1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
6 #include <linux/swap.h>
8 #include <linux/blkdev.h>
10 #include <linux/iocontext.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/export.h>
15 #include <linux/mempool.h>
16 #include <linux/workqueue.h>
17 #include <linux/cgroup.h>
18 #include <linux/blk-cgroup.h>
19 #include <linux/highmem.h>
20 #include <linux/sched/sysctl.h>
21 #include <linux/blk-crypto.h>
22 #include <linux/xarray.h>
24 #include <trace/events/block.h>
26 #include "blk-rq-qos.h"
28 static struct biovec_slab
{
31 struct kmem_cache
*slab
;
32 } bvec_slabs
[] __read_mostly
= {
33 { .nr_vecs
= 16, .name
= "biovec-16" },
34 { .nr_vecs
= 64, .name
= "biovec-64" },
35 { .nr_vecs
= 128, .name
= "biovec-128" },
36 { .nr_vecs
= BIO_MAX_VECS
, .name
= "biovec-max" },
39 static struct biovec_slab
*biovec_slab(unsigned short nr_vecs
)
42 /* smaller bios use inline vecs */
44 return &bvec_slabs
[0];
46 return &bvec_slabs
[1];
48 return &bvec_slabs
[2];
49 case 129 ... BIO_MAX_VECS
:
50 return &bvec_slabs
[3];
58 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
59 * IO code that does not need private memory pools.
61 struct bio_set fs_bio_set
;
62 EXPORT_SYMBOL(fs_bio_set
);
65 * Our slab pool management
68 struct kmem_cache
*slab
;
69 unsigned int slab_ref
;
70 unsigned int slab_size
;
73 static DEFINE_MUTEX(bio_slab_lock
);
74 static DEFINE_XARRAY(bio_slabs
);
76 static struct bio_slab
*create_bio_slab(unsigned int size
)
78 struct bio_slab
*bslab
= kzalloc(sizeof(*bslab
), GFP_KERNEL
);
83 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", size
);
84 bslab
->slab
= kmem_cache_create(bslab
->name
, size
,
85 ARCH_KMALLOC_MINALIGN
, SLAB_HWCACHE_ALIGN
, NULL
);
90 bslab
->slab_size
= size
;
92 if (!xa_err(xa_store(&bio_slabs
, size
, bslab
, GFP_KERNEL
)))
95 kmem_cache_destroy(bslab
->slab
);
102 static inline unsigned int bs_bio_slab_size(struct bio_set
*bs
)
104 return bs
->front_pad
+ sizeof(struct bio
) + bs
->back_pad
;
107 static struct kmem_cache
*bio_find_or_create_slab(struct bio_set
*bs
)
109 unsigned int size
= bs_bio_slab_size(bs
);
110 struct bio_slab
*bslab
;
112 mutex_lock(&bio_slab_lock
);
113 bslab
= xa_load(&bio_slabs
, size
);
117 bslab
= create_bio_slab(size
);
118 mutex_unlock(&bio_slab_lock
);
125 static void bio_put_slab(struct bio_set
*bs
)
127 struct bio_slab
*bslab
= NULL
;
128 unsigned int slab_size
= bs_bio_slab_size(bs
);
130 mutex_lock(&bio_slab_lock
);
132 bslab
= xa_load(&bio_slabs
, slab_size
);
133 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
136 WARN_ON_ONCE(bslab
->slab
!= bs
->bio_slab
);
138 WARN_ON(!bslab
->slab_ref
);
140 if (--bslab
->slab_ref
)
143 xa_erase(&bio_slabs
, slab_size
);
145 kmem_cache_destroy(bslab
->slab
);
149 mutex_unlock(&bio_slab_lock
);
152 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned short nr_vecs
)
154 BIO_BUG_ON(nr_vecs
> BIO_MAX_VECS
);
156 if (nr_vecs
== BIO_MAX_VECS
)
157 mempool_free(bv
, pool
);
158 else if (nr_vecs
> BIO_INLINE_VECS
)
159 kmem_cache_free(biovec_slab(nr_vecs
)->slab
, bv
);
163 * Make the first allocation restricted and don't dump info on allocation
164 * failures, since we'll fall back to the mempool in case of failure.
166 static inline gfp_t
bvec_alloc_gfp(gfp_t gfp
)
168 return (gfp
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
)) |
169 __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
172 struct bio_vec
*bvec_alloc(mempool_t
*pool
, unsigned short *nr_vecs
,
175 struct biovec_slab
*bvs
= biovec_slab(*nr_vecs
);
177 if (WARN_ON_ONCE(!bvs
))
181 * Upgrade the nr_vecs request to take full advantage of the allocation.
182 * We also rely on this in the bvec_free path.
184 *nr_vecs
= bvs
->nr_vecs
;
187 * Try a slab allocation first for all smaller allocations. If that
188 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
189 * The mempool is sized to handle up to BIO_MAX_VECS entries.
