Merge branch 'master' of git://git2.kernel.org/pub/scm/linux/kernel/git/torvalds...
[linux-2.6/linux-2.6-openrd.git] / fs / bio.c
blob88094afc29ea38c13906d861ca9bb7dd67c463a2
1 /*
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-
18 #include <linux/mm.h>
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/mempool.h>
27 #include <linux/workqueue.h>
28 #include <scsi/sg.h> /* for struct sg_iovec */
30 #include <trace/events/block.h>
33 * Test patch to inline a certain number of bi_io_vec's inside the bio
34 * itself, to shrink a bio data allocation from two mempool calls to one
36 #define BIO_INLINE_VECS 4
38 static mempool_t *bio_split_pool __read_mostly;
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
43 * unsigned short
45 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46 struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
47 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
49 #undef BV
52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
55 struct bio_set *fs_bio_set;
58 * Our slab pool management
60 struct bio_slab {
61 struct kmem_cache *slab;
62 unsigned int slab_ref;
63 unsigned int slab_size;
64 char name[8];
66 static DEFINE_MUTEX(bio_slab_lock);
67 static struct bio_slab *bio_slabs;
68 static unsigned int bio_slab_nr, bio_slab_max;
70 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72 unsigned int sz = sizeof(struct bio) + extra_size;
73 struct kmem_cache *slab = NULL;
74 struct bio_slab *bslab;
75 unsigned int i, entry = -1;
77 mutex_lock(&bio_slab_lock);
79 i = 0;
80 while (i < bio_slab_nr) {
81 bslab = &bio_slabs[i];
83 if (!bslab->slab && entry == -1)
84 entry = i;
85 else if (bslab->slab_size == sz) {
86 slab = bslab->slab;
87 bslab->slab_ref++;
88 break;
90 i++;
93 if (slab)
94 goto out_unlock;
96 if (bio_slab_nr == bio_slab_max && entry == -1) {
97 bio_slab_max <<= 1;
98 bio_slabs = krealloc(bio_slabs,
99 bio_slab_max * sizeof(struct bio_slab),
100 GFP_KERNEL);
101 if (!bio_slabs)
102 goto out_unlock;
104 if (entry == -1)
105 entry = bio_slab_nr++;
107 bslab = &bio_slabs[entry];
109 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
110 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
111 if (!slab)
112 goto out_unlock;
114 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
115 bslab->slab = slab;
116 bslab->slab_ref = 1;
117 bslab->slab_size = sz;
118 out_unlock:
119 mutex_unlock(&bio_slab_lock);
120 return slab;
123 static void bio_put_slab(struct bio_set *bs)
125 struct bio_slab *bslab = NULL;
126 unsigned int i;
128 mutex_lock(&bio_slab_lock);
130 for (i = 0; i < bio_slab_nr; i++) {
131 if (bs->bio_slab == bio_slabs[i].slab) {
132 bslab = &bio_slabs[i];
133 break;
137 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
138 goto out;
140 WARN_ON(!bslab->slab_ref);
142 if (--bslab->slab_ref)
143 goto out;
145 kmem_cache_destroy(bslab->slab);
146 bslab->slab = NULL;
148 out:
149 mutex_unlock(&bio_slab_lock);
152 unsigned int bvec_nr_vecs(unsigned short idx)
154 return bvec_slabs[idx].nr_vecs;
157 void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
159 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
161 if (idx == BIOVEC_MAX_IDX)
162 mempool_free(bv, bs->bvec_pool);
163 else {
164 struct biovec_slab *bvs = bvec_slabs + idx;
166 kmem_cache_free(bvs->slab, bv);
170 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
171 struct bio_set *bs)
173 struct bio_vec *bvl;
176 * see comment near bvec_array define!
178 switch (nr) {
179 case 1:
180 *idx = 0;
181 break;
182 case 2 ... 4:
183 *idx = 1;
184 break;
185 case 5 ... 16:
186 *idx = 2;
187 break;
188 case 17 ... 64:
189 *idx = 3;
190 break;
191 case 65 ... 128:
192 *idx = 4;
193 break;
194 case 129 ... BIO_MAX_PAGES:
195 *idx = 5;
196 break;
197 default:
198 return NULL;
202 * idx now points to the pool we want to allocate from. only the
203 * 1-vec entry pool is mempool backed.
205 if (*idx == BIOVEC_MAX_IDX) {
206 fallback:
207 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
208 } else {
209 struct biovec_slab *bvs = bvec_slabs + *idx;
210 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
213 * Make this allocation restricted and don't dump info on
214 * allocation failures, since we'll fallback to the mempool
215 * in case of failure.
217 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
220 * Try a slab allocation. If this fails and __GFP_WAIT
221 * is set, retry with the 1-entry mempool
223 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
224 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
225 *idx = BIOVEC_MAX_IDX;
226 goto fallback;
230 return bvl;
233 void bio_free(struct bio *bio, struct bio_set *bs)
235 void *p;
237 if (bio_has_allocated_vec(bio))
238 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
240 if (bio_integrity(bio))
241 bio_integrity_free(bio, bs);
244 * If we have front padding, adjust the bio pointer before freeing
246 p = bio;
247 if (bs->front_pad)
248 p -= bs->front_pad;
250 mempool_free(p, bs->bio_pool);
252 EXPORT_SYMBOL(bio_free);
254 void bio_init(struct bio *bio)
256 memset(bio, 0, sizeof(*bio));
257 bio->bi_flags = 1 << BIO_UPTODATE;
258 bio->bi_comp_cpu = -1;
259 atomic_set(&bio->bi_cnt, 1);
261 EXPORT_SYMBOL(bio_init);
264 * bio_alloc_bioset - allocate a bio for I/O
265 * @gfp_mask: the GFP_ mask given to the slab allocator
266 * @nr_iovecs: number of iovecs to pre-allocate
267 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
269 * Description:
270 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
271 * If %__GFP_WAIT is set then we will block on the internal pool waiting
272 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
273 * fall back to just using @kmalloc to allocate the required memory.
