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 mempool_t *bio_split_pool __read_mostly;
37 * if you change this list, also change bvec_alloc or things will
38 * break badly! cannot be bigger than what you can fit into an
41 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
42 struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
43 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
49 * IO code that does not need private memory pools.
51 struct bio_set *fs_bio_set;
54 * Our slab pool management
57 struct kmem_cache *slab;
58 unsigned int slab_ref;
59 unsigned int slab_size;
62 static DEFINE_MUTEX(bio_slab_lock);
63 static struct bio_slab *bio_slabs;
64 static unsigned int bio_slab_nr, bio_slab_max;
66 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
68 unsigned int sz = sizeof(struct bio) + extra_size;
69 struct kmem_cache *slab = NULL;
70 struct bio_slab *bslab;
71 unsigned int i, entry = -1;
73 mutex_lock(&bio_slab_lock);
76 while (i < bio_slab_nr) {
77 struct bio_slab *bslab = &bio_slabs[i];
79 if (!bslab->slab && entry == -1)
81 else if (bslab->slab_size == sz) {
92 if (bio_slab_nr == bio_slab_max && entry == -1) {
94 bio_slabs = krealloc(bio_slabs,
95 bio_slab_max * sizeof(struct bio_slab),
101 entry = bio_slab_nr++;
103 bslab = &bio_slabs[entry];
105 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
106 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
110 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
113 bslab->slab_size = sz;
115 mutex_unlock(&bio_slab_lock);
119 static void bio_put_slab(struct bio_set *bs)
121 struct bio_slab *bslab = NULL;
124 mutex_lock(&bio_slab_lock);
126 for (i = 0; i < bio_slab_nr; i++) {
127 if (bs->bio_slab == bio_slabs[i].slab) {
128 bslab = &bio_slabs[i];
133 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
136 WARN_ON(!bslab->slab_ref);
138 if (--bslab->slab_ref)
141 kmem_cache_destroy(bslab->slab);
145 mutex_unlock(&bio_slab_lock);
148 unsigned int bvec_nr_vecs(unsigned short idx)
150 return bvec_slabs[idx].nr_vecs;
153 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
155 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
157 if (idx == BIOVEC_MAX_IDX)
158 mempool_free(bv, bs->bvec_pool);
160 struct biovec_slab *bvs = bvec_slabs + idx;
162 kmem_cache_free(bvs->slab, bv);
166 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
172 * If 'bs' is given, lookup the pool and do the mempool alloc.
173 * If not, this is a bio_kmalloc() allocation and just do a
174 * kzalloc() for the exact number of vecs right away.
177 bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask);
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;
235 memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec));
240 void bio_free(struct bio *bio, struct bio_set *bs)
245 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
247 if (bio_integrity(bio))
248 bio_integrity_free(bio, bs);
251 * If we have front padding, adjust the bio pointer before freeing
257 mempool_free(p, bs->bio_pool);
261 * default destructor for a bio allocated with bio_alloc_bioset()
263 static void bio_fs_destructor(struct bio *bio)
265 bio_free(bio, fs_bio_set);
268 static void bio_kmalloc_destructor(struct bio *bio)
270 kfree(bio->bi_io_vec);
274 void bio_init(struct bio *bio)
276 memset(bio, 0, sizeof(*bio));
277 bio->bi_flags = 1 << BIO_UPTODATE;
278 bio->bi_comp_cpu = -1;
279 atomic_set(&bio->bi_cnt, 1);
283 * bio_alloc_bioset - allocate a bio for I/O
284 * @gfp_mask: the GFP_ mask given to the slab allocator
285 * @nr_iovecs: number of iovecs to pre-allocate
286 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
289 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
290 * If %__GFP_WAIT is set then we will block on the internal pool waiting
291 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
292 * fall back to just using @kmalloc to allocate the required memory.
