2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/mempool.h>
27 #include <linux/workqueue.h>
28 #include <linux/blktrace_api.h>
29 #include <trace/block.h>
30 #include <scsi/sg.h> /* for struct sg_iovec */
32 DEFINE_TRACE(block_split);
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
38 #define BIO_INLINE_VECS 4
40 static mempool_t *bio_split_pool __read_mostly;
43 * if you change this list, also change bvec_alloc or things will
44 * break badly! cannot be bigger than what you can fit into an
47 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
48 struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
49 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
54 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
55 * IO code that does not need private memory pools.
57 struct bio_set *fs_bio_set;
60 * Our slab pool management
63 struct kmem_cache *slab;
64 unsigned int slab_ref;
65 unsigned int slab_size;
68 static DEFINE_MUTEX(bio_slab_lock);
69 static struct bio_slab *bio_slabs;
70 static unsigned int bio_slab_nr, bio_slab_max;
72 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
74 unsigned int sz = sizeof(struct bio) + extra_size;
75 struct kmem_cache *slab = NULL;
76 struct bio_slab *bslab;
77 unsigned int i, entry = -1;
79 mutex_lock(&bio_slab_lock);
82 while (i < bio_slab_nr) {
83 struct bio_slab *bslab = &bio_slabs[i];
85 if (!bslab->slab && entry == -1)
87 else if (bslab->slab_size == sz) {
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
100 bio_slabs = krealloc(bio_slabs,
101 bio_slab_max * sizeof(struct bio_slab),
107 entry = bio_slab_nr++;
109 bslab = &bio_slabs[entry];
111 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
112 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
116 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
119 bslab->slab_size = sz;
121 mutex_unlock(&bio_slab_lock);
125 static void bio_put_slab(struct bio_set *bs)
127 struct bio_slab *bslab = NULL;
130 mutex_lock(&bio_slab_lock);
132 for (i = 0; i < bio_slab_nr; i++) {
133 if (bs->bio_slab == bio_slabs[i].slab) {
134 bslab = &bio_slabs[i];
139 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 WARN_ON(!bslab->slab_ref);
144 if (--bslab->slab_ref)
147 kmem_cache_destroy(bslab->slab);
151 mutex_unlock(&bio_slab_lock);
154 unsigned int bvec_nr_vecs(unsigned short idx)
156 return bvec_slabs[idx].nr_vecs;
159 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
161 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
163 if (idx == BIOVEC_MAX_IDX)
164 mempool_free(bv, bs->bvec_pool);
166 struct biovec_slab *bvs = bvec_slabs + idx;
168 kmem_cache_free(bvs->slab, bv);
172 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
178 * If 'bs' is given, lookup the pool and do the mempool alloc.
179 * If not, this is a bio_kmalloc() allocation and just do a
180 * kzalloc() for the exact number of vecs right away.
183 bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask);
186 * see comment near bvec_array define!
204 case 129 ... BIO_MAX_PAGES:
212 * idx now points to the pool we want to allocate from. only the
213 * 1-vec entry pool is mempool backed.
215 if (*idx == BIOVEC_MAX_IDX) {
217 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
219 struct biovec_slab *bvs = bvec_slabs + *idx;
220 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
223 * Make this allocation restricted and don't dump info on
224 * allocation failures, since we'll fallback to the mempool
225 * in case of failure.
227 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
230 * Try a slab allocation. If this fails and __GFP_WAIT
231 * is set, retry with the 1-entry mempool
233 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
234 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
235 *idx = BIOVEC_MAX_IDX;
241 memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec));
246 void bio_free(struct bio *bio, struct bio_set *bs)
250 if (bio_has_allocated_vec(bio))
251 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
253 if (bio_integrity(bio))
254 bio_integrity_free(bio, bs);
257 * If we have front padding, adjust the bio pointer before freeing
263 mempool_free(p, bs->bio_pool);
267 * default destructor for a bio allocated with bio_alloc_bioset()
269 static void bio_fs_destructor(struct bio *bio)
271 bio_free(bio, fs_bio_set);
274 static void bio_kmalloc_destructor(struct bio *bio)
276 if (bio_has_allocated_vec(bio))
277 kfree(bio->bi_io_vec);
281 void bio_init(struct bio *bio)
283 memset(bio, 0, sizeof(*bio));
284 bio->bi_flags = 1 << BIO_UPTODATE;
285 bio->bi_comp_cpu = -1;
286 atomic_set(&bio->bi_cnt, 1);
290 * bio_alloc_bioset - allocate a bio for I/O
291 * @gfp_mask: the GFP_ mask given to the slab allocator
292 * @nr_iovecs: number of iovecs to pre-allocate
293 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
296 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
297 * If %__GFP_WAIT is set then we will block on the internal pool waiting
298 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
299 * fall back to just using @kmalloc to allocate the required memory.
