Linux 2.6.36-rc6
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / bio.c
blob8abb2dfb2e7c8b006f9c80b716808a1ff07a0189
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.
269 * Description:
270 * bio_alloc_bioset will 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.
274 * Note that the caller must set ->bi_destructor on successful return
275 * of a bio, to do the appropriate freeing of the bio once the reference
276 * count drops to zero.
278 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
280 unsigned long idx = BIO_POOL_NONE;
281 struct bio_vec *bvl = NULL;
282 struct bio *bio;
283 void *p;
285 p = mempool_alloc(bs->bio_pool, gfp_mask);
286 if (unlikely(!p))
287 return NULL;
288 bio = p + bs->front_pad;
290 bio_init(bio);
292 if (unlikely(!nr_iovecs))
293 goto out_set;
295 if (nr_iovecs <= BIO_INLINE_VECS) {
296 bvl = bio->bi_inline_vecs;
297 nr_iovecs = BIO_INLINE_VECS;
298 } else {
299 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
300 if (unlikely(!bvl))
301 goto err_free;
303 nr_iovecs = bvec_nr_vecs(idx);
305 out_set:
306 bio->bi_flags |= idx << BIO_POOL_OFFSET;
307 bio->bi_max_vecs = nr_iovecs;
308 bio->bi_io_vec = bvl;
309 return bio;
311 err_free:
312 mempool_free(p, bs->bio_pool);
313 return NULL;
315 EXPORT_SYMBOL(bio_alloc_bioset);
317 static void bio_fs_destructor(struct bio *bio)
319 bio_free(bio, fs_bio_set);
323 * bio_alloc - allocate a new bio, memory pool backed
324 * @gfp_mask: allocation mask to use
325 * @nr_iovecs: number of iovecs
327 * bio_alloc will allocate a bio and associated bio_vec array that can hold
328 * at least @nr_iovecs entries. Allocations will be done from the
329 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
331 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
332 * a bio. This is due to the mempool guarantees. To make this work, callers
333 * must never allocate more than 1 bio at a time from this pool. Callers
334 * that need to allocate more than 1 bio must always submit the previously
335 * allocated bio for IO before attempting to allocate a new one. Failure to
336 * do so can cause livelocks under memory pressure.
338 * RETURNS:
339 * Pointer to new bio on success, NULL on failure.
341 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
343 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
345 if (bio)
346 bio->bi_destructor = bio_fs_destructor;
348 return bio;
350 EXPORT_SYMBOL(bio_alloc);
352 static void bio_kmalloc_destructor(struct bio *bio)
354 if (bio_integrity(bio))
355 bio_integrity_free(bio, fs_bio_set);
356 kfree(bio);
360 * bio_kmalloc - allocate a bio for I/O using kmalloc()
361 * @gfp_mask: the GFP_ mask given to the slab allocator
362 * @nr_iovecs: number of iovecs to pre-allocate
364 * Description:
365 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
366 * %__GFP_WAIT, the allocation is guaranteed to succeed.
369 struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
371 struct bio *bio;
373 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
374 gfp_mask);
375 if (unlikely(!bio))
376 return NULL;
378 bio_init(bio);
379 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
380 bio->bi_max_vecs = nr_iovecs;
381 bio->bi_io_vec = bio->bi_inline_vecs;
382 bio->bi_destructor = bio_kmalloc_destructor;
384 return bio;
386 EXPORT_SYMBOL(bio_kmalloc);
388 void zero_fill_bio(struct bio *bio)
390 unsigned long flags;
391 struct bio_vec *bv;
392 int i;
394 bio_for_each_segment(bv, bio, i) {
395 char *data = bvec_kmap_irq(bv, &flags);
396 memset(data, 0, bv->bv_len);
397 flush_dcache_page(bv->bv_page);
398 bvec_kunmap_irq(data, &flags);
401 EXPORT_SYMBOL(zero_fill_bio);
404 * bio_put - release a reference to a bio
405 * @bio: bio to release reference to
407 * Description:
408 * Put a reference to a &struct bio, either one you have gotten with
409 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
411 void bio_put(struct bio *bio)
413 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
416 * last put frees it
418 if (atomic_dec_and_test(&bio->bi_cnt)) {
419 bio->bi_next = NULL;
420 bio->bi_destructor(bio);
423 EXPORT_SYMBOL(bio_put);
425 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
427 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
428 blk_recount_segments(q, bio);
430 return bio->bi_phys_segments;
432 EXPORT_SYMBOL(bio_phys_segments);
435 * __bio_clone - clone a bio
436 * @bio: destination bio
437 * @bio_src: bio to clone
439 * Clone a &bio. Caller will own the returned bio, but not
440 * the actual data it points to. Reference count of returned
441 * bio will be one.
