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 <scsi/sg.h> /* for struct sg_iovec */
30 #include <trace/events/block.h>
33 * Test patch to inline a certain number of bi_io_vec's inside the bio
34 * itself, to shrink a bio data allocation from two mempool calls to one
36 #define BIO_INLINE_VECS 4
38 static mempool_t *bio_split_pool __read_mostly;
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
45 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
47 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
55 struct bio_set *fs_bio_set;
58 * Our slab pool management
61 struct kmem_cache *slab;
62 unsigned int slab_ref;
63 unsigned int slab_size;
66 static DEFINE_MUTEX(bio_slab_lock);
67 static struct bio_slab *bio_slabs;
68 static unsigned int bio_slab_nr, bio_slab_max;
70 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72 unsigned int sz = sizeof(struct bio) + extra_size;
73 struct kmem_cache *slab = NULL;
74 struct bio_slab *bslab;
75 unsigned int i, entry = -1;
77 mutex_lock(&bio_slab_lock);
80 while (i < bio_slab_nr) {
81 bslab = &bio_slabs[i];
83 if (!bslab->slab && entry == -1)
85 else if (bslab->slab_size == sz) {
96 if (bio_slab_nr == bio_slab_max && entry == -1) {
98 bio_slabs = krealloc(bio_slabs,
99 bio_slab_max * sizeof(struct bio_slab),
105 entry = bio_slab_nr++;
107 bslab = &bio_slabs[entry];
109 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
110 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
114 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
117 bslab->slab_size = sz;
119 mutex_unlock(&bio_slab_lock);
123 static void bio_put_slab(struct bio_set *bs)
125 struct bio_slab *bslab = NULL;
128 mutex_lock(&bio_slab_lock);
130 for (i = 0; i < bio_slab_nr; i++) {
131 if (bs->bio_slab == bio_slabs[i].slab) {
132 bslab = &bio_slabs[i];
137 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
140 WARN_ON(!bslab->slab_ref);
142 if (--bslab->slab_ref)
145 kmem_cache_destroy(bslab->slab);
149 mutex_unlock(&bio_slab_lock);
152 unsigned int bvec_nr_vecs(unsigned short idx)
154 return bvec_slabs[idx].nr_vecs;
157 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
159 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
161 if (idx == BIOVEC_MAX_IDX)
162 mempool_free(bv, bs->bvec_pool);
164 struct biovec_slab *bvs = bvec_slabs + idx;
166 kmem_cache_free(bvs->slab, bv);
170 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
176 * see comment near bvec_array define!
194 case 129 ... BIO_MAX_PAGES:
202 * idx now points to the pool we want to allocate from. only the
203 * 1-vec entry pool is mempool backed.
205 if (*idx == BIOVEC_MAX_IDX) {
207 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
209 struct biovec_slab *bvs = bvec_slabs + *idx;
210 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
213 * Make this allocation restricted and don't dump info on
214 * allocation failures, since we'll fallback to the mempool
215 * in case of failure.
217 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
220 * Try a slab allocation. If this fails and __GFP_WAIT
221 * is set, retry with the 1-entry mempool
223 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
224 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
225 *idx = BIOVEC_MAX_IDX;
233 void bio_free(struct bio *bio, struct bio_set *bs)
237 if (bio_has_allocated_vec(bio))
238 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
240 if (bio_integrity(bio))
241 bio_integrity_free(bio, bs);
244 * If we have front padding, adjust the bio pointer before freeing
250 mempool_free(p, bs->bio_pool);
252 EXPORT_SYMBOL(bio_free);
254 void bio_init(struct bio *bio)
256 memset(bio, 0, sizeof(*bio));
257 bio->bi_flags = 1 << BIO_UPTODATE;
258 bio->bi_comp_cpu = -1;
259 atomic_set(&bio->bi_cnt, 1);
261 EXPORT_SYMBOL(bio_init);
264 * bio_alloc_bioset - allocate a bio for I/O
265 * @gfp_mask: the GFP_ mask given to the slab allocator
266 * @nr_iovecs: number of iovecs to pre-allocate
267 * @bs: the bio_set to allocate from.
270 * bio_alloc_bioset will try its own mempool to satisfy the allocation.