294 * Note that the caller must set ->bi_destructor on succesful return
295 * of a bio, to do the appropriate freeing of the bio once the reference
296 * count drops to zero.
298 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
300 struct bio *bio = NULL;
303 void *p = mempool_alloc(bs->bio_pool, gfp_mask);
306 bio = p + bs->front_pad;
308 bio = kmalloc(sizeof(*bio), gfp_mask);
311 struct bio_vec *bvl = NULL;
314 if (likely(nr_iovecs)) {
315 unsigned long uninitialized_var(idx);
317 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
318 if (unlikely(!bvl)) {
320 mempool_free(bio, bs->bio_pool);
326 bio->bi_flags |= idx << BIO_POOL_OFFSET;
327 bio->bi_max_vecs = bvec_nr_vecs(idx);
329 bio->bi_io_vec = bvl;
335 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
337 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
340 bio->bi_destructor = bio_fs_destructor;
346 * Like bio_alloc(), but doesn't use a mempool backing. This means that
347 * it CAN fail, but while bio_alloc() can only be used for allocations
348 * that have a short (finite) life span, bio_kmalloc() should be used
349 * for more permanent bio allocations (like allocating some bio's for
350 * initalization or setup purposes).
352 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
354 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL);
357 bio->bi_destructor = bio_kmalloc_destructor;
362 void zero_fill_bio(struct bio *bio)
368 bio_for_each_segment(bv, bio, i) {
369 char *data = bvec_kmap_irq(bv, &flags);
370 memset(data, 0, bv->bv_len);
371 flush_dcache_page(bv->bv_page);
372 bvec_kunmap_irq(data, &flags);
375 EXPORT_SYMBOL(zero_fill_bio);
378 * bio_put - release a reference to a bio
379 * @bio: bio to release reference to
382 * Put a reference to a &struct bio, either one you have gotten with
383 * bio_alloc or bio_get. The last put of a bio will free it.
385 void bio_put(struct bio *bio)
387 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
392 if (atomic_dec_and_test(&bio->bi_cnt)) {
394 bio->bi_destructor(bio);
398 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
400 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
401 blk_recount_segments(q, bio);
403 return bio->bi_phys_segments;
407 * __bio_clone - clone a bio
408 * @bio: destination bio
409 * @bio_src: bio to clone
411 * Clone a &bio. Caller will own the returned bio, but not
412 * the actual data it points to. Reference count of returned
415 void __bio_clone(struct bio *bio, struct bio *bio_src)
417 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
418 bio_src->bi_max_vecs * sizeof(struct bio_vec));
421 * most users will be overriding ->bi_bdev with a new target,
422 * so we don't set nor calculate new physical/hw segment counts here
424 bio->bi_sector = bio_src->bi_sector;
425 bio->bi_bdev = bio_src->bi_bdev;
426 bio->bi_flags |= 1 << BIO_CLONED;
427 bio->bi_rw = bio_src->bi_rw;
428 bio->bi_vcnt = bio_src->bi_vcnt;
429 bio->bi_size = bio_src->bi_size;
430 bio->bi_idx = bio_src->bi_idx;
434 * bio_clone - clone a bio
436 * @gfp_mask: allocation priority
438 * Like __bio_clone, only also allocates the returned bio
440 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
442 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
447 b->bi_destructor = bio_fs_destructor;
450 if (bio_integrity(bio)) {
453 ret = bio_integrity_clone(b, bio, fs_bio_set);
463 * bio_get_nr_vecs - return approx number of vecs
466 * Return the approximate number of pages we can send to this target.
