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 <scsi/sg.h> /* for struct sg_iovec */
31 #include <trace/events/block.h>
34 * Test patch to inline a certain number of bi_io_vec's inside the bio
35 * itself, to shrink a bio data allocation from two mempool calls to one
37 #define BIO_INLINE_VECS 4
39 static mempool_t *bio_split_pool __read_mostly;
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
46 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
47 struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
48 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
56 struct bio_set *fs_bio_set;
59 * Our slab pool management
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
67 static DEFINE_MUTEX(bio_slab_lock);
68 static struct bio_slab *bio_slabs;
69 static unsigned int bio_slab_nr, bio_slab_max;
71 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
75 struct bio_slab *bslab;
76 unsigned int i, entry = -1;
78 mutex_lock(&bio_slab_lock);
81 while (i < bio_slab_nr) {
82 struct bio_slab *bslab = &bio_slabs[i];
84 if (!bslab->slab && entry == -1)
86 else if (bslab->slab_size == sz) {
97 if (bio_slab_nr == bio_slab_max && entry == -1) {
99 bio_slabs = krealloc(bio_slabs,
100 bio_slab_max * sizeof(struct bio_slab),
106 entry = bio_slab_nr++;
108 bslab = &bio_slabs[entry];
110 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
111 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
115 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
118 bslab->slab_size = sz;
120 mutex_unlock(&bio_slab_lock);
124 static void bio_put_slab(struct bio_set *bs)
126 struct bio_slab *bslab = NULL;
129 mutex_lock(&bio_slab_lock);
131 for (i = 0; i < bio_slab_nr; i++) {
132 if (bs->bio_slab == bio_slabs[i].slab) {
133 bslab = &bio_slabs[i];
138 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
141 WARN_ON(!bslab->slab_ref);
143 if (--bslab->slab_ref)
146 kmem_cache_destroy(bslab->slab);
150 mutex_unlock(&bio_slab_lock);
153 unsigned int bvec_nr_vecs(unsigned short idx)
155 return bvec_slabs[idx].nr_vecs;
158 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
160 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
162 if (idx == BIOVEC_MAX_IDX)
163 mempool_free(bv, bs->bvec_pool);
165 struct biovec_slab *bvs = bvec_slabs + idx;
167 kmem_cache_free(bvs->slab, bv);
171 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
177 * see comment near bvec_array define!
195 case 129 ... BIO_MAX_PAGES:
203 * idx now points to the pool we want to allocate from. only the
204 * 1-vec entry pool is mempool backed.
206 if (*idx == BIOVEC_MAX_IDX) {
208 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
210 struct biovec_slab *bvs = bvec_slabs + *idx;
211 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
214 * Make this allocation restricted and don't dump info on
215 * allocation failures, since we'll fallback to the mempool
216 * in case of failure.
218 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
221 * Try a slab allocation. If this fails and __GFP_WAIT
222 * is set, retry with the 1-entry mempool
224 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
225 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
226 *idx = BIOVEC_MAX_IDX;
234 void bio_free(struct bio *bio, struct bio_set *bs)
238 if (bio_has_allocated_vec(bio))
239 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
241 if (bio_integrity(bio))
242 bio_integrity_free(bio);
245 * If we have front padding, adjust the bio pointer before freeing
251 mempool_free(p, bs->bio_pool);
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);
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. If %NULL, just use kmalloc
269 * bio_alloc_bioset will first 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. If a %NULL @bs is passed in, we will
272 * fall back to just using @kmalloc to allocate the required memory.
274 * Note that the caller must set ->bi_destructor on succesful 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);
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 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask
327 * contains __GFP_WAIT, the allocation is guaranteed to succeed.
330 * Pointer to new bio on success, NULL on failure.
