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;
58 EXPORT_SYMBOL(fs_bio_set);
61 * Our slab pool management
64 struct kmem_cache *slab;
65 unsigned int slab_ref;
66 unsigned int slab_size;
69 static DEFINE_MUTEX(bio_slab_lock);
70 static struct bio_slab *bio_slabs;
71 static unsigned int bio_slab_nr, bio_slab_max;
73 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
75 unsigned int sz = sizeof(struct bio) + extra_size;
76 struct kmem_cache *slab = NULL;
77 struct bio_slab *bslab, *new_bio_slabs;
78 unsigned int new_bio_slab_max;
79 unsigned int i, entry = -1;
81 mutex_lock(&bio_slab_lock);
84 while (i < bio_slab_nr) {
85 bslab = &bio_slabs[i];
87 if (!bslab->slab && entry == -1)
89 else if (bslab->slab_size == sz) {
100 if (bio_slab_nr == bio_slab_max && entry == -1) {
101 new_bio_slab_max = bio_slab_max << 1;
102 new_bio_slabs = krealloc(bio_slabs,
103 new_bio_slab_max * sizeof(struct bio_slab),
107 bio_slab_max = new_bio_slab_max;
108 bio_slabs = new_bio_slabs;
111 entry = bio_slab_nr++;
113 bslab = &bio_slabs[entry];
115 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
116 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
120 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
123 bslab->slab_size = sz;
125 mutex_unlock(&bio_slab_lock);
129 static void bio_put_slab(struct bio_set *bs)
131 struct bio_slab *bslab = NULL;
134 mutex_lock(&bio_slab_lock);
136 for (i = 0; i < bio_slab_nr; i++) {
137 if (bs->bio_slab == bio_slabs[i].slab) {
138 bslab = &bio_slabs[i];
143 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
146 WARN_ON(!bslab->slab_ref);
148 if (--bslab->slab_ref)
151 kmem_cache_destroy(bslab->slab);
155 mutex_unlock(&bio_slab_lock);
158 unsigned int bvec_nr_vecs(unsigned short idx)
160 return bvec_slabs[idx].nr_vecs;
163 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
165 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
167 if (idx == BIOVEC_MAX_IDX)
168 mempool_free(bv, pool);
170 struct biovec_slab *bvs = bvec_slabs + idx;
172 kmem_cache_free(bvs->slab, bv);
176 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
182 * see comment near bvec_array define!
200 case 129 ... BIO_MAX_PAGES:
208 * idx now points to the pool we want to allocate from. only the
209 * 1-vec entry pool is mempool backed.
211 if (*idx == BIOVEC_MAX_IDX) {
213 bvl = mempool_alloc(pool, gfp_mask);
215 struct biovec_slab *bvs = bvec_slabs + *idx;
216 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
219 * Make this allocation restricted and don't dump info on
220 * allocation failures, since we'll fallback to the mempool
221 * in case of failure.
223 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
226 * Try a slab allocation. If this fails and __GFP_WAIT
227 * is set, retry with the 1-entry mempool
229 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
230 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
231 *idx = BIOVEC_MAX_IDX;
239 static void __bio_free(struct bio *bio)
241 bio_disassociate_task(bio);
243 if (bio_integrity(bio))
244 bio_integrity_free(bio);
247 static void bio_free(struct bio *bio)
249 struct bio_set *bs = bio->bi_pool;
255 if (bio_has_allocated_vec(bio))
256 bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
259 * If we have front padding, adjust the bio pointer before freeing
264 mempool_free(p, bs->bio_pool);
266 /* Bio was allocated by bio_kmalloc() */
271 void bio_init(struct bio *bio)
273 memset(bio, 0, sizeof(*bio));
274 bio->bi_flags = 1 << BIO_UPTODATE;
275 atomic_set(&bio->bi_cnt, 1);
277 EXPORT_SYMBOL(bio_init);
280 * bio_reset - reinitialize a bio
284 * After calling bio_reset(), @bio will be in the same state as a freshly
285 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
286 * preserved are the ones that are initialized by bio_alloc_bioset(). See
287 * comment in struct bio.
289 void bio_reset(struct bio *bio)
291 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
295 memset(bio, 0, BIO_RESET_BYTES);
296 bio->bi_flags = flags|(1 << BIO_UPTODATE);
298 EXPORT_SYMBOL(bio_reset);
300 static void bio_alloc_rescue(struct work_struct *work)
302 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
306 spin_lock(&bs->rescue_lock);
307 bio = bio_list_pop(&bs->rescue_list);
308 spin_unlock(&bs->rescue_lock);
313 generic_make_request(bio);
317 static void punt_bios_to_rescuer(struct bio_set *bs)
319 struct bio_list punt, nopunt;
323 * In order to guarantee forward progress we must punt only bios that
324 * were allocated from this bio_set; otherwise, if there was a bio on
325 * there for a stacking driver higher up in the stack, processing it
326 * could require allocating bios from this bio_set, and doing that from
327 * our own rescuer would be bad.