275 * Note that the caller must set ->bi_destructor on successful return
276 * of a bio, to do the appropriate freeing of the bio once the reference
277 * count drops to zero.
279 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
281 unsigned long idx = BIO_POOL_NONE;
282 struct bio_vec *bvl = NULL;
283 struct bio *bio;
284 void *p;
286 p = mempool_alloc(bs->bio_pool, gfp_mask);
287 if (unlikely(!p))
288 return NULL;
289 bio = p + bs->front_pad;
291 bio_init(bio);
293 if (unlikely(!nr_iovecs))
294 goto out_set;
296 if (nr_iovecs <= BIO_INLINE_VECS) {
297 bvl = bio->bi_inline_vecs;
298 nr_iovecs = BIO_INLINE_VECS;
299 } else {
300 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
301 if (unlikely(!bvl))
302 goto err_free;
304 nr_iovecs = bvec_nr_vecs(idx);
306 out_set:
307 bio->bi_flags |= idx << BIO_POOL_OFFSET;
308 bio->bi_max_vecs = nr_iovecs;
309 bio->bi_io_vec = bvl;
310 return bio;
312 err_free:
313 mempool_free(p, bs->bio_pool);
314 return NULL;
316 EXPORT_SYMBOL(bio_alloc_bioset);
318 static void bio_fs_destructor(struct bio *bio)
320 bio_free(bio, fs_bio_set);
324 * bio_alloc - allocate a new bio, memory pool backed
325 * @gfp_mask: allocation mask to use
326 * @nr_iovecs: number of iovecs
328 * bio_alloc will allocate a bio and associated bio_vec array that can hold
329 * at least @nr_iovecs entries. Allocations will be done from the
330 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
332 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
333 * a bio. This is due to the mempool guarantees. To make this work, callers
334 * must never allocate more than 1 bio at a time from this pool. Callers
335 * that need to allocate more than 1 bio must always submit the previously
336 * allocated bio for IO before attempting to allocate a new one. Failure to
337 * do so can cause livelocks under memory pressure.
339 * RETURNS:
340 * Pointer to new bio on success, NULL on failure.
342 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
344 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
346 if (bio)
347 bio->bi_destructor = bio_fs_destructor;
349 return bio;
351 EXPORT_SYMBOL(bio_alloc);
353 static void bio_kmalloc_destructor(struct bio *bio)
355 if (bio_integrity(bio))
356 bio_integrity_free(bio, fs_bio_set);
357 kfree(bio);
361 * bio_kmalloc - allocate a bio for I/O using kmalloc()
362 * @gfp_mask: the GFP_ mask given to the slab allocator
363 * @nr_iovecs: number of iovecs to pre-allocate
365 * Description:
366 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
367 * %__GFP_WAIT, the allocation is guaranteed to succeed.
370 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
372 struct bio *bio;
374 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
375 gfp_mask);
376 if (unlikely(!bio))
377 return NULL;
379 bio_init(bio);
380 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
381 bio->bi_max_vecs = nr_iovecs;
382 bio->bi_io_vec = bio->bi_inline_vecs;
383 bio->bi_destructor = bio_kmalloc_destructor;
385 return bio;
387 EXPORT_SYMBOL(bio_kmalloc);
389 void zero_fill_bio(struct bio *bio)
391 unsigned long flags;
392 struct bio_vec *bv;
393 int i;
395 bio_for_each_segment(bv, bio, i) {
396 char *data = bvec_kmap_irq(bv, &flags);
397 memset(data, 0, bv->bv_len);
398 flush_dcache_page(bv->bv_page);
399 bvec_kunmap_irq(data, &flags);
402 EXPORT_SYMBOL(zero_fill_bio);
405 * bio_put - release a reference to a bio
406 * @bio: bio to release reference to
408 * Description:
409 * Put a reference to a &struct bio, either one you have gotten with
410 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
412 void bio_put(struct bio *bio)
414 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
417 * last put frees it
419 if (atomic_dec_and_test(&bio->bi_cnt)) {
420 bio->bi_next = NULL;
421 bio->bi_destructor(bio);
424 EXPORT_SYMBOL(bio_put);
426 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
428 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
429 blk_recount_segments(q, bio);
431 return bio->bi_phys_segments;
433 EXPORT_SYMBOL(bio_phys_segments);
436 * __bio_clone - clone a bio
437 * @bio: destination bio
438 * @bio_src: bio to clone
440 * Clone a &bio. Caller will own the returned bio, but not
441 * the actual data it points to. Reference count of returned
442 * bio will be one.