271 * If %__GFP_WAIT is set then we will block on the internal pool waiting
272 * for a &struct bio to become free.
274 * Note that the caller must set ->bi_destructor on successful return
275 * of a bio, to do the appropriate freeing of the bio once the reference
276 * count drops to zero.
278 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
280 unsigned long idx = BIO_POOL_NONE;
281 struct bio_vec *bvl = NULL;
285 p = mempool_alloc(bs->bio_pool, gfp_mask);
288 bio = p + bs->front_pad;
292 if (unlikely(!nr_iovecs))
295 if (nr_iovecs <= BIO_INLINE_VECS) {
296 bvl = bio->bi_inline_vecs;
297 nr_iovecs = BIO_INLINE_VECS;
299 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
303 nr_iovecs = bvec_nr_vecs(idx);
306 bio->bi_flags |= idx << BIO_POOL_OFFSET;
307 bio->bi_max_vecs = nr_iovecs;
308 bio->bi_io_vec = bvl;
312 mempool_free(p, bs->bio_pool);
315 EXPORT_SYMBOL(bio_alloc_bioset);
317 static void bio_fs_destructor(struct bio *bio)
319 bio_free(bio, fs_bio_set);
323 * bio_alloc - allocate a new bio, memory pool backed
324 * @gfp_mask: allocation mask to use
325 * @nr_iovecs: number of iovecs
327 * bio_alloc will allocate a bio and associated bio_vec array that can hold
328 * at least @nr_iovecs entries. Allocations will be done from the
329 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
331 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
332 * a bio. This is due to the mempool guarantees. To make this work, callers
333 * must never allocate more than 1 bio at a time from this pool. Callers
334 * that need to allocate more than 1 bio must always submit the previously
335 * allocated bio for IO before attempting to allocate a new one. Failure to
336 * do so can cause livelocks under memory pressure.
339 * Pointer to new bio on success, NULL on failure.
341 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
343 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
346 bio->bi_destructor = bio_fs_destructor;
350 EXPORT_SYMBOL(bio_alloc);
352 static void bio_kmalloc_destructor(struct bio *bio)
354 if (bio_integrity(bio))
355 bio_integrity_free(bio, fs_bio_set);
360 * bio_kmalloc - allocate a bio for I/O using kmalloc()
361 * @gfp_mask: the GFP_ mask given to the slab allocator
362 * @nr_iovecs: number of iovecs to pre-allocate
365 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
366 * %__GFP_WAIT, the allocation is guaranteed to succeed.
369 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
373 if (nr_iovecs > UIO_MAXIOV)
376 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
382 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
383 bio->bi_max_vecs = nr_iovecs;
384 bio->bi_io_vec = bio->bi_inline_vecs;
385 bio->bi_destructor = bio_kmalloc_destructor;
389 EXPORT_SYMBOL(bio_kmalloc);
391 void zero_fill_bio(struct bio *bio)
397 bio_for_each_segment(bv, bio, i) {
398 char *data = bvec_kmap_irq(bv, &flags);
399 memset(data, 0, bv->bv_len);
400 flush_dcache_page(bv->bv_page);
401 bvec_kunmap_irq(data, &flags);
404 EXPORT_SYMBOL(zero_fill_bio);
407 * bio_put - release a reference to a bio
408 * @bio: bio to release reference to
411 * Put a reference to a &struct bio, either one you have gotten with
412 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
414 void bio_put(struct bio *bio)
416 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
421 if (atomic_dec_and_test(&bio->bi_cnt)) {
423 bio->bi_destructor(bio);
426 EXPORT_SYMBOL(bio_put);
428 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
430 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
431 blk_recount_segments(q, bio);
433 return bio->bi_phys_segments;
435 EXPORT_SYMBOL(bio_phys_segments);
438 * __bio_clone - clone a bio
439 * @bio: destination bio
440 * @bio_src: bio to clone
442 * Clone a &bio. Caller will own the returned bio, but not
443 * the actual data it points to. Reference count of returned
446 void __bio_clone(struct bio *bio, struct bio *bio_src)
448 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
449 bio_src->bi_max_vecs * sizeof(struct bio_vec));
452 * most users will be overriding ->bi_bdev with a new target,
453 * so we don't set nor calculate new physical/hw segment counts here
455 bio->bi_sector = bio_src->bi_sector;
456 bio->bi_bdev = bio_src->bi_bdev;
457 bio->bi_flags |= 1 << BIO_CLONED;
458 bio->bi_rw = bio_src->bi_rw;
459 bio->bi_vcnt = bio_src->bi_vcnt;
460 bio->bi_size = bio_src->bi_size;
461 bio->bi_idx = bio_src->bi_idx;
463 EXPORT_SYMBOL(__bio_clone);
466 * bio_clone - clone a bio
468 * @gfp_mask: allocation priority
470 * Like __bio_clone, only also allocates the returned bio
472 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
474 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
479 b->bi_destructor = bio_fs_destructor;
482 if (bio_integrity(bio)) {
485 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
495 EXPORT_SYMBOL(bio_clone);
498 * bio_get_nr_vecs - return approx number of vecs
501 * Return the approximate number of pages we can send to this target.