191 if (*nr_vecs
< BIO_MAX_VECS
) {
194 bvl
= kmem_cache_alloc(bvs
->slab
, bvec_alloc_gfp(gfp_mask
));
195 if (likely(bvl
) || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
197 *nr_vecs
= BIO_MAX_VECS
;
200 return mempool_alloc(pool
, gfp_mask
);
203 void bio_uninit(struct bio
*bio
)
205 #ifdef CONFIG_BLK_CGROUP
207 blkg_put(bio
->bi_blkg
);
211 if (bio_integrity(bio
))
212 bio_integrity_free(bio
);
214 bio_crypt_free_ctx(bio
);
216 EXPORT_SYMBOL(bio_uninit
);
218 static void bio_free(struct bio
*bio
)
220 struct bio_set
*bs
= bio
->bi_pool
;
226 bvec_free(&bs
->bvec_pool
, bio
->bi_io_vec
, bio
->bi_max_vecs
);
229 * If we have front padding, adjust the bio pointer before freeing
234 mempool_free(p
, &bs
->bio_pool
);
236 /* Bio was allocated by bio_kmalloc() */
242 * Users of this function have their own bio allocation. Subsequently,
243 * they must remember to pair any call to bio_init() with bio_uninit()
244 * when IO has completed, or when the bio is released.
246 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
247 unsigned short max_vecs
)
249 memset(bio
, 0, sizeof(*bio
));
250 atomic_set(&bio
->__bi_remaining
, 1);
251 atomic_set(&bio
->__bi_cnt
, 1);
253 bio
->bi_io_vec
= table
;
254 bio
->bi_max_vecs
= max_vecs
;
256 EXPORT_SYMBOL(bio_init
);
258 unsigned int bio_max_size(struct bio
*bio
)
260 struct block_device
*bdev
= bio
->bi_bdev
;
262 return bdev
? bdev
->bd_disk
->queue
->limits
.bio_max_bytes
: UINT_MAX
;
266 * bio_reset - reinitialize a bio
270 * After calling bio_reset(), @bio will be in the same state as a freshly
271 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
272 * preserved are the ones that are initialized by bio_alloc_bioset(). See
273 * comment in struct bio.
275 void bio_reset(struct bio
*bio
)
278 memset(bio
, 0, BIO_RESET_BYTES
);
279 atomic_set(&bio
->__bi_remaining
, 1);
281 EXPORT_SYMBOL(bio_reset
);
283 static struct bio
*__bio_chain_endio(struct bio
*bio
)
285 struct bio
*parent
= bio
->bi_private
;
287 if (bio
->bi_status
&& !parent
->bi_status
)
288 parent
->bi_status
= bio
->bi_status
;
293 static void bio_chain_endio(struct bio
*bio
)
295 bio_endio(__bio_chain_endio(bio
));
299 * bio_chain - chain bio completions
300 * @bio: the target bio
301 * @parent: the parent bio of @bio
303 * The caller won't have a bi_end_io called when @bio completes - instead,
304 * @parent's bi_end_io won't be called until both @parent and @bio have
305 * completed; the chained bio will also be freed when it completes.
307 * The caller must not set bi_private or bi_end_io in @bio.
309 void bio_chain(struct bio
*bio
, struct bio
*parent
)
311 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
313 bio
->bi_private
= parent
;
314 bio
->bi_end_io
= bio_chain_endio
;
315 bio_inc_remaining(parent
);
317 EXPORT_SYMBOL(bio_chain
);
319 static void bio_alloc_rescue(struct work_struct
*work
)
321 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
325 spin_lock(&bs
->rescue_lock
);
326 bio
= bio_list_pop(&bs
->rescue_list
);
327 spin_unlock(&bs
->rescue_lock
);
332 submit_bio_noacct(bio
);
336 static void punt_bios_to_rescuer(struct bio_set
*bs
)
338 struct bio_list punt
, nopunt
;
341 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
344 * In order to guarantee forward progress we must punt only bios that
345 * were allocated from this bio_set; otherwise, if there was a bio on
346 * there for a stacking driver higher up in the stack, processing it
347 * could require allocating bios from this bio_set, and doing that from
348 * our own rescuer would be bad.
350 * Since bio lists are singly linked, pop them all instead of trying to
351 * remove from the middle of the list:
354 bio_list_init(&punt
);
355 bio_list_init(&nopunt
);
357 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
358 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
359 current
->bio_list
[0] = nopunt
;
361 bio_list_init(&nopunt
);
362 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
363 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
364 current
->bio_list
[1] = nopunt
;
366 spin_lock(&bs
->rescue_lock
);
367 bio_list_merge(&bs
->rescue_list
, &punt
);
368 spin_unlock(&bs
->rescue_lock
);
370 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
374 * bio_alloc_bioset - allocate a bio for I/O
375 * @gfp_mask: the GFP_* mask given to the slab allocator
376 * @nr_iovecs: number of iovecs to pre-allocate
377 * @bs: the bio_set to allocate from.
379 * Allocate a bio from the mempools in @bs.
381 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
382 * allocate a bio. This is due to the mempool guarantees. To make this work,
383 * callers must never allocate more than 1 bio at a time from the general pool.
384 * Callers that need to allocate more than 1 bio must always submit the
385 * previously allocated bio for IO before attempting to allocate a new one.
386 * Failure to do so can cause deadlocks under memory pressure.
388 * Note that when running under submit_bio_noacct() (i.e. any block driver),
389 * bios are not submitted until after you return - see the code in
390 * submit_bio_noacct() that converts recursion into iteration, to prevent
393 * This would normally mean allocating multiple bios under submit_bio_noacct()
394 * would be susceptible to deadlocks, but we have
395 * deadlock avoidance code that resubmits any blocked bios from a rescuer
398 * However, we do not guarantee forward progress for allocations from other
399 * mempools. Doing multiple allocations from the same mempool under
400 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
401 * for per bio allocations.
403 * Returns: Pointer to new bio on success, NULL on failure.
