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/highmem.h>
19 #include <linux/sched/sysctl.h>
20 #include <linux/blk-crypto.h>
21 #include <linux/xarray.h>
23 #include <trace/events/block.h>
25 #include "blk-rq-qos.h"
26 #include "blk-cgroup.h"
28 struct bio_alloc_cache
{
29 struct bio
*free_list
;
33 static struct biovec_slab
{
36 struct kmem_cache
*slab
;
37 } bvec_slabs
[] __read_mostly
= {
38 { .nr_vecs
= 16, .name
= "biovec-16" },
39 { .nr_vecs
= 64, .name
= "biovec-64" },
40 { .nr_vecs
= 128, .name
= "biovec-128" },
41 { .nr_vecs
= BIO_MAX_VECS
, .name
= "biovec-max" },
44 static struct biovec_slab
*biovec_slab(unsigned short nr_vecs
)
47 /* smaller bios use inline vecs */
49 return &bvec_slabs
[0];
51 return &bvec_slabs
[1];
53 return &bvec_slabs
[2];
54 case 129 ... BIO_MAX_VECS
:
55 return &bvec_slabs
[3];
63 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
64 * IO code that does not need private memory pools.
66 struct bio_set fs_bio_set
;
67 EXPORT_SYMBOL(fs_bio_set
);
70 * Our slab pool management
73 struct kmem_cache
*slab
;
74 unsigned int slab_ref
;
75 unsigned int slab_size
;
78 static DEFINE_MUTEX(bio_slab_lock
);
79 static DEFINE_XARRAY(bio_slabs
);
81 static struct bio_slab
*create_bio_slab(unsigned int size
)
83 struct bio_slab
*bslab
= kzalloc(sizeof(*bslab
), GFP_KERNEL
);
88 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", size
);
89 bslab
->slab
= kmem_cache_create(bslab
->name
, size
,
90 ARCH_KMALLOC_MINALIGN
,
91 SLAB_HWCACHE_ALIGN
| SLAB_TYPESAFE_BY_RCU
, NULL
);
96 bslab
->slab_size
= size
;
98 if (!xa_err(xa_store(&bio_slabs
, size
, bslab
, GFP_KERNEL
)))
101 kmem_cache_destroy(bslab
->slab
);
108 static inline unsigned int bs_bio_slab_size(struct bio_set
*bs
)
110 return bs
->front_pad
+ sizeof(struct bio
) + bs
->back_pad
;
113 static struct kmem_cache
*bio_find_or_create_slab(struct bio_set
*bs
)
115 unsigned int size
= bs_bio_slab_size(bs
);
116 struct bio_slab
*bslab
;
118 mutex_lock(&bio_slab_lock
);
119 bslab
= xa_load(&bio_slabs
, size
);
123 bslab
= create_bio_slab(size
);
124 mutex_unlock(&bio_slab_lock
);
131 static void bio_put_slab(struct bio_set
*bs
)
133 struct bio_slab
*bslab
= NULL
;
134 unsigned int slab_size
= bs_bio_slab_size(bs
);
136 mutex_lock(&bio_slab_lock
);
138 bslab
= xa_load(&bio_slabs
, slab_size
);
139 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
142 WARN_ON_ONCE(bslab
->slab
!= bs
->bio_slab
);
144 WARN_ON(!bslab
->slab_ref
);
146 if (--bslab
->slab_ref
)
149 xa_erase(&bio_slabs
, slab_size
);
151 kmem_cache_destroy(bslab
->slab
);
155 mutex_unlock(&bio_slab_lock
);
158 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned short nr_vecs
)
160 BUG_ON(nr_vecs
> BIO_MAX_VECS
);
162 if (nr_vecs
== BIO_MAX_VECS
)
163 mempool_free(bv
, pool
);
164 else if (nr_vecs
> BIO_INLINE_VECS
)
165 kmem_cache_free(biovec_slab(nr_vecs
)->slab
, bv
);
169 * Make the first allocation restricted and don't dump info on allocation
170 * failures, since we'll fall back to the mempool in case of failure.