329 * Since bio lists are singly linked, pop them all instead of trying to
330 * remove from the middle of the list:
333 bio_list_init(&punt);
334 bio_list_init(&nopunt);
336 while ((bio = bio_list_pop(current->bio_list)))
337 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
339 *current->bio_list = nopunt;
341 spin_lock(&bs->rescue_lock);
342 bio_list_merge(&bs->rescue_list, &punt);
343 spin_unlock(&bs->rescue_lock);
345 queue_work(bs->rescue_workqueue, &bs->rescue_work);
349 * bio_alloc_bioset - allocate a bio for I/O
350 * @gfp_mask: the GFP_ mask given to the slab allocator
351 * @nr_iovecs: number of iovecs to pre-allocate
352 * @bs: the bio_set to allocate from.
355 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
356 * backed by the @bs's mempool.
358 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
359 * able to allocate a bio. This is due to the mempool guarantees. To make this
360 * work, callers must never allocate more than 1 bio at a time from this pool.
361 * Callers that need to allocate more than 1 bio must always submit the
362 * previously allocated bio for IO before attempting to allocate a new one.
363 * Failure to do so can cause deadlocks under memory pressure.
365 * Note that when running under generic_make_request() (i.e. any block
366 * driver), bios are not submitted until after you return - see the code in
367 * generic_make_request() that converts recursion into iteration, to prevent
370 * This would normally mean allocating multiple bios under
371 * generic_make_request() would be susceptible to deadlocks, but we have
372 * deadlock avoidance code that resubmits any blocked bios from a rescuer
375 * However, we do not guarantee forward progress for allocations from other
376 * mempools. Doing multiple allocations from the same mempool under
377 * generic_make_request() should be avoided - instead, use bio_set's front_pad
378 * for per bio allocations.
381 * Pointer to new bio on success, NULL on failure.
383 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
385 gfp_t saved_gfp = gfp_mask;
387 unsigned inline_vecs;
388 unsigned long idx = BIO_POOL_NONE;
389 struct bio_vec *bvl = NULL;
394 if (nr_iovecs > UIO_MAXIOV)
397 p = kmalloc(sizeof(struct bio) +
398 nr_iovecs * sizeof(struct bio_vec),
401 inline_vecs = nr_iovecs;
404 * generic_make_request() converts recursion to iteration; this
405 * means if we're running beneath it, any bios we allocate and
406 * submit will not be submitted (and thus freed) until after we
409 * This exposes us to a potential deadlock if we allocate
410 * multiple bios from the same bio_set() while running
411 * underneath generic_make_request(). If we were to allocate
412 * multiple bios (say a stacking block driver that was splitting
413 * bios), we would deadlock if we exhausted the mempool's
416 * We solve this, and guarantee forward progress, with a rescuer
417 * workqueue per bio_set. If we go to allocate and there are
418 * bios on current->bio_list, we first try the allocation
419 * without __GFP_WAIT; if that fails, we punt those bios we
420 * would be blocking to the rescuer workqueue before we retry
421 * with the original gfp_flags.
424 if (current->bio_list && !bio_list_empty(current->bio_list))
425 gfp_mask &= ~__GFP_WAIT;
427 p = mempool_alloc(bs->bio_pool, gfp_mask);
428 if (!p && gfp_mask != saved_gfp) {
429 punt_bios_to_rescuer(bs);
430 gfp_mask = saved_gfp;
431 p = mempool_alloc(bs->bio_pool, gfp_mask);
434 front_pad = bs->front_pad;
435 inline_vecs = BIO_INLINE_VECS;
444 if (nr_iovecs > inline_vecs) {
445 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
446 if (!bvl && gfp_mask != saved_gfp) {
447 punt_bios_to_rescuer(bs);
448 gfp_mask = saved_gfp;
449 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
454 } else if (nr_iovecs) {
455 bvl = bio->bi_inline_vecs;
459 bio->bi_flags |= idx << BIO_POOL_OFFSET;
460 bio->bi_max_vecs = nr_iovecs;
461 bio->bi_io_vec = bvl;
465 mempool_free(p, bs->bio_pool);
468 EXPORT_SYMBOL(bio_alloc_bioset);
470 void zero_fill_bio(struct bio *bio)
476 bio_for_each_segment(bv, bio, i) {
477 char *data = bvec_kmap_irq(bv, &flags);
478 memset(data, 0, bv->bv_len);
479 flush_dcache_page(bv->bv_page);
480 bvec_kunmap_irq(data, &flags);
483 EXPORT_SYMBOL(zero_fill_bio);
486 * bio_put - release a reference to a bio
487 * @bio: bio to release reference to
490 * Put a reference to a &struct bio, either one you have gotten with
491 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
493 void bio_put(struct bio *bio)
495 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
500 if (atomic_dec_and_test(&bio->bi_cnt))
503 EXPORT_SYMBOL(bio_put);
505 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