502 * There's no guarantee that you will be able to fit this number of pages
503 * into a bio, it does not account for dynamic restrictions that vary
506 int bio_get_nr_vecs(struct block_device *bdev)
508 struct request_queue *q = bdev_get_queue(bdev);
511 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
512 if (nr_pages > queue_max_segments(q))
513 nr_pages = queue_max_segments(q);
517 EXPORT_SYMBOL(bio_get_nr_vecs);
519 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
520 *page, unsigned int len, unsigned int offset,
521 unsigned short max_sectors)
523 int retried_segments = 0;
524 struct bio_vec *bvec;
527 * cloned bio must not modify vec list
529 if (unlikely(bio_flagged(bio, BIO_CLONED)))
532 if (((bio->bi_size + len) >> 9) > max_sectors)
536 * For filesystems with a blocksize smaller than the pagesize
537 * we will often be called with the same page as last time and
538 * a consecutive offset. Optimize this special case.
540 if (bio->bi_vcnt > 0) {
541 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
543 if (page == prev->bv_page &&
544 offset == prev->bv_offset + prev->bv_len) {
545 unsigned int prev_bv_len = prev->bv_len;
548 if (q->merge_bvec_fn) {
549 struct bvec_merge_data bvm = {
550 /* prev_bvec is already charged in
551 bi_size, discharge it in order to
552 simulate merging updated prev_bvec
554 .bi_bdev = bio->bi_bdev,
555 .bi_sector = bio->bi_sector,
556 .bi_size = bio->bi_size - prev_bv_len,
560 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
570 if (bio->bi_vcnt >= bio->bi_max_vecs)
574 * we might lose a segment or two here, but rather that than
575 * make this too complex.
578 while (bio->bi_phys_segments >= queue_max_segments(q)) {
580 if (retried_segments)
583 retried_segments = 1;
584 blk_recount_segments(q, bio);
588 * setup the new entry, we might clear it again later if we
589 * cannot add the page
591 bvec = &bio->bi_io_vec[bio->bi_vcnt];
592 bvec->bv_page = page;
594 bvec->bv_offset = offset;
597 * if queue has other restrictions (eg varying max sector size
598 * depending on offset), it can specify a merge_bvec_fn in the
599 * queue to get further control
601 if (q->merge_bvec_fn) {
602 struct bvec_merge_data bvm = {
603 .bi_bdev = bio->bi_bdev,
604 .bi_sector = bio->bi_sector,
605 .bi_size = bio->bi_size,
610 * merge_bvec_fn() returns number of bytes it can accept
613 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
614 bvec->bv_page = NULL;
621 /* If we may be able to merge these biovecs, force a recount */
622 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
623 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
626 bio->bi_phys_segments++;
633 * bio_add_pc_page - attempt to add page to bio
634 * @q: the target queue
635 * @bio: destination bio
637 * @len: vec entry length
638 * @offset: vec entry offset
640 * Attempt to add a page to the bio_vec maplist. This can fail for a
641 * number of reasons, such as the bio being full or target block
642 * device limitations. The target block device must allow bio's
643 * smaller than PAGE_SIZE, so it is always possible to add a single
644 * page to an empty bio. This should only be used by REQ_PC bios.