301 * Note that the caller must set ->bi_destructor on succesful return
302 * of a bio, to do the appropriate freeing of the bio once the reference
303 * count drops to zero.
305 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
307 struct bio *bio = NULL;
310 void *p = mempool_alloc(bs->bio_pool, gfp_mask);
313 bio = p + bs->front_pad;
315 bio = kmalloc(sizeof(*bio), gfp_mask);
318 struct bio_vec *bvl = NULL;
321 if (likely(nr_iovecs)) {
322 unsigned long uninitialized_var(idx);
324 if (nr_iovecs <= BIO_INLINE_VECS) {
326 bvl = bio->bi_inline_vecs;
327 nr_iovecs = BIO_INLINE_VECS;
328 memset(bvl, 0, BIO_INLINE_VECS * sizeof(*bvl));
330 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx,
332 nr_iovecs = bvec_nr_vecs(idx);
334 if (unlikely(!bvl)) {
336 mempool_free(bio, bs->bio_pool);
342 bio->bi_flags |= idx << BIO_POOL_OFFSET;
343 bio->bi_max_vecs = nr_iovecs;
345 bio->bi_io_vec = bvl;
351 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
353 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
356 bio->bi_destructor = bio_fs_destructor;
362 * Like bio_alloc(), but doesn't use a mempool backing. This means that
363 * it CAN fail, but while bio_alloc() can only be used for allocations
364 * that have a short (finite) life span, bio_kmalloc() should be used
365 * for more permanent bio allocations (like allocating some bio's for
366 * initalization or setup purposes).
368 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
370 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL);
373 bio->bi_destructor = bio_kmalloc_destructor;
378 void zero_fill_bio(struct bio *bio)
384 bio_for_each_segment(bv, bio, i) {
385 char *data = bvec_kmap_irq(bv, &flags);
386 memset(data, 0, bv->bv_len);
387 flush_dcache_page(bv->bv_page);
388 bvec_kunmap_irq(data, &flags);
391 EXPORT_SYMBOL(zero_fill_bio);
394 * bio_put - release a reference to a bio
395 * @bio: bio to release reference to
398 * Put a reference to a &struct bio, either one you have gotten with
399 * bio_alloc or bio_get. The last put of a bio will free it.
401 void bio_put(struct bio *bio)
403 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
408 if (atomic_dec_and_test(&bio->bi_cnt)) {
410 bio->bi_destructor(bio);
414 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
416 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
417 blk_recount_segments(q, bio);
419 return bio->bi_phys_segments;
423 * __bio_clone - clone a bio
424 * @bio: destination bio
425 * @bio_src: bio to clone
427 * Clone a &bio. Caller will own the returned bio, but not
428 * the actual data it points to. Reference count of returned
431 void __bio_clone(struct bio *bio, struct bio *bio_src)
433 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
434 bio_src->bi_max_vecs * sizeof(struct bio_vec));
437 * most users will be overriding ->bi_bdev with a new target,
438 * so we don't set nor calculate new physical/hw segment counts here
440 bio->bi_sector = bio_src->bi_sector;
441 bio->bi_bdev = bio_src->bi_bdev;
442 bio->bi_flags |= 1 << BIO_CLONED;
443 bio->bi_rw = bio_src->bi_rw;
444 bio->bi_vcnt = bio_src->bi_vcnt;
445 bio->bi_size = bio_src->bi_size;
446 bio->bi_idx = bio_src->bi_idx;
450 * bio_clone - clone a bio
452 * @gfp_mask: allocation priority
454 * Like __bio_clone, only also allocates the returned bio
456 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
458 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
463 b->bi_destructor = bio_fs_destructor;
466 if (bio_integrity(bio)) {
469 ret = bio_integrity_clone(b, bio, fs_bio_set);
479 * bio_get_nr_vecs - return approx number of vecs
482 * Return the approximate number of pages we can send to this target.