507 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
508 blk_recount_segments(q, bio);
510 return bio->bi_phys_segments;
512 EXPORT_SYMBOL(bio_phys_segments);
515 * __bio_clone - clone a bio
516 * @bio: destination bio
517 * @bio_src: bio to clone
519 * Clone a &bio. Caller will own the returned bio, but not
520 * the actual data it points to. Reference count of returned
523 void __bio_clone(struct bio *bio, struct bio *bio_src)
525 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
526 bio_src->bi_max_vecs * sizeof(struct bio_vec));
529 * most users will be overriding ->bi_bdev with a new target,
530 * so we don't set nor calculate new physical/hw segment counts here
532 bio->bi_sector = bio_src->bi_sector;
533 bio->bi_bdev = bio_src->bi_bdev;
534 bio->bi_flags |= 1 << BIO_CLONED;
535 bio->bi_rw = bio_src->bi_rw;
536 bio->bi_vcnt = bio_src->bi_vcnt;
537 bio->bi_size = bio_src->bi_size;
538 bio->bi_idx = bio_src->bi_idx;
540 EXPORT_SYMBOL(__bio_clone);
543 * bio_clone_bioset - clone a bio
545 * @gfp_mask: allocation priority
546 * @bs: bio_set to allocate from
548 * Like __bio_clone, only also allocates the returned bio
550 struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask,
555 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs);
561 if (bio_integrity(bio)) {
564 ret = bio_integrity_clone(b, bio, gfp_mask);
574 EXPORT_SYMBOL(bio_clone_bioset);
577 * bio_get_nr_vecs - return approx number of vecs
580 * Return the approximate number of pages we can send to this target.
581 * There's no guarantee that you will be able to fit this number of pages
582 * into a bio, it does not account for dynamic restrictions that vary
585 int bio_get_nr_vecs(struct block_device *bdev)
587 struct request_queue *q = bdev_get_queue(bdev);
590 nr_pages = min_t(unsigned,
591 queue_max_segments(q),
592 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
594 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
597 EXPORT_SYMBOL(bio_get_nr_vecs);
599 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
600 *page, unsigned int len, unsigned int offset,
601 unsigned short max_sectors)
603 int retried_segments = 0;
604 struct bio_vec *bvec;
607 * cloned bio must not modify vec list
609 if (unlikely(bio_flagged(bio, BIO_CLONED)))
612 if (((bio->bi_size + len) >> 9) > max_sectors)
616 * For filesystems with a blocksize smaller than the pagesize
617 * we will often be called with the same page as last time and
618 * a consecutive offset. Optimize this special case.
620 if (bio->bi_vcnt > 0) {
621 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
623 if (page == prev->bv_page &&
624 offset == prev->bv_offset + prev->bv_len) {
625 unsigned int prev_bv_len = prev->bv_len;
628 if (q->merge_bvec_fn) {
629 struct bvec_merge_data bvm = {
630 /* prev_bvec is already charged in
631 bi_size, discharge it in order to
632 simulate merging updated prev_bvec
634 .bi_bdev = bio->bi_bdev,
635 .bi_sector = bio->bi_sector,
636 .bi_size = bio->bi_size - prev_bv_len,
640 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
650 if (bio->bi_vcnt >= bio->bi_max_vecs)
654 * we might lose a segment or two here, but rather that than
655 * make this too complex.
658 while (bio->bi_phys_segments >= queue_max_segments(q)) {
660 if (retried_segments)
663 retried_segments = 1;
664 blk_recount_segments(q, bio);
668 * setup the new entry, we might clear it again later if we
669 * cannot add the page
671 bvec = &bio->bi_io_vec[bio->bi_vcnt];
672 bvec->bv_page = page;
674 bvec->bv_offset = offset;
677 * if queue has other restrictions (eg varying max sector size
678 * depending on offset), it can specify a merge_bvec_fn in the
679 * queue to get further control
681 if (q->merge_bvec_fn) {
682 struct bvec_merge_data bvm = {
683 .bi_bdev = bio->bi_bdev,
684 .bi_sector = bio->bi_sector,
685 .bi_size = bio->bi_size,
690 * merge_bvec_fn() returns number of bytes it can accept
693 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
694 bvec->bv_page = NULL;
701 /* If we may be able to merge these biovecs, force a recount */
702 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
703 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
706 bio->bi_phys_segments++;
713 * bio_add_pc_page - attempt to add page to bio
714 * @q: the target queue
715 * @bio: destination bio
717 * @len: vec entry length
718 * @offset: vec entry offset
720 * Attempt to add a page to the bio_vec maplist. This can fail for a
721 * number of reasons, such as the bio being full or target block device
722 * limitations. The target block device must allow bio's up to PAGE_SIZE,
723 * so it is always possible to add a single page to an empty bio.