646 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
647 unsigned int len, unsigned int offset)
649 return __bio_add_page(q, bio, page, len, offset,
650 queue_max_hw_sectors(q));
652 EXPORT_SYMBOL(bio_add_pc_page);
655 * bio_add_page - attempt to add page to bio
656 * @bio: destination bio
658 * @len: vec entry length
659 * @offset: vec entry offset
661 * Attempt to add a page to the bio_vec maplist. This can fail for a
662 * number of reasons, such as the bio being full or target block
663 * device limitations. The target block device must allow bio's
664 * smaller than PAGE_SIZE, so it is always possible to add a single
665 * page to an empty bio.
667 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
670 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
671 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
673 EXPORT_SYMBOL(bio_add_page);
675 struct bio_map_data {
676 struct bio_vec *iovecs;
677 struct sg_iovec *sgvecs;
682 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
683 struct sg_iovec *iov, int iov_count,
686 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
687 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
688 bmd->nr_sgvecs = iov_count;
689 bmd->is_our_pages = is_our_pages;
690 bio->bi_private = bmd;
693 static void bio_free_map_data(struct bio_map_data *bmd)
700 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
703 struct bio_map_data *bmd;
705 if (iov_count > UIO_MAXIOV)
708 bmd = kmalloc(sizeof(*bmd), gfp_mask);
712 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
718 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
727 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
728 struct sg_iovec *iov, int iov_count,
729 int to_user, int from_user, int do_free_page)
732 struct bio_vec *bvec;
734 unsigned int iov_off = 0;
736 __bio_for_each_segment(bvec, bio, i, 0) {
737 char *bv_addr = page_address(bvec->bv_page);
738 unsigned int bv_len = iovecs[i].bv_len;
740 while (bv_len && iov_idx < iov_count) {
742 char __user *iov_addr;
744 bytes = min_t(unsigned int,
745 iov[iov_idx].iov_len - iov_off, bv_len);
746 iov_addr = iov[iov_idx].iov_base + iov_off;
750 ret = copy_to_user(iov_addr, bv_addr,
754 ret = copy_from_user(bv_addr, iov_addr,
766 if (iov[iov_idx].iov_len == iov_off) {
773 __free_page(bvec->bv_page);
780 * bio_uncopy_user - finish previously mapped bio
781 * @bio: bio being terminated
783 * Free pages allocated from bio_copy_user() and write back data
784 * to user space in case of a read.
786 int bio_uncopy_user(struct bio *bio)
788 struct bio_map_data *bmd = bio->bi_private;
791 if (!bio_flagged(bio, BIO_NULL_MAPPED))
792 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
793 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
794 0, bmd->is_our_pages);
795 bio_free_map_data(bmd);
799 EXPORT_SYMBOL(bio_uncopy_user);
802 * bio_copy_user_iov - copy user data to bio
803 * @q: destination block queue
804 * @map_data: pointer to the rq_map_data holding pages (if necessary)
806 * @iov_count: number of elements in the iovec
807 * @write_to_vm: bool indicating writing to pages or not
808 * @gfp_mask: memory allocation flags
810 * Prepares and returns a bio for indirect user io, bouncing data
811 * to/from kernel pages as necessary. Must be paired with
812 * call bio_uncopy_user() on io completion.
814 struct bio *bio_copy_user_iov(struct request_queue *q,
815 struct rq_map_data *map_data,
816 struct sg_iovec *iov, int iov_count,
817 int write_to_vm, gfp_t gfp_mask)
819 struct bio_map_data *bmd;
820 struct bio_vec *bvec;
825 unsigned int len = 0;
826 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
828 for (i = 0; i < iov_count; i++) {
833 uaddr = (unsigned long)iov[i].iov_base;
834 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
835 start = uaddr >> PAGE_SHIFT;
841 return ERR_PTR(-EINVAL);
843 nr_pages += end - start;
844 len += iov[i].iov_len;
850 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
852 return ERR_PTR(-ENOMEM);
855 bio = bio_kmalloc(gfp_mask, nr_pages);
860 bio->bi_rw |= REQ_WRITE;
865 nr_pages = 1 << map_data->page_order;
866 i = map_data->offset / PAGE_SIZE;
869 unsigned int bytes = PAGE_SIZE;
877 if (i == map_data->nr_entries * nr_pages) {
882 page = map_data->pages[i / nr_pages];
883 page += (i % nr_pages);
887 page = alloc_page(q->bounce_gfp | gfp_mask);
894 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
907 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
908 (map_data && map_data->from_user)) {
909 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
914 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
918 bio_for_each_segment(bvec, bio, i)
919 __free_page(bvec->bv_page);
923 bio_free_map_data(bmd);
928 * bio_copy_user - copy user data to bio
929 * @q: destination block queue
930 * @map_data: pointer to the rq_map_data holding pages (if necessary)
931 * @uaddr: start of user address
932 * @len: length in bytes
933 * @write_to_vm: bool indicating writing to pages or not
934 * @gfp_mask: memory allocation flags
936 * Prepares and returns a bio for indirect user io, bouncing data
937 * to/from kernel pages as necessary. Must be paired with
938 * call bio_uncopy_user() on io completion.
