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/export.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 atomic_set(&bio->bi_cnt, 1);
260 EXPORT_SYMBOL(bio_init);
263 * bio_alloc_bioset - allocate a bio for I/O
264 * @gfp_mask: the GFP_ mask given to the slab allocator
265 * @nr_iovecs: number of iovecs to pre-allocate
266 * @bs: the bio_set to allocate from.
269 * bio_alloc_bioset will try its own mempool to satisfy the allocation.
270 * If %__GFP_WAIT is set then we will block on the internal pool waiting
271 * for a &struct bio to become free.
273 * Note that the caller must set ->bi_destructor on successful return
274 * of a bio, to do the appropriate freeing of the bio once the reference
275 * count drops to zero.
277 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
279 unsigned long idx = BIO_POOL_NONE;
280 struct bio_vec *bvl = NULL;
284 p = mempool_alloc(bs->bio_pool, gfp_mask);
287 bio = p + bs->front_pad;
291 if (unlikely(!nr_iovecs))
294 if (nr_iovecs <= BIO_INLINE_VECS) {
295 bvl = bio->bi_inline_vecs;
296 nr_iovecs = BIO_INLINE_VECS;
298 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
302 nr_iovecs = bvec_nr_vecs(idx);
305 bio->bi_flags |= idx << BIO_POOL_OFFSET;
306 bio->bi_max_vecs = nr_iovecs;
307 bio->bi_io_vec = bvl;
311 mempool_free(p, bs->bio_pool);
314 EXPORT_SYMBOL(bio_alloc_bioset);
316 static void bio_fs_destructor(struct bio *bio)
318 bio_free(bio, fs_bio_set);
322 * bio_alloc - allocate a new bio, memory pool backed
323 * @gfp_mask: allocation mask to use
324 * @nr_iovecs: number of iovecs
326 * bio_alloc will allocate a bio and associated bio_vec array that can hold
327 * at least @nr_iovecs entries. Allocations will be done from the
328 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
330 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
331 * a bio. This is due to the mempool guarantees. To make this work, callers
332 * must never allocate more than 1 bio at a time from this pool. Callers
333 * that need to allocate more than 1 bio must always submit the previously
334 * allocated bio for IO before attempting to allocate a new one. Failure to
335 * do so can cause livelocks under memory pressure.
338 * Pointer to new bio on success, NULL on failure.
340 struct bio *bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
342 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
345 bio->bi_destructor = bio_fs_destructor;
349 EXPORT_SYMBOL(bio_alloc);
351 static void bio_kmalloc_destructor(struct bio *bio)
353 if (bio_integrity(bio))
354 bio_integrity_free(bio, fs_bio_set);
359 * bio_kmalloc - allocate a bio for I/O using kmalloc()
360 * @gfp_mask: the GFP_ mask given to the slab allocator
361 * @nr_iovecs: number of iovecs to pre-allocate
364 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
365 * %__GFP_WAIT, the allocation is guaranteed to succeed.
368 struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned int nr_iovecs)
372 if (nr_iovecs > UIO_MAXIOV)
375 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
381 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
382 bio->bi_max_vecs = nr_iovecs;
383 bio->bi_io_vec = bio->bi_inline_vecs;
384 bio->bi_destructor = bio_kmalloc_destructor;
388 EXPORT_SYMBOL(bio_kmalloc);
390 void zero_fill_bio(struct bio *bio)
396 bio_for_each_segment(bv, bio, i) {
397 char *data = bvec_kmap_irq(bv, &flags);
398 memset(data, 0, bv->bv_len);
399 flush_dcache_page(bv->bv_page);
400 bvec_kunmap_irq(data, &flags);
403 EXPORT_SYMBOL(zero_fill_bio);
406 * bio_put - release a reference to a bio
407 * @bio: bio to release reference to
410 * Put a reference to a &struct bio, either one you have gotten with
411 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
413 void bio_put(struct bio *bio)
415 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
420 if (atomic_dec_and_test(&bio->bi_cnt)) {
422 bio->bi_destructor(bio);
425 EXPORT_SYMBOL(bio_put);
427 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
429 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
430 blk_recount_segments(q, bio);
432 return bio->bi_phys_segments;
434 EXPORT_SYMBOL(bio_phys_segments);
437 * __bio_clone - clone a bio
438 * @bio: destination bio
439 * @bio_src: bio to clone
441 * Clone a &bio. Caller will own the returned bio, but not
442 * the actual data it points to. Reference count of returned
445 void __bio_clone(struct bio *bio, struct bio *bio_src)
447 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
448 bio_src->bi_max_vecs * sizeof(struct bio_vec));
451 * most users will be overriding ->bi_bdev with a new target,
452 * so we don't set nor calculate new physical/hw segment counts here
454 bio->bi_sector = bio_src->bi_sector;
455 bio->bi_bdev = bio_src->bi_bdev;
456 bio->bi_flags |= 1 << BIO_CLONED;
457 bio->bi_rw = bio_src->bi_rw;
458 bio->bi_vcnt = bio_src->bi_vcnt;
459 bio->bi_size = bio_src->bi_size;
460 bio->bi_idx = bio_src->bi_idx;
462 EXPORT_SYMBOL(__bio_clone);
465 * bio_clone - clone a bio
467 * @gfp_mask: allocation priority
469 * Like __bio_clone, only also allocates the returned bio
471 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
473 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
478 b->bi_destructor = bio_fs_destructor;
481 if (bio_integrity(bio)) {
484 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
494 EXPORT_SYMBOL(bio_clone);
497 * bio_get_nr_vecs - return approx number of vecs
500 * Return the approximate number of pages we can send to this target.