725 * This should only be used by REQ_PC bios.
727 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
728 unsigned int len, unsigned int offset)
730 return __bio_add_page(q, bio, page, len, offset,
731 queue_max_hw_sectors(q));
733 EXPORT_SYMBOL(bio_add_pc_page);
736 * bio_add_page - attempt to add page to bio
737 * @bio: destination bio
739 * @len: vec entry length
740 * @offset: vec entry offset
742 * Attempt to add a page to the bio_vec maplist. This can fail for a
743 * number of reasons, such as the bio being full or target block device
744 * limitations. The target block device must allow bio's up to PAGE_SIZE,
745 * so it is always possible to add a single page to an empty bio.
747 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
750 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
751 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
753 EXPORT_SYMBOL(bio_add_page);
756 * bio_advance - increment/complete a bio by some number of bytes
757 * @bio: bio to advance
758 * @bytes: number of bytes to complete
760 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
761 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
762 * be updated on the last bvec as well.
764 * @bio will then represent the remaining, uncompleted portion of the io.
766 void bio_advance(struct bio *bio, unsigned bytes)
768 if (bio_integrity(bio))
769 bio_integrity_advance(bio, bytes);
771 bio->bi_sector += bytes >> 9;
772 bio->bi_size -= bytes;
774 if (bio->bi_rw & BIO_NO_ADVANCE_ITER_MASK)
778 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
779 WARN_ONCE(1, "bio idx %d >= vcnt %d\n",
780 bio->bi_idx, bio->bi_vcnt);
784 if (bytes >= bio_iovec(bio)->bv_len) {
785 bytes -= bio_iovec(bio)->bv_len;
788 bio_iovec(bio)->bv_len -= bytes;
789 bio_iovec(bio)->bv_offset += bytes;
794 EXPORT_SYMBOL(bio_advance);
796 struct bio_map_data {
797 struct bio_vec *iovecs;
798 struct sg_iovec *sgvecs;
803 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
804 struct sg_iovec *iov, int iov_count,
807 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
808 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
809 bmd->nr_sgvecs = iov_count;
810 bmd->is_our_pages = is_our_pages;
811 bio->bi_private = bmd;
814 static void bio_free_map_data(struct bio_map_data *bmd)
821 static struct bio_map_data *bio_alloc_map_data(int nr_segs,
822 unsigned int iov_count,
825 struct bio_map_data *bmd;
827 if (iov_count > UIO_MAXIOV)
830 bmd = kmalloc(sizeof(*bmd), gfp_mask);
834 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
840 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
849 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
850 struct sg_iovec *iov, int iov_count,
851 int to_user, int from_user, int do_free_page)
854 struct bio_vec *bvec;
856 unsigned int iov_off = 0;
858 __bio_for_each_segment(bvec, bio, i, 0) {
859 char *bv_addr = page_address(bvec->bv_page);
860 unsigned int bv_len = iovecs[i].bv_len;
862 while (bv_len && iov_idx < iov_count) {
864 char __user *iov_addr;
866 bytes = min_t(unsigned int,
867 iov[iov_idx].iov_len - iov_off, bv_len);
868 iov_addr = iov[iov_idx].iov_base + iov_off;
872 ret = copy_to_user(iov_addr, bv_addr,
876 ret = copy_from_user(bv_addr, iov_addr,
888 if (iov[iov_idx].iov_len == iov_off) {
895 __free_page(bvec->bv_page);
902 * bio_uncopy_user - finish previously mapped bio
903 * @bio: bio being terminated
905 * Free pages allocated from bio_copy_user() and write back data
906 * to user space in case of a read.
908 int bio_uncopy_user(struct bio *bio)
910 struct bio_map_data *bmd = bio->bi_private;
913 if (!bio_flagged(bio, BIO_NULL_MAPPED))
914 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
915 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
916 0, bmd->is_our_pages);
917 bio_free_map_data(bmd);
921 EXPORT_SYMBOL(bio_uncopy_user);
924 * bio_copy_user_iov - copy user data to bio
925 * @q: destination block queue
926 * @map_data: pointer to the rq_map_data holding pages (if necessary)
928 * @iov_count: number of elements in the iovec
929 * @write_to_vm: bool indicating writing to pages or not
930 * @gfp_mask: memory allocation flags
932 * Prepares and returns a bio for indirect user io, bouncing data
933 * to/from kernel pages as necessary. Must be paired with
934 * call bio_uncopy_user() on io completion.