940 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
941 unsigned long uaddr, unsigned int len,
942 int write_to_vm, gfp_t gfp_mask)
946 iov.iov_base = (void __user *)uaddr;
949 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
951 EXPORT_SYMBOL(bio_copy_user);
953 static struct bio *__bio_map_user_iov(struct request_queue *q,
954 struct block_device *bdev,
955 struct sg_iovec *iov, int iov_count,
956 int write_to_vm, gfp_t gfp_mask)
965 for (i = 0; i < iov_count; i++) {
966 unsigned long uaddr = (unsigned long)iov[i].iov_base;
967 unsigned long len = iov[i].iov_len;
968 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
969 unsigned long start = uaddr >> PAGE_SHIFT;
975 return ERR_PTR(-EINVAL);
977 nr_pages += end - start;
979 * buffer must be aligned to at least hardsector size for now
981 if (uaddr & queue_dma_alignment(q))
982 return ERR_PTR(-EINVAL);
986 return ERR_PTR(-EINVAL);
988 bio = bio_kmalloc(gfp_mask, nr_pages);
990 return ERR_PTR(-ENOMEM);
993 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
997 for (i = 0; i < iov_count; i++) {
998 unsigned long uaddr = (unsigned long)iov[i].iov_base;
999 unsigned long len = iov[i].iov_len;
1000 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1001 unsigned long start = uaddr >> PAGE_SHIFT;
1002 const int local_nr_pages = end - start;
1003 const int page_limit = cur_page + local_nr_pages;
1005 ret = get_user_pages_fast(uaddr, local_nr_pages,
1006 write_to_vm, &pages[cur_page]);
1007 if (ret < local_nr_pages) {
1012 offset = uaddr & ~PAGE_MASK;
1013 for (j = cur_page; j < page_limit; j++) {
1014 unsigned int bytes = PAGE_SIZE - offset;
1025 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1035 * release the pages we didn't map into the bio, if any
1037 while (j < page_limit)
1038 page_cache_release(pages[j++]);
1044 * set data direction, and check if mapped pages need bouncing
1047 bio->bi_rw |= REQ_WRITE;
1049 bio->bi_bdev = bdev;
1050 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1054 for (i = 0; i < nr_pages; i++) {
1057 page_cache_release(pages[i]);
1062 return ERR_PTR(ret);
1066 * bio_map_user - map user address into bio
1067 * @q: the struct request_queue for the bio
1068 * @bdev: destination block device
1069 * @uaddr: start of user address
1070 * @len: length in bytes
1071 * @write_to_vm: bool indicating writing to pages or not
1072 * @gfp_mask: memory allocation flags
1074 * Map the user space address into a bio suitable for io to a block
1075 * device. Returns an error pointer in case of error.
1077 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1078 unsigned long uaddr, unsigned int len, int write_to_vm,
1081 struct sg_iovec iov;
1083 iov.iov_base = (void __user *)uaddr;
1086 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1088 EXPORT_SYMBOL(bio_map_user);
1091 * bio_map_user_iov - map user sg_iovec table into bio
1092 * @q: the struct request_queue for the bio
1093 * @bdev: destination block device
1095 * @iov_count: number of elements in the iovec
1096 * @write_to_vm: bool indicating writing to pages or not
1097 * @gfp_mask: memory allocation flags
1099 * Map the user space address into a bio suitable for io to a block
1100 * device. Returns an error pointer in case of error.