444 void __bio_clone(struct bio *bio, struct bio *bio_src)
446 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
447 bio_src->bi_max_vecs * sizeof(struct bio_vec));
450 * most users will be overriding ->bi_bdev with a new target,
451 * so we don't set nor calculate new physical/hw segment counts here
453 bio->bi_sector = bio_src->bi_sector;
454 bio->bi_bdev = bio_src->bi_bdev;
455 bio->bi_flags |= 1 << BIO_CLONED;
456 bio->bi_rw = bio_src->bi_rw;
457 bio->bi_vcnt = bio_src->bi_vcnt;
458 bio->bi_size = bio_src->bi_size;
459 bio->bi_idx = bio_src->bi_idx;
461 EXPORT_SYMBOL(__bio_clone);
464 * bio_clone - clone a bio
465 * @bio: bio to clone
466 * @gfp_mask: allocation priority
468 * Like __bio_clone, only also allocates the returned bio
470 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
472 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
474 if (!b)
475 return NULL;
477 b->bi_destructor = bio_fs_destructor;
478 __bio_clone(b, bio);
480 if (bio_integrity(bio)) {
481 int ret;
483 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
485 if (ret < 0) {
486 bio_put(b);
487 return NULL;
491 return b;
493 EXPORT_SYMBOL(bio_clone);
496 * bio_get_nr_vecs - return approx number of vecs
497 * @bdev: I/O target
499 * Return the approximate number of pages we can send to this target.
500 * There's no guarantee that you will be able to fit this number of pages
501 * into a bio, it does not account for dynamic restrictions that vary
502 * on offset.
504 int bio_get_nr_vecs(struct block_device *bdev)
506 struct request_queue *q = bdev_get_queue(bdev);
507 int nr_pages;
509 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
510 if (nr_pages > queue_max_phys_segments(q))
511 nr_pages = queue_max_phys_segments(q);
512 if (nr_pages > queue_max_hw_segments(q))
513 nr_pages = queue_max_hw_segments(q);
515 return nr_pages;
517 EXPORT_SYMBOL(bio_get_nr_vecs);
519 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
520 *page, unsigned int len, unsigned int offset,
521 unsigned short max_sectors)
523 int retried_segments = 0;
524 struct bio_vec *bvec;
527 * cloned bio must not modify vec list
529 if (unlikely(bio_flagged(bio, BIO_CLONED)))
530 return 0;
532 if (((bio->bi_size + len) >> 9) > max_sectors)
533 return 0;
536 * For filesystems with a blocksize smaller than the pagesize
537 * we will often be called with the same page as last time and
538 * a consecutive offset. Optimize this special case.
540 if (bio->bi_vcnt > 0) {
541 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
543 if (page == prev->bv_page &&
544 offset == prev->bv_offset + prev->bv_len) {
545 unsigned int prev_bv_len = prev->bv_len;
546 prev->bv_len += len;
548 if (q->merge_bvec_fn) {
549 struct bvec_merge_data bvm = {
550 /* prev_bvec is already charged in
551 bi_size, discharge it in order to
552 simulate merging updated prev_bvec
553 as new bvec. */
554 .bi_bdev = bio->bi_bdev,
555 .bi_sector = bio->bi_sector,
556 .bi_size = bio->bi_size - prev_bv_len,
557 .bi_rw = bio->bi_rw,
560 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
561 prev->bv_len -= len;
562 return 0;
566 goto done;
570 if (bio->bi_vcnt >= bio->bi_max_vecs)
571 return 0;
574 * we might lose a segment or two here, but rather that than
575 * make this too complex.
578 while (bio->bi_phys_segments >= queue_max_phys_segments(q)
579 || bio->bi_phys_segments >= queue_max_hw_segments(q)) {
581 if (retried_segments)
582 return 0;
584 retried_segments = 1;
585 blk_recount_segments(q, bio);
589 * setup the new entry, we might clear it again later if we
590 * cannot add the page
592 bvec = &bio->bi_io_vec[bio->bi_vcnt];
593 bvec->bv_page = page;
594 bvec->bv_len = len;
595 bvec->bv_offset = offset;
598 * if queue has other restrictions (eg varying max sector size
599 * depending on offset), it can specify a merge_bvec_fn in the
600 * queue to get further control
602 if (q->merge_bvec_fn) {
603 struct bvec_merge_data bvm = {
604 .bi_bdev = bio->bi_bdev,
605 .bi_sector = bio->bi_sector,
606 .bi_size = bio->bi_size,
607 .bi_rw = bio->bi_rw,
611 * merge_bvec_fn() returns number of bytes it can accept
612 * at this offset
614 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
615 bvec->bv_page = NULL;
616 bvec->bv_len = 0;
617 bvec->bv_offset = 0;
618 return 0;
622 /* If we may be able to merge these biovecs, force a recount */
623 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
624 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
626 bio->bi_vcnt++;
627 bio->bi_phys_segments++;
628 done:
629 bio->bi_size += len;
630 return len;
634 * bio_add_pc_page - attempt to add page to bio
635 * @q: the target queue
636 * @bio: destination bio
637 * @page: page to add
638 * @len: vec entry length
639 * @offset: vec entry offset
641 * Attempt to add a page to the bio_vec maplist. This can fail for a
642 * number of reasons, such as the bio being full or target block
643 * device limitations. The target block device must allow bio's
644 * smaller than PAGE_SIZE, so it is always possible to add a single
645 * page to an empty bio. This should only be used by REQ_PC bios.