936 struct bio *bio_copy_user_iov(struct request_queue *q,
937 struct rq_map_data *map_data,
938 struct sg_iovec *iov, int iov_count,
939 int write_to_vm, gfp_t gfp_mask)
941 struct bio_map_data *bmd;
942 struct bio_vec *bvec;
947 unsigned int len = 0;
948 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
950 for (i = 0; i < iov_count; i++) {
955 uaddr = (unsigned long)iov[i].iov_base;
956 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
957 start = uaddr >> PAGE_SHIFT;
963 return ERR_PTR(-EINVAL);
965 nr_pages += end - start;
966 len += iov[i].iov_len;
972 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
974 return ERR_PTR(-ENOMEM);
977 bio = bio_kmalloc(gfp_mask, nr_pages);
982 bio->bi_rw |= REQ_WRITE;
987 nr_pages = 1 << map_data->page_order;
988 i = map_data->offset / PAGE_SIZE;
991 unsigned int bytes = PAGE_SIZE;
999 if (i == map_data->nr_entries * nr_pages) {
1004 page = map_data->pages[i / nr_pages];
1005 page += (i % nr_pages);
1009 page = alloc_page(q->bounce_gfp | gfp_mask);
1016 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1029 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
1030 (map_data && map_data->from_user)) {
1031 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
1036 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
1040 bio_for_each_segment(bvec, bio, i)
1041 __free_page(bvec->bv_page);
1045 bio_free_map_data(bmd);
1046 return ERR_PTR(ret);
1050 * bio_copy_user - copy user data to bio
1051 * @q: destination block queue
1052 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1053 * @uaddr: start of user address
1054 * @len: length in bytes
1055 * @write_to_vm: bool indicating writing to pages or not
1056 * @gfp_mask: memory allocation flags
1058 * Prepares and returns a bio for indirect user io, bouncing data
1059 * to/from kernel pages as necessary. Must be paired with
1060 * call bio_uncopy_user() on io completion.
1062 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
1063 unsigned long uaddr, unsigned int len,
1064 int write_to_vm, gfp_t gfp_mask)
1066 struct sg_iovec iov;
1068 iov.iov_base = (void __user *)uaddr;
1071 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
1073 EXPORT_SYMBOL(bio_copy_user);
1075 static struct bio *__bio_map_user_iov(struct request_queue *q,
1076 struct block_device *bdev,
1077 struct sg_iovec *iov, int iov_count,
1078 int write_to_vm, gfp_t gfp_mask)
1082 struct page **pages;
1087 for (i = 0; i < iov_count; i++) {
1088 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1089 unsigned long len = iov[i].iov_len;
1090 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1091 unsigned long start = uaddr >> PAGE_SHIFT;
1097 return ERR_PTR(-EINVAL);
1099 nr_pages += end - start;
1101 * buffer must be aligned to at least hardsector size for now
1103 if (uaddr & queue_dma_alignment(q))
1104 return ERR_PTR(-EINVAL);
1108 return ERR_PTR(-EINVAL);
1110 bio = bio_kmalloc(gfp_mask, nr_pages);
1112 return ERR_PTR(-ENOMEM);
1115 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1119 for (i = 0; i < iov_count; i++) {
1120 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1121 unsigned long len = iov[i].iov_len;
1122 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1123 unsigned long start = uaddr >> PAGE_SHIFT;
1124 const int local_nr_pages = end - start;
1125 const int page_limit = cur_page + local_nr_pages;
1127 ret = get_user_pages_fast(uaddr, local_nr_pages,
1128 write_to_vm, &pages[cur_page]);
1129 if (ret < local_nr_pages) {
1134 offset = uaddr & ~PAGE_MASK;
1135 for (j = cur_page; j < page_limit; j++) {
1136 unsigned int bytes = PAGE_SIZE - offset;
1147 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1157 * release the pages we didn't map into the bio, if any
1159 while (j < page_limit)
1160 page_cache_release(pages[j++]);
1166 * set data direction, and check if mapped pages need bouncing
1169 bio->bi_rw |= REQ_WRITE;
1171 bio->bi_bdev = bdev;
1172 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1176 for (i = 0; i < nr_pages; i++) {
1179 page_cache_release(pages[i]);
1184 return ERR_PTR(ret);
1188 * bio_map_user - map user address into bio
1189 * @q: the struct request_queue for the bio
1190 * @bdev: destination block device
1191 * @uaddr: start of user address
1192 * @len: length in bytes
1193 * @write_to_vm: bool indicating writing to pages or not
1194 * @gfp_mask: memory allocation flags
1196 * Map the user space address into a bio suitable for io to a block
1197 * device. Returns an error pointer in case of error.
1199 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1200 unsigned long uaddr, unsigned int len, int write_to_vm,
1203 struct sg_iovec iov;
1205 iov.iov_base = (void __user *)uaddr;
1208 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1210 EXPORT_SYMBOL(bio_map_user);
1213 * bio_map_user_iov - map user sg_iovec table into bio
1214 * @q: the struct request_queue for the bio
1215 * @bdev: destination block device
1217 * @iov_count: number of elements in the iovec
1218 * @write_to_vm: bool indicating writing to pages or not
1219 * @gfp_mask: memory allocation flags
1221 * Map the user space address into a bio suitable for io to a block
1222 * device. Returns an error pointer in case of error.