405 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned short nr_iovecs
,
408 gfp_t saved_gfp
= gfp_mask
;
412 /* should not use nobvec bioset for nr_iovecs > 0 */
413 if (WARN_ON_ONCE(!mempool_initialized(&bs
->bvec_pool
) && nr_iovecs
> 0))
417 * submit_bio_noacct() converts recursion to iteration; this means if
418 * we're running beneath it, any bios we allocate and submit will not be
419 * submitted (and thus freed) until after we return.
421 * This exposes us to a potential deadlock if we allocate multiple bios
422 * from the same bio_set() while running underneath submit_bio_noacct().
423 * If we were to allocate multiple bios (say a stacking block driver
424 * that was splitting bios), we would deadlock if we exhausted the
427 * We solve this, and guarantee forward progress, with a rescuer
428 * workqueue per bio_set. If we go to allocate and there are bios on
429 * current->bio_list, we first try the allocation without
430 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
431 * blocking to the rescuer workqueue before we retry with the original
434 if (current
->bio_list
&&
435 (!bio_list_empty(¤t
->bio_list
[0]) ||
436 !bio_list_empty(¤t
->bio_list
[1])) &&
437 bs
->rescue_workqueue
)
438 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
440 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
441 if (!p
&& gfp_mask
!= saved_gfp
) {
442 punt_bios_to_rescuer(bs
);
443 gfp_mask
= saved_gfp
;
444 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
449 bio
= p
+ bs
->front_pad
;
450 if (nr_iovecs
> BIO_INLINE_VECS
) {
451 struct bio_vec
*bvl
= NULL
;
453 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_iovecs
, gfp_mask
);
454 if (!bvl
&& gfp_mask
!= saved_gfp
) {
455 punt_bios_to_rescuer(bs
);
456 gfp_mask
= saved_gfp
;
457 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_iovecs
, gfp_mask
);
462 bio_init(bio
, bvl
, nr_iovecs
);
463 } else if (nr_iovecs
) {
464 bio_init(bio
, bio
->bi_inline_vecs
, BIO_INLINE_VECS
);
466 bio_init(bio
, NULL
, 0);
473 mempool_free(p
, &bs
->bio_pool
);
476 EXPORT_SYMBOL(bio_alloc_bioset
);
479 * bio_kmalloc - kmalloc a bio for I/O
480 * @gfp_mask: the GFP_* mask given to the slab allocator
481 * @nr_iovecs: number of iovecs to pre-allocate
483 * Use kmalloc to allocate and initialize a bio.
485 * Returns: Pointer to new bio on success, NULL on failure.
487 struct bio
*bio_kmalloc(gfp_t gfp_mask
, unsigned short nr_iovecs
)
491 if (nr_iovecs
> UIO_MAXIOV
)
494 bio
= kmalloc(struct_size(bio
, bi_inline_vecs
, nr_iovecs
), gfp_mask
);
497 bio_init(bio
, nr_iovecs
? bio
->bi_inline_vecs
: NULL
, nr_iovecs
);
501 EXPORT_SYMBOL(bio_kmalloc
);
503 void zero_fill_bio(struct bio
*bio
)
507 struct bvec_iter iter
;
509 bio_for_each_segment(bv
, bio
, iter
) {
510 char *data
= bvec_kmap_irq(&bv
, &flags
);
511 memset(data
, 0, bv
.bv_len
);
512 flush_dcache_page(bv
.bv_page
);
513 bvec_kunmap_irq(data
, &flags
);
516 EXPORT_SYMBOL(zero_fill_bio
);
519 * bio_truncate - truncate the bio to small size of @new_size
520 * @bio: the bio to be truncated
521 * @new_size: new size for truncating the bio
524 * Truncate the bio to new size of @new_size. If bio_op(bio) is
525 * REQ_OP_READ, zero the truncated part. This function should only
526 * be used for handling corner cases, such as bio eod.
528 void bio_truncate(struct bio
*bio
, unsigned new_size
)
531 struct bvec_iter iter
;
532 unsigned int done
= 0;
533 bool truncated
= false;
535 if (new_size
>= bio
->bi_iter
.bi_size
)
538 if (bio_op(bio
) != REQ_OP_READ
)
541 bio_for_each_segment(bv
, bio
, iter
) {
542 if (done
+ bv
.bv_len
> new_size
) {
546 offset
= new_size
- done
;
549 zero_user(bv
.bv_page
, offset
, bv
.bv_len
- offset
);
557 * Don't touch bvec table here and make it really immutable, since
558 * fs bio user has to retrieve all pages via bio_for_each_segment_all
559 * in its .end_bio() callback.
561 * It is enough to truncate bio by updating .bi_size since we can make
562 * correct bvec with the updated .bi_size for drivers.
564 bio
->bi_iter
.bi_size
= new_size
;
568 * guard_bio_eod - truncate a BIO to fit the block device
569 * @bio: bio to truncate
571 * This allows us to do IO even on the odd last sectors of a device, even if the
572 * block size is some multiple of the physical sector size.
574 * We'll just truncate the bio to the size of the device, and clear the end of
575 * the buffer head manually. Truly out-of-range accesses will turn into actual
576 * I/O errors, this only handles the "we need to be able to do I/O at the final
579 void guard_bio_eod(struct bio
*bio
)
581 sector_t maxsector
= bdev_nr_sectors(bio
->bi_bdev
);
587 * If the *whole* IO is past the end of the device,
588 * let it through, and the IO layer will turn it into
591 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
594 maxsector
-= bio
->bi_iter
.bi_sector
;
595 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
598 bio_truncate(bio
, maxsector
<< 9);
602 * bio_put - release a reference to a bio
603 * @bio: bio to release reference to
606 * Put a reference to a &struct bio, either one you have gotten with
607 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
609 void bio_put(struct bio
*bio
)
611 if (!bio_flagged(bio
, BIO_REFFED
))
614 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
619 if (atomic_dec_and_test(&bio
->__bi_cnt
))
623 EXPORT_SYMBOL(bio_put
);
626 * __bio_clone_fast - clone a bio that shares the original bio's biovec
627 * @bio: destination bio
628 * @bio_src: bio to clone
630 * Clone a &bio. Caller will own the returned bio, but not
631 * the actual data it points to. Reference count of returned
634 * Caller must ensure that @bio_src is not freed before @bio.