443 void __bio_clone(struct bio *bio, struct bio *bio_src)
445 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
446 bio_src->bi_max_vecs * sizeof(struct bio_vec));
449 * most users will be overriding ->bi_bdev with a new target,
450 * so we don't set nor calculate new physical/hw segment counts here
452 bio->bi_sector = bio_src->bi_sector;
453 bio->bi_bdev = bio_src->bi_bdev;
454 bio->bi_flags |= 1 << BIO_CLONED;
455 bio->bi_rw = bio_src->bi_rw;
456 bio->bi_vcnt = bio_src->bi_vcnt;
457 bio->bi_size = bio_src->bi_size;
458 bio->bi_idx = bio_src->bi_idx;
460 EXPORT_SYMBOL(__bio_clone);
463 * bio_clone - clone a bio
464 * @bio: bio to clone
465 * @gfp_mask: allocation priority
467 * Like __bio_clone, only also allocates the returned bio
469 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
471 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
473 if (!b)
474 return NULL;
476 b->bi_destructor = bio_fs_destructor;
477 __bio_clone(b, bio);
479 if (bio_integrity(bio)) {
480 int ret;
482 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
484 if (ret < 0) {
485 bio_put(b);
486 return NULL;
490 return b;
492 EXPORT_SYMBOL(bio_clone);
495 * bio_get_nr_vecs - return approx number of vecs
496 * @bdev: I/O target
498 * Return the approximate number of pages we can send to this target.
499 * There's no guarantee that you will be able to fit this number of pages
500 * into a bio, it does not account for dynamic restrictions that vary
501 * on offset.
503 int bio_get_nr_vecs(struct block_device *bdev)
505 struct request_queue *q = bdev_get_queue(bdev);
506 int nr_pages;
508 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
509 if (nr_pages > queue_max_segments(q))
510 nr_pages = queue_max_segments(q);
512 return nr_pages;
514 EXPORT_SYMBOL(bio_get_nr_vecs);
516 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
517 *page, unsigned int len, unsigned int offset,
518 unsigned short max_sectors)
520 int retried_segments = 0;
521 struct bio_vec *bvec;
524 * cloned bio must not modify vec list
526 if (unlikely(bio_flagged(bio, BIO_CLONED)))
527 return 0;
529 if (((bio->bi_size + len) >> 9) > max_sectors)
530 return 0;
533 * For filesystems with a blocksize smaller than the pagesize
534 * we will often be called with the same page as last time and
535 * a consecutive offset. Optimize this special case.
537 if (bio->bi_vcnt > 0) {
538 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
540 if (page == prev->bv_page &&
541 offset == prev->bv_offset + prev->bv_len) {
542 unsigned int prev_bv_len = prev->bv_len;
543 prev->bv_len += len;
545 if (q->merge_bvec_fn) {
546 struct bvec_merge_data bvm = {
547 /* prev_bvec is already charged in
548 bi_size, discharge it in order to
549 simulate merging updated prev_bvec
550 as new bvec. */
551 .bi_bdev = bio->bi_bdev,
552 .bi_sector = bio->bi_sector,
553 .bi_size = bio->bi_size - prev_bv_len,
554 .bi_rw = bio->bi_rw,
557 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
558 prev->bv_len -= len;
559 return 0;
563 goto done;
567 if (bio->bi_vcnt >= bio->bi_max_vecs)
568 return 0;
571 * we might lose a segment or two here, but rather that than
572 * make this too complex.
575 while (bio->bi_phys_segments >= queue_max_segments(q)) {
577 if (retried_segments)
578 return 0;
580 retried_segments = 1;
581 blk_recount_segments(q, bio);
585 * setup the new entry, we might clear it again later if we
586 * cannot add the page
588 bvec = &bio->bi_io_vec[bio->bi_vcnt];
589 bvec->bv_page = page;
590 bvec->bv_len = len;
591 bvec->bv_offset = offset;
594 * if queue has other restrictions (eg varying max sector size
595 * depending on offset), it can specify a merge_bvec_fn in the
596 * queue to get further control
598 if (q->merge_bvec_fn) {
599 struct bvec_merge_data bvm = {
600 .bi_bdev = bio->bi_bdev,
601 .bi_sector = bio->bi_sector,
602 .bi_size = bio->bi_size,
603 .bi_rw = bio->bi_rw,
607 * merge_bvec_fn() returns number of bytes it can accept
608 * at this offset
610 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
611 bvec->bv_page = NULL;
612 bvec->bv_len = 0;
613 bvec->bv_offset = 0;
614 return 0;
618 /* If we may be able to merge these biovecs, force a recount */
619 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
620 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
622 bio->bi_vcnt++;
623 bio->bi_phys_segments++;
624 done:
625 bio->bi_size += len;
626 return len;
630 * bio_add_pc_page - attempt to add page to bio
631 * @q: the target queue
632 * @bio: destination bio
633 * @page: page to add
634 * @len: vec entry length
635 * @offset: vec entry offset
637 * Attempt to add a page to the bio_vec maplist. This can fail for a
638 * number of reasons, such as the bio being full or target block
639 * device limitations. The target block device must allow bio's
640 * smaller than PAGE_SIZE, so it is always possible to add a single
641 * page to an empty bio. This should only be used by REQ_PC bios.