483 * There's no guarantee that you will be able to fit this number of pages
484 * into a bio, it does not account for dynamic restrictions that vary
487 int bio_get_nr_vecs(struct block_device *bdev)
489 struct request_queue *q = bdev_get_queue(bdev);
492 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
493 if (nr_pages > q->max_phys_segments)
494 nr_pages = q->max_phys_segments;
495 if (nr_pages > q->max_hw_segments)
496 nr_pages = q->max_hw_segments;
501 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
502 *page, unsigned int len, unsigned int offset,
503 unsigned short max_sectors)
505 int retried_segments = 0;
506 struct bio_vec *bvec;
509 * cloned bio must not modify vec list
511 if (unlikely(bio_flagged(bio, BIO_CLONED)))
514 if (((bio->bi_size + len) >> 9) > max_sectors)
518 * For filesystems with a blocksize smaller than the pagesize
519 * we will often be called with the same page as last time and
520 * a consecutive offset. Optimize this special case.
522 if (bio->bi_vcnt > 0) {
523 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
525 if (page == prev->bv_page &&
526 offset == prev->bv_offset + prev->bv_len) {
529 if (q->merge_bvec_fn) {
530 struct bvec_merge_data bvm = {
531 .bi_bdev = bio->bi_bdev,
532 .bi_sector = bio->bi_sector,
533 .bi_size = bio->bi_size,
537 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
547 if (bio->bi_vcnt >= bio->bi_max_vecs)
551 * we might lose a segment or two here, but rather that than
552 * make this too complex.
555 while (bio->bi_phys_segments >= q->max_phys_segments
556 || bio->bi_phys_segments >= q->max_hw_segments) {
558 if (retried_segments)
561 retried_segments = 1;
562 blk_recount_segments(q, bio);
566 * setup the new entry, we might clear it again later if we
567 * cannot add the page
569 bvec = &bio->bi_io_vec[bio->bi_vcnt];
570 bvec->bv_page = page;
572 bvec->bv_offset = offset;
575 * if queue has other restrictions (eg varying max sector size
576 * depending on offset), it can specify a merge_bvec_fn in the
577 * queue to get further control
579 if (q->merge_bvec_fn) {
580 struct bvec_merge_data bvm = {
581 .bi_bdev = bio->bi_bdev,
582 .bi_sector = bio->bi_sector,
583 .bi_size = bio->bi_size,
588 * merge_bvec_fn() returns number of bytes it can accept
591 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
592 bvec->bv_page = NULL;
599 /* If we may be able to merge these biovecs, force a recount */
600 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
601 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
604 bio->bi_phys_segments++;
611 * bio_add_pc_page - attempt to add page to bio
612 * @q: the target queue
613 * @bio: destination bio
615 * @len: vec entry length
616 * @offset: vec entry offset
618 * Attempt to add a page to the bio_vec maplist. This can fail for a
619 * number of reasons, such as the bio being full or target block
620 * device limitations. The target block device must allow bio's
621 * smaller than PAGE_SIZE, so it is always possible to add a single
622 * page to an empty bio. This should only be used by REQ_PC bios.
624 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
625 unsigned int len, unsigned int offset)
627 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
631 * bio_add_page - attempt to add page to bio
632 * @bio: destination bio
634 * @len: vec entry length
635 * @offset: vec entry offset
637 * Attempt to add a page to the bio_vec maplist. This can fail for a
638 * number of reasons, such as the bio being full or target block
639 * device limitations. The target block device must allow bio's
640 * smaller than PAGE_SIZE, so it is always possible to add a single
641 * page to an empty bio.