647 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
648 unsigned int len, unsigned int offset)
650 return __bio_add_page(q, bio, page, len, offset,
651 queue_max_hw_sectors(q));
653 EXPORT_SYMBOL(bio_add_pc_page);
656 * bio_add_page - attempt to add page to bio
657 * @bio: destination bio
658 * @page: page to add
659 * @len: vec entry length
660 * @offset: vec entry offset
662 * Attempt to add a page to the bio_vec maplist. This can fail for a
663 * number of reasons, such as the bio being full or target block
664 * device limitations. The target block device must allow bio's
665 * smaller than PAGE_SIZE, so it is always possible to add a single
666 * page to an empty bio.
668 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
669 unsigned int offset)
671 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
672 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
674 EXPORT_SYMBOL(bio_add_page);
676 struct bio_map_data {
677 struct bio_vec *iovecs;
678 struct sg_iovec *sgvecs;
679 int nr_sgvecs;
680 int is_our_pages;
683 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
684 struct sg_iovec *iov, int iov_count,
685 int is_our_pages)
687 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
688 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
689 bmd->nr_sgvecs = iov_count;
690 bmd->is_our_pages = is_our_pages;
691 bio->bi_private = bmd;
694 static void bio_free_map_data(struct bio_map_data *bmd)
696 kfree(bmd->iovecs);
697 kfree(bmd->sgvecs);
698 kfree(bmd);
701 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
702 gfp_t gfp_mask)
704 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
706 if (!bmd)
707 return NULL;
709 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
710 if (!bmd->iovecs) {
711 kfree(bmd);
712 return NULL;
715 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
716 if (bmd->sgvecs)
717 return bmd;
719 kfree(bmd->iovecs);
720 kfree(bmd);
721 return NULL;
724 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
725 struct sg_iovec *iov, int iov_count,
726 int to_user, int from_user, int do_free_page)
728 int ret = 0, i;
729 struct bio_vec *bvec;
730 int iov_idx = 0;
731 unsigned int iov_off = 0;
733 __bio_for_each_segment(bvec, bio, i, 0) {
734 char *bv_addr = page_address(bvec->bv_page);
735 unsigned int bv_len = iovecs[i].bv_len;
737 while (bv_len && iov_idx < iov_count) {
738 unsigned int bytes;
739 char __user *iov_addr;
741 bytes = min_t(unsigned int,
742 iov[iov_idx].iov_len - iov_off, bv_len);
743 iov_addr = iov[iov_idx].iov_base + iov_off;
745 if (!ret) {
746 if (to_user)
747 ret = copy_to_user(iov_addr, bv_addr,
748 bytes);
750 if (from_user)
751 ret = copy_from_user(bv_addr, iov_addr,
752 bytes);
754 if (ret)
755 ret = -EFAULT;
758 bv_len -= bytes;
759 bv_addr += bytes;
760 iov_addr += bytes;
761 iov_off += bytes;
763 if (iov[iov_idx].iov_len == iov_off) {
764 iov_idx++;
765 iov_off = 0;
769 if (do_free_page)
770 __free_page(bvec->bv_page);
773 return ret;
777 * bio_uncopy_user - finish previously mapped bio
778 * @bio: bio being terminated
780 * Free pages allocated from bio_copy_user() and write back data
781 * to user space in case of a read.
783 int bio_uncopy_user(struct bio *bio)
785 struct bio_map_data *bmd = bio->bi_private;
786 int ret = 0;
788 if (!bio_flagged(bio, BIO_NULL_MAPPED))
789 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
790 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
791 0, bmd->is_our_pages);
792 bio_free_map_data(bmd);
793 bio_put(bio);
794 return ret;
796 EXPORT_SYMBOL(bio_uncopy_user);
799 * bio_copy_user_iov - copy user data to bio
800 * @q: destination block queue
801 * @map_data: pointer to the rq_map_data holding pages (if necessary)
802 * @iov: the iovec.
803 * @iov_count: number of elements in the iovec
804 * @write_to_vm: bool indicating writing to pages or not
805 * @gfp_mask: memory allocation flags
807 * Prepares and returns a bio for indirect user io, bouncing data
808 * to/from kernel pages as necessary. Must be paired with
809 * call bio_uncopy_user() on io completion.