643 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
644 unsigned int len, unsigned int offset)
646 return __bio_add_page(q, bio, page, len, offset,
647 queue_max_hw_sectors(q));
649 EXPORT_SYMBOL(bio_add_pc_page);
652 * bio_add_page - attempt to add page to bio
653 * @bio: destination bio
654 * @page: page to add
655 * @len: vec entry length
656 * @offset: vec entry offset
658 * Attempt to add a page to the bio_vec maplist. This can fail for a
659 * number of reasons, such as the bio being full or target block
660 * device limitations. The target block device must allow bio's
661 * smaller than PAGE_SIZE, so it is always possible to add a single
662 * page to an empty bio.
664 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
665 unsigned int offset)
667 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
668 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
670 EXPORT_SYMBOL(bio_add_page);
672 struct bio_map_data {
673 struct bio_vec *iovecs;
674 struct sg_iovec *sgvecs;
675 int nr_sgvecs;
676 int is_our_pages;
679 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
680 struct sg_iovec *iov, int iov_count,
681 int is_our_pages)
683 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
684 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
685 bmd->nr_sgvecs = iov_count;
686 bmd->is_our_pages = is_our_pages;
687 bio->bi_private = bmd;
690 static void bio_free_map_data(struct bio_map_data *bmd)
692 kfree(bmd->iovecs);
693 kfree(bmd->sgvecs);
694 kfree(bmd);
697 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
698 gfp_t gfp_mask)
700 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
702 if (!bmd)
703 return NULL;
705 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
706 if (!bmd->iovecs) {
707 kfree(bmd);
708 return NULL;
711 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
712 if (bmd->sgvecs)
713 return bmd;
715 kfree(bmd->iovecs);
716 kfree(bmd);
717 return NULL;
720 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
721 struct sg_iovec *iov, int iov_count,
722 int to_user, int from_user, int do_free_page)
724 int ret = 0, i;
725 struct bio_vec *bvec;
726 int iov_idx = 0;
727 unsigned int iov_off = 0;
729 __bio_for_each_segment(bvec, bio, i, 0) {
730 char *bv_addr = page_address(bvec->bv_page);
731 unsigned int bv_len = iovecs[i].bv_len;
733 while (bv_len && iov_idx < iov_count) {
734 unsigned int bytes;
735 char __user *iov_addr;
737 bytes = min_t(unsigned int,
738 iov[iov_idx].iov_len - iov_off, bv_len);
739 iov_addr = iov[iov_idx].iov_base + iov_off;
741 if (!ret) {
742 if (to_user)
743 ret = copy_to_user(iov_addr, bv_addr,
744 bytes);
746 if (from_user)
747 ret = copy_from_user(bv_addr, iov_addr,
748 bytes);
750 if (ret)
751 ret = -EFAULT;
754 bv_len -= bytes;
755 bv_addr += bytes;
756 iov_addr += bytes;
757 iov_off += bytes;
759 if (iov[iov_idx].iov_len == iov_off) {
760 iov_idx++;
761 iov_off = 0;
765 if (do_free_page)
766 __free_page(bvec->bv_page);
769 return ret;
773 * bio_uncopy_user - finish previously mapped bio
774 * @bio: bio being terminated
776 * Free pages allocated from bio_copy_user() and write back data
777 * to user space in case of a read.
779 int bio_uncopy_user(struct bio *bio)
781 struct bio_map_data *bmd = bio->bi_private;
782 int ret = 0;
784 if (!bio_flagged(bio, BIO_NULL_MAPPED))
785 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
786 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
787 0, bmd->is_our_pages);
788 bio_free_map_data(bmd);
789 bio_put(bio);
790 return ret;
792 EXPORT_SYMBOL(bio_uncopy_user);
795 * bio_copy_user_iov - copy user data to bio
796 * @q: destination block queue
797 * @map_data: pointer to the rq_map_data holding pages (if necessary)
798 * @iov: the iovec.