501 * There's no guarantee that you will be able to fit this number of pages
502 * into a bio, it does not account for dynamic restrictions that vary
505 int bio_get_nr_vecs(struct block_device *bdev)
507 struct request_queue *q = bdev_get_queue(bdev);
510 nr_pages = min_t(unsigned,
511 queue_max_segments(q),
512 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
514 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
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 device
642 * limitations. The target block device must allow bio's up to PAGE_SIZE,
643 * so it is always possible to add a single page to an empty bio.
645 * This should only be used by REQ_PC bios.
647 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
648 unsigned int len, unsigned int offset)
650 return __bio_add_page(q, bio, page, len, offset,
651 queue_max_hw_sectors(q));
653 EXPORT_SYMBOL(bio_add_pc_page);
656 * bio_add_page - attempt to add page to bio
657 * @bio: destination bio
659 * @len: vec entry length
660 * @offset: vec entry offset
662 * Attempt to add a page to the bio_vec maplist. This can fail for a
663 * number of reasons, such as the bio being full or target block device
664 * limitations. The target block device must allow bio's up to PAGE_SIZE,
665 * so it is always possible to add a single 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,
701 unsigned int iov_count,
704 struct bio_map_data *bmd;
706 if (iov_count > UIO_MAXIOV)
709 bmd = kmalloc(sizeof(*bmd), gfp_mask);
713 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
719 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
728 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
729 struct sg_iovec *iov, int iov_count,
730 int to_user, int from_user, int do_free_page)
733 struct bio_vec *bvec;
735 unsigned int iov_off = 0;
737 __bio_for_each_segment(bvec, bio, i, 0) {
738 char *bv_addr = page_address(bvec->bv_page);
739 unsigned int bv_len = iovecs[i].bv_len;
741 while (bv_len && iov_idx < iov_count) {
743 char __user *iov_addr;
745 bytes = min_t(unsigned int,
746 iov[iov_idx].iov_len - iov_off, bv_len);
747 iov_addr = iov[iov_idx].iov_base + iov_off;
751 ret = copy_to_user(iov_addr, bv_addr,
755 ret = copy_from_user(bv_addr, iov_addr,
767 if (iov[iov_idx].iov_len == iov_off) {
774 __free_page(bvec->bv_page);
781 * bio_uncopy_user - finish previously mapped bio
782 * @bio: bio being terminated
784 * Free pages allocated from bio_copy_user() and write back data
785 * to user space in case of a read.
787 int bio_uncopy_user(struct bio *bio)
789 struct bio_map_data *bmd = bio->bi_private;
792 if (!bio_flagged(bio, BIO_NULL_MAPPED))
793 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
794 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
795 0, bmd->is_our_pages);
796 bio_free_map_data(bmd);
800 EXPORT_SYMBOL(bio_uncopy_user);
803 * bio_copy_user_iov - copy user data to bio
804 * @q: destination block queue
805 * @map_data: pointer to the rq_map_data holding pages (if necessary)
807 * @iov_count: number of elements in the iovec
808 * @write_to_vm: bool indicating writing to pages or not
809 * @gfp_mask: memory allocation flags
811 * Prepares and returns a bio for indirect user io, bouncing data
812 * to/from kernel pages as necessary. Must be paired with
813 * call bio_uncopy_user() on io completion.