332 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
334 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
337 bio->bi_destructor = bio_fs_destructor;
342 static void bio_kmalloc_destructor(struct bio *bio)
344 if (bio_integrity(bio))
345 bio_integrity_free(bio);
350 * bio_alloc - allocate a bio for I/O
351 * @gfp_mask: the GFP_ mask given to the slab allocator
352 * @nr_iovecs: number of iovecs to pre-allocate
355 * bio_alloc will allocate a bio and associated bio_vec array that can hold
356 * at least @nr_iovecs entries. Allocations will be done from the
357 * fs_bio_set. Also see @bio_alloc_bioset.
359 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
360 * a bio. This is due to the mempool guarantees. To make this work, callers
361 * must never allocate more than 1 bio at a time from this pool. Callers
362 * that need to allocate more than 1 bio must always submit the previously
363 * allocated bio for IO before attempting to allocate a new one. Failure to
364 * do so can cause livelocks under memory pressure.
367 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
371 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
377 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
378 bio->bi_max_vecs = nr_iovecs;
379 bio->bi_io_vec = bio->bi_inline_vecs;
380 bio->bi_destructor = bio_kmalloc_destructor;
385 void zero_fill_bio(struct bio *bio)
391 bio_for_each_segment(bv, bio, i) {
392 char *data = bvec_kmap_irq(bv, &flags);
393 memset(data, 0, bv->bv_len);
394 flush_dcache_page(bv->bv_page);
395 bvec_kunmap_irq(data, &flags);
398 EXPORT_SYMBOL(zero_fill_bio);
401 * bio_put - release a reference to a bio
402 * @bio: bio to release reference to
405 * Put a reference to a &struct bio, either one you have gotten with
406 * bio_alloc or bio_get. The last put of a bio will free it.
408 void bio_put(struct bio *bio)
410 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
415 if (atomic_dec_and_test(&bio->bi_cnt)) {
417 bio->bi_destructor(bio);
421 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
423 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
424 blk_recount_segments(q, bio);
426 return bio->bi_phys_segments;
430 * __bio_clone - clone a bio
431 * @bio: destination bio
432 * @bio_src: bio to clone
434 * Clone a &bio. Caller will own the returned bio, but not
435 * the actual data it points to. Reference count of returned
438 void __bio_clone(struct bio *bio, struct bio *bio_src)
440 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
441 bio_src->bi_max_vecs * sizeof(struct bio_vec));
444 * most users will be overriding ->bi_bdev with a new target,
445 * so we don't set nor calculate new physical/hw segment counts here
447 bio->bi_sector = bio_src->bi_sector;
448 bio->bi_bdev = bio_src->bi_bdev;
449 bio->bi_flags |= 1 << BIO_CLONED;
450 bio->bi_rw = bio_src->bi_rw;
451 bio->bi_vcnt = bio_src->bi_vcnt;
452 bio->bi_size = bio_src->bi_size;
453 bio->bi_idx = bio_src->bi_idx;
457 * bio_clone - clone a bio
459 * @gfp_mask: allocation priority
461 * Like __bio_clone, only also allocates the returned bio
463 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
465 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
470 b->bi_destructor = bio_fs_destructor;
473 if (bio_integrity(bio)) {
476 ret = bio_integrity_clone(b, bio, gfp_mask);
488 * bio_get_nr_vecs - return approx number of vecs
491 * Return the approximate number of pages we can send to this target.
492 * There's no guarantee that you will be able to fit this number of pages
493 * into a bio, it does not account for dynamic restrictions that vary
496 int bio_get_nr_vecs(struct block_device *bdev)
498 struct request_queue *q = bdev_get_queue(bdev);
501 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
502 if (nr_pages > queue_max_phys_segments(q))
503 nr_pages = queue_max_phys_segments(q);
504 if (nr_pages > queue_max_hw_segments(q))
505 nr_pages = queue_max_hw_segments(q);
510 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
511 *page, unsigned int len, unsigned int offset,
512 unsigned short max_sectors)
514 int retried_segments = 0;
515 struct bio_vec *bvec;
518 * cloned bio must not modify vec list
520 if (unlikely(bio_flagged(bio, BIO_CLONED)))
523 if (((bio->bi_size + len) >> 9) > max_sectors)
527 * For filesystems with a blocksize smaller than the pagesize
528 * we will often be called with the same page as last time and
529 * a consecutive offset. Optimize this special case.