799 * @iov_count: number of elements in the iovec
800 * @write_to_vm: bool indicating writing to pages or not
801 * @gfp_mask: memory allocation flags
803 * Prepares and returns a bio for indirect user io, bouncing data
804 * to/from kernel pages as necessary. Must be paired with
805 * call bio_uncopy_user() on io completion.
807 struct bio *bio_copy_user_iov(struct request_queue *q,
808 struct rq_map_data *map_data,
809 struct sg_iovec *iov, int iov_count,
810 int write_to_vm, gfp_t gfp_mask)
812 struct bio_map_data *bmd;
813 struct bio_vec *bvec;
814 struct page *page;
815 struct bio *bio;
816 int i, ret;
817 int nr_pages = 0;
818 unsigned int len = 0;
819 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
821 for (i = 0; i < iov_count; i++) {
822 unsigned long uaddr;
823 unsigned long end;
824 unsigned long start;
826 uaddr = (unsigned long)iov[i].iov_base;
827 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
828 start = uaddr >> PAGE_SHIFT;
830 nr_pages += end - start;
831 len += iov[i].iov_len;
834 if (offset)
835 nr_pages++;
837 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
838 if (!bmd)
839 return ERR_PTR(-ENOMEM);
841 ret = -ENOMEM;
842 bio = bio_kmalloc(gfp_mask, nr_pages);
843 if (!bio)
844 goto out_bmd;
846 if (!write_to_vm)
847 bio->bi_rw |= REQ_WRITE;
849 ret = 0;
851 if (map_data) {
852 nr_pages = 1 << map_data->page_order;
853 i = map_data->offset / PAGE_SIZE;
855 while (len) {
856 unsigned int bytes = PAGE_SIZE;
858 bytes -= offset;
860 if (bytes > len)
861 bytes = len;
863 if (map_data) {
864 if (i == map_data->nr_entries * nr_pages) {
865 ret = -ENOMEM;
866 break;
869 page = map_data->pages[i / nr_pages];
870 page += (i % nr_pages);
872 i++;
873 } else {
874 page = alloc_page(q->bounce_gfp | gfp_mask);
875 if (!page) {
876 ret = -ENOMEM;
877 break;
881 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
882 break;
884 len -= bytes;
885 offset = 0;
888 if (ret)
889 goto cleanup;
892 * success
894 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
895 (map_data && map_data->from_user)) {
896 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
897 if (ret)
898 goto cleanup;
901 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
902 return bio;
903 cleanup:
904 if (!map_data)
905 bio_for_each_segment(bvec, bio, i)
906 __free_page(bvec->bv_page);
908 bio_put(bio);
909 out_bmd:
910 bio_free_map_data(bmd);
911 return ERR_PTR(ret);
915 * bio_copy_user - copy user data to bio
916 * @q: destination block queue
917 * @map_data: pointer to the rq_map_data holding pages (if necessary)
918 * @uaddr: start of user address
919 * @len: length in bytes
920 * @write_to_vm: bool indicating writing to pages or not
921 * @gfp_mask: memory allocation flags
923 * Prepares and returns a bio for indirect user io, bouncing data
924 * to/from kernel pages as necessary. Must be paired with
925 * call bio_uncopy_user() on io completion.
927 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
928 unsigned long uaddr, unsigned int len,
929 int write_to_vm, gfp_t gfp_mask)
931 struct sg_iovec iov;
933 iov.iov_base = (void __user *)uaddr;
934 iov.iov_len = len;
936 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
938 EXPORT_SYMBOL(bio_copy_user);
940 static struct bio *__bio_map_user_iov(struct request_queue *q,
941 struct block_device *bdev,
942 struct sg_iovec *iov, int iov_count,
943 int write_to_vm, gfp_t gfp_mask)
945 int i, j;
946 int nr_pages = 0;
947 struct page **pages;
948 struct bio *bio;
949 int cur_page = 0;
950 int ret, offset;
952 for (i = 0; i < iov_count; i++) {
953 unsigned long uaddr = (unsigned long)iov[i].iov_base;
954 unsigned long len = iov[i].iov_len;
955 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
956 unsigned long start = uaddr >> PAGE_SHIFT;
958 nr_pages += end - start;
960 * buffer must be aligned to at least hardsector size for now
962 if (uaddr & queue_dma_alignment(q))
963 return ERR_PTR(-EINVAL);
966 if (!nr_pages)
967 return ERR_PTR(-EINVAL);
969 bio = bio_kmalloc(gfp_mask, nr_pages);
970 if (!bio)
971 return ERR_PTR(-ENOMEM);
973 ret = -ENOMEM;
974 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
975 if (!pages)
976 goto out;
978 for (i = 0; i < iov_count; i++) {
979 unsigned long uaddr = (unsigned long)iov[i].iov_base;
980 unsigned long len = iov[i].iov_len;
981 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
982 unsigned long start = uaddr >> PAGE_SHIFT;
983 const int local_nr_pages = end - start;
984 const int page_limit = cur_page + local_nr_pages;
986 ret = get_user_pages_fast(uaddr, local_nr_pages,
987 write_to_vm, &pages[cur_page]);
988 if (ret < local_nr_pages) {
989 ret = -EFAULT;
990 goto out_unmap;
993 offset = uaddr & ~PAGE_MASK;
994 for (j = cur_page; j < page_limit; j++) {
995 unsigned int bytes = PAGE_SIZE - offset;
997 if (len <= 0)
998 break;
1000 if (bytes > len)
1001 bytes = len;
1004 * sorry...
