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/iocontext.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/kernel.h>
26 #include <linux/export.h>
27 #include <linux/mempool.h>
28 #include <linux/workqueue.h>
29 #include <linux/cgroup.h>
30 #include <scsi/sg.h> /* for struct sg_iovec */
32 #include <trace/events/block.h>
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 static 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, *new_bio_slabs;
77 unsigned int i, entry = -1;
79 mutex_lock(&bio_slab_lock);
82 while (i < bio_slab_nr) {
83 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 new_bio_slabs = krealloc(bio_slabs,
101 bio_slab_max * sizeof(struct bio_slab),
105 bio_slabs = new_bio_slabs;
108 entry = bio_slab_nr++;
110 bslab = &bio_slabs[entry];
112 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
113 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
117 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
120 bslab->slab_size = sz;
122 mutex_unlock(&bio_slab_lock);
126 static void bio_put_slab(struct bio_set *bs)
128 struct bio_slab *bslab = NULL;
131 mutex_lock(&bio_slab_lock);
133 for (i = 0; i < bio_slab_nr; i++) {
134 if (bs->bio_slab == bio_slabs[i].slab) {
135 bslab = &bio_slabs[i];
140 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
143 WARN_ON(!bslab->slab_ref);
145 if (--bslab->slab_ref)
148 kmem_cache_destroy(bslab->slab);
152 mutex_unlock(&bio_slab_lock);
155 unsigned int bvec_nr_vecs(unsigned short idx)
157 return bvec_slabs[idx].nr_vecs;
160 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
162 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
164 if (idx == BIOVEC_MAX_IDX)
165 mempool_free(bv, bs->bvec_pool);
167 struct biovec_slab *bvs = bvec_slabs + idx;
169 kmem_cache_free(bvs->slab, bv);
173 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
179 * see comment near bvec_array define!
197 case 129 ... BIO_MAX_PAGES:
205 * idx now points to the pool we want to allocate from. only the
206 * 1-vec entry pool is mempool backed.
208 if (*idx == BIOVEC_MAX_IDX) {
210 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
212 struct biovec_slab *bvs = bvec_slabs + *idx;
213 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
216 * Make this allocation restricted and don't dump info on
217 * allocation failures, since we'll fallback to the mempool
218 * in case of failure.
220 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
223 * Try a slab allocation. If this fails and __GFP_WAIT
224 * is set, retry with the 1-entry mempool
226 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
227 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
228 *idx = BIOVEC_MAX_IDX;
236 void bio_free(struct bio *bio, struct bio_set *bs)
240 if (bio_has_allocated_vec(bio))
241 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
243 if (bio_integrity(bio))
244 bio_integrity_free(bio, bs);
247 * If we have front padding, adjust the bio pointer before freeing
253 mempool_free(p, bs->bio_pool);
255 EXPORT_SYMBOL(bio_free);
257 void bio_init(struct bio *bio)
259 memset(bio, 0, sizeof(*bio));
260 bio->bi_flags = 1 << BIO_UPTODATE;
261 atomic_set(&bio->bi_cnt, 1);
263 EXPORT_SYMBOL(bio_init);
266 * bio_alloc_bioset - allocate a bio for I/O
267 * @gfp_mask: the GFP_ mask given to the slab allocator
268 * @nr_iovecs: number of iovecs to pre-allocate
269 * @bs: the bio_set to allocate from.
272 * bio_alloc_bioset will try its own mempool to satisfy the allocation.
273 * If %__GFP_WAIT is set then we will block on the internal pool waiting
274 * for a &struct bio to become free.
276 * Note that the caller must set ->bi_destructor on successful return
277 * of a bio, to do the appropriate freeing of the bio once the reference
278 * count drops to zero.
