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);
755 struct submit_bio_ret {
756 struct completion event;
760 static void submit_bio_wait_endio(struct bio *bio, int error)
762 struct submit_bio_ret *ret = bio->bi_private;
765 complete(&ret->event);
769 * submit_bio_wait - submit a bio, and wait until it completes
770 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
771 * @bio: The &struct bio which describes the I/O
773 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
774 * bio_endio() on failure.
776 int submit_bio_wait(int rw, struct bio *bio)
778 struct submit_bio_ret ret;
781 init_completion(&ret.event);
782 bio->bi_private = &ret;
783 bio->bi_end_io = submit_bio_wait_endio;
785 wait_for_completion(&ret.event);
789 EXPORT_SYMBOL(submit_bio_wait);
792 * bio_advance - increment/complete a bio by some number of bytes
793 * @bio: bio to advance
794 * @bytes: number of bytes to complete
796 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
797 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
798 * be updated on the last bvec as well.
800 * @bio will then represent the remaining, uncompleted portion of the io.
802 void bio_advance(struct bio *bio, unsigned bytes)
804 if (bio_integrity(bio))
805 bio_integrity_advance(bio, bytes);
807 bio->bi_sector += bytes >> 9;
808 bio->bi_size -= bytes;
810 if (bio->bi_rw & BIO_NO_ADVANCE_ITER_MASK)
814 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
815 WARN_ONCE(1, "bio idx %d >= vcnt %d\n",
816 bio->bi_idx, bio->bi_vcnt);
820 if (bytes >= bio_iovec(bio)->bv_len) {
821 bytes -= bio_iovec(bio)->bv_len;
824 bio_iovec(bio)->bv_len -= bytes;
825 bio_iovec(bio)->bv_offset += bytes;
830 EXPORT_SYMBOL(bio_advance);
832 struct bio_map_data {
833 struct bio_vec *iovecs;
834 struct sg_iovec *sgvecs;
839 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
840 struct sg_iovec *iov, int iov_count,
843 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
844 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
845 bmd->nr_sgvecs = iov_count;
846 bmd->is_our_pages = is_our_pages;
847 bio->bi_private = bmd;
850 static void bio_free_map_data(struct bio_map_data *bmd)
857 static struct bio_map_data *bio_alloc_map_data(int nr_segs,
858 unsigned int iov_count,
861 struct bio_map_data *bmd;
863 if (iov_count > UIO_MAXIOV)
866 bmd = kmalloc(sizeof(*bmd), gfp_mask);
870 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
876 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
885 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
886 struct sg_iovec *iov, int iov_count,
887 int to_user, int from_user, int do_free_page)
890 struct bio_vec *bvec;
892 unsigned int iov_off = 0;
894 __bio_for_each_segment(bvec, bio, i, 0) {
895 char *bv_addr = page_address(bvec->bv_page);
896 unsigned int bv_len = iovecs[i].bv_len;
898 while (bv_len && iov_idx < iov_count) {
900 char __user *iov_addr;
902 bytes = min_t(unsigned int,
903 iov[iov_idx].iov_len - iov_off, bv_len);
904 iov_addr = iov[iov_idx].iov_base + iov_off;
908 ret = copy_to_user(iov_addr, bv_addr,
912 ret = copy_from_user(bv_addr, iov_addr,
924 if (iov[iov_idx].iov_len == iov_off) {
931 __free_page(bvec->bv_page);
938 * bio_uncopy_user - finish previously mapped bio
939 * @bio: bio being terminated
941 * Free pages allocated from bio_copy_user() and write back data
942 * to user space in case of a read.