467 * There's no guarantee that you will be able to fit this number of pages
468 * into a bio, it does not account for dynamic restrictions that vary
471 int bio_get_nr_vecs(struct block_device *bdev)
473 struct request_queue *q = bdev_get_queue(bdev);
476 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
477 if (nr_pages > q->max_phys_segments)
478 nr_pages = q->max_phys_segments;
479 if (nr_pages > q->max_hw_segments)
480 nr_pages = q->max_hw_segments;
485 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
486 *page, unsigned int len, unsigned int offset,
487 unsigned short max_sectors)
489 int retried_segments = 0;
490 struct bio_vec *bvec;
493 * cloned bio must not modify vec list
495 if (unlikely(bio_flagged(bio, BIO_CLONED)))
498 if (((bio->bi_size + len) >> 9) > max_sectors)
502 * For filesystems with a blocksize smaller than the pagesize
503 * we will often be called with the same page as last time and
504 * a consecutive offset. Optimize this special case.
506 if (bio->bi_vcnt > 0) {
507 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
509 if (page == prev->bv_page &&
510 offset == prev->bv_offset + prev->bv_len) {
513 if (q->merge_bvec_fn) {
514 struct bvec_merge_data bvm = {
515 .bi_bdev = bio->bi_bdev,
516 .bi_sector = bio->bi_sector,
517 .bi_size = bio->bi_size,
521 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
531 if (bio->bi_vcnt >= bio->bi_max_vecs)
535 * we might lose a segment or two here, but rather that than
536 * make this too complex.
539 while (bio->bi_phys_segments >= q->max_phys_segments
540 || bio->bi_phys_segments >= q->max_hw_segments) {
542 if (retried_segments)
545 retried_segments = 1;
546 blk_recount_segments(q, bio);
550 * setup the new entry, we might clear it again later if we
551 * cannot add the page
553 bvec = &bio->bi_io_vec[bio->bi_vcnt];
554 bvec->bv_page = page;
556 bvec->bv_offset = offset;
559 * if queue has other restrictions (eg varying max sector size
560 * depending on offset), it can specify a merge_bvec_fn in the
561 * queue to get further control
563 if (q->merge_bvec_fn) {
564 struct bvec_merge_data bvm = {
565 .bi_bdev = bio->bi_bdev,
566 .bi_sector = bio->bi_sector,
567 .bi_size = bio->bi_size,
572 * merge_bvec_fn() returns number of bytes it can accept
575 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
576 bvec->bv_page = NULL;
583 /* If we may be able to merge these biovecs, force a recount */
584 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
585 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
588 bio->bi_phys_segments++;
595 * bio_add_pc_page - attempt to add page to bio
596 * @q: the target queue
597 * @bio: destination bio
599 * @len: vec entry length
600 * @offset: vec entry offset
602 * Attempt to add a page to the bio_vec maplist. This can fail for a
603 * number of reasons, such as the bio being full or target block
604 * device limitations. The target block device must allow bio's
605 * smaller than PAGE_SIZE, so it is always possible to add a single
606 * page to an empty bio. This should only be used by REQ_PC bios.
608 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
609 unsigned int len, unsigned int offset)
611 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
615 * bio_add_page - attempt to add page to bio
616 * @bio: destination bio
618 * @len: vec entry length
619 * @offset: vec entry offset
621 * Attempt to add a page to the bio_vec maplist. This can fail for a
622 * number of reasons, such as the bio being full or target block
623 * device limitations. The target block device must allow bio's
624 * smaller than PAGE_SIZE, so it is always possible to add a single
625 * page to an empty bio.