280 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
282 unsigned long idx = BIO_POOL_NONE;
283 struct bio_vec *bvl = NULL;
287 p = mempool_alloc(bs->bio_pool, gfp_mask);
290 bio = p + bs->front_pad;
294 if (unlikely(!nr_iovecs))
297 if (nr_iovecs <= BIO_INLINE_VECS) {
298 bvl = bio->bi_inline_vecs;
299 nr_iovecs = BIO_INLINE_VECS;
301 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
305 nr_iovecs = bvec_nr_vecs(idx);
308 bio->bi_flags |= idx << BIO_POOL_OFFSET;
309 bio->bi_max_vecs = nr_iovecs;
310 bio->bi_io_vec = bvl;
314 mempool_free(p, bs->bio_pool);
317 EXPORT_SYMBOL(bio_alloc_bioset);
319 static void bio_fs_destructor(struct bio *bio)
321 bio_free(bio, fs_bio_set);
325 * bio_alloc - allocate a new bio, memory pool backed
326 * @gfp_mask: allocation mask to use
327 * @nr_iovecs: number of iovecs
329 * bio_alloc will allocate a bio and associated bio_vec array that can hold
330 * at least @nr_iovecs entries. Allocations will be done from the
331 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
333 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
334 * a bio. This is due to the mempool guarantees. To make this work, callers
335 * must never allocate more than 1 bio at a time from this pool. Callers
336 * that need to allocate more than 1 bio must always submit the previously
337 * allocated bio for IO before attempting to allocate a new one. Failure to
338 * do so can cause livelocks under memory pressure.
341 * Pointer to new bio on success, NULL on failure.
343 struct bio *bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
345 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
348 bio->bi_destructor = bio_fs_destructor;
352 EXPORT_SYMBOL(bio_alloc);
354 static void bio_kmalloc_destructor(struct bio *bio)
356 if (bio_integrity(bio))
357 bio_integrity_free(bio, fs_bio_set);
362 * bio_kmalloc - allocate a bio for I/O using kmalloc()
363 * @gfp_mask: the GFP_ mask given to the slab allocator
364 * @nr_iovecs: number of iovecs to pre-allocate
367 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
368 * %__GFP_WAIT, the allocation is guaranteed to succeed.
371 struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned int nr_iovecs)
375 if (nr_iovecs > UIO_MAXIOV)
378 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
384 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
385 bio->bi_max_vecs = nr_iovecs;
386 bio->bi_io_vec = bio->bi_inline_vecs;
387 bio->bi_destructor = bio_kmalloc_destructor;
391 EXPORT_SYMBOL(bio_kmalloc);
393 void zero_fill_bio(struct bio *bio)
399 bio_for_each_segment(bv, bio, i) {
400 char *data = bvec_kmap_irq(bv, &flags);
401 memset(data, 0, bv->bv_len);
402 flush_dcache_page(bv->bv_page);
403 bvec_kunmap_irq(data, &flags);
406 EXPORT_SYMBOL(zero_fill_bio);
409 * bio_put - release a reference to a bio
410 * @bio: bio to release reference to
413 * Put a reference to a &struct bio, either one you have gotten with
414 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
416 void bio_put(struct bio *bio)
418 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
423 if (atomic_dec_and_test(&bio->bi_cnt)) {
424 bio_disassociate_task(bio);
426 bio->bi_destructor(bio);
429 EXPORT_SYMBOL(bio_put);
431 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
433 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
434 blk_recount_segments(q, bio);
436 return bio->bi_phys_segments;
438 EXPORT_SYMBOL(bio_phys_segments);
441 * __bio_clone - clone a bio
442 * @bio: destination bio
443 * @bio_src: bio to clone
445 * Clone a &bio. Caller will own the returned bio, but not
446 * the actual data it points to. Reference count of returned
449 void __bio_clone(struct bio *bio, struct bio *bio_src)
451 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
452 bio_src->bi_max_vecs * sizeof(struct bio_vec));
455 * most users will be overriding ->bi_bdev with a new target,
456 * so we don't set nor calculate new physical/hw segment counts here
458 bio->bi_sector = bio_src->bi_sector;
459 bio->bi_bdev = bio_src->bi_bdev;
460 bio->bi_flags |= 1 << BIO_CLONED;
461 bio->bi_rw = bio_src->bi_rw;
462 bio->bi_vcnt = bio_src->bi_vcnt;
463 bio->bi_size = bio_src->bi_size;
464 bio->bi_idx = bio_src->bi_idx;
466 EXPORT_SYMBOL(__bio_clone);
469 * bio_clone - clone a bio
471 * @gfp_mask: allocation priority
473 * Like __bio_clone, only also allocates the returned bio
475 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
477 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
482 b->bi_destructor = bio_fs_destructor;
485 if (bio_integrity(bio)) {
488 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
498 EXPORT_SYMBOL(bio_clone);
501 * bio_get_nr_vecs - return approx number of vecs
504 * Return the approximate number of pages we can send to this target.