636 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
638 WARN_ON_ONCE(bio
->bi_pool
&& bio
->bi_max_vecs
);
641 * most users will be overriding ->bi_bdev with a new target,
642 * so we don't set nor calculate new physical/hw segment counts here
644 bio
->bi_bdev
= bio_src
->bi_bdev
;
645 bio_set_flag(bio
, BIO_CLONED
);
646 if (bio_flagged(bio_src
, BIO_THROTTLED
))
647 bio_set_flag(bio
, BIO_THROTTLED
);
648 if (bio_flagged(bio_src
, BIO_REMAPPED
))
649 bio_set_flag(bio
, BIO_REMAPPED
);
650 bio
->bi_opf
= bio_src
->bi_opf
;
651 bio
->bi_ioprio
= bio_src
->bi_ioprio
;
652 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
653 bio
->bi_iter
= bio_src
->bi_iter
;
654 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
656 bio_clone_blkg_association(bio
, bio_src
);
657 blkcg_bio_issue_init(bio
);
659 EXPORT_SYMBOL(__bio_clone_fast
);
662 * bio_clone_fast - clone a bio that shares the original bio's biovec
664 * @gfp_mask: allocation priority
665 * @bs: bio_set to allocate from
667 * Like __bio_clone_fast, only also allocates the returned bio
669 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
673 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
677 __bio_clone_fast(b
, bio
);
679 if (bio_crypt_clone(b
, bio
, gfp_mask
) < 0)
682 if (bio_integrity(bio
) &&
683 bio_integrity_clone(b
, bio
, gfp_mask
) < 0)
692 EXPORT_SYMBOL(bio_clone_fast
);
694 const char *bio_devname(struct bio
*bio
, char *buf
)
696 return bdevname(bio
->bi_bdev
, buf
);
698 EXPORT_SYMBOL(bio_devname
);
700 static inline bool page_is_mergeable(const struct bio_vec
*bv
,
701 struct page
*page
, unsigned int len
, unsigned int off
,
704 size_t bv_end
= bv
->bv_offset
+ bv
->bv_len
;
705 phys_addr_t vec_end_addr
= page_to_phys(bv
->bv_page
) + bv_end
- 1;
706 phys_addr_t page_addr
= page_to_phys(page
);
708 if (vec_end_addr
+ 1 != page_addr
+ off
)
710 if (xen_domain() && !xen_biovec_phys_mergeable(bv
, page
))
713 *same_page
= ((vec_end_addr
& PAGE_MASK
) == page_addr
);
716 return (bv
->bv_page
+ bv_end
/ PAGE_SIZE
) == (page
+ off
/ PAGE_SIZE
);
720 * Try to merge a page into a segment, while obeying the hardware segment
721 * size limit. This is not for normal read/write bios, but for passthrough
722 * or Zone Append operations that we can't split.
724 static bool bio_try_merge_hw_seg(struct request_queue
*q
, struct bio
*bio
,
725 struct page
*page
, unsigned len
,
726 unsigned offset
, bool *same_page
)
728 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
729 unsigned long mask
= queue_segment_boundary(q
);
730 phys_addr_t addr1
= page_to_phys(bv
->bv_page
) + bv
->bv_offset
;
731 phys_addr_t addr2
= page_to_phys(page
) + offset
+ len
- 1;
733 if ((addr1
| mask
) != (addr2
| mask
))
735 if (bv
->bv_len
+ len
> queue_max_segment_size(q
))
737 return __bio_try_merge_page(bio
, page
, len
, offset
, same_page
);
741 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
742 * @q: the target queue
743 * @bio: destination bio
745 * @len: vec entry length
746 * @offset: vec entry offset
747 * @max_sectors: maximum number of sectors that can be added
748 * @same_page: return if the segment has been merged inside the same page
750 * Add a page to a bio while respecting the hardware max_sectors, max_segment
751 * and gap limitations.
753 int bio_add_hw_page(struct request_queue
*q
, struct bio
*bio
,
754 struct page
*page
, unsigned int len
, unsigned int offset
,
755 unsigned int max_sectors
, bool *same_page
)
757 struct bio_vec
*bvec
;
759 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
762 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
765 if (bio
->bi_vcnt
> 0) {
766 if (bio_try_merge_hw_seg(q
, bio
, page
, len
, offset
, same_page
))
770 * If the queue doesn't support SG gaps and adding this segment
771 * would create a gap, disallow it.
773 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
774 if (bvec_gap_to_prev(q
, bvec
, offset
))
778 if (bio_full(bio
, len
))
781 if (bio
->bi_vcnt
>= queue_max_segments(q
))
784 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
785 bvec
->bv_page
= page
;
787 bvec
->bv_offset
= offset
;
789 bio
->bi_iter
.bi_size
+= len
;
794 * bio_add_pc_page - attempt to add page to passthrough bio
795 * @q: the target queue
796 * @bio: destination bio
798 * @len: vec entry length
799 * @offset: vec entry offset
801 * Attempt to add a page to the bio_vec maplist. This can fail for a
802 * number of reasons, such as the bio being full or target block device
803 * limitations. The target block device must allow bio's up to PAGE_SIZE,
804 * so it is always possible to add a single page to an empty bio.