1006 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1007 bytes)
1008 break;
1010 len -= bytes;
1011 offset = 0;
1014 cur_page = j;
1016 * release the pages we didn't map into the bio, if any
1018 while (j < page_limit)
1019 page_cache_release(pages[j++]);
1022 kfree(pages);
1025 * set data direction, and check if mapped pages need bouncing
1027 if (!write_to_vm)
1028 bio->bi_rw |= REQ_WRITE;
1030 bio->bi_bdev = bdev;
1031 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1032 return bio;
1034 out_unmap:
1035 for (i = 0; i < nr_pages; i++) {
1036 if(!pages[i])
1037 break;
1038 page_cache_release(pages[i]);
1040 out:
1041 kfree(pages);
1042 bio_put(bio);
1043 return ERR_PTR(ret);
1047 * bio_map_user - map user address into bio
1048 * @q: the struct request_queue for the bio
1049 * @bdev: destination block device
1050 * @uaddr: start of user address
1051 * @len: length in bytes
1052 * @write_to_vm: bool indicating writing to pages or not
1053 * @gfp_mask: memory allocation flags
1055 * Map the user space address into a bio suitable for io to a block
1056 * device. Returns an error pointer in case of error.
1058 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1059 unsigned long uaddr, unsigned int len, int write_to_vm,
1060 gfp_t gfp_mask)
1062 struct sg_iovec iov;
1064 iov.iov_base = (void __user *)uaddr;
1065 iov.iov_len = len;
1067 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1069 EXPORT_SYMBOL(bio_map_user);
1072 * bio_map_user_iov - map user sg_iovec table into bio
1073 * @q: the struct request_queue for the bio
1074 * @bdev: destination block device
1075 * @iov: the iovec.
1076 * @iov_count: number of elements in the iovec
1077 * @write_to_vm: bool indicating writing to pages or not
1078 * @gfp_mask: memory allocation flags
1080 * Map the user space address into a bio suitable for io to a block
1081 * device. Returns an error pointer in case of error.
1083 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1084 struct sg_iovec *iov, int iov_count,
1085 int write_to_vm, gfp_t gfp_mask)
1087 struct bio *bio;
1089 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1090 gfp_mask);
1091 if (IS_ERR(bio))
1092 return bio;
1095 * subtle -- if __bio_map_user() ended up bouncing a bio,
1096 * it would normally disappear when its bi_end_io is run.
1097 * however, we need it for the unmap, so grab an extra
1098 * reference to it
1100 bio_get(bio);
1102 return bio;
1105 static void __bio_unmap_user(struct bio *bio)
1107 struct bio_vec *bvec;
1108 int i;
1111 * make sure we dirty pages we wrote to
1113 __bio_for_each_segment(bvec, bio, i, 0) {
1114 if (bio_data_dir(bio) == READ)
1115 set_page_dirty_lock(bvec->bv_page);
1117 page_cache_release(bvec->bv_page);
1120 bio_put(bio);
1124 * bio_unmap_user - unmap a bio
1125 * @bio: the bio being unmapped
1127 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1128 * a process context.
1130 * bio_unmap_user() may sleep.