505 * There's no guarantee that you will be able to fit this number of pages
506 * into a bio, it does not account for dynamic restrictions that vary
509 int bio_get_nr_vecs(struct block_device *bdev)
511 struct request_queue *q = bdev_get_queue(bdev);
514 nr_pages = min_t(unsigned,
515 queue_max_segments(q),
516 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
518 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
521 EXPORT_SYMBOL(bio_get_nr_vecs);
523 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
524 *page, unsigned int len, unsigned int offset,
525 unsigned short max_sectors)
527 int retried_segments = 0;
528 struct bio_vec *bvec;
531 * cloned bio must not modify vec list
533 if (unlikely(bio_flagged(bio, BIO_CLONED)))
536 if (((bio->bi_size + len) >> 9) > max_sectors)
540 * For filesystems with a blocksize smaller than the pagesize
541 * we will often be called with the same page as last time and
542 * a consecutive offset. Optimize this special case.
544 if (bio->bi_vcnt > 0) {
545 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
547 if (page == prev->bv_page &&
548 offset == prev->bv_offset + prev->bv_len) {
549 unsigned int prev_bv_len = prev->bv_len;
552 if (q->merge_bvec_fn) {
553 struct bvec_merge_data bvm = {
554 /* prev_bvec is already charged in
555 bi_size, discharge it in order to
556 simulate merging updated prev_bvec
558 .bi_bdev = bio->bi_bdev,
559 .bi_sector = bio->bi_sector,
560 .bi_size = bio->bi_size - prev_bv_len,
564 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
574 if (bio->bi_vcnt >= bio->bi_max_vecs)
578 * we might lose a segment or two here, but rather that than
579 * make this too complex.
582 while (bio->bi_phys_segments >= queue_max_segments(q)) {
584 if (retried_segments)
587 retried_segments = 1;
588 blk_recount_segments(q, bio);
592 * setup the new entry, we might clear it again later if we
593 * cannot add the page
595 bvec = &bio->bi_io_vec[bio->bi_vcnt];
596 bvec->bv_page = page;
598 bvec->bv_offset = offset;
601 * if queue has other restrictions (eg varying max sector size
602 * depending on offset), it can specify a merge_bvec_fn in the
603 * queue to get further control
605 if (q->merge_bvec_fn) {
606 struct bvec_merge_data bvm = {
607 .bi_bdev = bio->bi_bdev,
608 .bi_sector = bio->bi_sector,
609 .bi_size = bio->bi_size,
614 * merge_bvec_fn() returns number of bytes it can accept
617 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
618 bvec->bv_page = NULL;
625 /* If we may be able to merge these biovecs, force a recount */
626 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
627 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
630 bio->bi_phys_segments++;
637 * bio_add_pc_page - attempt to add page to bio
638 * @q: the target queue
639 * @bio: destination bio
641 * @len: vec entry length
642 * @offset: vec entry offset
644 * Attempt to add a page to the bio_vec maplist. This can fail for a
645 * number of reasons, such as the bio being full or target block device
646 * limitations. The target block device must allow bio's up to PAGE_SIZE,
647 * so it is always possible to add a single page to an empty bio.
649 * This should only be used by REQ_PC bios.
651 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
652 unsigned int len, unsigned int offset)
654 return __bio_add_page(q, bio, page, len, offset,
655 queue_max_hw_sectors(q));
657 EXPORT_SYMBOL(bio_add_pc_page);
660 * bio_add_page - attempt to add page to bio
661 * @bio: destination bio
663 * @len: vec entry length
664 * @offset: vec entry offset
666 * Attempt to add a page to the bio_vec maplist. This can fail for a
667 * number of reasons, such as the bio being full or target block device
668 * limitations. The target block device must allow bio's up to PAGE_SIZE,
669 * so it is always possible to add a single page to an empty bio.