815 struct bio *bio_copy_user_iov(struct request_queue *q,
816 struct rq_map_data *map_data,
817 struct sg_iovec *iov, int iov_count,
818 int write_to_vm, gfp_t gfp_mask)
820 struct bio_map_data *bmd;
821 struct bio_vec *bvec;
826 unsigned int len = 0;
827 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
829 for (i = 0; i < iov_count; i++) {
834 uaddr = (unsigned long)iov[i].iov_base;
835 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
836 start = uaddr >> PAGE_SHIFT;
842 return ERR_PTR(-EINVAL);
844 nr_pages += end - start;
845 len += iov[i].iov_len;
851 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
853 return ERR_PTR(-ENOMEM);
856 bio = bio_kmalloc(gfp_mask, nr_pages);
861 bio->bi_rw |= REQ_WRITE;
866 nr_pages = 1 << map_data->page_order;
867 i = map_data->offset / PAGE_SIZE;
870 unsigned int bytes = PAGE_SIZE;
878 if (i == map_data->nr_entries * nr_pages) {
883 page = map_data->pages[i / nr_pages];
884 page += (i % nr_pages);
888 page = alloc_page(q->bounce_gfp | gfp_mask);
895 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
908 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
909 (map_data && map_data->from_user)) {
910 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
915 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
919 bio_for_each_segment(bvec, bio, i)
920 __free_page(bvec->bv_page);
924 bio_free_map_data(bmd);
929 * bio_copy_user - copy user data to bio
930 * @q: destination block queue
931 * @map_data: pointer to the rq_map_data holding pages (if necessary)
932 * @uaddr: start of user address
933 * @len: length in bytes
934 * @write_to_vm: bool indicating writing to pages or not
935 * @gfp_mask: memory allocation flags
937 * Prepares and returns a bio for indirect user io, bouncing data
938 * to/from kernel pages as necessary. Must be paired with
939 * call bio_uncopy_user() on io completion.
941 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
942 unsigned long uaddr, unsigned int len,
943 int write_to_vm, gfp_t gfp_mask)
947 iov.iov_base = (void __user *)uaddr;
950 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
952 EXPORT_SYMBOL(bio_copy_user);
954 static struct bio *__bio_map_user_iov(struct request_queue *q,
955 struct block_device *bdev,
956 struct sg_iovec *iov, int iov_count,
957 int write_to_vm, gfp_t gfp_mask)
966 for (i = 0; i < iov_count; i++) {
967 unsigned long uaddr = (unsigned long)iov[i].iov_base;
968 unsigned long len = iov[i].iov_len;
969 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
970 unsigned long start = uaddr >> PAGE_SHIFT;
976 return ERR_PTR(-EINVAL);
978 nr_pages += end - start;
980 * buffer must be aligned to at least hardsector size for now
982 if (uaddr & queue_dma_alignment(q))
983 return ERR_PTR(-EINVAL);
987 return ERR_PTR(-EINVAL);
989 bio = bio_kmalloc(gfp_mask, nr_pages);
991 return ERR_PTR(-ENOMEM);
994 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
998 for (i = 0; i < iov_count; i++) {
999 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1000 unsigned long len = iov[i].iov_len;
1001 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1002 unsigned long start = uaddr >> PAGE_SHIFT;
1003 const int local_nr_pages = end - start;
1004 const int page_limit = cur_page + local_nr_pages;
1006 ret = get_user_pages_fast(uaddr, local_nr_pages,
1007 write_to_vm, &pages[cur_page]);
1008 if (ret < local_nr_pages) {
1013 offset = uaddr & ~PAGE_MASK;
1014 for (j = cur_page; j < page_limit; j++) {
1015 unsigned int bytes = PAGE_SIZE - offset;
1026 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1036 * release the pages we didn't map into the bio, if any
1038 while (j < page_limit)
1039 page_cache_release(pages[j++]);
1045 * set data direction, and check if mapped pages need bouncing
1048 bio->bi_rw |= REQ_WRITE;
1050 bio->bi_bdev = bdev;
1051 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1055 for (i = 0; i < nr_pages; i++) {
1058 page_cache_release(pages[i]);
1063 return ERR_PTR(ret);
1067 * bio_map_user - map user address into bio
1068 * @q: the struct request_queue for the bio
1069 * @bdev: destination block device
1070 * @uaddr: start of user address
1071 * @len: length in bytes
1072 * @write_to_vm: bool indicating writing to pages or not
1073 * @gfp_mask: memory allocation flags
1075 * Map the user space address into a bio suitable for io to a block
1076 * device. Returns an error pointer in case of error.