1132 void bio_unmap_user(struct bio *bio)
1134 __bio_unmap_user(bio);
1135 bio_put(bio);
1137 EXPORT_SYMBOL(bio_unmap_user);
1139 static void bio_map_kern_endio(struct bio *bio, int err)
1141 bio_put(bio);
1144 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1145 unsigned int len, gfp_t gfp_mask)
1147 unsigned long kaddr = (unsigned long)data;
1148 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1149 unsigned long start = kaddr >> PAGE_SHIFT;
1150 const int nr_pages = end - start;
1151 int offset, i;
1152 struct bio *bio;
1154 bio = bio_kmalloc(gfp_mask, nr_pages);
1155 if (!bio)
1156 return ERR_PTR(-ENOMEM);
1158 offset = offset_in_page(kaddr);
1159 for (i = 0; i < nr_pages; i++) {
1160 unsigned int bytes = PAGE_SIZE - offset;
1162 if (len <= 0)
1163 break;
1165 if (bytes > len)
1166 bytes = len;
1168 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1169 offset) < bytes)
1170 break;
1172 data += bytes;
1173 len -= bytes;
1174 offset = 0;
1177 bio->bi_end_io = bio_map_kern_endio;
1178 return bio;
1182 * bio_map_kern - map kernel address into bio
1183 * @q: the struct request_queue for the bio
1184 * @data: pointer to buffer to map
1185 * @len: length in bytes
1186 * @gfp_mask: allocation flags for bio allocation
1188 * Map the kernel address into a bio suitable for io to a block
1189 * device. Returns an error pointer in case of error.
1191 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1192 gfp_t gfp_mask)
1194 struct bio *bio;
1196 bio = __bio_map_kern(q, data, len, gfp_mask);
1197 if (IS_ERR(bio))
1198 return bio;
1200 if (bio->bi_size == len)
1201 return bio;
1204 * Don't support partial mappings.
1206 bio_put(bio);
1207 return ERR_PTR(-EINVAL);
1209 EXPORT_SYMBOL(bio_map_kern);
1211 static void bio_copy_kern_endio(struct bio *bio, int err)
1213 struct bio_vec *bvec;
1214 const int read = bio_data_dir(bio) == READ;
1215 struct bio_map_data *bmd = bio->bi_private;
1216 int i;
1217 char *p = bmd->sgvecs[0].iov_base;
1219 __bio_for_each_segment(bvec, bio, i, 0) {
1220 char *addr = page_address(bvec->bv_page);
1221 int len = bmd->iovecs[i].bv_len;
1223 if (read)
1224 memcpy(p, addr, len);
1226 __free_page(bvec->bv_page);
1227 p += len;
1230 bio_free_map_data(bmd);
1231 bio_put(bio);
1235 * bio_copy_kern - copy kernel address into bio
1236 * @q: the struct request_queue for the bio
1237 * @data: pointer to buffer to copy
1238 * @len: length in bytes
1239 * @gfp_mask: allocation flags for bio and page allocation
1240 * @reading: data direction is READ
1242 * copy the kernel address into a bio suitable for io to a block
1243 * device. Returns an error pointer in case of error.
1245 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1246 gfp_t gfp_mask, int reading)
1248 struct bio *bio;
1249 struct bio_vec *bvec;
1250 int i;
1252 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1253 if (IS_ERR(bio))
1254 return bio;
1256 if (!reading) {
1257 void *p = data;
1259 bio_for_each_segment(bvec, bio, i) {
1260 char *addr = page_address(bvec->bv_page);
1262 memcpy(addr, p, bvec->bv_len);
1263 p += bvec->bv_len;
1267 bio->bi_end_io = bio_copy_kern_endio;
1269 return bio;
1271 EXPORT_SYMBOL(bio_copy_kern);
1274 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1275 * for performing direct-IO in BIOs.
1277 * The problem is that we cannot run set_page_dirty() from interrupt context
1278 * because the required locks are not interrupt-safe. So what we can do is to
1279 * mark the pages dirty _before_ performing IO. And in interrupt context,
1280 * check that the pages are still dirty. If so, fine. If not, redirty them
1281 * in process context.
1283 * We special-case compound pages here: normally this means reads into hugetlb
1284 * pages. The logic in here doesn't really work right for compound pages
1285 * because the VM does not uniformly chase down the head page in all cases.
1286 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1287 * handle them at all. So we skip compound pages here at an early stage.
1289 * Note that this code is very hard to test under normal circumstances because
1290 * direct-io pins the pages with get_user_pages(). This makes
1291 * is_page_cache_freeable return false, and the VM will not clean the pages.
1292 * But other code (eg, pdflush) could clean the pages if they are mapped
1293 * pagecache.
1295 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1296 * deferred bio dirtying paths.
1300 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1302 void bio_set_pages_dirty(struct bio *bio)
1304 struct bio_vec *bvec = bio->bi_io_vec;
1305 int i;
1307 for (i = 0; i < bio->bi_vcnt; i++) {
1308 struct page *page = bvec[i].bv_page;
1310 if (page && !PageCompound(page))
1311 set_page_dirty_lock(page);
1315 static void bio_release_pages(struct bio *bio)
1317 struct bio_vec *bvec = bio->bi_io_vec;
1318 int i;
1320 for (i = 0; i < bio->bi_vcnt; i++) {
1321 struct page *page = bvec[i].bv_page;
1323 if (page)
1324 put_page(page);
1329 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1330 * If they are, then fine. If, however, some pages are clean then they must
1331 * have been written out during the direct-IO read. So we take another ref on
1332 * the BIO and the offending pages and re-dirty the pages in process context.