1102 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1103 struct sg_iovec *iov, int iov_count,
1104 int write_to_vm, gfp_t gfp_mask)
1108 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1114 * subtle -- if __bio_map_user() ended up bouncing a bio,
1115 * it would normally disappear when its bi_end_io is run.
1116 * however, we need it for the unmap, so grab an extra
1124 static void __bio_unmap_user(struct bio *bio)
1126 struct bio_vec *bvec;
1130 * make sure we dirty pages we wrote to
1132 __bio_for_each_segment(bvec, bio, i, 0) {
1133 if (bio_data_dir(bio) == READ)
1134 set_page_dirty_lock(bvec->bv_page);
1136 page_cache_release(bvec->bv_page);
1143 * bio_unmap_user - unmap a bio
1144 * @bio: the bio being unmapped
1146 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1147 * a process context.
1149 * bio_unmap_user() may sleep.
1151 void bio_unmap_user(struct bio *bio)
1153 __bio_unmap_user(bio);
1156 EXPORT_SYMBOL(bio_unmap_user);
1158 static void bio_map_kern_endio(struct bio *bio, int err)
1163 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1164 unsigned int len, gfp_t gfp_mask)
1166 unsigned long kaddr = (unsigned long)data;
1167 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1168 unsigned long start = kaddr >> PAGE_SHIFT;
1169 const int nr_pages = end - start;
1173 bio = bio_kmalloc(gfp_mask, nr_pages);
1175 return ERR_PTR(-ENOMEM);
1177 offset = offset_in_page(kaddr);
1178 for (i = 0; i < nr_pages; i++) {
1179 unsigned int bytes = PAGE_SIZE - offset;
1187 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1196 bio->bi_end_io = bio_map_kern_endio;
1201 * bio_map_kern - map kernel address into bio
1202 * @q: the struct request_queue for the bio
1203 * @data: pointer to buffer to map
1204 * @len: length in bytes
1205 * @gfp_mask: allocation flags for bio allocation
1207 * Map the kernel address into a bio suitable for io to a block
1208 * device. Returns an error pointer in case of error.
1210 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1215 bio = __bio_map_kern(q, data, len, gfp_mask);
1219 if (bio->bi_size == len)
1223 * Don't support partial mappings.
1226 return ERR_PTR(-EINVAL);
1228 EXPORT_SYMBOL(bio_map_kern);
1230 static void bio_copy_kern_endio(struct bio *bio, int err)
1232 struct bio_vec *bvec;
1233 const int read = bio_data_dir(bio) == READ;
1234 struct bio_map_data *bmd = bio->bi_private;
1236 char *p = bmd->sgvecs[0].iov_base;
1238 __bio_for_each_segment(bvec, bio, i, 0) {
1239 char *addr = page_address(bvec->bv_page);
1240 int len = bmd->iovecs[i].bv_len;
1243 memcpy(p, addr, len);
1245 __free_page(bvec->bv_page);
1249 bio_free_map_data(bmd);
1254 * bio_copy_kern - copy kernel address into bio
1255 * @q: the struct request_queue for the bio
1256 * @data: pointer to buffer to copy
1257 * @len: length in bytes
1258 * @gfp_mask: allocation flags for bio and page allocation
1259 * @reading: data direction is READ
1261 * copy the kernel address into a bio suitable for io to a block
1262 * device. Returns an error pointer in case of error.
1264 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1265 gfp_t gfp_mask, int reading)
1268 struct bio_vec *bvec;
1271 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1278 bio_for_each_segment(bvec, bio, i) {
1279 char *addr = page_address(bvec->bv_page);
1281 memcpy(addr, p, bvec->bv_len);
1286 bio->bi_end_io = bio_copy_kern_endio;
1290 EXPORT_SYMBOL(bio_copy_kern);
1293 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1294 * for performing direct-IO in BIOs.
1296 * The problem is that we cannot run set_page_dirty() from interrupt context
1297 * because the required locks are not interrupt-safe. So what we can do is to
1298 * mark the pages dirty _before_ performing IO. And in interrupt context,
1299 * check that the pages are still dirty. If so, fine. If not, redirty them
1300 * in process context.