806 * This should only be used by passthrough bios.
808 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
,
809 struct page
*page
, unsigned int len
, unsigned int offset
)
811 bool same_page
= false;
812 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
813 queue_max_hw_sectors(q
), &same_page
);
815 EXPORT_SYMBOL(bio_add_pc_page
);
818 * bio_add_zone_append_page - attempt to add page to zone-append bio
819 * @bio: destination bio
821 * @len: vec entry length
822 * @offset: vec entry offset
824 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
825 * for a zone-append request. This can fail for a number of reasons, such as the
826 * bio being full or the target block device is not a zoned block device or
827 * other limitations of the target block device. The target block device must
828 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
831 * Returns: number of bytes added to the bio, or 0 in case of a failure.
833 int bio_add_zone_append_page(struct bio
*bio
, struct page
*page
,
834 unsigned int len
, unsigned int offset
)
836 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
837 bool same_page
= false;
839 if (WARN_ON_ONCE(bio_op(bio
) != REQ_OP_ZONE_APPEND
))
842 if (WARN_ON_ONCE(!blk_queue_is_zoned(q
)))
845 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
846 queue_max_zone_append_sectors(q
), &same_page
);
848 EXPORT_SYMBOL_GPL(bio_add_zone_append_page
);
851 * __bio_try_merge_page - try appending data to an existing bvec.
852 * @bio: destination bio
853 * @page: start page to add
854 * @len: length of the data to add
855 * @off: offset of the data relative to @page
856 * @same_page: return if the segment has been merged inside the same page
858 * Try to add the data at @page + @off to the last bvec of @bio. This is a
859 * useful optimisation for file systems with a block size smaller than the
862 * Warn if (@len, @off) crosses pages in case that @same_page is true.
864 * Return %true on success or %false on failure.
866 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
867 unsigned int len
, unsigned int off
, bool *same_page
)
869 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
872 if (bio
->bi_vcnt
> 0) {
873 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
875 if (page_is_mergeable(bv
, page
, len
, off
, same_page
)) {
876 if (bio
->bi_iter
.bi_size
> bio_max_size(bio
) - len
) {
881 bio
->bi_iter
.bi_size
+= len
;
887 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
890 * __bio_add_page - add page(s) to a bio in a new segment
891 * @bio: destination bio
892 * @page: start page to add
893 * @len: length of the data to add, may cross pages
894 * @off: offset of the data relative to @page, may cross pages
896 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
897 * that @bio has space for another bvec.
899 void __bio_add_page(struct bio
*bio
, struct page
*page
,
900 unsigned int len
, unsigned int off
)
902 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
904 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
905 WARN_ON_ONCE(bio_full(bio
, len
));
911 bio
->bi_iter
.bi_size
+= len
;
914 if (!bio_flagged(bio
, BIO_WORKINGSET
) && unlikely(PageWorkingset(page
)))
915 bio_set_flag(bio
, BIO_WORKINGSET
);
917 EXPORT_SYMBOL_GPL(__bio_add_page
);
920 * bio_add_page - attempt to add page(s) to bio
921 * @bio: destination bio
922 * @page: start page to add
923 * @len: vec entry length, may cross pages
924 * @offset: vec entry offset relative to @page, may cross pages
926 * Attempt to add page(s) to the bio_vec maplist. This will only fail
927 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
929 int bio_add_page(struct bio
*bio
, struct page
*page
,
930 unsigned int len
, unsigned int offset
)
932 bool same_page
= false;
934 if (!__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
935 if (bio_full(bio
, len
))
937 __bio_add_page(bio
, page
, len
, offset
);
941 EXPORT_SYMBOL(bio_add_page
);
943 void bio_release_pages(struct bio
*bio
, bool mark_dirty
)
945 struct bvec_iter_all iter_all
;
946 struct bio_vec
*bvec
;
948 if (bio_flagged(bio
, BIO_NO_PAGE_REF
))
951 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
952 if (mark_dirty
&& !PageCompound(bvec
->bv_page
))
953 set_page_dirty_lock(bvec
->bv_page
);
954 put_page(bvec
->bv_page
);
957 EXPORT_SYMBOL_GPL(bio_release_pages
);
959 static void __bio_iov_bvec_set(struct bio
*bio
, struct iov_iter
*iter
)
961 WARN_ON_ONCE(bio
->bi_max_vecs
);
963 bio
->bi_vcnt
= iter
->nr_segs
;
964 bio
->bi_io_vec
= (struct bio_vec
*)iter
->bvec
;
965 bio
->bi_iter
.bi_bvec_done
= iter
->iov_offset
;
966 bio
->bi_iter
.bi_size
= iter
->count
;
967 bio_set_flag(bio
, BIO_NO_PAGE_REF
);
968 bio_set_flag(bio
, BIO_CLONED
);
971 static int bio_iov_bvec_set(struct bio
*bio
, struct iov_iter
*iter
)
973 __bio_iov_bvec_set(bio
, iter
);
974 iov_iter_advance(iter
, iter
->count
);
978 static int bio_iov_bvec_set_append(struct bio
*bio
, struct iov_iter
*iter
)
980 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
981 struct iov_iter i
= *iter
;
983 iov_iter_truncate(&i
, queue_max_zone_append_sectors(q
) << 9);
984 __bio_iov_bvec_set(bio
, &i
);
985 iov_iter_advance(iter
, i
.count
);
989 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
992 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
993 * @bio: bio to add pages to
994 * @iter: iov iterator describing the region to be mapped
996 * Pins pages from *iter and appends them to @bio's bvec array. The
997 * pages will have to be released using put_page() when done.