1334 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1335 * here on. It will run one page_cache_release() against each page and will
1336 * run one bio_put() against the BIO.
1339 static void bio_dirty_fn(struct work_struct *work);
1341 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1342 static DEFINE_SPINLOCK(bio_dirty_lock);
1343 static struct bio *bio_dirty_list;
1346 * This runs in process context
1348 static void bio_dirty_fn(struct work_struct *work)
1350 unsigned long flags;
1351 struct bio *bio;
1353 spin_lock_irqsave(&bio_dirty_lock, flags);
1354 bio = bio_dirty_list;
1355 bio_dirty_list = NULL;
1356 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1358 while (bio) {
1359 struct bio *next = bio->bi_private;
1361 bio_set_pages_dirty(bio);
1362 bio_release_pages(bio);
1363 bio_put(bio);
1364 bio = next;
1368 void bio_check_pages_dirty(struct bio *bio)
1370 struct bio_vec *bvec = bio->bi_io_vec;
1371 int nr_clean_pages = 0;
1372 int i;
1374 for (i = 0; i < bio->bi_vcnt; i++) {
1375 struct page *page = bvec[i].bv_page;
1377 if (PageDirty(page) || PageCompound(page)) {
1378 page_cache_release(page);
1379 bvec[i].bv_page = NULL;
1380 } else {
1381 nr_clean_pages++;
1385 if (nr_clean_pages) {
1386 unsigned long flags;
1388 spin_lock_irqsave(&bio_dirty_lock, flags);
1389 bio->bi_private = bio_dirty_list;
1390 bio_dirty_list = bio;
1391 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1392 schedule_work(&bio_dirty_work);
1393 } else {
1394 bio_put(bio);
1398 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1399 void bio_flush_dcache_pages(struct bio *bi)
1401 int i;
1402 struct bio_vec *bvec;
1404 bio_for_each_segment(bvec, bi, i)
1405 flush_dcache_page(bvec->bv_page);
1407 EXPORT_SYMBOL(bio_flush_dcache_pages);
1408 #endif
1411 * bio_endio - end I/O on a bio
1412 * @bio: bio
1413 * @error: error, if any
1415 * Description:
1416 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1417 * preferred way to end I/O on a bio, it takes care of clearing
1418 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1419 * established -Exxxx (-EIO, for instance) error values in case
1420 * something went wrong. Noone should call bi_end_io() directly on a
1421 * bio unless they own it and thus know that it has an end_io
1422 * function.
1424 void bio_endio(struct bio *bio, int error)
1426 if (error)
1427 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1428 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1429 error = -EIO;
1431 if (bio->bi_end_io)
1432 bio->bi_end_io(bio, error);
1434 EXPORT_SYMBOL(bio_endio);
1436 void bio_pair_release(struct bio_pair *bp)
1438 if (atomic_dec_and_test(&bp->cnt)) {
1439 struct bio *master = bp->bio1.bi_private;
1441 bio_endio(master, bp->error);
1442 mempool_free(bp, bp->bio2.bi_private);
1445 EXPORT_SYMBOL(bio_pair_release);
1447 static void bio_pair_end_1(struct bio *bi, int err)
1449 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1451 if (err)
1452 bp->error = err;
1454 bio_pair_release(bp);
1457 static void bio_pair_end_2(struct bio *bi, int err)
1459 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1461 if (err)
1462 bp->error = err;
1464 bio_pair_release(bp);
1468 * split a bio - only worry about a bio with a single page in its iovec
1470 struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1472 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1474 if (!bp)
1475 return bp;
1477 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
1478 bi->bi_sector + first_sectors);
1480 BUG_ON(bi->bi_vcnt != 1);
1481 BUG_ON(bi->bi_idx != 0);
1482 atomic_set(&bp->cnt, 3);
1483 bp->error = 0;
1484 bp->bio1 = *bi;
1485 bp->bio2 = *bi;
1486 bp->bio2.bi_sector += first_sectors;
1487 bp->bio2.bi_size -= first_sectors << 9;
1488 bp->bio1.bi_size = first_sectors << 9;
1490 bp->bv1 = bi->bi_io_vec[0];
1491 bp->bv2 = bi->bi_io_vec[0];
1492 bp->bv2.bv_offset += first_sectors << 9;
1493 bp->bv2.bv_len -= first_sectors << 9;
1494 bp->bv1.bv_len = first_sectors << 9;
1496 bp->bio1.bi_io_vec = &bp->bv1;
1497 bp->bio2.bi_io_vec = &bp->bv2;
1499 bp->bio1.bi_max_vecs = 1;
1500 bp->bio2.bi_max_vecs = 1;
1502 bp->bio1.bi_end_io = bio_pair_end_1;
1503 bp->bio2.bi_end_io = bio_pair_end_2;
1505 bp->bio1.bi_private = bi;
1506 bp->bio2.bi_private = bio_split_pool;
1508 if (bio_integrity(bi))
1509 bio_integrity_split(bi, bp, first_sectors);
1511 return bp;
1513 EXPORT_SYMBOL(bio_split);
1516 * bio_sector_offset - Find hardware sector offset in bio
1517 * @bio: bio to inspect
1518 * @index: bio_vec index
1519 * @offset: offset in bv_page
1521 * Return the number of hardware sectors between beginning of bio
1522 * and an end point indicated by a bio_vec index and an offset
1523 * within that vector's page.