643 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
646 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
647 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
650 struct bio_map_data {
651 struct bio_vec *iovecs;
652 struct sg_iovec *sgvecs;
657 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
658 struct sg_iovec *iov, int iov_count,
661 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
662 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
663 bmd->nr_sgvecs = iov_count;
664 bmd->is_our_pages = is_our_pages;
665 bio->bi_private = bmd;
668 static void bio_free_map_data(struct bio_map_data *bmd)
675 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
678 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
683 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
689 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
698 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
699 struct sg_iovec *iov, int iov_count, int uncopy,
703 struct bio_vec *bvec;
705 unsigned int iov_off = 0;
706 int read = bio_data_dir(bio) == READ;
708 __bio_for_each_segment(bvec, bio, i, 0) {
709 char *bv_addr = page_address(bvec->bv_page);
710 unsigned int bv_len = iovecs[i].bv_len;
712 while (bv_len && iov_idx < iov_count) {
716 bytes = min_t(unsigned int,
717 iov[iov_idx].iov_len - iov_off, bv_len);
718 iov_addr = iov[iov_idx].iov_base + iov_off;
721 if (!read && !uncopy)
722 ret = copy_from_user(bv_addr, iov_addr,
725 ret = copy_to_user(iov_addr, bv_addr,
737 if (iov[iov_idx].iov_len == iov_off) {
744 __free_page(bvec->bv_page);
751 * bio_uncopy_user - finish previously mapped bio
752 * @bio: bio being terminated
754 * Free pages allocated from bio_copy_user() and write back data
755 * to user space in case of a read.
757 int bio_uncopy_user(struct bio *bio)
759 struct bio_map_data *bmd = bio->bi_private;
762 if (!bio_flagged(bio, BIO_NULL_MAPPED))
763 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
764 bmd->nr_sgvecs, 1, bmd->is_our_pages);
765 bio_free_map_data(bmd);
771 * bio_copy_user_iov - copy user data to bio
772 * @q: destination block queue
773 * @map_data: pointer to the rq_map_data holding pages (if necessary)
775 * @iov_count: number of elements in the iovec
776 * @write_to_vm: bool indicating writing to pages or not
777 * @gfp_mask: memory allocation flags
779 * Prepares and returns a bio for indirect user io, bouncing data
780 * to/from kernel pages as necessary. Must be paired with
781 * call bio_uncopy_user() on io completion.
783 struct bio *bio_copy_user_iov(struct request_queue *q,
784 struct rq_map_data *map_data,
785 struct sg_iovec *iov, int iov_count,
786 int write_to_vm, gfp_t gfp_mask)
788 struct bio_map_data *bmd;
789 struct bio_vec *bvec;
794 unsigned int len = 0;
796 for (i = 0; i < iov_count; i++) {
801 uaddr = (unsigned long)iov[i].iov_base;
802 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
803 start = uaddr >> PAGE_SHIFT;
805 nr_pages += end - start;
806 len += iov[i].iov_len;
809 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
811 return ERR_PTR(-ENOMEM);
814 bio = bio_alloc(gfp_mask, nr_pages);
818 bio->bi_rw |= (!write_to_vm << BIO_RW);
826 bytes = 1U << (PAGE_SHIFT + map_data->page_order);
834 if (i == map_data->nr_entries) {
838 page = map_data->pages[i++];
840 page = alloc_page(q->bounce_gfp | gfp_mask);
846 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
859 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
864 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
868 bio_for_each_segment(bvec, bio, i)
869 __free_page(bvec->bv_page);
873 bio_free_map_data(bmd);
878 * bio_copy_user - copy user data to bio
879 * @q: destination block queue
880 * @map_data: pointer to the rq_map_data holding pages (if necessary)
881 * @uaddr: start of user address
882 * @len: length in bytes
883 * @write_to_vm: bool indicating writing to pages or not
884 * @gfp_mask: memory allocation flags
886 * Prepares and returns a bio for indirect user io, bouncing data
887 * to/from kernel pages as necessary. Must be paired with
888 * call bio_uncopy_user() on io completion.