1302 * We special-case compound pages here: normally this means reads into hugetlb
1303 * pages. The logic in here doesn't really work right for compound pages
1304 * because the VM does not uniformly chase down the head page in all cases.
1305 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1306 * handle them at all. So we skip compound pages here at an early stage.
1308 * Note that this code is very hard to test under normal circumstances because
1309 * direct-io pins the pages with get_user_pages(). This makes
1310 * is_page_cache_freeable return false, and the VM will not clean the pages.
1311 * But other code (eg, pdflush) could clean the pages if they are mapped
1314 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1315 * deferred bio dirtying paths.
1319 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1321 void bio_set_pages_dirty(struct bio *bio)
1323 struct bio_vec *bvec = bio->bi_io_vec;
1326 for (i = 0; i < bio->bi_vcnt; i++) {
1327 struct page *page = bvec[i].bv_page;
1329 if (page && !PageCompound(page))
1330 set_page_dirty_lock(page);
1334 static void bio_release_pages(struct bio *bio)
1336 struct bio_vec *bvec = bio->bi_io_vec;
1339 for (i = 0; i < bio->bi_vcnt; i++) {
1340 struct page *page = bvec[i].bv_page;
1348 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1349 * If they are, then fine. If, however, some pages are clean then they must
1350 * have been written out during the direct-IO read. So we take another ref on
1351 * the BIO and the offending pages and re-dirty the pages in process context.
1353 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1354 * here on. It will run one page_cache_release() against each page and will
1355 * run one bio_put() against the BIO.
1358 static void bio_dirty_fn(struct work_struct *work);
1360 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1361 static DEFINE_SPINLOCK(bio_dirty_lock);
1362 static struct bio *bio_dirty_list;
1365 * This runs in process context
1367 static void bio_dirty_fn(struct work_struct *work)
1369 unsigned long flags;
1372 spin_lock_irqsave(&bio_dirty_lock, flags);
1373 bio = bio_dirty_list;
1374 bio_dirty_list = NULL;
1375 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1378 struct bio *next = bio->bi_private;
1380 bio_set_pages_dirty(bio);
1381 bio_release_pages(bio);
1387 void bio_check_pages_dirty(struct bio *bio)
1389 struct bio_vec *bvec = bio->bi_io_vec;
1390 int nr_clean_pages = 0;
1393 for (i = 0; i < bio->bi_vcnt; i++) {
1394 struct page *page = bvec[i].bv_page;
1396 if (PageDirty(page) || PageCompound(page)) {
1397 page_cache_release(page);
1398 bvec[i].bv_page = NULL;
1404 if (nr_clean_pages) {
1405 unsigned long flags;
1407 spin_lock_irqsave(&bio_dirty_lock, flags);
1408 bio->bi_private = bio_dirty_list;
1409 bio_dirty_list = bio;
1410 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1411 schedule_work(&bio_dirty_work);
1417 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1418 void bio_flush_dcache_pages(struct bio *bi)
1421 struct bio_vec *bvec;
1423 bio_for_each_segment(bvec, bi, i)
1424 flush_dcache_page(bvec->bv_page);
1426 EXPORT_SYMBOL(bio_flush_dcache_pages);
1430 * bio_endio - end I/O on a bio
1432 * @error: error, if any
1435 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1436 * preferred way to end I/O on a bio, it takes care of clearing
1437 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1438 * established -Exxxx (-EIO, for instance) error values in case
1439 * something went wrong. Noone should call bi_end_io() directly on a
1440 * bio unless they own it and thus know that it has an end_io
1443 void bio_endio(struct bio *bio, int error)
1446 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1447 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1451 bio->bi_end_io(bio, error);
1453 EXPORT_SYMBOL(bio_endio);
1455 void bio_pair_release(struct bio_pair *bp)
1457 if (atomic_dec_and_test(&bp->cnt)) {
1458 struct bio *master = bp->bio1.bi_private;
1460 bio_endio(master, bp->error);
1461 mempool_free(bp, bp->bio2.bi_private);
1464 EXPORT_SYMBOL(bio_pair_release);
1466 static void bio_pair_end_1(struct bio *bi, int err)
1468 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1473 bio_pair_release(bp);
1476 static void bio_pair_end_2(struct bio *bi, int err)
1478 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1483 bio_pair_release(bp);