531 if (bio->bi_vcnt > 0) {
532 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
534 if (page == prev->bv_page &&
535 offset == prev->bv_offset + prev->bv_len) {
538 if (q->merge_bvec_fn) {
539 struct bvec_merge_data bvm = {
540 .bi_bdev = bio->bi_bdev,
541 .bi_sector = bio->bi_sector,
542 .bi_size = bio->bi_size,
546 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
556 if (bio->bi_vcnt >= bio->bi_max_vecs)
560 * we might lose a segment or two here, but rather that than
561 * make this too complex.
564 while (bio->bi_phys_segments >= queue_max_phys_segments(q)
565 || bio->bi_phys_segments >= queue_max_hw_segments(q)) {
567 if (retried_segments)
570 retried_segments = 1;
571 blk_recount_segments(q, bio);
575 * setup the new entry, we might clear it again later if we
576 * cannot add the page
578 bvec = &bio->bi_io_vec[bio->bi_vcnt];
579 bvec->bv_page = page;
581 bvec->bv_offset = offset;
584 * if queue has other restrictions (eg varying max sector size
585 * depending on offset), it can specify a merge_bvec_fn in the
586 * queue to get further control
588 if (q->merge_bvec_fn) {
589 struct bvec_merge_data bvm = {
590 .bi_bdev = bio->bi_bdev,
591 .bi_sector = bio->bi_sector,
592 .bi_size = bio->bi_size,
597 * merge_bvec_fn() returns number of bytes it can accept
600 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
601 bvec->bv_page = NULL;
608 /* If we may be able to merge these biovecs, force a recount */
609 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
610 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
613 bio->bi_phys_segments++;
620 * bio_add_pc_page - attempt to add page to bio
621 * @q: the target queue
622 * @bio: destination bio
624 * @len: vec entry length
625 * @offset: vec entry offset
627 * Attempt to add a page to the bio_vec maplist. This can fail for a
628 * number of reasons, such as the bio being full or target block
629 * device limitations. The target block device must allow bio's
630 * smaller than PAGE_SIZE, so it is always possible to add a single
631 * page to an empty bio. This should only be used by REQ_PC bios.
633 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
634 unsigned int len, unsigned int offset)
636 return __bio_add_page(q, bio, page, len, offset,
637 queue_max_hw_sectors(q));
641 * bio_add_page - attempt to add page to bio
642 * @bio: destination bio
644 * @len: vec entry length
645 * @offset: vec entry offset
647 * Attempt to add a page to the bio_vec maplist. This can fail for a
648 * number of reasons, such as the bio being full or target block
649 * device limitations. The target block device must allow bio's
650 * smaller than PAGE_SIZE, so it is always possible to add a single
651 * page to an empty bio.
653 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
656 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
657 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
660 struct bio_map_data {
661 struct bio_vec *iovecs;
662 struct sg_iovec *sgvecs;
667 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
668 struct sg_iovec *iov, int iov_count,
671 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
672 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
673 bmd->nr_sgvecs = iov_count;
674 bmd->is_our_pages = is_our_pages;
675 bio->bi_private = bmd;
678 static void bio_free_map_data(struct bio_map_data *bmd)
685 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
688 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
693 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
699 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
708 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
709 struct sg_iovec *iov, int iov_count, int uncopy,
713 struct bio_vec *bvec;
715 unsigned int iov_off = 0;
716 int read = bio_data_dir(bio) == READ;
718 __bio_for_each_segment(bvec, bio, i, 0) {
719 char *bv_addr = page_address(bvec->bv_page);
720 unsigned int bv_len = iovecs[i].bv_len;
722 while (bv_len && iov_idx < iov_count) {
724 char __user *iov_addr;
726 bytes = min_t(unsigned int,
727 iov[iov_idx].iov_len - iov_off, bv_len);
728 iov_addr = iov[iov_idx].iov_base + iov_off;
731 if (!read && !uncopy)
732 ret = copy_from_user(bv_addr, iov_addr,
735 ret = copy_to_user(iov_addr, bv_addr,
747 if (iov[iov_idx].iov_len == iov_off) {
754 __free_page(bvec->bv_page);
761 * bio_uncopy_user - finish previously mapped bio
762 * @bio: bio being terminated
764 * Free pages allocated from bio_copy_user() and write back data
765 * to user space in case of a read.