890 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
891 unsigned long uaddr, unsigned int len,
892 int write_to_vm, gfp_t gfp_mask)
896 iov.iov_base = (void __user *)uaddr;
899 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
902 static struct bio *__bio_map_user_iov(struct request_queue *q,
903 struct block_device *bdev,
904 struct sg_iovec *iov, int iov_count,
905 int write_to_vm, gfp_t gfp_mask)
914 for (i = 0; i < iov_count; i++) {
915 unsigned long uaddr = (unsigned long)iov[i].iov_base;
916 unsigned long len = iov[i].iov_len;
917 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
918 unsigned long start = uaddr >> PAGE_SHIFT;
920 nr_pages += end - start;
922 * buffer must be aligned to at least hardsector size for now
924 if (uaddr & queue_dma_alignment(q))
925 return ERR_PTR(-EINVAL);
929 return ERR_PTR(-EINVAL);
931 bio = bio_alloc(gfp_mask, nr_pages);
933 return ERR_PTR(-ENOMEM);
936 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
940 for (i = 0; i < iov_count; i++) {
941 unsigned long uaddr = (unsigned long)iov[i].iov_base;
942 unsigned long len = iov[i].iov_len;
943 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
944 unsigned long start = uaddr >> PAGE_SHIFT;
945 const int local_nr_pages = end - start;
946 const int page_limit = cur_page + local_nr_pages;
948 ret = get_user_pages_fast(uaddr, local_nr_pages,
949 write_to_vm, &pages[cur_page]);
950 if (ret < local_nr_pages) {
955 offset = uaddr & ~PAGE_MASK;
956 for (j = cur_page; j < page_limit; j++) {
957 unsigned int bytes = PAGE_SIZE - offset;
968 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
978 * release the pages we didn't map into the bio, if any
980 while (j < page_limit)
981 page_cache_release(pages[j++]);
987 * set data direction, and check if mapped pages need bouncing
990 bio->bi_rw |= (1 << BIO_RW);
993 bio->bi_flags |= (1 << BIO_USER_MAPPED);
997 for (i = 0; i < nr_pages; i++) {
1000 page_cache_release(pages[i]);
1005 return ERR_PTR(ret);
1009 * bio_map_user - map user address into bio
1010 * @q: the struct request_queue for the bio
1011 * @bdev: destination block device
1012 * @uaddr: start of user address
1013 * @len: length in bytes
1014 * @write_to_vm: bool indicating writing to pages or not
1015 * @gfp_mask: memory allocation flags
1017 * Map the user space address into a bio suitable for io to a block
1018 * device. Returns an error pointer in case of error.
1020 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1021 unsigned long uaddr, unsigned int len, int write_to_vm,
1024 struct sg_iovec iov;
1026 iov.iov_base = (void __user *)uaddr;
1029 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1033 * bio_map_user_iov - map user sg_iovec table into bio
1034 * @q: the struct request_queue for the bio
1035 * @bdev: destination block device
1037 * @iov_count: number of elements in the iovec
1038 * @write_to_vm: bool indicating writing to pages or not
1039 * @gfp_mask: memory allocation flags
1041 * Map the user space address into a bio suitable for io to a block
1042 * device. Returns an error pointer in case of error.
1044 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1045 struct sg_iovec *iov, int iov_count,
1046 int write_to_vm, gfp_t gfp_mask)
1050 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1056 * subtle -- if __bio_map_user() ended up bouncing a bio,
1057 * it would normally disappear when its bi_end_io is run.
1058 * however, we need it for the unmap, so grab an extra
1066 static void __bio_unmap_user(struct bio *bio)
1068 struct bio_vec *bvec;
1072 * make sure we dirty pages we wrote to
1074 __bio_for_each_segment(bvec, bio, i, 0) {
1075 if (bio_data_dir(bio) == READ)
1076 set_page_dirty_lock(bvec->bv_page);
1078 page_cache_release(bvec->bv_page);
1085 * bio_unmap_user - unmap a bio
1086 * @bio: the bio being unmapped
1088 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1089 * a process context.
1091 * bio_unmap_user() may sleep.