172 static inline gfp_t
bvec_alloc_gfp(gfp_t gfp
)
174 return (gfp
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
)) |
175 __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
178 struct bio_vec
*bvec_alloc(mempool_t
*pool
, unsigned short *nr_vecs
,
181 struct biovec_slab
*bvs
= biovec_slab(*nr_vecs
);
183 if (WARN_ON_ONCE(!bvs
))
187 * Upgrade the nr_vecs request to take full advantage of the allocation.
188 * We also rely on this in the bvec_free path.
190 *nr_vecs
= bvs
->nr_vecs
;
193 * Try a slab allocation first for all smaller allocations. If that
194 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
195 * The mempool is sized to handle up to BIO_MAX_VECS entries.
197 if (*nr_vecs
< BIO_MAX_VECS
) {
200 bvl
= kmem_cache_alloc(bvs
->slab
, bvec_alloc_gfp(gfp_mask
));
201 if (likely(bvl
) || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
203 *nr_vecs
= BIO_MAX_VECS
;
206 return mempool_alloc(pool
, gfp_mask
);
209 void bio_uninit(struct bio
*bio
)
211 #ifdef CONFIG_BLK_CGROUP
213 blkg_put(bio
->bi_blkg
);
217 if (bio_integrity(bio
))
218 bio_integrity_free(bio
);
220 bio_crypt_free_ctx(bio
);
222 EXPORT_SYMBOL(bio_uninit
);
224 static void bio_free(struct bio
*bio
)
226 struct bio_set
*bs
= bio
->bi_pool
;
232 bvec_free(&bs
->bvec_pool
, bio
->bi_io_vec
, bio
->bi_max_vecs
);
233 mempool_free(p
- bs
->front_pad
, &bs
->bio_pool
);
237 * Users of this function have their own bio allocation. Subsequently,
238 * they must remember to pair any call to bio_init() with bio_uninit()
239 * when IO has completed, or when the bio is released.
241 void bio_init(struct bio
*bio
, struct block_device
*bdev
, struct bio_vec
*table
,
242 unsigned short max_vecs
, unsigned int opf
)
250 bio
->bi_iter
.bi_sector
= 0;
251 bio
->bi_iter
.bi_size
= 0;
252 bio
->bi_iter
.bi_idx
= 0;
253 bio
->bi_iter
.bi_bvec_done
= 0;
254 bio
->bi_end_io
= NULL
;
255 bio
->bi_private
= NULL
;
256 #ifdef CONFIG_BLK_CGROUP
258 bio
->bi_issue
.value
= 0;
260 bio_associate_blkg(bio
);
261 #ifdef CONFIG_BLK_CGROUP_IOCOST
262 bio
->bi_iocost_cost
= 0;
265 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
266 bio
->bi_crypt_context
= NULL
;
268 #ifdef CONFIG_BLK_DEV_INTEGRITY
269 bio
->bi_integrity
= NULL
;
273 atomic_set(&bio
->__bi_remaining
, 1);
274 atomic_set(&bio
->__bi_cnt
, 1);
275 bio
->bi_cookie
= BLK_QC_T_NONE
;
277 bio
->bi_max_vecs
= max_vecs
;
278 bio
->bi_io_vec
= table
;
281 EXPORT_SYMBOL(bio_init
);
284 * bio_reset - reinitialize a bio
286 * @bdev: block device to use the bio for
287 * @opf: operation and flags for bio
290 * After calling bio_reset(), @bio will be in the same state as a freshly
291 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
292 * preserved are the ones that are initialized by bio_alloc_bioset(). See
293 * comment in struct bio.