767 int bio_uncopy_user(struct bio *bio)
769 struct bio_map_data *bmd = bio->bi_private;
772 if (!bio_flagged(bio, BIO_NULL_MAPPED))
773 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
774 bmd->nr_sgvecs, 1, bmd->is_our_pages);
775 bio_free_map_data(bmd);
781 * bio_copy_user_iov - copy user data to bio
782 * @q: destination block queue
783 * @map_data: pointer to the rq_map_data holding pages (if necessary)
785 * @iov_count: number of elements in the iovec
786 * @write_to_vm: bool indicating writing to pages or not
787 * @gfp_mask: memory allocation flags
789 * Prepares and returns a bio for indirect user io, bouncing data
790 * to/from kernel pages as necessary. Must be paired with
791 * call bio_uncopy_user() on io completion.
793 struct bio *bio_copy_user_iov(struct request_queue *q,
794 struct rq_map_data *map_data,
795 struct sg_iovec *iov, int iov_count,
796 int write_to_vm, gfp_t gfp_mask)
798 struct bio_map_data *bmd;
799 struct bio_vec *bvec;
804 unsigned int len = 0;
805 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
807 for (i = 0; i < iov_count; i++) {
812 uaddr = (unsigned long)iov[i].iov_base;
813 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
814 start = uaddr >> PAGE_SHIFT;
816 nr_pages += end - start;
817 len += iov[i].iov_len;
823 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
825 return ERR_PTR(-ENOMEM);
828 bio = bio_kmalloc(gfp_mask, nr_pages);
832 bio->bi_rw |= (!write_to_vm << BIO_RW);
837 nr_pages = 1 << map_data->page_order;
838 i = map_data->offset / PAGE_SIZE;
841 unsigned int bytes = PAGE_SIZE;
849 if (i == map_data->nr_entries * nr_pages) {
854 page = map_data->pages[i / nr_pages];
855 page += (i % nr_pages);
859 page = alloc_page(q->bounce_gfp | gfp_mask);
866 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
879 if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
880 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
885 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
889 bio_for_each_segment(bvec, bio, i)
890 __free_page(bvec->bv_page);
894 bio_free_map_data(bmd);
899 * bio_copy_user - copy user data to bio
900 * @q: destination block queue
901 * @map_data: pointer to the rq_map_data holding pages (if necessary)
902 * @uaddr: start of user address
903 * @len: length in bytes
904 * @write_to_vm: bool indicating writing to pages or not
905 * @gfp_mask: memory allocation flags
907 * Prepares and returns a bio for indirect user io, bouncing data
908 * to/from kernel pages as necessary. Must be paired with
909 * call bio_uncopy_user() on io completion.