671 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
674 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
675 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
677 EXPORT_SYMBOL(bio_add_page);
679 struct bio_map_data {
680 struct bio_vec *iovecs;
681 struct sg_iovec *sgvecs;
686 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
687 struct sg_iovec *iov, int iov_count,
690 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
691 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
692 bmd->nr_sgvecs = iov_count;
693 bmd->is_our_pages = is_our_pages;
694 bio->bi_private = bmd;
697 static void bio_free_map_data(struct bio_map_data *bmd)
704 static struct bio_map_data *bio_alloc_map_data(int nr_segs,
705 unsigned int iov_count,
708 struct bio_map_data *bmd;
710 if (iov_count > UIO_MAXIOV)
713 bmd = kmalloc(sizeof(*bmd), gfp_mask);
717 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
723 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
732 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
733 struct sg_iovec *iov, int iov_count,
734 int to_user, int from_user, int do_free_page)
737 struct bio_vec *bvec;
739 unsigned int iov_off = 0;
741 __bio_for_each_segment(bvec, bio, i, 0) {
742 char *bv_addr = page_address(bvec->bv_page);
743 unsigned int bv_len = iovecs[i].bv_len;
745 while (bv_len && iov_idx < iov_count) {
747 char __user *iov_addr;
749 bytes = min_t(unsigned int,
750 iov[iov_idx].iov_len - iov_off, bv_len);
751 iov_addr = iov[iov_idx].iov_base + iov_off;
755 ret = copy_to_user(iov_addr, bv_addr,
759 ret = copy_from_user(bv_addr, iov_addr,
771 if (iov[iov_idx].iov_len == iov_off) {
778 __free_page(bvec->bv_page);
785 * bio_uncopy_user - finish previously mapped bio
786 * @bio: bio being terminated
788 * Free pages allocated from bio_copy_user() and write back data
789 * to user space in case of a read.
791 int bio_uncopy_user(struct bio *bio)
793 struct bio_map_data *bmd = bio->bi_private;
796 if (!bio_flagged(bio, BIO_NULL_MAPPED))
797 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
798 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
799 0, bmd->is_our_pages);
800 bio_free_map_data(bmd);
804 EXPORT_SYMBOL(bio_uncopy_user);
807 * bio_copy_user_iov - copy user data to bio
808 * @q: destination block queue
809 * @map_data: pointer to the rq_map_data holding pages (if necessary)
811 * @iov_count: number of elements in the iovec
812 * @write_to_vm: bool indicating writing to pages or not
813 * @gfp_mask: memory allocation flags
815 * Prepares and returns a bio for indirect user io, bouncing data
816 * to/from kernel pages as necessary. Must be paired with
817 * call bio_uncopy_user() on io completion.
819 struct bio *bio_copy_user_iov(struct request_queue *q,
820 struct rq_map_data *map_data,
821 struct sg_iovec *iov, int iov_count,
822 int write_to_vm, gfp_t gfp_mask)
824 struct bio_map_data *bmd;
825 struct bio_vec *bvec;
830 unsigned int len = 0;
831 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
833 for (i = 0; i < iov_count; i++) {
838 uaddr = (unsigned long)iov[i].iov_base;
839 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
840 start = uaddr >> PAGE_SHIFT;
846 return ERR_PTR(-EINVAL);
848 nr_pages += end - start;
849 len += iov[i].iov_len;
855 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
857 return ERR_PTR(-ENOMEM);
860 bio = bio_kmalloc(gfp_mask, nr_pages);
865 bio->bi_rw |= REQ_WRITE;
870 nr_pages = 1 << map_data->page_order;
871 i = map_data->offset / PAGE_SIZE;
874 unsigned int bytes = PAGE_SIZE;
882 if (i == map_data->nr_entries * nr_pages) {
887 page = map_data->pages[i / nr_pages];
888 page += (i % nr_pages);
892 page = alloc_page(q->bounce_gfp | gfp_mask);
899 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
912 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
913 (map_data && map_data->from_user)) {
914 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
919 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
923 bio_for_each_segment(bvec, bio, i)
924 __free_page(bvec->bv_page);
928 bio_free_map_data(bmd);
933 * bio_copy_user - copy user data to bio
934 * @q: destination block queue
935 * @map_data: pointer to the rq_map_data holding pages (if necessary)
936 * @uaddr: start of user address
937 * @len: length in bytes
938 * @write_to_vm: bool indicating writing to pages or not
939 * @gfp_mask: memory allocation flags
941 * Prepares and returns a bio for indirect user io, bouncing data
942 * to/from kernel pages as necessary. Must be paired with
943 * call bio_uncopy_user() on io completion.