627 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
630 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
631 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
634 struct bio_map_data {
635 struct bio_vec *iovecs;
636 struct sg_iovec *sgvecs;
641 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
642 struct sg_iovec *iov, int iov_count,
645 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
646 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
647 bmd->nr_sgvecs = iov_count;
648 bmd->is_our_pages = is_our_pages;
649 bio->bi_private = bmd;
652 static void bio_free_map_data(struct bio_map_data *bmd)
659 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
662 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
667 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
673 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
682 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
683 struct sg_iovec *iov, int iov_count, int uncopy,
687 struct bio_vec *bvec;
689 unsigned int iov_off = 0;
690 int read = bio_data_dir(bio) == READ;
692 __bio_for_each_segment(bvec, bio, i, 0) {
693 char *bv_addr = page_address(bvec->bv_page);
694 unsigned int bv_len = iovecs[i].bv_len;
696 while (bv_len && iov_idx < iov_count) {
700 bytes = min_t(unsigned int,
701 iov[iov_idx].iov_len - iov_off, bv_len);
702 iov_addr = iov[iov_idx].iov_base + iov_off;
705 if (!read && !uncopy)
706 ret = copy_from_user(bv_addr, iov_addr,
709 ret = copy_to_user(iov_addr, bv_addr,
721 if (iov[iov_idx].iov_len == iov_off) {
728 __free_page(bvec->bv_page);
735 * bio_uncopy_user - finish previously mapped bio
736 * @bio: bio being terminated
738 * Free pages allocated from bio_copy_user() and write back data
739 * to user space in case of a read.
741 int bio_uncopy_user(struct bio *bio)
743 struct bio_map_data *bmd = bio->bi_private;
746 if (!bio_flagged(bio, BIO_NULL_MAPPED))
747 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
748 bmd->nr_sgvecs, 1, bmd->is_our_pages);
749 bio_free_map_data(bmd);
755 * bio_copy_user_iov - copy user data to bio
756 * @q: destination block queue
757 * @map_data: pointer to the rq_map_data holding pages (if necessary)
759 * @iov_count: number of elements in the iovec
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_iov(struct request_queue *q,
768 struct rq_map_data *map_data,
769 struct sg_iovec *iov, int iov_count,
770 int write_to_vm, gfp_t gfp_mask)
772 struct bio_map_data *bmd;
773 struct bio_vec *bvec;
778 unsigned int len = 0;
780 for (i = 0; i < iov_count; i++) {
785 uaddr = (unsigned long)iov[i].iov_base;
786 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
787 start = uaddr >> PAGE_SHIFT;
789 nr_pages += end - start;
790 len += iov[i].iov_len;
793 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
795 return ERR_PTR(-ENOMEM);
798 bio = bio_alloc(gfp_mask, nr_pages);
802 bio->bi_rw |= (!write_to_vm << BIO_RW);
810 bytes = 1U << (PAGE_SHIFT + map_data->page_order);
818 if (i == map_data->nr_entries) {
822 page = map_data->pages[i++];
824 page = alloc_page(q->bounce_gfp | gfp_mask);
830 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
843 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
848 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
852 bio_for_each_segment(bvec, bio, i)
853 __free_page(bvec->bv_page);
857 bio_free_map_data(bmd);
862 * bio_copy_user - copy user data to bio
863 * @q: destination block queue
864 * @map_data: pointer to the rq_map_data holding pages (if necessary)
865 * @uaddr: start of user address
866 * @len: length in bytes
867 * @write_to_vm: bool indicating writing to pages or not
868 * @gfp_mask: memory allocation flags
870 * Prepares and returns a bio for indirect user io, bouncing data
871 * to/from kernel pages as necessary. Must be paired with
872 * call bio_uncopy_user() on io completion.