1093 void bio_unmap_user(struct bio *bio)
1095 __bio_unmap_user(bio);
1099 static void bio_map_kern_endio(struct bio *bio, int err)
1105 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1106 unsigned int len, gfp_t gfp_mask)
1108 unsigned long kaddr = (unsigned long)data;
1109 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1110 unsigned long start = kaddr >> PAGE_SHIFT;
1111 const int nr_pages = end - start;
1115 bio = bio_alloc(gfp_mask, nr_pages);
1117 return ERR_PTR(-ENOMEM);
1119 offset = offset_in_page(kaddr);
1120 for (i = 0; i < nr_pages; i++) {
1121 unsigned int bytes = PAGE_SIZE - offset;
1129 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1138 bio->bi_end_io = bio_map_kern_endio;
1143 * bio_map_kern - map kernel address into bio
1144 * @q: the struct request_queue for the bio
1145 * @data: pointer to buffer to map
1146 * @len: length in bytes
1147 * @gfp_mask: allocation flags for bio allocation
1149 * Map the kernel address into a bio suitable for io to a block
1150 * device. Returns an error pointer in case of error.
1152 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1157 bio = __bio_map_kern(q, data, len, gfp_mask);
1161 if (bio->bi_size == len)
1165 * Don't support partial mappings.
1168 return ERR_PTR(-EINVAL);
1171 static void bio_copy_kern_endio(struct bio *bio, int err)
1173 struct bio_vec *bvec;
1174 const int read = bio_data_dir(bio) == READ;
1175 struct bio_map_data *bmd = bio->bi_private;
1177 char *p = bmd->sgvecs[0].iov_base;
1179 __bio_for_each_segment(bvec, bio, i, 0) {
1180 char *addr = page_address(bvec->bv_page);
1181 int len = bmd->iovecs[i].bv_len;
1184 memcpy(p, addr, len);
1186 __free_page(bvec->bv_page);
1190 bio_free_map_data(bmd);
1195 * bio_copy_kern - copy kernel address into bio
1196 * @q: the struct request_queue for the bio
1197 * @data: pointer to buffer to copy
1198 * @len: length in bytes
1199 * @gfp_mask: allocation flags for bio and page allocation
1200 * @reading: data direction is READ
1202 * copy the kernel address into a bio suitable for io to a block
1203 * device. Returns an error pointer in case of error.
1205 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1206 gfp_t gfp_mask, int reading)
1209 struct bio_vec *bvec;
1212 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1219 bio_for_each_segment(bvec, bio, i) {
1220 char *addr = page_address(bvec->bv_page);
1222 memcpy(addr, p, bvec->bv_len);
1227 bio->bi_end_io = bio_copy_kern_endio;
1233 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1234 * for performing direct-IO in BIOs.
1236 * The problem is that we cannot run set_page_dirty() from interrupt context
1237 * because the required locks are not interrupt-safe. So what we can do is to
1238 * mark the pages dirty _before_ performing IO. And in interrupt context,
1239 * check that the pages are still dirty. If so, fine. If not, redirty them
1240 * in process context.
1242 * We special-case compound pages here: normally this means reads into hugetlb
1243 * pages. The logic in here doesn't really work right for compound pages
1244 * because the VM does not uniformly chase down the head page in all cases.
1245 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1246 * handle them at all. So we skip compound pages here at an early stage.
1248 * Note that this code is very hard to test under normal circumstances because
1249 * direct-io pins the pages with get_user_pages(). This makes
1250 * is_page_cache_freeable return false, and the VM will not clean the pages.
1251 * But other code (eg, pdflush) could clean the pages if they are mapped
1254 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1255 * deferred bio dirtying paths.
1259 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1261 void bio_set_pages_dirty(struct bio *bio)
1263 struct bio_vec *bvec = bio->bi_io_vec;
1266 for (i = 0; i < bio->bi_vcnt; i++) {
1267 struct page *page = bvec[i].bv_page;
1269 if (page && !PageCompound(page))
1270 set_page_dirty_lock(page);
1274 static void bio_release_pages(struct bio *bio)
1276 struct bio_vec *bvec = bio->bi_io_vec;
1279 for (i = 0; i < bio->bi_vcnt; i++) {
1280 struct page *page = bvec[i].bv_page;
1288 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1289 * If they are, then fine. If, however, some pages are clean then they must
1290 * have been written out during the direct-IO read. So we take another ref on
1291 * the BIO and the offending pages and re-dirty the pages in process context.