811 struct bio *bio_copy_user_iov(struct request_queue *q,
812 struct rq_map_data *map_data,
813 struct sg_iovec *iov, int iov_count,
814 int write_to_vm, gfp_t gfp_mask)
816 struct bio_map_data *bmd;
817 struct bio_vec *bvec;
818 struct page *page;
819 struct bio *bio;
820 int i, ret;
821 int nr_pages = 0;
822 unsigned int len = 0;
823 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
825 for (i = 0; i < iov_count; i++) {
826 unsigned long uaddr;
827 unsigned long end;
828 unsigned long start;
830 uaddr = (unsigned long)iov[i].iov_base;
831 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
832 start = uaddr >> PAGE_SHIFT;
834 nr_pages += end - start;
835 len += iov[i].iov_len;
838 if (offset)
839 nr_pages++;
841 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
842 if (!bmd)
843 return ERR_PTR(-ENOMEM);
845 ret = -ENOMEM;
846 bio = bio_kmalloc(gfp_mask, nr_pages);
847 if (!bio)
848 goto out_bmd;
850 bio->bi_rw |= (!write_to_vm << BIO_RW);
852 ret = 0;
854 if (map_data) {
855 nr_pages = 1 << map_data->page_order;
856 i = map_data->offset / PAGE_SIZE;
858 while (len) {
859 unsigned int bytes = PAGE_SIZE;
861 bytes -= offset;
863 if (bytes > len)
864 bytes = len;
866 if (map_data) {
867 if (i == map_data->nr_entries * nr_pages) {
868 ret = -ENOMEM;
869 break;
872 page = map_data->pages[i / nr_pages];
873 page += (i % nr_pages);
875 i++;
876 } else {
877 page = alloc_page(q->bounce_gfp | gfp_mask);
878 if (!page) {
879 ret = -ENOMEM;
880 break;
884 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
885 break;
887 len -= bytes;
888 offset = 0;
891 if (ret)
892 goto cleanup;
895 * success
897 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
898 (map_data && map_data->from_user)) {
899 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
900 if (ret)
901 goto cleanup;
904 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
905 return bio;
906 cleanup:
907 if (!map_data)
908 bio_for_each_segment(bvec, bio, i)
909 __free_page(bvec->bv_page);
911 bio_put(bio);
912 out_bmd:
913 bio_free_map_data(bmd);
914 return ERR_PTR(ret);
918 * bio_copy_user - copy user data to bio
919 * @q: destination block queue
920 * @map_data: pointer to the rq_map_data holding pages (if necessary)
921 * @uaddr: start of user address
922 * @len: length in bytes
923 * @write_to_vm: bool indicating writing to pages or not
924 * @gfp_mask: memory allocation flags
926 * Prepares and returns a bio for indirect user io, bouncing data
927 * to/from kernel pages as necessary. Must be paired with
928 * call bio_uncopy_user() on io completion.
930 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
931 unsigned long uaddr, unsigned int len,
932 int write_to_vm, gfp_t gfp_mask)
934 struct sg_iovec iov;
936 iov.iov_base = (void __user *)uaddr;
937 iov.iov_len = len;
939 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
941 EXPORT_SYMBOL(bio_copy_user);
943 static struct bio *__bio_map_user_iov(struct request_queue *q,
944 struct block_device *bdev,
945 struct sg_iovec *iov, int iov_count,
946 int write_to_vm, gfp_t gfp_mask)
948 int i, j;
949 int nr_pages = 0;
950 struct page **pages;
951 struct bio *bio;
952 int cur_page = 0;
953 int ret, offset;
955 for (i = 0; i < iov_count; i++) {
956 unsigned long uaddr = (unsigned long)iov[i].iov_base;
957 unsigned long len = iov[i].iov_len;
958 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
959 unsigned long start = uaddr >> PAGE_SHIFT;
961 nr_pages += end - start;
963 * buffer must be aligned to at least hardsector size for now
965 if (uaddr & queue_dma_alignment(q))
966 return ERR_PTR(-EINVAL);
969 if (!nr_pages)
970 return ERR_PTR(-EINVAL);
972 bio = bio_kmalloc(gfp_mask, nr_pages);
973 if (!bio)
974 return ERR_PTR(-ENOMEM);
976 ret = -ENOMEM;
977 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
978 if (!pages)
979 goto out;
981 for (i = 0; i < iov_count; i++) {
982 unsigned long uaddr = (unsigned long)iov[i].iov_base;
983 unsigned long len = iov[i].iov_len;
984 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
985 unsigned long start = uaddr >> PAGE_SHIFT;
986 const int local_nr_pages = end - start;
987 const int page_limit = cur_page + local_nr_pages;
989 ret = get_user_pages_fast(uaddr, local_nr_pages,
990 write_to_vm, &pages[cur_page]);
991 if (ret < local_nr_pages) {
992 ret = -EFAULT;
993 goto out_unmap;
996 offset = uaddr & ~PAGE_MASK;
997 for (j = cur_page; j < page_limit; j++) {
998 unsigned int bytes = PAGE_SIZE - offset;
1000 if (len <= 0)
1001 break;
1003 if (bytes > len)
1004 bytes = len;
1007 * sorry...
1009 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1010 bytes)
1011 break;
1013 len -= bytes;
1014 offset = 0;
1017 cur_page = j;
1019 * release the pages we didn't map into the bio, if any
1021 while (j < page_limit)
1022 page_cache_release(pages[j++]);
1025 kfree(pages);
1028 * set data direction, and check if mapped pages need bouncing
1030 if (!write_to_vm)
1031 bio->bi_rw |= (1 << BIO_RW);
1033 bio->bi_bdev = bdev;
1034 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1035 return bio;
1037 out_unmap:
1038 for (i = 0; i < nr_pages; i++) {
1039 if(!pages[i])
1040 break;
1041 page_cache_release(pages[i]);
1043 out:
1044 kfree(pages);
1045 bio_put(bio);
1046 return ERR_PTR(ret);
1050 * bio_map_user - map user address into bio
1051 * @q: the struct request_queue for the bio
1052 * @bdev: destination block device
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 * Map the user space address into a bio suitable for io to a block
1059 * device. Returns an error pointer in case of error.
1061 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1062 unsigned long uaddr, unsigned int len, int write_to_vm,
1063 gfp_t gfp_mask)
1065 struct sg_iovec iov;
1067 iov.iov_base = (void __user *)uaddr;
1068 iov.iov_len = len;
1070 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1072 EXPORT_SYMBOL(bio_map_user);
1075 * bio_map_user_iov - map user sg_iovec table into bio
1076 * @q: the struct request_queue for the bio
1077 * @bdev: destination block device
1078 * @iov: the iovec.