911 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
912 unsigned long uaddr, unsigned int len,
913 int write_to_vm, gfp_t gfp_mask)
917 iov.iov_base = (void __user *)uaddr;
920 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
923 static struct bio *__bio_map_user_iov(struct request_queue *q,
924 struct block_device *bdev,
925 struct sg_iovec *iov, int iov_count,
926 int write_to_vm, gfp_t gfp_mask)
935 for (i = 0; i < iov_count; i++) {
936 unsigned long uaddr = (unsigned long)iov[i].iov_base;
937 unsigned long len = iov[i].iov_len;
938 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
939 unsigned long start = uaddr >> PAGE_SHIFT;
941 nr_pages += end - start;
943 * buffer must be aligned to at least hardsector size for now
945 if (uaddr & queue_dma_alignment(q))
946 return ERR_PTR(-EINVAL);
950 return ERR_PTR(-EINVAL);
952 bio = bio_kmalloc(gfp_mask, nr_pages);
954 return ERR_PTR(-ENOMEM);
957 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
961 for (i = 0; i < iov_count; i++) {
962 unsigned long uaddr = (unsigned long)iov[i].iov_base;
963 unsigned long len = iov[i].iov_len;
964 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
965 unsigned long start = uaddr >> PAGE_SHIFT;
966 const int local_nr_pages = end - start;
967 const int page_limit = cur_page + local_nr_pages;
969 ret = get_user_pages_fast(uaddr, local_nr_pages,
970 write_to_vm, &pages[cur_page]);
971 if (ret < local_nr_pages) {
976 offset = uaddr & ~PAGE_MASK;
977 for (j = cur_page; j < page_limit; j++) {
978 unsigned int bytes = PAGE_SIZE - offset;
989 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
999 * release the pages we didn't map into the bio, if any
1001 while (j < page_limit)
1002 page_cache_release(pages[j++]);
1008 * set data direction, and check if mapped pages need bouncing
1011 bio->bi_rw |= (1 << BIO_RW);
1013 bio->bi_bdev = bdev;
1014 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1018 for (i = 0; i < nr_pages; i++) {
1021 page_cache_release(pages[i]);
1026 return ERR_PTR(ret);
1030 * bio_map_user - map user address into bio
1031 * @q: the struct request_queue for the bio
1032 * @bdev: destination block device
1033 * @uaddr: start of user address
1034 * @len: length in bytes
1035 * @write_to_vm: bool indicating writing to pages or not
1036 * @gfp_mask: memory allocation flags
1038 * Map the user space address into a bio suitable for io to a block
1039 * device. Returns an error pointer in case of error.
1041 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1042 unsigned long uaddr, unsigned int len, int write_to_vm,
1045 struct sg_iovec iov;
1047 iov.iov_base = (void __user *)uaddr;
1050 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1054 * bio_map_user_iov - map user sg_iovec table into bio
1055 * @q: the struct request_queue for the bio
1056 * @bdev: destination block device
1058 * @iov_count: number of elements in the iovec
1059 * @write_to_vm: bool indicating writing to pages or not
1060 * @gfp_mask: memory allocation flags
1062 * Map the user space address into a bio suitable for io to a block
1063 * device. Returns an error pointer in case of error.
1065 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1066 struct sg_iovec *iov, int iov_count,
1067 int write_to_vm, gfp_t gfp_mask)
1071 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1077 * subtle -- if __bio_map_user() ended up bouncing a bio,
1078 * it would normally disappear when its bi_end_io is run.
1079 * however, we need it for the unmap, so grab an extra
1087 static void __bio_unmap_user(struct bio *bio)
1089 struct bio_vec *bvec;
1093 * make sure we dirty pages we wrote to
1095 __bio_for_each_segment(bvec, bio, i, 0) {
1096 if (bio_data_dir(bio) == READ)
1097 set_page_dirty_lock(bvec->bv_page);
1099 page_cache_release(bvec->bv_page);
1106 * bio_unmap_user - unmap a bio
1107 * @bio: the bio being unmapped
1109 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1110 * a process context.
1112 * bio_unmap_user() may sleep.