945 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
946 unsigned long uaddr, unsigned int len,
947 int write_to_vm, gfp_t gfp_mask)
951 iov.iov_base = (void __user *)uaddr;
954 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
956 EXPORT_SYMBOL(bio_copy_user);
958 static struct bio *__bio_map_user_iov(struct request_queue *q,
959 struct block_device *bdev,
960 struct sg_iovec *iov, int iov_count,
961 int write_to_vm, gfp_t gfp_mask)
970 for (i = 0; i < iov_count; i++) {
971 unsigned long uaddr = (unsigned long)iov[i].iov_base;
972 unsigned long len = iov[i].iov_len;
973 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
974 unsigned long start = uaddr >> PAGE_SHIFT;
980 return ERR_PTR(-EINVAL);
982 nr_pages += end - start;
984 * buffer must be aligned to at least hardsector size for now
986 if (uaddr & queue_dma_alignment(q))
987 return ERR_PTR(-EINVAL);
991 return ERR_PTR(-EINVAL);
993 bio = bio_kmalloc(gfp_mask, nr_pages);
995 return ERR_PTR(-ENOMEM);
998 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1002 for (i = 0; i < iov_count; i++) {
1003 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1004 unsigned long len = iov[i].iov_len;
1005 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1006 unsigned long start = uaddr >> PAGE_SHIFT;
1007 const int local_nr_pages = end - start;
1008 const int page_limit = cur_page + local_nr_pages;
1010 ret = get_user_pages_fast(uaddr, local_nr_pages,
1011 write_to_vm, &pages[cur_page]);
1012 if (ret < local_nr_pages) {
1017 offset = uaddr & ~PAGE_MASK;
1018 for (j = cur_page; j < page_limit; j++) {
1019 unsigned int bytes = PAGE_SIZE - offset;
1030 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1040 * release the pages we didn't map into the bio, if any
1042 while (j < page_limit)
1043 page_cache_release(pages[j++]);
1049 * set data direction, and check if mapped pages need bouncing
1052 bio->bi_rw |= REQ_WRITE;
1054 bio->bi_bdev = bdev;
1055 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1059 for (i = 0; i < nr_pages; i++) {
1062 page_cache_release(pages[i]);
1067 return ERR_PTR(ret);
1071 * bio_map_user - map user address into bio
1072 * @q: the struct request_queue for the bio
1073 * @bdev: destination block device
1074 * @uaddr: start of user address
1075 * @len: length in bytes
1076 * @write_to_vm: bool indicating writing to pages or not
1077 * @gfp_mask: memory allocation flags
1079 * Map the user space address into a bio suitable for io to a block
1080 * device. Returns an error pointer in case of error.
1082 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1083 unsigned long uaddr, unsigned int len, int write_to_vm,
1086 struct sg_iovec iov;
1088 iov.iov_base = (void __user *)uaddr;
1091 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1093 EXPORT_SYMBOL(bio_map_user);
1096 * bio_map_user_iov - map user sg_iovec table into bio
1097 * @q: the struct request_queue for the bio
1098 * @bdev: destination block device
1100 * @iov_count: number of elements in the iovec
1101 * @write_to_vm: bool indicating writing to pages or not
1102 * @gfp_mask: memory allocation flags
1104 * Map the user space address into a bio suitable for io to a block
1105 * device. Returns an error pointer in case of error.
1107 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1108 struct sg_iovec *iov, int iov_count,
1109 int write_to_vm, gfp_t gfp_mask)
1113 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1119 * subtle -- if __bio_map_user() ended up bouncing a bio,
1120 * it would normally disappear when its bi_end_io is run.
1121 * however, we need it for the unmap, so grab an extra
1129 static void __bio_unmap_user(struct bio *bio)
1131 struct bio_vec *bvec;
1135 * make sure we dirty pages we wrote to
1137 __bio_for_each_segment(bvec, bio, i, 0) {
1138 if (bio_data_dir(bio) == READ)
1139 set_page_dirty_lock(bvec->bv_page);
1141 page_cache_release(bvec->bv_page);
1148 * bio_unmap_user - unmap a bio
1149 * @bio: the bio being unmapped
1151 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1152 * a process context.
1154 * bio_unmap_user() may sleep.