1078 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1079 unsigned long uaddr, unsigned int len, int write_to_vm,
1082 struct sg_iovec iov;
1084 iov.iov_base = (void __user *)uaddr;
1087 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1089 EXPORT_SYMBOL(bio_map_user);
1092 * bio_map_user_iov - map user sg_iovec table into bio
1093 * @q: the struct request_queue for the bio
1094 * @bdev: destination block device
1096 * @iov_count: number of elements in the iovec
1097 * @write_to_vm: bool indicating writing to pages or not
1098 * @gfp_mask: memory allocation flags
1100 * Map the user space address into a bio suitable for io to a block
1101 * device. Returns an error pointer in case of error.
1103 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1104 struct sg_iovec *iov, int iov_count,
1105 int write_to_vm, gfp_t gfp_mask)
1109 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1115 * subtle -- if __bio_map_user() ended up bouncing a bio,
1116 * it would normally disappear when its bi_end_io is run.
1117 * however, we need it for the unmap, so grab an extra
1125 static void __bio_unmap_user(struct bio *bio)
1127 struct bio_vec *bvec;
1131 * make sure we dirty pages we wrote to
1133 __bio_for_each_segment(bvec, bio, i, 0) {
1134 if (bio_data_dir(bio) == READ)
1135 set_page_dirty_lock(bvec->bv_page);
1137 page_cache_release(bvec->bv_page);
1144 * bio_unmap_user - unmap a bio
1145 * @bio: the bio being unmapped
1147 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1148 * a process context.
1150 * bio_unmap_user() may sleep.
1152 void bio_unmap_user(struct bio *bio)
1154 __bio_unmap_user(bio);
1157 EXPORT_SYMBOL(bio_unmap_user);
1159 static void bio_map_kern_endio(struct bio *bio, int err)
1164 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1165 unsigned int len, gfp_t gfp_mask)
1167 unsigned long kaddr = (unsigned long)data;
1168 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1169 unsigned long start = kaddr >> PAGE_SHIFT;
1170 const int nr_pages = end - start;
1174 bio = bio_kmalloc(gfp_mask, nr_pages);
1176 return ERR_PTR(-ENOMEM);
1178 offset = offset_in_page(kaddr);
1179 for (i = 0; i < nr_pages; i++) {
1180 unsigned int bytes = PAGE_SIZE - offset;
1188 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1197 bio->bi_end_io = bio_map_kern_endio;
1202 * bio_map_kern - map kernel address into bio
1203 * @q: the struct request_queue for the bio
1204 * @data: pointer to buffer to map
1205 * @len: length in bytes
1206 * @gfp_mask: allocation flags for bio allocation
1208 * Map the kernel address into a bio suitable for io to a block
1209 * device. Returns an error pointer in case of error.
1211 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1216 bio = __bio_map_kern(q, data, len, gfp_mask);
1220 if (bio->bi_size == len)
1224 * Don't support partial mappings.
1227 return ERR_PTR(-EINVAL);
1229 EXPORT_SYMBOL(bio_map_kern);
1231 static void bio_copy_kern_endio(struct bio *bio, int err)
1233 struct bio_vec *bvec;
1234 const int read = bio_data_dir(bio) == READ;
1235 struct bio_map_data *bmd = bio->bi_private;
1237 char *p = bmd->sgvecs[0].iov_base;
1239 __bio_for_each_segment(bvec, bio, i, 0) {
1240 char *addr = page_address(bvec->bv_page);
1241 int len = bmd->iovecs[i].bv_len;
1244 memcpy(p, addr, len);
1246 __free_page(bvec->bv_page);
1250 bio_free_map_data(bmd);
1255 * bio_copy_kern - copy kernel address into bio
1256 * @q: the struct request_queue for the bio
1257 * @data: pointer to buffer to copy
1258 * @len: length in bytes
1259 * @gfp_mask: allocation flags for bio and page allocation
1260 * @reading: data direction is READ
1262 * copy the kernel address into a bio suitable for io to a block
1263 * device. Returns an error pointer in case of error.
1265 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1266 gfp_t gfp_mask, int reading)
1269 struct bio_vec *bvec;
1272 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1279 bio_for_each_segment(bvec, bio, i) {
1280 char *addr = page_address(bvec->bv_page);
1282 memcpy(addr, p, bvec->bv_len);
1287 bio->bi_end_io = bio_copy_kern_endio;
1291 EXPORT_SYMBOL(bio_copy_kern);
1294 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1295 * for performing direct-IO in BIOs.