944 int bio_uncopy_user(struct bio *bio)
946 struct bio_map_data *bmd = bio->bi_private;
949 if (!bio_flagged(bio, BIO_NULL_MAPPED))
950 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
951 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
952 0, bmd->is_our_pages);
953 bio_free_map_data(bmd);
957 EXPORT_SYMBOL(bio_uncopy_user);
960 * bio_copy_user_iov - copy user data to bio
961 * @q: destination block queue
962 * @map_data: pointer to the rq_map_data holding pages (if necessary)
964 * @iov_count: number of elements in the iovec
965 * @write_to_vm: bool indicating writing to pages or not
966 * @gfp_mask: memory allocation flags
968 * Prepares and returns a bio for indirect user io, bouncing data
969 * to/from kernel pages as necessary. Must be paired with
970 * call bio_uncopy_user() on io completion.
972 struct bio *bio_copy_user_iov(struct request_queue *q,
973 struct rq_map_data *map_data,
974 struct sg_iovec *iov, int iov_count,
975 int write_to_vm, gfp_t gfp_mask)
977 struct bio_map_data *bmd;
978 struct bio_vec *bvec;
983 unsigned int len = 0;
984 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
986 for (i = 0; i < iov_count; i++) {
991 uaddr = (unsigned long)iov[i].iov_base;
992 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
993 start = uaddr >> PAGE_SHIFT;
999 return ERR_PTR(-EINVAL);
1001 nr_pages += end - start;
1002 len += iov[i].iov_len;
1008 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
1010 return ERR_PTR(-ENOMEM);
1013 bio = bio_kmalloc(gfp_mask, nr_pages);
1018 bio->bi_rw |= REQ_WRITE;
1023 nr_pages = 1 << map_data->page_order;
1024 i = map_data->offset / PAGE_SIZE;
1027 unsigned int bytes = PAGE_SIZE;
1035 if (i == map_data->nr_entries * nr_pages) {
1040 page = map_data->pages[i / nr_pages];
1041 page += (i % nr_pages);
1045 page = alloc_page(q->bounce_gfp | gfp_mask);
1052 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1065 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
1066 (map_data && map_data->from_user)) {
1067 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
1072 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
1076 bio_for_each_segment(bvec, bio, i)
1077 __free_page(bvec->bv_page);
1081 bio_free_map_data(bmd);
1082 return ERR_PTR(ret);
1086 * bio_copy_user - copy user data to bio
1087 * @q: destination block queue
1088 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1089 * @uaddr: start of user address
1090 * @len: length in bytes
1091 * @write_to_vm: bool indicating writing to pages or not
1092 * @gfp_mask: memory allocation flags
1094 * Prepares and returns a bio for indirect user io, bouncing data
1095 * to/from kernel pages as necessary. Must be paired with
1096 * call bio_uncopy_user() on io completion.
1098 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
1099 unsigned long uaddr, unsigned int len,
1100 int write_to_vm, gfp_t gfp_mask)
1102 struct sg_iovec iov;
1104 iov.iov_base = (void __user *)uaddr;
1107 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
1109 EXPORT_SYMBOL(bio_copy_user);
1111 static struct bio *__bio_map_user_iov(struct request_queue *q,
1112 struct block_device *bdev,
1113 struct sg_iovec *iov, int iov_count,
1114 int write_to_vm, gfp_t gfp_mask)
1118 struct page **pages;
1123 for (i = 0; i < iov_count; i++) {
1124 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1125 unsigned long len = iov[i].iov_len;
1126 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1127 unsigned long start = uaddr >> PAGE_SHIFT;
1133 return ERR_PTR(-EINVAL);
1135 nr_pages += end - start;
1137 * buffer must be aligned to at least hardsector size for now
1139 if (uaddr & queue_dma_alignment(q))
1140 return ERR_PTR(-EINVAL);
1144 return ERR_PTR(-EINVAL);
1146 bio = bio_kmalloc(gfp_mask, nr_pages);
1148 return ERR_PTR(-ENOMEM);
1151 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1155 for (i = 0; i < iov_count; i++) {
1156 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1157 unsigned long len = iov[i].iov_len;
1158 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1159 unsigned long start = uaddr >> PAGE_SHIFT;
1160 const int local_nr_pages = end - start;
1161 const int page_limit = cur_page + local_nr_pages;
1163 ret = get_user_pages_fast(uaddr, local_nr_pages,
1164 write_to_vm, &pages[cur_page]);
1165 if (ret < local_nr_pages) {
1170 offset = uaddr & ~PAGE_MASK;
1171 for (j = cur_page; j < page_limit; j++) {
1172 unsigned int bytes = PAGE_SIZE - offset;
1183 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1193 * release the pages we didn't map into the bio, if any
1195 while (j < page_limit)
1196 page_cache_release(pages[j++]);
1202 * set data direction, and check if mapped pages need bouncing
1205 bio->bi_rw |= REQ_WRITE;
1207 bio->bi_bdev = bdev;
1208 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1212 for (i = 0; i < nr_pages; i++) {
1215 page_cache_release(pages[i]);
1220 return ERR_PTR(ret);
1224 * bio_map_user - map user address into bio
1225 * @q: the struct request_queue for the bio
1226 * @bdev: destination block device
1227 * @uaddr: start of user address
1228 * @len: length in bytes
1229 * @write_to_vm: bool indicating writing to pages or not
1230 * @gfp_mask: memory allocation flags
1232 * Map the user space address into a bio suitable for io to a block
1233 * device. Returns an error pointer in case of error.