1114 void bio_unmap_user(struct bio *bio)
1116 __bio_unmap_user(bio);
1120 static void bio_map_kern_endio(struct bio *bio, int err)
1126 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1127 unsigned int len, gfp_t gfp_mask)
1129 unsigned long kaddr = (unsigned long)data;
1130 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1131 unsigned long start = kaddr >> PAGE_SHIFT;
1132 const int nr_pages = end - start;
1136 bio = bio_kmalloc(gfp_mask, nr_pages);
1138 return ERR_PTR(-ENOMEM);
1140 offset = offset_in_page(kaddr);
1141 for (i = 0; i < nr_pages; i++) {
1142 unsigned int bytes = PAGE_SIZE - offset;
1150 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1159 bio->bi_end_io = bio_map_kern_endio;
1164 * bio_map_kern - map kernel address into bio
1165 * @q: the struct request_queue for the bio
1166 * @data: pointer to buffer to map
1167 * @len: length in bytes
1168 * @gfp_mask: allocation flags for bio allocation
1170 * Map the kernel address into a bio suitable for io to a block
1171 * device. Returns an error pointer in case of error.
1173 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1178 bio = __bio_map_kern(q, data, len, gfp_mask);
1182 if (bio->bi_size == len)
1186 * Don't support partial mappings.
1189 return ERR_PTR(-EINVAL);
1192 static void bio_copy_kern_endio(struct bio *bio, int err)
1194 struct bio_vec *bvec;
1195 const int read = bio_data_dir(bio) == READ;
1196 struct bio_map_data *bmd = bio->bi_private;
1198 char *p = bmd->sgvecs[0].iov_base;
1200 __bio_for_each_segment(bvec, bio, i, 0) {
1201 char *addr = page_address(bvec->bv_page);
1202 int len = bmd->iovecs[i].bv_len;
1205 memcpy(p, addr, len);
1207 __free_page(bvec->bv_page);
1211 bio_free_map_data(bmd);
1216 * bio_copy_kern - copy kernel address into bio
1217 * @q: the struct request_queue for the bio
1218 * @data: pointer to buffer to copy
1219 * @len: length in bytes
1220 * @gfp_mask: allocation flags for bio and page allocation
1221 * @reading: data direction is READ
1223 * copy the kernel address into a bio suitable for io to a block
1224 * device. Returns an error pointer in case of error.
1226 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1227 gfp_t gfp_mask, int reading)
1230 struct bio_vec *bvec;
1233 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1240 bio_for_each_segment(bvec, bio, i) {
1241 char *addr = page_address(bvec->bv_page);
1243 memcpy(addr, p, bvec->bv_len);
1248 bio->bi_end_io = bio_copy_kern_endio;
1254 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1255 * for performing direct-IO in BIOs.
1257 * The problem is that we cannot run set_page_dirty() from interrupt context
1258 * because the required locks are not interrupt-safe. So what we can do is to
1259 * mark the pages dirty _before_ performing IO. And in interrupt context,
1260 * check that the pages are still dirty. If so, fine. If not, redirty them
1261 * in process context.
1263 * We special-case compound pages here: normally this means reads into hugetlb
1264 * pages. The logic in here doesn't really work right for compound pages
1265 * because the VM does not uniformly chase down the head page in all cases.
1266 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1267 * handle them at all. So we skip compound pages here at an early stage.
1269 * Note that this code is very hard to test under normal circumstances because
1270 * direct-io pins the pages with get_user_pages(). This makes
1271 * is_page_cache_freeable return false, and the VM will not clean the pages.
1272 * But other code (eg, pdflush) could clean the pages if they are mapped
1275 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1276 * deferred bio dirtying paths.
1280 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1282 void bio_set_pages_dirty(struct bio *bio)
1284 struct bio_vec *bvec = bio->bi_io_vec;
1287 for (i = 0; i < bio->bi_vcnt; i++) {
1288 struct page *page = bvec[i].bv_page;
1290 if (page && !PageCompound(page))
1291 set_page_dirty_lock(page);
1295 static void bio_release_pages(struct bio *bio)
1297 struct bio_vec *bvec = bio->bi_io_vec;
1300 for (i = 0; i < bio->bi_vcnt; i++) {
1301 struct page *page = bvec[i].bv_page;
1309 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1310 * If they are, then fine. If, however, some pages are clean then they must
1311 * have been written out during the direct-IO read. So we take another ref on
1312 * the BIO and the offending pages and re-dirty the pages in process context.