1297 * The problem is that we cannot run set_page_dirty() from interrupt context
1298 * because the required locks are not interrupt-safe. So what we can do is to
1299 * mark the pages dirty _before_ performing IO. And in interrupt context,
1300 * check that the pages are still dirty. If so, fine. If not, redirty them
1301 * in process context.
1303 * We special-case compound pages here: normally this means reads into hugetlb
1304 * pages. The logic in here doesn't really work right for compound pages
1305 * because the VM does not uniformly chase down the head page in all cases.
1306 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1307 * handle them at all. So we skip compound pages here at an early stage.
1309 * Note that this code is very hard to test under normal circumstances because
1310 * direct-io pins the pages with get_user_pages(). This makes
1311 * is_page_cache_freeable return false, and the VM will not clean the pages.
1312 * But other code (eg, pdflush) could clean the pages if they are mapped
1315 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1316 * deferred bio dirtying paths.
1320 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1322 void bio_set_pages_dirty(struct bio *bio)
1324 struct bio_vec *bvec = bio->bi_io_vec;
1327 for (i = 0; i < bio->bi_vcnt; i++) {
1328 struct page *page = bvec[i].bv_page;
1330 if (page && !PageCompound(page))
1331 set_page_dirty_lock(page);
1335 static void bio_release_pages(struct bio *bio)
1337 struct bio_vec *bvec = bio->bi_io_vec;
1340 for (i = 0; i < bio->bi_vcnt; i++) {
1341 struct page *page = bvec[i].bv_page;
1349 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1350 * If they are, then fine. If, however, some pages are clean then they must
1351 * have been written out during the direct-IO read. So we take another ref on
1352 * the BIO and the offending pages and re-dirty the pages in process context.
1354 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1355 * here on. It will run one page_cache_release() against each page and will
1356 * run one bio_put() against the BIO.
1359 static void bio_dirty_fn(struct work_struct *work);
1361 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1362 static DEFINE_SPINLOCK(bio_dirty_lock);
1363 static struct bio *bio_dirty_list;
1366 * This runs in process context
1368 static void bio_dirty_fn(struct work_struct *work)
1370 unsigned long flags;
1373 spin_lock_irqsave(&bio_dirty_lock, flags);
1374 bio = bio_dirty_list;
1375 bio_dirty_list = NULL;
1376 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1379 struct bio *next = bio->bi_private;
1381 bio_set_pages_dirty(bio);
1382 bio_release_pages(bio);
1388 void bio_check_pages_dirty(struct bio *bio)
1390 struct bio_vec *bvec = bio->bi_io_vec;
1391 int nr_clean_pages = 0;
1394 for (i = 0; i < bio->bi_vcnt; i++) {
1395 struct page *page = bvec[i].bv_page;
1397 if (PageDirty(page) || PageCompound(page)) {
1398 page_cache_release(page);
1399 bvec[i].bv_page = NULL;
1405 if (nr_clean_pages) {
1406 unsigned long flags;
1408 spin_lock_irqsave(&bio_dirty_lock, flags);
1409 bio->bi_private = bio_dirty_list;
1410 bio_dirty_list = bio;
1411 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1412 schedule_work(&bio_dirty_work);
1418 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1419 void bio_flush_dcache_pages(struct bio *bi)
1422 struct bio_vec *bvec;
1424 bio_for_each_segment(bvec, bi, i)
1425 flush_dcache_page(bvec->bv_page);
1427 EXPORT_SYMBOL(bio_flush_dcache_pages);
1431 * bio_endio - end I/O on a bio
1433 * @error: error, if any
1436 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1437 * preferred way to end I/O on a bio, it takes care of clearing
1438 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1439 * established -Exxxx (-EIO, for instance) error values in case
1440 * something went wrong. No one should call bi_end_io() directly on a
1441 * bio unless they own it and thus know that it has an end_io
1444 void bio_endio(struct bio *bio, int error)
1447 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1448 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1452 bio->bi_end_io(bio, error);
1454 EXPORT_SYMBOL(bio_endio);
1456 void bio_pair_release(struct bio_pair *bp)
1458 if (atomic_dec_and_test(&bp->cnt)) {
1459 struct bio *master = bp->bio1.bi_private;
1461 bio_endio(master, bp->error);
1462 mempool_free(bp, bp->bio2.bi_private);
1465 EXPORT_SYMBOL(bio_pair_release);
1467 static void bio_pair_end_1(struct bio *bi, int err)
1469 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1474 bio_pair_release(bp);