1235 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1236 unsigned long uaddr, unsigned int len, int write_to_vm,
1239 struct sg_iovec iov;
1241 iov.iov_base = (void __user *)uaddr;
1244 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1246 EXPORT_SYMBOL(bio_map_user);
1249 * bio_map_user_iov - map user sg_iovec table into bio
1250 * @q: the struct request_queue for the bio
1251 * @bdev: destination block device
1253 * @iov_count: number of elements in the iovec
1254 * @write_to_vm: bool indicating writing to pages or not
1255 * @gfp_mask: memory allocation flags
1257 * Map the user space address into a bio suitable for io to a block
1258 * device. Returns an error pointer in case of error.
1260 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1261 struct sg_iovec *iov, int iov_count,
1262 int write_to_vm, gfp_t gfp_mask)
1266 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1272 * subtle -- if __bio_map_user() ended up bouncing a bio,
1273 * it would normally disappear when its bi_end_io is run.
1274 * however, we need it for the unmap, so grab an extra
1282 static void __bio_unmap_user(struct bio *bio)
1284 struct bio_vec *bvec;
1288 * make sure we dirty pages we wrote to
1290 __bio_for_each_segment(bvec, bio, i, 0) {
1291 if (bio_data_dir(bio) == READ)
1292 set_page_dirty_lock(bvec->bv_page);
1294 page_cache_release(bvec->bv_page);
1301 * bio_unmap_user - unmap a bio
1302 * @bio: the bio being unmapped
1304 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1305 * a process context.
1307 * bio_unmap_user() may sleep.
1309 void bio_unmap_user(struct bio *bio)
1311 __bio_unmap_user(bio);
1314 EXPORT_SYMBOL(bio_unmap_user);
1316 static void bio_map_kern_endio(struct bio *bio, int err)
1321 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1322 unsigned int len, gfp_t gfp_mask)
1324 unsigned long kaddr = (unsigned long)data;
1325 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1326 unsigned long start = kaddr >> PAGE_SHIFT;
1327 const int nr_pages = end - start;
1331 bio = bio_kmalloc(gfp_mask, nr_pages);
1333 return ERR_PTR(-ENOMEM);
1335 offset = offset_in_page(kaddr);
1336 for (i = 0; i < nr_pages; i++) {
1337 unsigned int bytes = PAGE_SIZE - offset;
1345 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1354 bio->bi_end_io = bio_map_kern_endio;
1359 * bio_map_kern - map kernel address into bio
1360 * @q: the struct request_queue for the bio
1361 * @data: pointer to buffer to map
1362 * @len: length in bytes
1363 * @gfp_mask: allocation flags for bio allocation
1365 * Map the kernel address into a bio suitable for io to a block
1366 * device. Returns an error pointer in case of error.
1368 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1373 bio = __bio_map_kern(q, data, len, gfp_mask);
1377 if (bio->bi_size == len)
1381 * Don't support partial mappings.