874 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
875 unsigned long uaddr, unsigned int len,
876 int write_to_vm, gfp_t gfp_mask)
880 iov.iov_base = (void __user *)uaddr;
883 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
886 static struct bio *__bio_map_user_iov(struct request_queue *q,
887 struct block_device *bdev,
888 struct sg_iovec *iov, int iov_count,
889 int write_to_vm, gfp_t gfp_mask)
898 for (i = 0; i < iov_count; i++) {
899 unsigned long uaddr = (unsigned long)iov[i].iov_base;
900 unsigned long len = iov[i].iov_len;
901 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
902 unsigned long start = uaddr >> PAGE_SHIFT;
904 nr_pages += end - start;
906 * buffer must be aligned to at least hardsector size for now
908 if (uaddr & queue_dma_alignment(q))
909 return ERR_PTR(-EINVAL);
913 return ERR_PTR(-EINVAL);
915 bio = bio_alloc(gfp_mask, nr_pages);
917 return ERR_PTR(-ENOMEM);
920 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
924 for (i = 0; i < iov_count; i++) {
925 unsigned long uaddr = (unsigned long)iov[i].iov_base;
926 unsigned long len = iov[i].iov_len;
927 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
928 unsigned long start = uaddr >> PAGE_SHIFT;
929 const int local_nr_pages = end - start;
930 const int page_limit = cur_page + local_nr_pages;
932 ret = get_user_pages_fast(uaddr, local_nr_pages,
933 write_to_vm, &pages[cur_page]);
934 if (ret < local_nr_pages) {
939 offset = uaddr & ~PAGE_MASK;
940 for (j = cur_page; j < page_limit; j++) {
941 unsigned int bytes = PAGE_SIZE - offset;
952 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
962 * release the pages we didn't map into the bio, if any
964 while (j < page_limit)
965 page_cache_release(pages[j++]);
971 * set data direction, and check if mapped pages need bouncing
974 bio->bi_rw |= (1 << BIO_RW);
977 bio->bi_flags |= (1 << BIO_USER_MAPPED);
981 for (i = 0; i < nr_pages; i++) {
984 page_cache_release(pages[i]);
993 * bio_map_user - map user address into bio
994 * @q: the struct request_queue for the bio
995 * @bdev: destination block device
996 * @uaddr: start of user address
997 * @len: length in bytes
998 * @write_to_vm: bool indicating writing to pages or not
999 * @gfp_mask: memory allocation flags
1001 * Map the user space address into a bio suitable for io to a block
1002 * device. Returns an error pointer in case of error.
1004 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1005 unsigned long uaddr, unsigned int len, int write_to_vm,
1008 struct sg_iovec iov;
1010 iov.iov_base = (void __user *)uaddr;
1013 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1017 * bio_map_user_iov - map user sg_iovec table into bio
1018 * @q: the struct request_queue for the bio
1019 * @bdev: destination block device
1021 * @iov_count: number of elements in the iovec
1022 * @write_to_vm: bool indicating writing to pages or not
1023 * @gfp_mask: memory allocation flags
1025 * Map the user space address into a bio suitable for io to a block
1026 * device. Returns an error pointer in case of error.
1028 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1029 struct sg_iovec *iov, int iov_count,
1030 int write_to_vm, gfp_t gfp_mask)
1034 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1040 * subtle -- if __bio_map_user() ended up bouncing a bio,
1041 * it would normally disappear when its bi_end_io is run.
1042 * however, we need it for the unmap, so grab an extra
1050 static void __bio_unmap_user(struct bio *bio)
1052 struct bio_vec *bvec;
1056 * make sure we dirty pages we wrote to
1058 __bio_for_each_segment(bvec, bio, i, 0) {
1059 if (bio_data_dir(bio) == READ)
1060 set_page_dirty_lock(bvec->bv_page);
1062 page_cache_release(bvec->bv_page);
1069 * bio_unmap_user - unmap a bio
1070 * @bio: the bio being unmapped
1072 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1073 * a process context.
1075 * bio_unmap_user() may sleep.
1077 void bio_unmap_user(struct bio *bio)
1079 __bio_unmap_user(bio);
1083 static void bio_map_kern_endio(struct bio *bio, int err)
1089 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1090 unsigned int len, gfp_t gfp_mask)
1092 unsigned long kaddr = (unsigned long)data;
1093 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1094 unsigned long start = kaddr >> PAGE_SHIFT;
1095 const int nr_pages = end - start;
1099 bio = bio_alloc(gfp_mask, nr_pages);
1101 return ERR_PTR(-ENOMEM);
1103 offset = offset_in_page(kaddr);
1104 for (i = 0; i < nr_pages; i++) {
1105 unsigned int bytes = PAGE_SIZE - offset;
1113 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1122 bio->bi_end_io = bio_map_kern_endio;
1127 * bio_map_kern - map kernel address into bio
1128 * @q: the struct request_queue for the bio
1129 * @data: pointer to buffer to map
1130 * @len: length in bytes
1131 * @gfp_mask: allocation flags for bio allocation
1133 * Map the kernel address into a bio suitable for io to a block
1134 * device. Returns an error pointer in case of error.