998 * For multi-segment *iter, this function only adds pages from the
999 * next non-empty segment of the iov iterator.
1001 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1003 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1004 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1005 unsigned int bytes_left
= bio_max_size(bio
) - bio
->bi_iter
.bi_size
;
1006 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1007 struct page
**pages
= (struct page
**)bv
;
1008 bool same_page
= false;
1014 * Move page array up in the allocated memory for the bio vecs as far as
1015 * possible so that we can start filling biovecs from the beginning
1016 * without overwriting the temporary page array.
1018 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1019 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1021 size
= iov_iter_get_pages(iter
, pages
, bytes_left
, nr_pages
,
1023 if (unlikely(size
<= 0))
1024 return size
? size
: -EFAULT
;
1026 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1027 struct page
*page
= pages
[i
];
1029 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1031 if (__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1035 if (WARN_ON_ONCE(bio_full(bio
, len
)))
1037 __bio_add_page(bio
, page
, len
, offset
);
1042 iov_iter_advance(iter
, size
);
1046 static int __bio_iov_append_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1048 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1049 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1050 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
1051 unsigned int max_append_sectors
= queue_max_zone_append_sectors(q
);
1052 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1053 struct page
**pages
= (struct page
**)bv
;
1059 if (WARN_ON_ONCE(!max_append_sectors
))
1063 * Move page array up in the allocated memory for the bio vecs as far as
1064 * possible so that we can start filling biovecs from the beginning
1065 * without overwriting the temporary page array.
1067 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1068 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1070 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1071 if (unlikely(size
<= 0))
1072 return size
? size
: -EFAULT
;
1074 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1075 struct page
*page
= pages
[i
];
1076 bool same_page
= false;
1078 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1079 if (bio_add_hw_page(q
, bio
, page
, len
, offset
,
1080 max_append_sectors
, &same_page
) != len
) {
1089 iov_iter_advance(iter
, size
- left
);
1094 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1095 * @bio: bio to add pages to
1096 * @iter: iov iterator describing the region to be added
1098 * This takes either an iterator pointing to user memory, or one pointing to
1099 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1100 * map them into the kernel. On IO completion, the caller should put those
1101 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1102 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1103 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1104 * completed by a call to ->ki_complete() or returns with an error other than
1105 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1106 * on IO completion. If it isn't, then pages should be released.
1108 * The function tries, but does not guarantee, to pin as many pages as
1109 * fit into the bio, or are requested in @iter, whatever is smaller. If
1110 * MM encounters an error pinning the requested pages, it stops. Error
1111 * is returned only if 0 pages could be pinned.
1113 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1114 * responsible for setting BIO_WORKINGSET if necessary.
1116 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1120 if (iov_iter_is_bvec(iter
)) {
1121 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
)
1122 return bio_iov_bvec_set_append(bio
, iter
);
1123 return bio_iov_bvec_set(bio
, iter
);
1127 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
)
1128 ret
= __bio_iov_append_get_pages(bio
, iter
);
1130 ret
= __bio_iov_iter_get_pages(bio
, iter
);
1131 } while (!ret
&& iov_iter_count(iter
) && !bio_full(bio
, 0));
1133 /* don't account direct I/O as memory stall */
1134 bio_clear_flag(bio
, BIO_WORKINGSET
);
1135 return bio
->bi_vcnt
? 0 : ret
;
1137 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
1139 static void submit_bio_wait_endio(struct bio
*bio
)
1141 complete(bio
->bi_private
);
1145 * submit_bio_wait - submit a bio, and wait until it completes
1146 * @bio: The &struct bio which describes the I/O
1148 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1149 * bio_endio() on failure.
1151 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1152 * result in bio reference to be consumed. The caller must drop the reference
1155 int submit_bio_wait(struct bio
*bio
)
1157 DECLARE_COMPLETION_ONSTACK_MAP(done
,
1158 bio
->bi_bdev
->bd_disk
->lockdep_map
);
1159 unsigned long hang_check
;
1161 bio
->bi_private
= &done
;
1162 bio
->bi_end_io
= submit_bio_wait_endio
;
1163 bio
->bi_opf
|= REQ_SYNC
;
1166 /* Prevent hang_check timer from firing at us during very long I/O */
1167 hang_check
= sysctl_hung_task_timeout_secs
;
1169 while (!wait_for_completion_io_timeout(&done
,
1170 hang_check
* (HZ
/2)))
1173 wait_for_completion_io(&done
);
1175 return blk_status_to_errno(bio
->bi_status
);
1177 EXPORT_SYMBOL(submit_bio_wait
);
1180 * bio_advance - increment/complete a bio by some number of bytes
1181 * @bio: bio to advance
1182 * @bytes: number of bytes to complete
1184 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1185 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1186 * be updated on the last bvec as well.
1188 * @bio will then represent the remaining, uncompleted portion of the io.