1384 return ERR_PTR(-EINVAL);
1386 EXPORT_SYMBOL(bio_map_kern);
1388 static void bio_copy_kern_endio(struct bio *bio, int err)
1390 struct bio_vec *bvec;
1391 const int read = bio_data_dir(bio) == READ;
1392 struct bio_map_data *bmd = bio->bi_private;
1394 char *p = bmd->sgvecs[0].iov_base;
1396 __bio_for_each_segment(bvec, bio, i, 0) {
1397 char *addr = page_address(bvec->bv_page);
1398 int len = bmd->iovecs[i].bv_len;
1401 memcpy(p, addr, len);
1403 __free_page(bvec->bv_page);
1407 bio_free_map_data(bmd);
1412 * bio_copy_kern - copy kernel address into bio
1413 * @q: the struct request_queue for the bio
1414 * @data: pointer to buffer to copy
1415 * @len: length in bytes
1416 * @gfp_mask: allocation flags for bio and page allocation
1417 * @reading: data direction is READ
1419 * copy the kernel address into a bio suitable for io to a block
1420 * device. Returns an error pointer in case of error.
1422 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1423 gfp_t gfp_mask, int reading)
1426 struct bio_vec *bvec;
1429 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1436 bio_for_each_segment(bvec, bio, i) {
1437 char *addr = page_address(bvec->bv_page);
1439 memcpy(addr, p, bvec->bv_len);
1444 bio->bi_end_io = bio_copy_kern_endio;
1448 EXPORT_SYMBOL(bio_copy_kern);
1451 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1452 * for performing direct-IO in BIOs.
1454 * The problem is that we cannot run set_page_dirty() from interrupt context
1455 * because the required locks are not interrupt-safe. So what we can do is to
1456 * mark the pages dirty _before_ performing IO. And in interrupt context,
1457 * check that the pages are still dirty. If so, fine. If not, redirty them
1458 * in process context.
1460 * We special-case compound pages here: normally this means reads into hugetlb
1461 * pages. The logic in here doesn't really work right for compound pages
1462 * because the VM does not uniformly chase down the head page in all cases.
1463 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1464 * handle them at all. So we skip compound pages here at an early stage.
1466 * Note that this code is very hard to test under normal circumstances because
1467 * direct-io pins the pages with get_user_pages(). This makes
1468 * is_page_cache_freeable return false, and the VM will not clean the pages.
1469 * But other code (eg, flusher threads) could clean the pages if they are mapped
1472 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1473 * deferred bio dirtying paths.
1477 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1479 void bio_set_pages_dirty(struct bio *bio)
1481 struct bio_vec *bvec = bio->bi_io_vec;
1484 for (i = 0; i < bio->bi_vcnt; i++) {
1485 struct page *page = bvec[i].bv_page;
1487 if (page && !PageCompound(page))
1488 set_page_dirty_lock(page);
1492 static void bio_release_pages(struct bio *bio)
1494 struct bio_vec *bvec = bio->bi_io_vec;
1497 for (i = 0; i < bio->bi_vcnt; i++) {
1498 struct page *page = bvec[i].bv_page;
1506 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1507 * If they are, then fine. If, however, some pages are clean then they must
1508 * have been written out during the direct-IO read. So we take another ref on
1509 * the BIO and the offending pages and re-dirty the pages in process context.
1511 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1512 * here on. It will run one page_cache_release() against each page and will
1513 * run one bio_put() against the BIO.