1477 static void bio_pair_end_2(struct bio *bi, int err)
1479 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1484 bio_pair_release(bp);
1488 * split a bio - only worry about a bio with a single page in its iovec
1490 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1492 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1497 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1498 bi->bi_sector + first_sectors);
1500 BUG_ON(bi->bi_vcnt != 1);
1501 BUG_ON(bi->bi_idx != 0);
1502 atomic_set(&bp->cnt, 3);
1506 bp->bio2.bi_sector += first_sectors;
1507 bp->bio2.bi_size -= first_sectors << 9;
1508 bp->bio1.bi_size = first_sectors << 9;
1510 bp->bv1 = bi->bi_io_vec[0];
1511 bp->bv2 = bi->bi_io_vec[0];
1512 bp->bv2.bv_offset += first_sectors << 9;
1513 bp->bv2.bv_len -= first_sectors << 9;
1514 bp->bv1.bv_len = first_sectors << 9;
1516 bp->bio1.bi_io_vec = &bp->bv1;
1517 bp->bio2.bi_io_vec = &bp->bv2;
1519 bp->bio1.bi_max_vecs = 1;
1520 bp->bio2.bi_max_vecs = 1;
1522 bp->bio1.bi_end_io = bio_pair_end_1;
1523 bp->bio2.bi_end_io = bio_pair_end_2;
1525 bp->bio1.bi_private = bi;
1526 bp->bio2.bi_private = bio_split_pool;
1528 if (bio_integrity(bi))
1529 bio_integrity_split(bi, bp, first_sectors);
1533 EXPORT_SYMBOL(bio_split);
1536 * bio_sector_offset - Find hardware sector offset in bio
1537 * @bio: bio to inspect
1538 * @index: bio_vec index
1539 * @offset: offset in bv_page
1541 * Return the number of hardware sectors between beginning of bio
1542 * and an end point indicated by a bio_vec index and an offset
1543 * within that vector's page.
1545 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1546 unsigned int offset)
1548 unsigned int sector_sz;
1553 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1556 if (index >= bio->bi_idx)
1557 index = bio->bi_vcnt - 1;
1559 __bio_for_each_segment(bv, bio, i, 0) {
1561 if (offset > bv->bv_offset)
1562 sectors += (offset - bv->bv_offset) / sector_sz;
1566 sectors += bv->bv_len / sector_sz;
1571 EXPORT_SYMBOL(bio_sector_offset);
1574 * create memory pools for biovec's in a bio_set.
1575 * use the global biovec slabs created for general use.
1577 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1579 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1581 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1588 static void biovec_free_pools(struct bio_set *bs)
1590 mempool_destroy(bs->bvec_pool);
1593 void bioset_free(struct bio_set *bs)
1596 mempool_destroy(bs->bio_pool);
1598 bioset_integrity_free(bs);
1599 biovec_free_pools(bs);
1604 EXPORT_SYMBOL(bioset_free);
1607 * bioset_create - Create a bio_set
1608 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1609 * @front_pad: Number of bytes to allocate in front of the returned bio
1612 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1613 * to ask for a number of bytes to be allocated in front of the bio.
1614 * Front pad allocation is useful for embedding the bio inside
1615 * another structure, to avoid allocating extra data to go with the bio.
1616 * Note that the bio must be embedded at the END of that structure always,
1617 * or things will break badly.
1619 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1621 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1624 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1628 bs->front_pad = front_pad;
1630 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1631 if (!bs->bio_slab) {
1636 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1640 if (!biovec_create_pools(bs, pool_size))
1647 EXPORT_SYMBOL(bioset_create);
1649 static void __init biovec_init_slabs(void)
1653 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1655 struct biovec_slab *bvs = bvec_slabs + i;
1657 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1662 size = bvs->nr_vecs * sizeof(struct bio_vec);
1663 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1664 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1668 static int __init init_bio(void)
1672 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1674 panic("bio: can't allocate bios\n");
1676 bio_integrity_init();
1677 biovec_init_slabs();
1679 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1681 panic("bio: can't allocate bios\n");
1683 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1684 panic("bio: can't create integrity pool\n");
1686 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1687 sizeof(struct bio_pair));
1688 if (!bio_split_pool)
1689 panic("bio: can't create split pool\n");
1693 subsys_initcall(init_bio);