1156 void bio_unmap_user(struct bio *bio)
1158 __bio_unmap_user(bio);
1161 EXPORT_SYMBOL(bio_unmap_user);
1163 static void bio_map_kern_endio(struct bio *bio, int err)
1168 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1169 unsigned int len, gfp_t gfp_mask)
1171 unsigned long kaddr = (unsigned long)data;
1172 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1173 unsigned long start = kaddr >> PAGE_SHIFT;
1174 const int nr_pages = end - start;
1178 bio = bio_kmalloc(gfp_mask, nr_pages);
1180 return ERR_PTR(-ENOMEM);
1182 offset = offset_in_page(kaddr);
1183 for (i = 0; i < nr_pages; i++) {
1184 unsigned int bytes = PAGE_SIZE - offset;
1192 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1201 bio->bi_end_io = bio_map_kern_endio;
1206 * bio_map_kern - map kernel address into bio
1207 * @q: the struct request_queue for the bio
1208 * @data: pointer to buffer to map
1209 * @len: length in bytes
1210 * @gfp_mask: allocation flags for bio allocation
1212 * Map the kernel address into a bio suitable for io to a block
1213 * device. Returns an error pointer in case of error.
1215 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1220 bio = __bio_map_kern(q, data, len, gfp_mask);
1224 if (bio->bi_size == len)
1228 * Don't support partial mappings.
1231 return ERR_PTR(-EINVAL);
1233 EXPORT_SYMBOL(bio_map_kern);
1235 static void bio_copy_kern_endio(struct bio *bio, int err)
1237 struct bio_vec *bvec;
1238 const int read = bio_data_dir(bio) == READ;
1239 struct bio_map_data *bmd = bio->bi_private;
1241 char *p = bmd->sgvecs[0].iov_base;
1243 __bio_for_each_segment(bvec, bio, i, 0) {
1244 char *addr = page_address(bvec->bv_page);
1245 int len = bmd->iovecs[i].bv_len;
1248 memcpy(p, addr, len);
1250 __free_page(bvec->bv_page);
1254 bio_free_map_data(bmd);
1259 * bio_copy_kern - copy kernel address into bio
1260 * @q: the struct request_queue for the bio
1261 * @data: pointer to buffer to copy
1262 * @len: length in bytes
1263 * @gfp_mask: allocation flags for bio and page allocation
1264 * @reading: data direction is READ
1266 * copy the kernel address into a bio suitable for io to a block
1267 * device. Returns an error pointer in case of error.
1269 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1270 gfp_t gfp_mask, int reading)
1273 struct bio_vec *bvec;
1276 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1283 bio_for_each_segment(bvec, bio, i) {
1284 char *addr = page_address(bvec->bv_page);
1286 memcpy(addr, p, bvec->bv_len);
1291 bio->bi_end_io = bio_copy_kern_endio;
1295 EXPORT_SYMBOL(bio_copy_kern);
1298 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1299 * for performing direct-IO in BIOs.
1301 * The problem is that we cannot run set_page_dirty() from interrupt context
1302 * because the required locks are not interrupt-safe. So what we can do is to
1303 * mark the pages dirty _before_ performing IO. And in interrupt context,
1304 * check that the pages are still dirty. If so, fine. If not, redirty them
1305 * in process context.
1307 * We special-case compound pages here: normally this means reads into hugetlb
1308 * pages. The logic in here doesn't really work right for compound pages
1309 * because the VM does not uniformly chase down the head page in all cases.
1310 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1311 * handle them at all. So we skip compound pages here at an early stage.
1313 * Note that this code is very hard to test under normal circumstances because
1314 * direct-io pins the pages with get_user_pages(). This makes
1315 * is_page_cache_freeable return false, and the VM will not clean the pages.
1316 * But other code (eg, flusher threads) could clean the pages if they are mapped
1319 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1320 * deferred bio dirtying paths.
1324 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1326 void bio_set_pages_dirty(struct bio *bio)
1328 struct bio_vec *bvec = bio->bi_io_vec;
1331 for (i = 0; i < bio->bi_vcnt; i++) {
1332 struct page *page = bvec[i].bv_page;
1334 if (page && !PageCompound(page))
1335 set_page_dirty_lock(page);
1339 static void bio_release_pages(struct bio *bio)
1341 struct bio_vec *bvec = bio->bi_io_vec;
1344 for (i = 0; i < bio->bi_vcnt; i++) {
1345 struct page *page = bvec[i].bv_page;
1353 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1354 * If they are, then fine. If, however, some pages are clean then they must
1355 * have been written out during the direct-IO read. So we take another ref on
1356 * the BIO and the offending pages and re-dirty the pages in process context.