1487 * split a bio - only worry about a bio with a single page in its iovec
1489 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1491 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1496 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1497 bi->bi_sector + first_sectors);
1499 BUG_ON(bi->bi_vcnt != 1);
1500 BUG_ON(bi->bi_idx != 0);
1501 atomic_set(&bp->cnt, 3);
1505 bp->bio2.bi_sector += first_sectors;
1506 bp->bio2.bi_size -= first_sectors << 9;
1507 bp->bio1.bi_size = first_sectors << 9;
1509 bp->bv1 = bi->bi_io_vec[0];
1510 bp->bv2 = bi->bi_io_vec[0];
1511 bp->bv2.bv_offset += first_sectors << 9;
1512 bp->bv2.bv_len -= first_sectors << 9;
1513 bp->bv1.bv_len = first_sectors << 9;
1515 bp->bio1.bi_io_vec = &bp->bv1;
1516 bp->bio2.bi_io_vec = &bp->bv2;
1518 bp->bio1.bi_max_vecs = 1;
1519 bp->bio2.bi_max_vecs = 1;
1521 bp->bio1.bi_end_io = bio_pair_end_1;
1522 bp->bio2.bi_end_io = bio_pair_end_2;
1524 bp->bio1.bi_private = bi;
1525 bp->bio2.bi_private = bio_split_pool;
1527 if (bio_integrity(bi))
1528 bio_integrity_split(bi, bp, first_sectors);
1532 EXPORT_SYMBOL(bio_split);
1535 * bio_sector_offset - Find hardware sector offset in bio
1536 * @bio: bio to inspect
1537 * @index: bio_vec index
1538 * @offset: offset in bv_page
1540 * Return the number of hardware sectors between beginning of bio
1541 * and an end point indicated by a bio_vec index and an offset
1542 * within that vector's page.
1544 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1545 unsigned int offset)
1547 unsigned int sector_sz;
1552 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1555 if (index >= bio->bi_idx)
1556 index = bio->bi_vcnt - 1;
1558 __bio_for_each_segment(bv, bio, i, 0) {
1560 if (offset > bv->bv_offset)
1561 sectors += (offset - bv->bv_offset) / sector_sz;
1565 sectors += bv->bv_len / sector_sz;
1570 EXPORT_SYMBOL(bio_sector_offset);
1573 * create memory pools for biovec's in a bio_set.
1574 * use the global biovec slabs created for general use.
1576 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1578 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1580 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1587 static void biovec_free_pools(struct bio_set *bs)
1589 mempool_destroy(bs->bvec_pool);
1592 void bioset_free(struct bio_set *bs)
1595 mempool_destroy(bs->bio_pool);
1597 bioset_integrity_free(bs);
1598 biovec_free_pools(bs);
1603 EXPORT_SYMBOL(bioset_free);
1606 * bioset_create - Create a bio_set
1607 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1608 * @front_pad: Number of bytes to allocate in front of the returned bio
1611 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1612 * to ask for a number of bytes to be allocated in front of the bio.
1613 * Front pad allocation is useful for embedding the bio inside
1614 * another structure, to avoid allocating extra data to go with the bio.
1615 * Note that the bio must be embedded at the END of that structure always,
1616 * or things will break badly.
1618 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1620 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1623 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1627 bs->front_pad = front_pad;
1629 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1630 if (!bs->bio_slab) {
1635 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1639 if (!biovec_create_pools(bs, pool_size))
1646 EXPORT_SYMBOL(bioset_create);
1648 static void __init biovec_init_slabs(void)
1652 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1654 struct biovec_slab *bvs = bvec_slabs + i;
1656 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1661 size = bvs->nr_vecs * sizeof(struct bio_vec);
1662 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1663 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1667 static int __init init_bio(void)
1671 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1673 panic("bio: can't allocate bios\n");
1675 bio_integrity_init();
1676 biovec_init_slabs();
1678 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1680 panic("bio: can't allocate bios\n");
1682 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1683 panic("bio: can't create integrity pool\n");
1685 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1686 sizeof(struct bio_pair));
1687 if (!bio_split_pool)
1688 panic("bio: can't create split pool\n");
1692 subsys_initcall(init_bio);