1136 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1141 bio = __bio_map_kern(q, data, len, gfp_mask);
1145 if (bio->bi_size == len)
1149 * Don't support partial mappings.
1152 return ERR_PTR(-EINVAL);
1155 static void bio_copy_kern_endio(struct bio *bio, int err)
1157 struct bio_vec *bvec;
1158 const int read = bio_data_dir(bio) == READ;
1159 struct bio_map_data *bmd = bio->bi_private;
1161 char *p = bmd->sgvecs[0].iov_base;
1163 __bio_for_each_segment(bvec, bio, i, 0) {
1164 char *addr = page_address(bvec->bv_page);
1165 int len = bmd->iovecs[i].bv_len;
1168 memcpy(p, addr, len);
1170 __free_page(bvec->bv_page);
1174 bio_free_map_data(bmd);
1179 * bio_copy_kern - copy kernel address into bio
1180 * @q: the struct request_queue for the bio
1181 * @data: pointer to buffer to copy
1182 * @len: length in bytes
1183 * @gfp_mask: allocation flags for bio and page allocation
1184 * @reading: data direction is READ
1186 * copy 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_copy_kern(struct request_queue *q, void *data, unsigned int len,
1190 gfp_t gfp_mask, int reading)
1193 struct bio_vec *bvec;
1196 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1203 bio_for_each_segment(bvec, bio, i) {
1204 char *addr = page_address(bvec->bv_page);
1206 memcpy(addr, p, bvec->bv_len);
1211 bio->bi_end_io = bio_copy_kern_endio;
1217 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1218 * for performing direct-IO in BIOs.
1220 * The problem is that we cannot run set_page_dirty() from interrupt context
1221 * because the required locks are not interrupt-safe. So what we can do is to
1222 * mark the pages dirty _before_ performing IO. And in interrupt context,
1223 * check that the pages are still dirty. If so, fine. If not, redirty them
1224 * in process context.
1226 * We special-case compound pages here: normally this means reads into hugetlb
1227 * pages. The logic in here doesn't really work right for compound pages
1228 * because the VM does not uniformly chase down the head page in all cases.
1229 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1230 * handle them at all. So we skip compound pages here at an early stage.
1232 * Note that this code is very hard to test under normal circumstances because
1233 * direct-io pins the pages with get_user_pages(). This makes
1234 * is_page_cache_freeable return false, and the VM will not clean the pages.
1235 * But other code (eg, pdflush) could clean the pages if they are mapped
1238 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1239 * deferred bio dirtying paths.
1243 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1245 void bio_set_pages_dirty(struct bio *bio)
1247 struct bio_vec *bvec = bio->bi_io_vec;
1250 for (i = 0; i < bio->bi_vcnt; i++) {
1251 struct page *page = bvec[i].bv_page;
1253 if (page && !PageCompound(page))
1254 set_page_dirty_lock(page);
1258 static void bio_release_pages(struct bio *bio)
1260 struct bio_vec *bvec = bio->bi_io_vec;
1263 for (i = 0; i < bio->bi_vcnt; i++) {
1264 struct page *page = bvec[i].bv_page;
1272 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1273 * If they are, then fine. If, however, some pages are clean then they must
1274 * have been written out during the direct-IO read. So we take another ref on
1275 * the BIO and the offending pages and re-dirty the pages in process context.
1277 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1278 * here on. It will run one page_cache_release() against each page and will
1279 * run one bio_put() against the BIO.