1314 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1315 * here on. It will run one page_cache_release() against each page and will
1316 * run one bio_put() against the BIO.
1319 static void bio_dirty_fn(struct work_struct *work);
1321 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1322 static DEFINE_SPINLOCK(bio_dirty_lock);
1323 static struct bio *bio_dirty_list;
1326 * This runs in process context
1328 static void bio_dirty_fn(struct work_struct *work)
1330 unsigned long flags;
1333 spin_lock_irqsave(&bio_dirty_lock, flags);
1334 bio = bio_dirty_list;
1335 bio_dirty_list = NULL;
1336 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1339 struct bio *next = bio->bi_private;
1341 bio_set_pages_dirty(bio);
1342 bio_release_pages(bio);
1348 void bio_check_pages_dirty(struct bio *bio)
1350 struct bio_vec *bvec = bio->bi_io_vec;
1351 int nr_clean_pages = 0;
1354 for (i = 0; i < bio->bi_vcnt; i++) {
1355 struct page *page = bvec[i].bv_page;
1357 if (PageDirty(page) || PageCompound(page)) {
1358 page_cache_release(page);
1359 bvec[i].bv_page = NULL;
1365 if (nr_clean_pages) {
1366 unsigned long flags;
1368 spin_lock_irqsave(&bio_dirty_lock, flags);
1369 bio->bi_private = bio_dirty_list;
1370 bio_dirty_list = bio;
1371 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1372 schedule_work(&bio_dirty_work);
1379 * bio_endio - end I/O on a bio
1381 * @error: error, if any
1384 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1385 * preferred way to end I/O on a bio, it takes care of clearing
1386 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1387 * established -Exxxx (-EIO, for instance) error values in case
1388 * something went wrong. Noone should call bi_end_io() directly on a
1389 * bio unless they own it and thus know that it has an end_io
1392 void bio_endio(struct bio *bio, int error)
1395 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1396 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1400 bio->bi_end_io(bio, error);
1403 void bio_pair_release(struct bio_pair *bp)
1405 if (atomic_dec_and_test(&bp->cnt)) {
1406 struct bio *master = bp->bio1.bi_private;
1408 bio_endio(master, bp->error);
1409 mempool_free(bp, bp->bio2.bi_private);
1413 static void bio_pair_end_1(struct bio *bi, int err)
1415 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1420 bio_pair_release(bp);
1423 static void bio_pair_end_2(struct bio *bi, int err)
1425 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1430 bio_pair_release(bp);
1434 * split a bio - only worry about a bio with a single page in its iovec
1436 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1438 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1443 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1444 bi->bi_sector + first_sectors);
1446 BUG_ON(bi->bi_vcnt != 1);
1447 BUG_ON(bi->bi_idx != 0);
1448 atomic_set(&bp->cnt, 3);
1452 bp->bio2.bi_sector += first_sectors;
1453 bp->bio2.bi_size -= first_sectors << 9;
1454 bp->bio1.bi_size = first_sectors << 9;
1456 bp->bv1 = bi->bi_io_vec[0];
1457 bp->bv2 = bi->bi_io_vec[0];
1458 bp->bv2.bv_offset += first_sectors << 9;
1459 bp->bv2.bv_len -= first_sectors << 9;
1460 bp->bv1.bv_len = first_sectors << 9;
1462 bp->bio1.bi_io_vec = &bp->bv1;
1463 bp->bio2.bi_io_vec = &bp->bv2;
1465 bp->bio1.bi_max_vecs = 1;
1466 bp->bio2.bi_max_vecs = 1;
1468 bp->bio1.bi_end_io = bio_pair_end_1;
1469 bp->bio2.bi_end_io = bio_pair_end_2;
1471 bp->bio1.bi_private = bi;
1472 bp->bio2.bi_private = bio_split_pool;
1474 if (bio_integrity(bi))
1475 bio_integrity_split(bi, bp, first_sectors);
1481 * bio_sector_offset - Find hardware sector offset in bio
1482 * @bio: bio to inspect
1483 * @index: bio_vec index
1484 * @offset: offset in bv_page
1486 * Return the number of hardware sectors between beginning of bio
1487 * and an end point indicated by a bio_vec index and an offset
1488 * within that vector's page.