1079 * @iov_count: number of elements in the iovec
1080 * @write_to_vm: bool indicating writing to pages or not
1081 * @gfp_mask: memory allocation flags
1083 * Map the user space address into a bio suitable for io to a block
1084 * device. Returns an error pointer in case of error.
1086 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1087 struct sg_iovec *iov, int iov_count,
1088 int write_to_vm, gfp_t gfp_mask)
1090 struct bio *bio;
1092 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1093 gfp_mask);
1094 if (IS_ERR(bio))
1095 return bio;
1098 * subtle -- if __bio_map_user() ended up bouncing a bio,
1099 * it would normally disappear when its bi_end_io is run.
1100 * however, we need it for the unmap, so grab an extra
1101 * reference to it
1103 bio_get(bio);
1105 return bio;
1108 static void __bio_unmap_user(struct bio *bio)
1110 struct bio_vec *bvec;
1111 int i;
1114 * make sure we dirty pages we wrote to
1116 __bio_for_each_segment(bvec, bio, i, 0) {
1117 if (bio_data_dir(bio) == READ)
1118 set_page_dirty_lock(bvec->bv_page);
1120 page_cache_release(bvec->bv_page);
1123 bio_put(bio);
1127 * bio_unmap_user - unmap a bio
1128 * @bio: the bio being unmapped
1130 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1131 * a process context.
1133 * bio_unmap_user() may sleep.
1135 void bio_unmap_user(struct bio *bio)
1137 __bio_unmap_user(bio);
1138 bio_put(bio);
1140 EXPORT_SYMBOL(bio_unmap_user);
1142 static void bio_map_kern_endio(struct bio *bio, int err)
1144 bio_put(bio);
1147 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1148 unsigned int len, gfp_t gfp_mask)
1150 unsigned long kaddr = (unsigned long)data;
1151 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1152 unsigned long start = kaddr >> PAGE_SHIFT;
1153 const int nr_pages = end - start;
1154 int offset, i;
1155 struct bio *bio;
1157 bio = bio_kmalloc(gfp_mask, nr_pages);
1158 if (!bio)
1159 return ERR_PTR(-ENOMEM);
1161 offset = offset_in_page(kaddr);
1162 for (i = 0; i < nr_pages; i++) {
1163 unsigned int bytes = PAGE_SIZE - offset;
1165 if (len <= 0)
1166 break;
1168 if (bytes > len)
1169 bytes = len;
1171 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1172 offset) < bytes)
1173 break;
1175 data += bytes;
1176 len -= bytes;
1177 offset = 0;
1180 bio->bi_end_io = bio_map_kern_endio;
1181 return bio;
1185 * bio_map_kern - map kernel address into bio
1186 * @q: the struct request_queue for the bio
1187 * @data: pointer to buffer to map
1188 * @len: length in bytes
1189 * @gfp_mask: allocation flags for bio allocation
1191 * Map the kernel address into a bio suitable for io to a block
1192 * device. Returns an error pointer in case of error.
1194 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1195 gfp_t gfp_mask)
1197 struct bio *bio;
1199 bio = __bio_map_kern(q, data, len, gfp_mask);
1200 if (IS_ERR(bio))
1201 return bio;
1203 if (bio->bi_size == len)
1204 return bio;
1207 * Don't support partial mappings.
1209 bio_put(bio);
1210 return ERR_PTR(-EINVAL);
1212 EXPORT_SYMBOL(bio_map_kern);
1214 static void bio_copy_kern_endio(struct bio *bio, int err)
1216 struct bio_vec *bvec;
1217 const int read = bio_data_dir(bio) == READ;
1218 struct bio_map_data *bmd = bio->bi_private;
1219 int i;
1220 char *p = bmd->sgvecs[0].iov_base;
1222 __bio_for_each_segment(bvec, bio, i, 0) {
1223 char *addr = page_address(bvec->bv_page);
1224 int len = bmd->iovecs[i].bv_len;
1226 if (read)
1227 memcpy(p, addr, len);
1229 __free_page(bvec->bv_page);
1230 p += len;
1233 bio_free_map_data(bmd);
1234 bio_put(bio);
1238 * bio_copy_kern - copy kernel address into bio
1239 * @q: the struct request_queue for the bio
1240 * @data: pointer to buffer to copy
1241 * @len: length in bytes
1242 * @gfp_mask: allocation flags for bio and page allocation
1243 * @reading: data direction is READ
1245 * copy the kernel address into a bio suitable for io to a block
1246 * device. Returns an error pointer in case of error.
1248 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1249 gfp_t gfp_mask, int reading)
1251 struct bio *bio;
1252 struct bio_vec *bvec;
1253 int i;
1255 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1256 if (IS_ERR(bio))
1257 return bio;
1259 if (!reading) {
1260 void *p = data;
1262 bio_for_each_segment(bvec, bio, i) {
1263 char *addr = page_address(bvec->bv_page);
1265 memcpy(addr, p, bvec->bv_len);
1266 p += bvec->bv_len;
1270 bio->bi_end_io = bio_copy_kern_endio;
1272 return bio;
1274 EXPORT_SYMBOL(bio_copy_kern);
1277 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1278 * for performing direct-IO in BIOs.