1293 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1294 * here on. It will run one page_cache_release() against each page and will
1295 * run one bio_put() against the BIO.
1298 static void bio_dirty_fn(struct work_struct *work);
1300 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1301 static DEFINE_SPINLOCK(bio_dirty_lock);
1302 static struct bio *bio_dirty_list;
1305 * This runs in process context
1307 static void bio_dirty_fn(struct work_struct *work)
1309 unsigned long flags;
1312 spin_lock_irqsave(&bio_dirty_lock, flags);
1313 bio = bio_dirty_list;
1314 bio_dirty_list = NULL;
1315 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1318 struct bio *next = bio->bi_private;
1320 bio_set_pages_dirty(bio);
1321 bio_release_pages(bio);
1327 void bio_check_pages_dirty(struct bio *bio)
1329 struct bio_vec *bvec = bio->bi_io_vec;
1330 int nr_clean_pages = 0;
1333 for (i = 0; i < bio->bi_vcnt; i++) {
1334 struct page *page = bvec[i].bv_page;
1336 if (PageDirty(page) || PageCompound(page)) {
1337 page_cache_release(page);
1338 bvec[i].bv_page = NULL;
1344 if (nr_clean_pages) {
1345 unsigned long flags;
1347 spin_lock_irqsave(&bio_dirty_lock, flags);
1348 bio->bi_private = bio_dirty_list;
1349 bio_dirty_list = bio;
1350 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1351 schedule_work(&bio_dirty_work);
1358 * bio_endio - end I/O on a bio
1360 * @error: error, if any
1363 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1364 * preferred way to end I/O on a bio, it takes care of clearing
1365 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1366 * established -Exxxx (-EIO, for instance) error values in case
1367 * something went wrong. Noone should call bi_end_io() directly on a
1368 * bio unless they own it and thus know that it has an end_io
1371 void bio_endio(struct bio *bio, int error)
1374 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1375 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1379 bio->bi_end_io(bio, error);
1382 void bio_pair_release(struct bio_pair *bp)
1384 if (atomic_dec_and_test(&bp->cnt)) {
1385 struct bio *master = bp->bio1.bi_private;
1387 bio_endio(master, bp->error);
1388 mempool_free(bp, bp->bio2.bi_private);
1392 static void bio_pair_end_1(struct bio *bi, int err)
1394 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1399 bio_pair_release(bp);
1402 static void bio_pair_end_2(struct bio *bi, int err)
1404 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1409 bio_pair_release(bp);
1413 * split a bio - only worry about a bio with a single page
1416 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1418 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1423 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1424 bi->bi_sector + first_sectors);
1426 BUG_ON(bi->bi_vcnt != 1);
1427 BUG_ON(bi->bi_idx != 0);
1428 atomic_set(&bp->cnt, 3);
1432 bp->bio2.bi_sector += first_sectors;
1433 bp->bio2.bi_size -= first_sectors << 9;
1434 bp->bio1.bi_size = first_sectors << 9;
1436 bp->bv1 = bi->bi_io_vec[0];
1437 bp->bv2 = bi->bi_io_vec[0];
1438 bp->bv2.bv_offset += first_sectors << 9;
1439 bp->bv2.bv_len -= first_sectors << 9;
1440 bp->bv1.bv_len = first_sectors << 9;
1442 bp->bio1.bi_io_vec = &bp->bv1;
1443 bp->bio2.bi_io_vec = &bp->bv2;
1445 bp->bio1.bi_max_vecs = 1;
1446 bp->bio2.bi_max_vecs = 1;
1448 bp->bio1.bi_end_io = bio_pair_end_1;
1449 bp->bio2.bi_end_io = bio_pair_end_2;
1451 bp->bio1.bi_private = bi;
1452 bp->bio2.bi_private = bio_split_pool;
1454 if (bio_integrity(bi))
1455 bio_integrity_split(bi, bp, first_sectors);
1461 * bio_sector_offset - Find hardware sector offset in bio
1462 * @bio: bio to inspect
1463 * @index: bio_vec index
1464 * @offset: offset in bv_page
1466 * Return the number of hardware sectors between beginning of bio
1467 * and an end point indicated by a bio_vec index and an offset
1468 * within that vector's page.