1516 static void bio_dirty_fn(struct work_struct *work);
1518 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1519 static DEFINE_SPINLOCK(bio_dirty_lock);
1520 static struct bio *bio_dirty_list;
1523 * This runs in process context
1525 static void bio_dirty_fn(struct work_struct *work)
1527 unsigned long flags;
1530 spin_lock_irqsave(&bio_dirty_lock, flags);
1531 bio = bio_dirty_list;
1532 bio_dirty_list = NULL;
1533 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1536 struct bio *next = bio->bi_private;
1538 bio_set_pages_dirty(bio);
1539 bio_release_pages(bio);
1545 void bio_check_pages_dirty(struct bio *bio)
1547 struct bio_vec *bvec = bio->bi_io_vec;
1548 int nr_clean_pages = 0;
1551 for (i = 0; i < bio->bi_vcnt; i++) {
1552 struct page *page = bvec[i].bv_page;
1554 if (PageDirty(page) || PageCompound(page)) {
1555 page_cache_release(page);
1556 bvec[i].bv_page = NULL;
1562 if (nr_clean_pages) {
1563 unsigned long flags;
1565 spin_lock_irqsave(&bio_dirty_lock, flags);
1566 bio->bi_private = bio_dirty_list;
1567 bio_dirty_list = bio;
1568 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1569 schedule_work(&bio_dirty_work);
1575 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1576 void bio_flush_dcache_pages(struct bio *bi)
1579 struct bio_vec *bvec;
1581 bio_for_each_segment(bvec, bi, i)
1582 flush_dcache_page(bvec->bv_page);
1584 EXPORT_SYMBOL(bio_flush_dcache_pages);
1588 * bio_endio - end I/O on a bio
1590 * @error: error, if any
1593 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1594 * preferred way to end I/O on a bio, it takes care of clearing
1595 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1596 * established -Exxxx (-EIO, for instance) error values in case
1597 * something went wrong. No one should call bi_end_io() directly on a
1598 * bio unless they own it and thus know that it has an end_io
1601 void bio_endio(struct bio *bio, int error)
1604 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1605 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1608 trace_block_bio_complete(bio, error);
1611 bio->bi_end_io(bio, error);
1613 EXPORT_SYMBOL(bio_endio);
1615 void bio_pair_release(struct bio_pair *bp)
1617 if (atomic_dec_and_test(&bp->cnt)) {
1618 struct bio *master = bp->bio1.bi_private;
1620 bio_endio(master, bp->error);
1621 mempool_free(bp, bp->bio2.bi_private);
1624 EXPORT_SYMBOL(bio_pair_release);
1626 static void bio_pair_end_1(struct bio *bi, int err)
1628 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1633 bio_pair_release(bp);
1636 static void bio_pair_end_2(struct bio *bi, int err)
1638 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1643 bio_pair_release(bp);
1647 * split a bio - only worry about a bio with a single page in its iovec
1649 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1651 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1656 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1657 bi->bi_sector + first_sectors);
1659 BUG_ON(bio_segments(bi) > 1);
1660 atomic_set(&bp->cnt, 3);
1664 bp->bio2.bi_sector += first_sectors;
1665 bp->bio2.bi_size -= first_sectors << 9;
1666 bp->bio1.bi_size = first_sectors << 9;
1668 if (bi->bi_vcnt != 0) {
1669 bp->bv1 = *bio_iovec(bi);
1670 bp->bv2 = *bio_iovec(bi);
1672 if (bio_is_rw(bi)) {
1673 bp->bv2.bv_offset += first_sectors << 9;
1674 bp->bv2.bv_len -= first_sectors << 9;
1675 bp->bv1.bv_len = first_sectors << 9;
1678 bp->bio1.bi_io_vec = &bp->bv1;
1679 bp->bio2.bi_io_vec = &bp->bv2;
1681 bp->bio1.bi_max_vecs = 1;
1682 bp->bio2.bi_max_vecs = 1;
1685 bp->bio1.bi_end_io = bio_pair_end_1;
1686 bp->bio2.bi_end_io = bio_pair_end_2;
1688 bp->bio1.bi_private = bi;
1689 bp->bio2.bi_private = bio_split_pool;
1691 if (bio_integrity(bi))
1692 bio_integrity_split(bi, bp, first_sectors);
1696 EXPORT_SYMBOL(bio_split);
1699 * bio_sector_offset - Find hardware sector offset in bio
1700 * @bio: bio to inspect
1701 * @index: bio_vec index
1702 * @offset: offset in bv_page
1704 * Return the number of hardware sectors between beginning of bio
1705 * and an end point indicated by a bio_vec index and an offset
1706 * within that vector's page.