1190 void bio_advance(struct bio
*bio
, unsigned bytes
)
1192 if (bio_integrity(bio
))
1193 bio_integrity_advance(bio
, bytes
);
1195 bio_crypt_advance(bio
, bytes
);
1196 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
1198 EXPORT_SYMBOL(bio_advance
);
1200 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
1201 struct bio
*src
, struct bvec_iter
*src_iter
)
1203 struct bio_vec src_bv
, dst_bv
;
1204 void *src_p
, *dst_p
;
1207 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
1208 src_bv
= bio_iter_iovec(src
, *src_iter
);
1209 dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
1211 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1213 src_p
= kmap_atomic(src_bv
.bv_page
);
1214 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1216 memcpy(dst_p
+ dst_bv
.bv_offset
,
1217 src_p
+ src_bv
.bv_offset
,
1220 kunmap_atomic(dst_p
);
1221 kunmap_atomic(src_p
);
1223 flush_dcache_page(dst_bv
.bv_page
);
1225 bio_advance_iter_single(src
, src_iter
, bytes
);
1226 bio_advance_iter_single(dst
, dst_iter
, bytes
);
1229 EXPORT_SYMBOL(bio_copy_data_iter
);
1232 * bio_copy_data - copy contents of data buffers from one bio to another
1234 * @dst: destination bio
1236 * Stops when it reaches the end of either @src or @dst - that is, copies
1237 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1239 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1241 struct bvec_iter src_iter
= src
->bi_iter
;
1242 struct bvec_iter dst_iter
= dst
->bi_iter
;
1244 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1246 EXPORT_SYMBOL(bio_copy_data
);
1248 void bio_free_pages(struct bio
*bio
)
1250 struct bio_vec
*bvec
;
1251 struct bvec_iter_all iter_all
;
1253 bio_for_each_segment_all(bvec
, bio
, iter_all
)
1254 __free_page(bvec
->bv_page
);
1256 EXPORT_SYMBOL(bio_free_pages
);
1259 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1260 * for performing direct-IO in BIOs.
1262 * The problem is that we cannot run set_page_dirty() from interrupt context
1263 * because the required locks are not interrupt-safe. So what we can do is to
1264 * mark the pages dirty _before_ performing IO. And in interrupt context,
1265 * check that the pages are still dirty. If so, fine. If not, redirty them
1266 * in process context.
1268 * We special-case compound pages here: normally this means reads into hugetlb
1269 * pages. The logic in here doesn't really work right for compound pages
1270 * because the VM does not uniformly chase down the head page in all cases.
1271 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1272 * handle them at all. So we skip compound pages here at an early stage.
1274 * Note that this code is very hard to test under normal circumstances because
1275 * direct-io pins the pages with get_user_pages(). This makes
1276 * is_page_cache_freeable return false, and the VM will not clean the pages.
1277 * But other code (eg, flusher threads) could clean the pages if they are mapped
1280 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1281 * deferred bio dirtying paths.
1285 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1287 void bio_set_pages_dirty(struct bio
*bio
)
1289 struct bio_vec
*bvec
;
1290 struct bvec_iter_all iter_all
;
1292 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1293 if (!PageCompound(bvec
->bv_page
))
1294 set_page_dirty_lock(bvec
->bv_page
);
1299 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1300 * If they are, then fine. If, however, some pages are clean then they must
1301 * have been written out during the direct-IO read. So we take another ref on
1302 * the BIO and re-dirty the pages in process context.
1304 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1305 * here on. It will run one put_page() against each page and will run one
1306 * bio_put() against the BIO.
1309 static void bio_dirty_fn(struct work_struct
*work
);
1311 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1312 static DEFINE_SPINLOCK(bio_dirty_lock
);
1313 static struct bio
*bio_dirty_list
;
1316 * This runs in process context
1318 static void bio_dirty_fn(struct work_struct
*work
)
1320 struct bio
*bio
, *next
;
1322 spin_lock_irq(&bio_dirty_lock
);
1323 next
= bio_dirty_list
;
1324 bio_dirty_list
= NULL
;
1325 spin_unlock_irq(&bio_dirty_lock
);
1327 while ((bio
= next
) != NULL
) {
1328 next
= bio
->bi_private
;
1330 bio_release_pages(bio
, true);
1335 void bio_check_pages_dirty(struct bio
*bio
)
1337 struct bio_vec
*bvec
;
1338 unsigned long flags
;
1339 struct bvec_iter_all iter_all
;
1341 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1342 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1346 bio_release_pages(bio
, false);
1350 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1351 bio
->bi_private
= bio_dirty_list
;
1352 bio_dirty_list
= bio
;
1353 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1354 schedule_work(&bio_dirty_work
);
1357 static inline bool bio_remaining_done(struct bio
*bio
)
1360 * If we're not chaining, then ->__bi_remaining is always 1 and
1361 * we always end io on the first invocation.
1363 if (!bio_flagged(bio
, BIO_CHAIN
))
1366 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1368 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1369 bio_clear_flag(bio
, BIO_CHAIN
);
1377 * bio_endio - end I/O on a bio
1381 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1382 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1383 * bio unless they own it and thus know that it has an end_io function.
1385 * bio_endio() can be called several times on a bio that has been chained
1386 * using bio_chain(). The ->bi_end_io() function will only be called the
1387 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1388 * generated if BIO_TRACE_COMPLETION is set.
1390 void bio_endio(struct bio
*bio
)
1393 if (!bio_remaining_done(bio
))
1395 if (!bio_integrity_endio(bio
))
1399 rq_qos_done_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
);
1402 * Need to have a real endio function for chained bios, otherwise
1403 * various corner cases will break (like stacking block devices that
1404 * save/restore bi_end_io) - however, we want to avoid unbounded
1405 * recursion and blowing the stack. Tail call optimization would
1406 * handle this, but compiling with frame pointers also disables
1407 * gcc's sibling call optimization.
