc);
1467 EXPORT_SYMBOL(bio_put);
1468 EXPORT_SYMBOL(bio_free);
1469 EXPORT_SYMBOL(bio_endio);
1470 EXPORT_SYMBOL(bio_init);
1471 EXPORT_SYMBOL(__bio_clone);
1472 EXPORT_SYMBOL(bio_clone);
1473 EXPORT_SYMBOL(bio_phys_segments);
1474 EXPORT_SYMBOL(bio_add_page);
1475 EXPORT_SYMBOL(bio_add_pc_page);
1476 EXPORT_SYMBOL(bio_get_nr_vecs);
1477 EXPORT_SYMBOL(bio_map_user);
1478 EXPORT_SYMBOL(bio_unmap_user);
1479 EXPORT_SYMBOL(bio_map_kern);
1480 EXPORT_SYMBOL(bio_copy_kern);
1481 EXPORT_SYMBOL(bio_pair_release);
1482 EXPORT_SYMBOL(bio_split);
1483 EXPORT_SYMBOL(bio_copy_user);
1484 EXPORT_SYMBOL(bio_uncopy_user);
1485 EXPORT_SYMBOL(bioset_create);
1486 EXPORT_SYMBOL(bioset_free);
1487 EXPORT_SYMBOL(bio_alloc_bioset);