1280 * The problem is that we cannot run set_page_dirty() from interrupt context
1281 * because the required locks are not interrupt-safe. So what we can do is to
1282 * mark the pages dirty _before_ performing IO. And in interrupt context,
1283 * check that the pages are still dirty. If so, fine. If not, redirty them
1284 * in process context.
1286 * We special-case compound pages here: normally this means reads into hugetlb
1287 * pages. The logic in here doesn't really work right for compound pages
1288 * because the VM does not uniformly chase down the head page in all cases.
1289 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1290 * handle them at all. So we skip compound pages here at an early stage.
1292 * Note that this code is very hard to test under normal circumstances because
1293 * direct-io pins the pages with get_user_pages(). This makes
1294 * is_page_cache_freeable return false, and the VM will not clean the pages.
1295 * But other code (eg, pdflush) could clean the pages if they are mapped
1296 * pagecache.
1298 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1299 * deferred bio dirtying paths.
1303 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1305 void bio_set_pages_dirty(struct bio *bio)
1307 struct bio_vec *bvec = bio->bi_io_vec;
1308 int i;
1310 for (i = 0; i < bio->bi_vcnt; i++) {
1311 struct page *page = bvec[i].bv_page;
1313 if (page && !PageCompound(page))
1314 set_page_dirty_lock(page);
1318 static void bio_release_pages(struct bio *bio)
1320 struct bio_vec *bvec = bio->bi_io_vec;
1321 int i;
1323 for (i = 0; i < bio->bi_vcnt; i++) {
1324 struct page *page = bvec[i].bv_page;
1326 if (page)
1327 put_page(page);
1332 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1333 * If they are, then fine. If, however, some pages are clean then they must
1334 * have been written out during the direct-IO read. So we take another ref on
1335 * the BIO and the offending pages and re-dirty the pages in process context.
1337 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1338 * here on. It will run one page_cache_release() against each page and will
1339 * run one bio_put() against the BIO.
1342 static void bio_dirty_fn(struct work_struct *work);
1344 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1345 static DEFINE_SPINLOCK(bio_dirty_lock);
1346 static struct bio *bio_dirty_list;
1349 * This runs in process context
1351 static void bio_dirty_fn(struct work_struct *work)
1353 unsigned long flags;
1354 struct bio *bio;
1356 spin_lock_irqsave(&bio_dirty_lock, flags);
1357 bio = bio_dirty_list;
1358 bio_dirty_list = NULL;
1359 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1361 while (bio) {
1362 struct bio *next = bio->bi_private;
1364 bio_set_pages_dirty(bio);
1365 bio_release_pages(bio);
1366 bio_put(bio);
1367 bio = next;
1371 void bio_check_pages_dirty(struct bio *bio)
1373 struct bio_vec *bvec = bio->bi_io_vec;
1374 int nr_clean_pages = 0;
1375 int i;
1377 for (i = 0; i < bio->bi_vcnt; i++) {
1378 struct page *page = bvec[i].bv_page;
1380 if (PageDirty(page) || PageCompound(page)) {
1381 page_cache_release(page);
1382 bvec[i].bv_page = NULL;
1383 } else {
1384 nr_clean_pages++;
1388 if (nr_clean_pages) {
1389 unsigned long flags;
1391 spin_lock_irqsave(&bio_dirty_lock, flags);
1392 bio->bi_private = bio_dirty_list;
1393 bio_dirty_list = bio;
1394 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1395 schedule_work(&bio_dirty_work);
1396 } else {
1397 bio_put(bio);
1401 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1402 void bio_flush_dcache_pages(struct bio *bi)
1404 int i;
1405 struct bio_vec *bvec;
1407 bio_for_each_segment(bvec, bi, i)
1408 flush_dcache_page(bvec->bv_page);
1410 EXPORT_SYMBOL(bio_flush_dcache_pages);
1411 #endif
1414 * bio_endio - end I/O on a bio
1415 * @bio: bio
1416 * @error: error, if any
1418 * Description:
1419 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1420 * preferred way to end I/O on a bio, it takes care of clearing
1421 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1422 * established -Exxxx (-EIO, for instance) error values in case
1423 * something went wrong. Noone should call bi_end_io() directly on a
1424 * bio unless they own it and thus know that it has an end_io
1425 * function.