1409 if (bio
->bi_end_io
== bio_chain_endio
) {
1410 bio
= __bio_chain_endio(bio
);
1414 if (bio
->bi_bdev
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1415 trace_block_bio_complete(bio
->bi_bdev
->bd_disk
->queue
, bio
);
1416 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1419 blk_throtl_bio_endio(bio
);
1420 /* release cgroup info */
1423 bio
->bi_end_io(bio
);
1425 EXPORT_SYMBOL(bio_endio
);
1428 * bio_split - split a bio
1429 * @bio: bio to split
1430 * @sectors: number of sectors to split from the front of @bio
1432 * @bs: bio set to allocate from
1434 * Allocates and returns a new bio which represents @sectors from the start of
1435 * @bio, and updates @bio to represent the remaining sectors.
1437 * Unless this is a discard request the newly allocated bio will point
1438 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1439 * neither @bio nor @bs are freed before the split bio.
1441 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1442 gfp_t gfp
, struct bio_set
*bs
)
1446 BUG_ON(sectors
<= 0);
1447 BUG_ON(sectors
>= bio_sectors(bio
));
1449 /* Zone append commands cannot be split */
1450 if (WARN_ON_ONCE(bio_op(bio
) == REQ_OP_ZONE_APPEND
))
1453 split
= bio_clone_fast(bio
, gfp
, bs
);
1457 split
->bi_iter
.bi_size
= sectors
<< 9;
1459 if (bio_integrity(split
))
1460 bio_integrity_trim(split
);
1462 bio_advance(bio
, split
->bi_iter
.bi_size
);
1464 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1465 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1469 EXPORT_SYMBOL(bio_split
);
1472 * bio_trim - trim a bio
1474 * @offset: number of sectors to trim from the front of @bio
1475 * @size: size we want to trim @bio to, in sectors
1477 void bio_trim(struct bio
*bio
, int offset
, int size
)
1479 /* 'bio' is a cloned bio which we need to trim to match
1480 * the given offset and size.
1484 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1487 bio_advance(bio
, offset
<< 9);
1488 bio
->bi_iter
.bi_size
= size
;
1490 if (bio_integrity(bio
))
1491 bio_integrity_trim(bio
);
1494 EXPORT_SYMBOL_GPL(bio_trim
);
1497 * create memory pools for biovec's in a bio_set.
1498 * use the global biovec slabs created for general use.
1500 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1502 struct biovec_slab
*bp
= bvec_slabs
+ ARRAY_SIZE(bvec_slabs
) - 1;
1504 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1508 * bioset_exit - exit a bioset initialized with bioset_init()
1510 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1513 void bioset_exit(struct bio_set
*bs
)
1515 if (bs
->rescue_workqueue
)
1516 destroy_workqueue(bs
->rescue_workqueue
);
1517 bs
->rescue_workqueue
= NULL
;
1519 mempool_exit(&bs
->bio_pool
);
1520 mempool_exit(&bs
->bvec_pool
);
1522 bioset_integrity_free(bs
);
1525 bs
->bio_slab
= NULL
;
1527 EXPORT_SYMBOL(bioset_exit
);
1530 * bioset_init - Initialize a bio_set
1531 * @bs: pool to initialize
1532 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1533 * @front_pad: Number of bytes to allocate in front of the returned bio
1534 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1535 * and %BIOSET_NEED_RESCUER
1538 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1539 * to ask for a number of bytes to be allocated in front of the bio.
1540 * Front pad allocation is useful for embedding the bio inside
1541 * another structure, to avoid allocating extra data to go with the bio.
1542 * Note that the bio must be embedded at the END of that structure always,
1543 * or things will break badly.
1544 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1545 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1546 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1547 * dispatch queued requests when the mempool runs out of space.
1550 int bioset_init(struct bio_set
*bs
,
1551 unsigned int pool_size
,
1552 unsigned int front_pad
,
1555 bs
->front_pad
= front_pad
;
1556 if (flags
& BIOSET_NEED_BVECS
)
1557 bs
->back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1561 spin_lock_init(&bs
->rescue_lock
);
1562 bio_list_init(&bs
->rescue_list
);
1563 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1565 bs
->bio_slab
= bio_find_or_create_slab(bs
);
1569 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
1572 if ((flags
& BIOSET_NEED_BVECS
) &&
1573 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
1576 if (!(flags
& BIOSET_NEED_RESCUER
))
1579 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1580 if (!bs
->rescue_workqueue
)
1588 EXPORT_SYMBOL(bioset_init
);
1591 * Initialize and setup a new bio_set, based on the settings from
1594 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
1599 if (src
->bvec_pool
.min_nr
)
1600 flags
|= BIOSET_NEED_BVECS
;
1601 if (src
->rescue_workqueue
)
1602 flags
|= BIOSET_NEED_RESCUER
;
1604 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
1606 EXPORT_SYMBOL(bioset_init_from_src
);
1608 static int __init
init_bio(void)
1612 bio_integrity_init();
1614 for (i
= 0; i
< ARRAY_SIZE(bvec_slabs
); i
++) {
1615 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1617 bvs
->slab
= kmem_cache_create(bvs
->name
,
1618 bvs
->nr_vecs
* sizeof(struct bio_vec
), 0,
1619 SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1622 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
1623 panic("bio: can't allocate bios\n");
1625 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
1626 panic("bio: can't create integrity pool\n");
1630 subsys_initcall(init_bio
);