1282 static void bio_dirty_fn(struct work_struct *work);
1284 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1285 static DEFINE_SPINLOCK(bio_dirty_lock);
1286 static struct bio *bio_dirty_list;
1289 * This runs in process context
1291 static void bio_dirty_fn(struct work_struct *work)
1293 unsigned long flags;
1296 spin_lock_irqsave(&bio_dirty_lock, flags);
1297 bio = bio_dirty_list;
1298 bio_dirty_list = NULL;
1299 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1302 struct bio *next = bio->bi_private;
1304 bio_set_pages_dirty(bio);
1305 bio_release_pages(bio);
1311 void bio_check_pages_dirty(struct bio *bio)
1313 struct bio_vec *bvec = bio->bi_io_vec;
1314 int nr_clean_pages = 0;
1317 for (i = 0; i < bio->bi_vcnt; i++) {
1318 struct page *page = bvec[i].bv_page;
1320 if (PageDirty(page) || PageCompound(page)) {
1321 page_cache_release(page);
1322 bvec[i].bv_page = NULL;
1328 if (nr_clean_pages) {
1329 unsigned long flags;
1331 spin_lock_irqsave(&bio_dirty_lock, flags);
1332 bio->bi_private = bio_dirty_list;
1333 bio_dirty_list = bio;
1334 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1335 schedule_work(&bio_dirty_work);
1342 * bio_endio - end I/O on a bio
1344 * @error: error, if any
1347 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1348 * preferred way to end I/O on a bio, it takes care of clearing
1349 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1350 * established -Exxxx (-EIO, for instance) error values in case
1351 * something went wrong. Noone should call bi_end_io() directly on a
1352 * bio unless they own it and thus know that it has an end_io
1355 void bio_endio(struct bio *bio, int error)
1358 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1359 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1363 bio->bi_end_io(bio, error);
1366 void bio_pair_release(struct bio_pair *bp)
1368 if (atomic_dec_and_test(&bp->cnt)) {
1369 struct bio *master = bp->bio1.bi_private;
1371 bio_endio(master, bp->error);
1372 mempool_free(bp, bp->bio2.bi_private);
1376 static void bio_pair_end_1(struct bio *bi, int err)
1378 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1383 bio_pair_release(bp);
1386 static void bio_pair_end_2(struct bio *bi, int err)
1388 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1393 bio_pair_release(bp);
1397 * split a bio - only worry about a bio with a single page
1400 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1402 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1407 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1408 bi->bi_sector + first_sectors);
1410 BUG_ON(bi->bi_vcnt != 1);
1411 BUG_ON(bi->bi_idx != 0);
1412 atomic_set(&bp->cnt, 3);
1416 bp->bio2.bi_sector += first_sectors;
1417 bp->bio2.bi_size -= first_sectors << 9;
1418 bp->bio1.bi_size = first_sectors << 9;
1420 bp->bv1 = bi->bi_io_vec[0];
1421 bp->bv2 = bi->bi_io_vec[0];
1422 bp->bv2.bv_offset += first_sectors << 9;
1423 bp->bv2.bv_len -= first_sectors << 9;
1424 bp->bv1.bv_len = first_sectors << 9;
1426 bp->bio1.bi_io_vec = &bp->bv1;
1427 bp->bio2.bi_io_vec = &bp->bv2;
1429 bp->bio1.bi_max_vecs = 1;
1430 bp->bio2.bi_max_vecs = 1;
1432 bp->bio1.bi_end_io = bio_pair_end_1;
1433 bp->bio2.bi_end_io = bio_pair_end_2;
1435 bp->bio1.bi_private = bi;
1436 bp->bio2.bi_private = bio_split_pool;
1438 if (bio_integrity(bi))
1439 bio_integrity_split(bi, bp, first_sectors);
1445 * bio_sector_offset - Find hardware sector offset in bio
1446 * @bio: bio to inspect
1447 * @index: bio_vec index
1448 * @offset: offset in bv_page
1450 * Return the number of hardware sectors between beginning of bio
1451 * and an end point indicated by a bio_vec index and an offset
1452 * within that vector's page.