1224 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1225 struct sg_iovec *iov, int iov_count,
1226 int write_to_vm, gfp_t gfp_mask)
1230 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1236 * subtle -- if __bio_map_user() ended up bouncing a bio,
1237 * it would normally disappear when its bi_end_io is run.
1238 * however, we need it for the unmap, so grab an extra
1246 static void __bio_unmap_user(struct bio *bio)
1248 struct bio_vec *bvec;
1252 * make sure we dirty pages we wrote to
1254 __bio_for_each_segment(bvec, bio, i, 0) {
1255 if (bio_data_dir(bio) == READ)
1256 set_page_dirty_lock(bvec->bv_page);
1258 page_cache_release(bvec->bv_page);
1265 * bio_unmap_user - unmap a bio
1266 * @bio: the bio being unmapped
1268 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1269 * a process context.
1271 * bio_unmap_user() may sleep.
1273 void bio_unmap_user(struct bio *bio)
1275 __bio_unmap_user(bio);
1278 EXPORT_SYMBOL(bio_unmap_user);
1280 static void bio_map_kern_endio(struct bio *bio, int err)
1285 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1286 unsigned int len, gfp_t gfp_mask)
1288 unsigned long kaddr = (unsigned long)data;
1289 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1290 unsigned long start = kaddr >> PAGE_SHIFT;
1291 const int nr_pages = end - start;
1295 bio = bio_kmalloc(gfp_mask, nr_pages);
1297 return ERR_PTR(-ENOMEM);
1299 offset = offset_in_page(kaddr);
1300 for (i = 0; i < nr_pages; i++) {
1301 unsigned int bytes = PAGE_SIZE - offset;
1309 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1318 bio->bi_end_io = bio_map_kern_endio;
1323 * bio_map_kern - map kernel address into bio
1324 * @q: the struct request_queue for the bio
1325 * @data: pointer to buffer to map
1326 * @len: length in bytes
1327 * @gfp_mask: allocation flags for bio allocation
1329 * Map the kernel address into a bio suitable for io to a block
1330 * device. Returns an error pointer in case of error.
1332 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1337 bio = __bio_map_kern(q, data, len, gfp_mask);
1341 if (bio->bi_size == len)
1345 * Don't support partial mappings.
1348 return ERR_PTR(-EINVAL);
1350 EXPORT_SYMBOL(bio_map_kern);
1352 static void bio_copy_kern_endio(struct bio *bio, int err)
1354 struct bio_vec *bvec;
1355 const int read = bio_data_dir(bio) == READ;
1356 struct bio_map_data *bmd = bio->bi_private;
1358 char *p = bmd->sgvecs[0].iov_base;
1360 __bio_for_each_segment(bvec, bio, i, 0) {
1361 char *addr = page_address(bvec->bv_page);
1362 int len = bmd->iovecs[i].bv_len;
1365 memcpy(p, addr, len);
1367 __free_page(bvec->bv_page);
1371 bio_free_map_data(bmd);
1376 * bio_copy_kern - copy kernel address into bio
1377 * @q: the struct request_queue for the bio
1378 * @data: pointer to buffer to copy
1379 * @len: length in bytes
1380 * @gfp_mask: allocation flags for bio and page allocation
1381 * @reading: data direction is READ
1383 * copy the kernel address into a bio suitable for io to a block
1384 * device. Returns an error pointer in case of error.
1386 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1387 gfp_t gfp_mask, int reading)
1390 struct bio_vec *bvec;
1393 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1400 bio_for_each_segment(bvec, bio, i) {
1401 char *addr = page_address(bvec->bv_page);
1403 memcpy(addr, p, bvec->bv_len);
1408 bio->bi_end_io = bio_copy_kern_endio;
1412 EXPORT_SYMBOL(bio_copy_kern);
1415 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1416 * for performing direct-IO in BIOs.
1418 * The problem is that we cannot run set_page_dirty() from interrupt context
1419 * because the required locks are not interrupt-safe. So what we can do is to
1420 * mark the pages dirty _before_ performing IO. And in interrupt context,
1421 * check that the pages are still dirty. If so, fine. If not, redirty them
1422 * in process context.
1424 * We special-case compound pages here: normally this means reads into hugetlb
1425 * pages. The logic in here doesn't really work right for compound pages
1426 * because the VM does not uniformly chase down the head page in all cases.
1427 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1428 * handle them at all. So we skip compound pages here at an early stage.
1430 * Note that this code is very hard to test under normal circumstances because
1431 * direct-io pins the pages with get_user_pages(). This makes
1432 * is_page_cache_freeable return false, and the VM will not clean the pages.