1708 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1709 unsigned int offset)
1711 unsigned int sector_sz;
1716 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1719 if (index >= bio->bi_idx)
1720 index = bio->bi_vcnt - 1;
1722 __bio_for_each_segment(bv, bio, i, 0) {
1724 if (offset > bv->bv_offset)
1725 sectors += (offset - bv->bv_offset) / sector_sz;
1729 sectors += bv->bv_len / sector_sz;
1734 EXPORT_SYMBOL(bio_sector_offset);
1737 * create memory pools for biovec's in a bio_set.
1738 * use the global biovec slabs created for general use.
1740 mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries)
1742 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1744 return mempool_create_slab_pool(pool_entries, bp->slab);
1747 void bioset_free(struct bio_set *bs)
1749 if (bs->rescue_workqueue)
1750 destroy_workqueue(bs->rescue_workqueue);
1753 mempool_destroy(bs->bio_pool);
1756 mempool_destroy(bs->bvec_pool);
1758 bioset_integrity_free(bs);
1763 EXPORT_SYMBOL(bioset_free);
1766 * bioset_create - Create a bio_set
1767 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1768 * @front_pad: Number of bytes to allocate in front of the returned bio
1771 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1772 * to ask for a number of bytes to be allocated in front of the bio.
1773 * Front pad allocation is useful for embedding the bio inside
1774 * another structure, to avoid allocating extra data to go with the bio.
1775 * Note that the bio must be embedded at the END of that structure always,
1776 * or things will break badly.
1778 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1780 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1783 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1787 bs->front_pad = front_pad;
1789 spin_lock_init(&bs->rescue_lock);
1790 bio_list_init(&bs->rescue_list);
1791 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1793 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1794 if (!bs->bio_slab) {
1799 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1803 bs->bvec_pool = biovec_create_pool(bs, pool_size);
1807 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1808 if (!bs->rescue_workqueue)
1816 EXPORT_SYMBOL(bioset_create);
1818 #ifdef CONFIG_BLK_CGROUP
1820 * bio_associate_current - associate a bio with %current
1823 * Associate @bio with %current if it hasn't been associated yet. Block
1824 * layer will treat @bio as if it were issued by %current no matter which
1825 * task actually issues it.
1827 * This function takes an extra reference of @task's io_context and blkcg
1828 * which will be put when @bio is released. The caller must own @bio,
1829 * ensure %current->io_context exists, and is responsible for synchronizing
1830 * calls to this function.
1832 int bio_associate_current(struct bio *bio)
1834 struct io_context *ioc;
1835 struct cgroup_subsys_state *css;
1840 ioc = current->io_context;
1844 /* acquire active ref on @ioc and associate */
1845 get_io_context_active(ioc);
1848 /* associate blkcg if exists */
1850 css = task_subsys_state(current, blkio_subsys_id);
1851 if (css && css_tryget(css))
1859 * bio_disassociate_task - undo bio_associate_current()
1862 void bio_disassociate_task(struct bio *bio)
1865 put_io_context(bio->bi_ioc);
1869 css_put(bio->bi_css);
1874 #endif /* CONFIG_BLK_CGROUP */
1876 static void __init biovec_init_slabs(void)
1880 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1882 struct biovec_slab *bvs = bvec_slabs + i;
1884 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1889 size = bvs->nr_vecs * sizeof(struct bio_vec);
1890 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1891 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1895 static int __init init_bio(void)
1899 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1901 panic("bio: can't allocate bios\n");
1903 bio_integrity_init();
1904 biovec_init_slabs();
1906 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1908 panic("bio: can't allocate bios\n");
1910 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1911 panic("bio: can't create integrity pool\n");
1913 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1914 sizeof(struct bio_pair));
1915 if (!bio_split_pool)
1916 panic("bio: can't create split pool\n");
1920 subsys_initcall(init_bio);