1427 void bio_endio(struct bio *bio, int error)
1429 if (error)
1430 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1431 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1432 error = -EIO;
1434 if (bio->bi_end_io)
1435 bio->bi_end_io(bio, error);
1437 EXPORT_SYMBOL(bio_endio);
1439 void bio_pair_release(struct bio_pair *bp)
1441 if (atomic_dec_and_test(&bp->cnt)) {
1442 struct bio *master = bp->bio1.bi_private;
1444 bio_endio(master, bp->error);
1445 mempool_free(bp, bp->bio2.bi_private);
1448 EXPORT_SYMBOL(bio_pair_release);
1450 static void bio_pair_end_1(struct bio *bi, int err)
1452 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1454 if (err)
1455 bp->error = err;
1457 bio_pair_release(bp);
1460 static void bio_pair_end_2(struct bio *bi, int err)
1462 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1464 if (err)
1465 bp->error = err;
1467 bio_pair_release(bp);
1471 * split a bio - only worry about a bio with a single page in its iovec
1473 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1475 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1477 if (!bp)
1478 return bp;
1480 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1481 bi->bi_sector + first_sectors);
1483 BUG_ON(bi->bi_vcnt != 1);
1484 BUG_ON(bi->bi_idx != 0);
1485 atomic_set(&bp->cnt, 3);
1486 bp->error = 0;
1487 bp->bio1 = *bi;
1488 bp->bio2 = *bi;
1489 bp->bio2.bi_sector += first_sectors;
1490 bp->bio2.bi_size -= first_sectors << 9;
1491 bp->bio1.bi_size = first_sectors << 9;
1493 bp->bv1 = bi->bi_io_vec[0];
1494 bp->bv2 = bi->bi_io_vec[0];
1495 bp->bv2.bv_offset += first_sectors << 9;
1496 bp->bv2.bv_len -= first_sectors << 9;
1497 bp->bv1.bv_len = first_sectors << 9;
1499 bp->bio1.bi_io_vec = &bp->bv1;
1500 bp->bio2.bi_io_vec = &bp->bv2;
1502 bp->bio1.bi_max_vecs = 1;
1503 bp->bio2.bi_max_vecs = 1;
1505 bp->bio1.bi_end_io = bio_pair_end_1;
1506 bp->bio2.bi_end_io = bio_pair_end_2;
1508 bp->bio1.bi_private = bi;
1509 bp->bio2.bi_private = bio_split_pool;
1511 if (bio_integrity(bi))
1512 bio_integrity_split(bi, bp, first_sectors);
1514 return bp;
1516 EXPORT_SYMBOL(bio_split);
1519 * bio_sector_offset - Find hardware sector offset in bio
1520 * @bio: bio to inspect
1521 * @index: bio_vec index
1522 * @offset: offset in bv_page
1524 * Return the number of hardware sectors between beginning of bio
1525 * and an end point indicated by a bio_vec index and an offset
1526 * within that vector's page.
1528 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1529 unsigned int offset)
1531 unsigned int sector_sz;
1532 struct bio_vec *bv;
1533 sector_t sectors;
1534 int i;
1536 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1537 sectors = 0;
1539 if (index >= bio->bi_idx)
1540 index = bio->bi_vcnt - 1;
1542 __bio_for_each_segment(bv, bio, i, 0) {
1543 if (i == index) {
1544 if (offset > bv->bv_offset)
1545 sectors += (offset - bv->bv_offset) / sector_sz;
1546 break;
1549 sectors += bv->bv_len / sector_sz;
1552 return sectors;
1554 EXPORT_SYMBOL(bio_sector_offset);
1557 * create memory pools for biovec's in a bio_set.
1558 * use the global biovec slabs created for general use.
1560 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1562 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1564 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1565 if (!bs->bvec_pool)
1566 return -ENOMEM;
1568 return 0;
1571 static void biovec_free_pools(struct bio_set *bs)
1573 mempool_destroy(bs->bvec_pool);
1576 void bioset_free(struct bio_set *bs)
1578 if (bs->bio_pool)
1579 mempool_destroy(bs->bio_pool);
1581 bioset_integrity_free(bs);
1582 biovec_free_pools(bs);
1583 bio_put_slab(bs);
1585 kfree(bs);
1587 EXPORT_SYMBOL(bioset_free);
1590 * bioset_create - Create a bio_set
1591 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1592 * @front_pad: Number of bytes to allocate in front of the returned bio
1594 * Description:
1595 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1596 * to ask for a number of bytes to be allocated in front of the bio.
1597 * Front pad allocation is useful for embedding the bio inside
1598 * another structure, to avoid allocating extra data to go with the bio.
1599 * Note that the bio must be embedded at the END of that structure always,
1600 * or things will break badly.
1602 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1604 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1605 struct bio_set *bs;
1607 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1608 if (!bs)
1609 return NULL;
1611 bs->front_pad = front_pad;
1613 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1614 if (!bs->bio_slab) {
1615 kfree(bs);
1616 return NULL;
1619 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1620 if (!bs->bio_pool)
1621 goto bad;
1623 if (bioset_integrity_create(bs, pool_size))
1624 goto bad;
1626 if (!biovec_create_pools(bs, pool_size))
1627 return bs;
1629 bad:
1630 bioset_free(bs);
1631 return NULL;
1633 EXPORT_SYMBOL(bioset_create);
1635 static void __init biovec_init_slabs(void)
1637 int i;
1639 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1640 int size;
1641 struct biovec_slab *bvs = bvec_slabs + i;
1643 #ifndef CONFIG_BLK_DEV_INTEGRITY
1644 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1645 bvs->slab = NULL;
1646 continue;
1648 #endif
1650 size = bvs->nr_vecs * sizeof(struct bio_vec);
1651 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1652 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1656 static int __init init_bio(void)
1658 bio_slab_max = 2;
1659 bio_slab_nr = 0;
1660 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1661 if (!bio_slabs)
1662 panic("bio: can't allocate bios\n");
1664 bio_integrity_init();
1665 biovec_init_slabs();
1667 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1668 if (!fs_bio_set)
1669 panic("bio: can't allocate bios\n");
1671 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1672 sizeof(struct bio_pair));
1673 if (!bio_split_pool)
1674 panic("bio: can't create split pool\n");
1676 return 0;
1678 subsys_initcall(init_bio);