1433 * But other code (eg, flusher threads) could clean the pages if they are mapped
1436 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1437 * deferred bio dirtying paths.
1441 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1443 void bio_set_pages_dirty(struct bio *bio)
1445 struct bio_vec *bvec = bio->bi_io_vec;
1448 for (i = 0; i < bio->bi_vcnt; i++) {
1449 struct page *page = bvec[i].bv_page;
1451 if (page && !PageCompound(page))
1452 set_page_dirty_lock(page);
1456 static void bio_release_pages(struct bio *bio)
1458 struct bio_vec *bvec = bio->bi_io_vec;
1461 for (i = 0; i < bio->bi_vcnt; i++) {
1462 struct page *page = bvec[i].bv_page;
1470 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1471 * If they are, then fine. If, however, some pages are clean then they must
1472 * have been written out during the direct-IO read. So we take another ref on
1473 * the BIO and the offending pages and re-dirty the pages in process context.
1475 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1476 * here on. It will run one page_cache_release() against each page and will
1477 * run one bio_put() against the BIO.
1480 static void bio_dirty_fn(struct work_struct *work);
1482 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1483 static DEFINE_SPINLOCK(bio_dirty_lock);
1484 static struct bio *bio_dirty_list;
1487 * This runs in process context
1489 static void bio_dirty_fn(struct work_struct *work)
1491 unsigned long flags;
1494 spin_lock_irqsave(&bio_dirty_lock, flags);
1495 bio = bio_dirty_list;
1496 bio_dirty_list = NULL;
1497 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1500 struct bio *next = bio->bi_private;
1502 bio_set_pages_dirty(bio);
1503 bio_release_pages(bio);
1509 void bio_check_pages_dirty(struct bio *bio)
1511 struct bio_vec *bvec = bio->bi_io_vec;
1512 int nr_clean_pages = 0;
1515 for (i = 0; i < bio->bi_vcnt; i++) {
1516 struct page *page = bvec[i].bv_page;
1518 if (PageDirty(page) || PageCompound(page)) {
1519 page_cache_release(page);
1520 bvec[i].bv_page = NULL;
1526 if (nr_clean_pages) {
1527 unsigned long flags;
1529 spin_lock_irqsave(&bio_dirty_lock, flags);
1530 bio->bi_private = bio_dirty_list;
1531 bio_dirty_list = bio;
1532 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1533 schedule_work(&bio_dirty_work);
1539 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1540 void bio_flush_dcache_pages(struct bio *bi)
1543 struct bio_vec *bvec;
1545 bio_for_each_segment(bvec, bi, i)
1546 flush_dcache_page(bvec->bv_page);
1548 EXPORT_SYMBOL(bio_flush_dcache_pages);
1552 * bio_endio - end I/O on a bio
1554 * @error: error, if any
1557 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1558 * preferred way to end I/O on a bio, it takes care of clearing
1559 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1560 * established -Exxxx (-EIO, for instance) error values in case
1561 * something went wrong. No one should call bi_end_io() directly on a
1562 * bio unless they own it and thus know that it has an end_io
1565 void bio_endio(struct bio *bio, int error)
1568 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1569 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1572 trace_block_bio_complete(bio, error);
1575 bio->bi_end_io(bio, error);
1577 EXPORT_SYMBOL(bio_endio);
1579 void bio_pair_release(struct bio_pair *bp)
1581 if (atomic_dec_and_test(&bp->cnt)) {
1582 struct bio *master = bp->bio1.bi_private;
1584 bio_endio(master, bp->error);
1585 mempool_free(bp, bp->bio2.bi_private);
1588 EXPORT_SYMBOL(bio_pair_release);
1590 static void bio_pair_end_1(struct bio *bi, int err)
1592 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1597 bio_pair_release(bp);
1600 static void bio_pair_end_2(struct bio *bi, int err)
1602 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1607 bio_pair_release(bp);
1611 * split a bio - only worry about a bio with a single page in its iovec
1613 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1615 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1620 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1621 bi->bi_sector + first_sectors);
1623 BUG_ON(bi->bi_vcnt != 1 && bi->bi_vcnt != 0);
1624 BUG_ON(bi->bi_idx != 0);
1625 atomic_set(&bp->cnt, 3);
1629 bp->bio2.bi_sector += first_sectors;
1630 bp->bio2.bi_size -= first_sectors << 9;
1631 bp->bio1.bi_size = first_sectors << 9;
1633 if (bi->bi_vcnt != 0) {
1634 bp->bv1 = bi->bi_io_vec[0];
1635 bp->bv2 = bi->bi_io_vec[0];
1637 if (bio_is_rw(bi)) {
1638 bp->bv2.bv_offset += first_sectors << 9;
1639 bp->bv2.bv_len -= first_sectors << 9;
1640 bp->bv1.bv_len = first_sectors << 9;
1643 bp->bio1.bi_io_vec = &bp->bv1;
1644 bp->bio2.bi_io_vec = &bp->bv2;
1646 bp->bio1.bi_max_vecs = 1;
1647 bp->bio2.bi_max_vecs = 1;
1650 bp->bio1.bi_end_io = bio_pair_end_1;
1651 bp->bio2.bi_end_io = bio_pair_end_2;
1653 bp->bio1.bi_private = bi;
1654 bp->bio2.bi_private = bio_split_pool;
1656 if (bio_integrity(bi))
1657 bio_integrity_split(bi, bp, first_sectors);
1661 EXPORT_SYMBOL(bio_split);
1664 * bio_sector_offset - Find hardware sector offset in bio
1665 * @bio: bio to inspect
1666 * @index: bio_vec index
1667 * @offset: offset in bv_page
1669 * Return the number of hardware sectors between beginning of bio
1670 * and an end point indicated by a bio_vec index and an offset
1671 * within that vector's page.