1358 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1359 * here on. It will run one page_cache_release() against each page and will
1360 * run one bio_put() against the BIO.
1363 static void bio_dirty_fn(struct work_struct *work);
1365 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1366 static DEFINE_SPINLOCK(bio_dirty_lock);
1367 static struct bio *bio_dirty_list;
1370 * This runs in process context
1372 static void bio_dirty_fn(struct work_struct *work)
1374 unsigned long flags;
1377 spin_lock_irqsave(&bio_dirty_lock, flags);
1378 bio = bio_dirty_list;
1379 bio_dirty_list = NULL;
1380 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1383 struct bio *next = bio->bi_private;
1385 bio_set_pages_dirty(bio);
1386 bio_release_pages(bio);
1392 void bio_check_pages_dirty(struct bio *bio)
1394 struct bio_vec *bvec = bio->bi_io_vec;
1395 int nr_clean_pages = 0;
1398 for (i = 0; i < bio->bi_vcnt; i++) {
1399 struct page *page = bvec[i].bv_page;
1401 if (PageDirty(page) || PageCompound(page)) {
1402 page_cache_release(page);
1403 bvec[i].bv_page = NULL;
1409 if (nr_clean_pages) {
1410 unsigned long flags;
1412 spin_lock_irqsave(&bio_dirty_lock, flags);
1413 bio->bi_private = bio_dirty_list;
1414 bio_dirty_list = bio;
1415 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1416 schedule_work(&bio_dirty_work);
1422 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1423 void bio_flush_dcache_pages(struct bio *bi)
1426 struct bio_vec *bvec;
1428 bio_for_each_segment(bvec, bi, i)
1429 flush_dcache_page(bvec->bv_page);
1431 EXPORT_SYMBOL(bio_flush_dcache_pages);
1435 * bio_endio - end I/O on a bio
1437 * @error: error, if any
1440 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1441 * preferred way to end I/O on a bio, it takes care of clearing
1442 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1443 * established -Exxxx (-EIO, for instance) error values in case
1444 * something went wrong. No one should call bi_end_io() directly on a
1445 * bio unless they own it and thus know that it has an end_io
1448 void bio_endio(struct bio *bio, int error)
1451 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1452 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1456 bio->bi_end_io(bio, error);
1458 EXPORT_SYMBOL(bio_endio);
1460 void bio_pair_release(struct bio_pair *bp)
1462 if (atomic_dec_and_test(&bp->cnt)) {
1463 struct bio *master = bp->bio1.bi_private;
1465 bio_endio(master, bp->error);
1466 mempool_free(bp, bp->bio2.bi_private);
1469 EXPORT_SYMBOL(bio_pair_release);
1471 static void bio_pair_end_1(struct bio *bi, int err)
1473 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1478 bio_pair_release(bp);
1481 static void bio_pair_end_2(struct bio *bi, int err)
1483 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1488 bio_pair_release(bp);
1492 * split a bio - only worry about a bio with a single page in its iovec
1494 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1496 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1501 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1502 bi->bi_sector + first_sectors);
1504 BUG_ON(bi->bi_vcnt != 1);
1505 BUG_ON(bi->bi_idx != 0);
1506 atomic_set(&bp->cnt, 3);
1510 bp->bio2.bi_sector += first_sectors;
1511 bp->bio2.bi_size -= first_sectors << 9;
1512 bp->bio1.bi_size = first_sectors << 9;
1514 bp->bv1 = bi->bi_io_vec[0];
1515 bp->bv2 = bi->bi_io_vec[0];
1516 bp->bv2.bv_offset += first_sectors << 9;
1517 bp->bv2.bv_len -= first_sectors << 9;
1518 bp->bv1.bv_len = first_sectors << 9;
1520 bp->bio1.bi_io_vec = &bp->bv1;
1521 bp->bio2.bi_io_vec = &bp->bv2;
1523 bp->bio1.bi_max_vecs = 1;
1524 bp->bio2.bi_max_vecs = 1;
1526 bp->bio1.bi_end_io = bio_pair_end_1;
1527 bp->bio2.bi_end_io = bio_pair_end_2;
1529 bp->bio1.bi_private = bi;
1530 bp->bio2.bi_private = bio_split_pool;
1532 if (bio_integrity(bi))
1533 bio_integrity_split(bi, bp, first_sectors);
1537 EXPORT_SYMBOL(bio_split);
1540 * bio_sector_offset - Find hardware sector offset in bio
1541 * @bio: bio to inspect
1542 * @index: bio_vec index
1543 * @offset: offset in bv_page
1545 * Return the number of hardware sectors between beginning of bio
1546 * and an end point indicated by a bio_vec index and an offset
1547 * within that vector's page.