1454 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1455 unsigned int offset)
1457 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1464 if (index >= bio->bi_idx)
1465 index = bio->bi_vcnt - 1;
1467 __bio_for_each_segment(bv, bio, i, 0) {
1469 if (offset > bv->bv_offset)
1470 sectors += (offset - bv->bv_offset) / sector_sz;
1474 sectors += bv->bv_len / sector_sz;
1479 EXPORT_SYMBOL(bio_sector_offset);
1482 * create memory pools for biovec's in a bio_set.
1483 * use the global biovec slabs created for general use.
1485 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1487 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1489 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1496 static void biovec_free_pools(struct bio_set *bs)
1498 mempool_destroy(bs->bvec_pool);
1501 void bioset_free(struct bio_set *bs)
1504 mempool_destroy(bs->bio_pool);
1506 bioset_integrity_free(bs);
1507 biovec_free_pools(bs);
1514 * bioset_create - Create a bio_set
1515 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1516 * @front_pad: Number of bytes to allocate in front of the returned bio
1519 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1520 * to ask for a number of bytes to be allocated in front of the bio.
1521 * Front pad allocation is useful for embedding the bio inside
1522 * another structure, to avoid allocating extra data to go with the bio.
1523 * Note that the bio must be embedded at the END of that structure always,
1524 * or things will break badly.
1526 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1530 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1534 bs->front_pad = front_pad;
1536 bs->bio_slab = bio_find_or_create_slab(front_pad);
1537 if (!bs->bio_slab) {
1542 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1546 if (bioset_integrity_create(bs, pool_size))
1549 if (!biovec_create_pools(bs, pool_size))
1557 static void __init biovec_init_slabs(void)
1561 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1563 struct biovec_slab *bvs = bvec_slabs + i;
1565 size = bvs->nr_vecs * sizeof(struct bio_vec);
1566 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1567 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1571 static int __init init_bio(void)
1575 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1577 panic("bio: can't allocate bios\n");
1579 bio_integrity_init_slab();
1580 biovec_init_slabs();
1582 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1584 panic("bio: can't allocate bios\n");
1586 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1587 sizeof(struct bio_pair));
1588 if (!bio_split_pool)
1589 panic("bio: can't create split pool\n");
1594 subsys_initcall(init_bio);
1596 EXPORT_SYMBOL(bio_alloc);
1597 EXPORT_SYMBOL(bio_kmalloc);
1598 EXPORT_SYMBOL(bio_put);
1599 EXPORT_SYMBOL(bio_free);
1600 EXPORT_SYMBOL(bio_endio);
1601 EXPORT_SYMBOL(bio_init);
1602 EXPORT_SYMBOL(__bio_clone);
1603 EXPORT_SYMBOL(bio_clone);
1604 EXPORT_SYMBOL(bio_phys_segments);
1605 EXPORT_SYMBOL(bio_add_page);
1606 EXPORT_SYMBOL(bio_add_pc_page);
1607 EXPORT_SYMBOL(bio_get_nr_vecs);
1608 EXPORT_SYMBOL(bio_map_user);
1609 EXPORT_SYMBOL(bio_unmap_user);
1610 EXPORT_SYMBOL(bio_map_kern);
1611 EXPORT_SYMBOL(bio_copy_kern);
1612 EXPORT_SYMBOL(bio_pair_release);
1613 EXPORT_SYMBOL(bio_split);
1614 EXPORT_SYMBOL(bio_copy_user);
1615 EXPORT_SYMBOL(bio_uncopy_user);
1616 EXPORT_SYMBOL(bioset_create);
1617 EXPORT_SYMBOL(bioset_free);
1618 EXPORT_SYMBOL(bio_alloc_bioset);