1490 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1491 unsigned int offset)
1493 unsigned int sector_sz;
1498 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1501 if (index >= bio->bi_idx)
1502 index = bio->bi_vcnt - 1;
1504 __bio_for_each_segment(bv, bio, i, 0) {
1506 if (offset > bv->bv_offset)
1507 sectors += (offset - bv->bv_offset) / sector_sz;
1511 sectors += bv->bv_len / sector_sz;
1516 EXPORT_SYMBOL(bio_sector_offset);
1519 * create memory pools for biovec's in a bio_set.
1520 * use the global biovec slabs created for general use.
1522 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1524 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1526 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1533 static void biovec_free_pools(struct bio_set *bs)
1535 mempool_destroy(bs->bvec_pool);
1538 void bioset_free(struct bio_set *bs)
1541 mempool_destroy(bs->bio_pool);
1543 biovec_free_pools(bs);
1550 * bioset_create - Create a bio_set
1551 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1552 * @front_pad: Number of bytes to allocate in front of the returned bio
1555 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1556 * to ask for a number of bytes to be allocated in front of the bio.
1557 * Front pad allocation is useful for embedding the bio inside
1558 * another structure, to avoid allocating extra data to go with the bio.
1559 * Note that the bio must be embedded at the END of that structure always,
1560 * or things will break badly.
1562 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1564 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1567 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1571 bs->front_pad = front_pad;
1573 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1574 if (!bs->bio_slab) {
1579 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1583 if (!biovec_create_pools(bs, pool_size))
1591 static void __init biovec_init_slabs(void)
1595 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1597 struct biovec_slab *bvs = bvec_slabs + i;
1599 #ifndef CONFIG_BLK_DEV_INTEGRITY
1600 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1606 size = bvs->nr_vecs * sizeof(struct bio_vec);
1607 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1608 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1612 static int __init init_bio(void)
1616 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1618 panic("bio: can't allocate bios\n");
1620 biovec_init_slabs();
1622 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1624 panic("bio: can't allocate bios\n");
1626 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1627 sizeof(struct bio_pair));
1628 if (!bio_split_pool)
1629 panic("bio: can't create split pool\n");
1634 subsys_initcall(init_bio);
1636 EXPORT_SYMBOL(bio_alloc);
1637 EXPORT_SYMBOL(bio_kmalloc);
1638 EXPORT_SYMBOL(bio_put);
1639 EXPORT_SYMBOL(bio_free);
1640 EXPORT_SYMBOL(bio_endio);
1641 EXPORT_SYMBOL(bio_init);
1642 EXPORT_SYMBOL(__bio_clone);
1643 EXPORT_SYMBOL(bio_clone);
1644 EXPORT_SYMBOL(bio_phys_segments);
1645 EXPORT_SYMBOL(bio_add_page);
1646 EXPORT_SYMBOL(bio_add_pc_page);
1647 EXPORT_SYMBOL(bio_get_nr_vecs);
1648 EXPORT_SYMBOL(bio_map_user);
1649 EXPORT_SYMBOL(bio_unmap_user);
1650 EXPORT_SYMBOL(bio_map_kern);
1651 EXPORT_SYMBOL(bio_copy_kern);
1652 EXPORT_SYMBOL(bio_pair_release);
1653 EXPORT_SYMBOL(bio_split);
1654 EXPORT_SYMBOL(bio_copy_user);
1655 EXPORT_SYMBOL(bio_uncopy_user);
1656 EXPORT_SYMBOL(bioset_create);
1657 EXPORT_SYMBOL(bioset_free);
1658 EXPORT_SYMBOL(bio_alloc_bioset);