1549 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1550 unsigned int offset)
1552 unsigned int sector_sz;
1557 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1560 if (index >= bio->bi_idx)
1561 index = bio->bi_vcnt - 1;
1563 __bio_for_each_segment(bv, bio, i, 0) {
1565 if (offset > bv->bv_offset)
1566 sectors += (offset - bv->bv_offset) / sector_sz;
1570 sectors += bv->bv_len / sector_sz;
1575 EXPORT_SYMBOL(bio_sector_offset);
1578 * create memory pools for biovec's in a bio_set.
1579 * use the global biovec slabs created for general use.
1581 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1583 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1585 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1592 static void biovec_free_pools(struct bio_set *bs)
1594 mempool_destroy(bs->bvec_pool);
1597 void bioset_free(struct bio_set *bs)
1600 mempool_destroy(bs->bio_pool);
1602 bioset_integrity_free(bs);
1603 biovec_free_pools(bs);
1608 EXPORT_SYMBOL(bioset_free);
1611 * bioset_create - Create a bio_set
1612 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1613 * @front_pad: Number of bytes to allocate in front of the returned bio
1616 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1617 * to ask for a number of bytes to be allocated in front of the bio.
1618 * Front pad allocation is useful for embedding the bio inside
1619 * another structure, to avoid allocating extra data to go with the bio.
1620 * Note that the bio must be embedded at the END of that structure always,
1621 * or things will break badly.
1623 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1625 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1628 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1632 bs->front_pad = front_pad;
1634 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1635 if (!bs->bio_slab) {
1640 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1644 if (!biovec_create_pools(bs, pool_size))
1651 EXPORT_SYMBOL(bioset_create);
1653 #ifdef CONFIG_BLK_CGROUP
1655 * bio_associate_current - associate a bio with %current
1658 * Associate @bio with %current if it hasn't been associated yet. Block
1659 * layer will treat @bio as if it were issued by %current no matter which
1660 * task actually issues it.
1662 * This function takes an extra reference of @task's io_context and blkcg
1663 * which will be put when @bio is released. The caller must own @bio,
1664 * ensure %current->io_context exists, and is responsible for synchronizing
1665 * calls to this function.
1667 int bio_associate_current(struct bio *bio)
1669 struct io_context *ioc;
1670 struct cgroup_subsys_state *css;
1675 ioc = current->io_context;
1679 /* acquire active ref on @ioc and associate */
1680 get_io_context_active(ioc);
1683 /* associate blkcg if exists */
1685 css = task_subsys_state(current, blkio_subsys_id);
1686 if (css && css_tryget(css))
1694 * bio_disassociate_task - undo bio_associate_current()
1697 void bio_disassociate_task(struct bio *bio)
1700 put_io_context(bio->bi_ioc);
1704 css_put(bio->bi_css);
1709 #endif /* CONFIG_BLK_CGROUP */
1711 static void __init biovec_init_slabs(void)
1715 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1717 struct biovec_slab *bvs = bvec_slabs + i;
1719 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1724 size = bvs->nr_vecs * sizeof(struct bio_vec);
1725 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1726 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1730 static int __init init_bio(void)
1734 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1736 panic("bio: can't allocate bios\n");
1738 bio_integrity_init();
1739 biovec_init_slabs();
1741 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1743 panic("bio: can't allocate bios\n");
1745 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1746 panic("bio: can't create integrity pool\n");
1748 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1749 sizeof(struct bio_pair));
1750 if (!bio_split_pool)
1751 panic("bio: can't create split pool\n");
1755 subsys_initcall(init_bio);