1525 sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1526 unsigned int offset)
1528 unsigned int sector_sz;
1529 struct bio_vec *bv;
1530 sector_t sectors;
1531 int i;
1533 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
1534 sectors = 0;
1536 if (index >= bio->bi_idx)
1537 index = bio->bi_vcnt - 1;
1539 __bio_for_each_segment(bv, bio, i, 0) {
1540 if (i == index) {
1541 if (offset > bv->bv_offset)
1542 sectors += (offset - bv->bv_offset) / sector_sz;
1543 break;
1546 sectors += bv->bv_len / sector_sz;
1549 return sectors;
1551 EXPORT_SYMBOL(bio_sector_offset);
1554 * create memory pools for biovec's in a bio_set.
1555 * use the global biovec slabs created for general use.
1557 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1559 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1561 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1562 if (!bs->bvec_pool)
1563 return -ENOMEM;
1565 return 0;
1568 static void biovec_free_pools(struct bio_set *bs)
1570 mempool_destroy(bs->bvec_pool);
1573 void bioset_free(struct bio_set *bs)
1575 if (bs->bio_pool)
1576 mempool_destroy(bs->bio_pool);
1578 bioset_integrity_free(bs);
1579 biovec_free_pools(bs);
1580 bio_put_slab(bs);
1582 kfree(bs);
1584 EXPORT_SYMBOL(bioset_free);
1587 * bioset_create - Create a bio_set
1588 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1589 * @front_pad: Number of bytes to allocate in front of the returned bio
1591 * Description:
1592 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1593 * to ask for a number of bytes to be allocated in front of the bio.
1594 * Front pad allocation is useful for embedding the bio inside
1595 * another structure, to avoid allocating extra data to go with the bio.
1596 * Note that the bio must be embedded at the END of that structure always,
1597 * or things will break badly.
1599 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1601 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1602 struct bio_set *bs;
1604 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1605 if (!bs)
1606 return NULL;
1608 bs->front_pad = front_pad;
1610 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1611 if (!bs->bio_slab) {
1612 kfree(bs);
1613 return NULL;
1616 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1617 if (!bs->bio_pool)
1618 goto bad;
1620 if (bioset_integrity_create(bs, pool_size))
1621 goto bad;
1623 if (!biovec_create_pools(bs, pool_size))
1624 return bs;
1626 bad:
1627 bioset_free(bs);
1628 return NULL;
1630 EXPORT_SYMBOL(bioset_create);
1632 static void __init biovec_init_slabs(void)
1634 int i;
1636 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1637 int size;
1638 struct biovec_slab *bvs = bvec_slabs + i;
1640 #ifndef CONFIG_BLK_DEV_INTEGRITY
1641 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1642 bvs->slab = NULL;
1643 continue;
1645 #endif
1647 size = bvs->nr_vecs * sizeof(struct bio_vec);
1648 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1649 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1653 static int __init init_bio(void)
1655 bio_slab_max = 2;
1656 bio_slab_nr = 0;
1657 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1658 if (!bio_slabs)
1659 panic("bio: can't allocate bios\n");
1661 bio_integrity_init();
1662 biovec_init_slabs();
1664 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1665 if (!fs_bio_set)
1666 panic("bio: can't allocate bios\n");
1668 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1669 sizeof(struct bio_pair));
1670 if (!bio_split_pool)
1671 panic("bio: can't create split pool\n");
1673 return 0;
1675 subsys_initcall(init_bio);