1470 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1471 unsigned int offset)
1473 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1480 if (index >= bio->bi_idx)
1481 index = bio->bi_vcnt - 1;
1483 __bio_for_each_segment(bv, bio, i, 0) {
1485 if (offset > bv->bv_offset)
1486 sectors += (offset - bv->bv_offset) / sector_sz;
1490 sectors += bv->bv_len / sector_sz;
1495 EXPORT_SYMBOL(bio_sector_offset);
1498 * create memory pools for biovec's in a bio_set.
1499 * use the global biovec slabs created for general use.
1501 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1503 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1505 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1512 static void biovec_free_pools(struct bio_set *bs)
1514 mempool_destroy(bs->bvec_pool);
1517 void bioset_free(struct bio_set *bs)
1520 mempool_destroy(bs->bio_pool);
1522 bioset_integrity_free(bs);
1523 biovec_free_pools(bs);
1530 * bioset_create - Create a bio_set
1531 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1532 * @front_pad: Number of bytes to allocate in front of the returned bio
1535 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1536 * to ask for a number of bytes to be allocated in front of the bio.
1537 * Front pad allocation is useful for embedding the bio inside
1538 * another structure, to avoid allocating extra data to go with the bio.
1539 * Note that the bio must be embedded at the END of that structure always,
1540 * or things will break badly.
1542 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1544 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1547 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1551 bs->front_pad = front_pad;
1553 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1554 if (!bs->bio_slab) {
1559 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1563 if (bioset_integrity_create(bs, pool_size))
1566 if (!biovec_create_pools(bs, pool_size))
1574 static void __init biovec_init_slabs(void)
1578 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1580 struct biovec_slab *bvs = bvec_slabs + i;
1582 size = bvs->nr_vecs * sizeof(struct bio_vec);
1583 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1584 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1588 static int __init init_bio(void)
1592 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1594 panic("bio: can't allocate bios\n");
1596 bio_integrity_init_slab();
1597 biovec_init_slabs();
1599 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1601 panic("bio: can't allocate bios\n");
1603 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1604 sizeof(struct bio_pair));
1605 if (!bio_split_pool)
1606 panic("bio: can't create split pool\n");
1611 subsys_initcall(init_bio);
1613 EXPORT_SYMBOL(bio_alloc);
1614 EXPORT_SYMBOL(bio_kmalloc);
1615 EXPORT_SYMBOL(bio_put);
1616 EXPORT_SYMBOL(bio_free);
1617 EXPORT_SYMBOL(bio_endio);
1618 EXPORT_SYMBOL(bio_init);
1619 EXPORT_SYMBOL(__bio_clone);
1620 EXPORT_SYMBOL(bio_clone);
1621 EXPORT_SYMBOL(bio_phys_segments);
1622 EXPORT_SYMBOL(bio_add_page);
1623 EXPORT_SYMBOL(bio_add_pc_page);
1624 EXPORT_SYMBOL(bio_get_nr_vecs);
1625 EXPORT_SYMBOL(bio_map_user);
1626 EXPORT_SYMBOL(bio_unmap_user);
1627 EXPORT_SYMBOL(bio_map_kern);
1628 EXPORT_SYMBOL(bio_copy_kern);
1629 EXPORT_SYMBOL(bio_pair_release);
1630 EXPORT_SYMBOL(bio_split);
1631 EXPORT_SYMBOL(bio_copy_user);
1632 EXPORT_SYMBOL(bio_uncopy_user);
1633 EXPORT_SYMBOL(bioset_create);
1634 EXPORT_SYMBOL(bioset_free);
1635 EXPORT_SYMBOL(bio_alloc_bioset);