1673 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1674 unsigned int offset)
1676 unsigned int sector_sz;
1681 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1684 if (index >= bio->bi_idx)
1685 index = bio->bi_vcnt - 1;
1687 __bio_for_each_segment(bv, bio, i, 0) {
1689 if (offset > bv->bv_offset)
1690 sectors += (offset - bv->bv_offset) / sector_sz;
1694 sectors += bv->bv_len / sector_sz;
1699 EXPORT_SYMBOL(bio_sector_offset);
1702 * create memory pools for biovec's in a bio_set.
1703 * use the global biovec slabs created for general use.
1705 mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries)
1707 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1709 return mempool_create_slab_pool(pool_entries, bp->slab);
1712 void bioset_free(struct bio_set *bs)
1714 if (bs->rescue_workqueue)
1715 destroy_workqueue(bs->rescue_workqueue);
1718 mempool_destroy(bs->bio_pool);
1721 mempool_destroy(bs->bvec_pool);
1723 bioset_integrity_free(bs);
1728 EXPORT_SYMBOL(bioset_free);
1731 * bioset_create - Create a bio_set
1732 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1733 * @front_pad: Number of bytes to allocate in front of the returned bio
1736 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1737 * to ask for a number of bytes to be allocated in front of the bio.
1738 * Front pad allocation is useful for embedding the bio inside
1739 * another structure, to avoid allocating extra data to go with the bio.
1740 * Note that the bio must be embedded at the END of that structure always,
1741 * or things will break badly.
1743 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1745 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1748 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1752 bs->front_pad = front_pad;
1754 spin_lock_init(&bs->rescue_lock);
1755 bio_list_init(&bs->rescue_list);
1756 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1758 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1759 if (!bs->bio_slab) {
1764 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1768 bs->bvec_pool = biovec_create_pool(bs, pool_size);
1772 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1773 if (!bs->rescue_workqueue)
1781 EXPORT_SYMBOL(bioset_create);
1783 #ifdef CONFIG_BLK_CGROUP
1785 * bio_associate_current - associate a bio with %current
1788 * Associate @bio with %current if it hasn't been associated yet. Block
1789 * layer will treat @bio as if it were issued by %current no matter which
1790 * task actually issues it.
1792 * This function takes an extra reference of @task's io_context and blkcg
1793 * which will be put when @bio is released. The caller must own @bio,
1794 * ensure %current->io_context exists, and is responsible for synchronizing
1795 * calls to this function.
1797 int bio_associate_current(struct bio *bio)
1799 struct io_context *ioc;
1800 struct cgroup_subsys_state *css;
1805 ioc = current->io_context;
1809 /* acquire active ref on @ioc and associate */
1810 get_io_context_active(ioc);
1813 /* associate blkcg if exists */
1815 css = task_subsys_state(current, blkio_subsys_id);
1816 if (css && css_tryget(css))
1824 * bio_disassociate_task - undo bio_associate_current()
1827 void bio_disassociate_task(struct bio *bio)
1830 put_io_context(bio->bi_ioc);
1834 css_put(bio->bi_css);
1839 #endif /* CONFIG_BLK_CGROUP */
1841 static void __init biovec_init_slabs(void)
1845 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1847 struct biovec_slab *bvs = bvec_slabs + i;
1849 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1854 size = bvs->nr_vecs * sizeof(struct bio_vec);
1855 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1856 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1860 static int __init init_bio(void)
1864 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1866 panic("bio: can't allocate bios\n");
1868 bio_integrity_init();
1869 biovec_init_slabs();
1871 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1873 panic("bio: can't allocate bios\n");
1875 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1876 panic("bio: can't create integrity pool\n");
1878 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1879 sizeof(struct bio_pair));
1880 if (!bio_split_pool)
1881 panic("bio: can't create split pool\n");
1885 subsys_initcall(init_bio);