295 void bio_reset(struct bio
*bio
, struct block_device
*bdev
, unsigned int opf
)
298 memset(bio
, 0, BIO_RESET_BYTES
);
299 atomic_set(&bio
->__bi_remaining
, 1);
302 bio_associate_blkg(bio
);
305 EXPORT_SYMBOL(bio_reset
);
307 static struct bio
*__bio_chain_endio(struct bio
*bio
)
309 struct bio
*parent
= bio
->bi_private
;
311 if (bio
->bi_status
&& !parent
->bi_status
)
312 parent
->bi_status
= bio
->bi_status
;
317 static void bio_chain_endio(struct bio
*bio
)
319 bio_endio(__bio_chain_endio(bio
));
323 * bio_chain - chain bio completions
324 * @bio: the target bio
325 * @parent: the parent bio of @bio
327 * The caller won't have a bi_end_io called when @bio completes - instead,
328 * @parent's bi_end_io won't be called until both @parent and @bio have
329 * completed; the chained bio will also be freed when it completes.
331 * The caller must not set bi_private or bi_end_io in @bio.
333 void bio_chain(struct bio
*bio
, struct bio
*parent
)
335 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
337 bio
->bi_private
= parent
;
338 bio
->bi_end_io
= bio_chain_endio
;
339 bio_inc_remaining(parent
);
341 EXPORT_SYMBOL(bio_chain
);
343 struct bio
*blk_next_bio(struct bio
*bio
, struct block_device
*bdev
,
344 unsigned int nr_pages
, unsigned int opf
, gfp_t gfp
)
346 struct bio
*new = bio_alloc(bdev
, nr_pages
, opf
, gfp
);
355 EXPORT_SYMBOL_GPL(blk_next_bio
);
357 static void bio_alloc_rescue(struct work_struct
*work
)
359 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
363 spin_lock(&bs
->rescue_lock
);
364 bio
= bio_list_pop(&bs
->rescue_list
);
365 spin_unlock(&bs
->rescue_lock
);
370 submit_bio_noacct(bio
);
374 static void punt_bios_to_rescuer(struct bio_set
*bs
)
376 struct bio_list punt
, nopunt
;
379 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
382 * In order to guarantee forward progress we must punt only bios that
383 * were allocated from this bio_set; otherwise, if there was a bio on
384 * there for a stacking driver higher up in the stack, processing it
385 * could require allocating bios from this bio_set, and doing that from
386 * our own rescuer would be bad.
388 * Since bio lists are singly linked, pop them all instead of trying to
389 * remove from the middle of the list:
392 bio_list_init(&punt
);
393 bio_list_init(&nopunt
);
395 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
396 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
397 current
->bio_list
[0] = nopunt
;
399 bio_list_init(&nopunt
);
400 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
401 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
402 current
->bio_list
[1] = nopunt
;
404 spin_lock(&bs
->rescue_lock
);
405 bio_list_merge(&bs
->rescue_list
, &punt
);
406 spin_unlock(&bs
->rescue_lock
);
408 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
411 static struct bio
*bio_alloc_percpu_cache(struct block_device
*bdev
,
412 unsigned short nr_vecs
, unsigned int opf
, gfp_t gfp
,
415 struct bio_alloc_cache
*cache
;
418 cache
= per_cpu_ptr(bs
->cache
, get_cpu());
419 if (!cache
->free_list
) {
423 bio
= cache
->free_list
;
424 cache
->free_list
= bio
->bi_next
;
428 bio_init(bio
, bdev
, nr_vecs
? bio
->bi_inline_vecs
: NULL
, nr_vecs
, opf
);
434 * bio_alloc_bioset - allocate a bio for I/O
435 * @bdev: block device to allocate the bio for (can be %NULL)
436 * @nr_vecs: number of bvecs to pre-allocate
437 * @opf: operation and flags for bio
438 * @gfp_mask: the GFP_* mask given to the slab allocator
439 * @bs: the bio_set to allocate from.
441 * Allocate a bio from the mempools in @bs.
443 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
444 * allocate a bio. This is due to the mempool guarantees. To make this work,
445 * callers must never allocate more than 1 bio at a time from the general pool.
446 * Callers that need to allocate more than 1 bio must always submit the
447 * previously allocated bio for IO before attempting to allocate a new one.
448 * Failure to do so can cause deadlocks under memory pressure.
450 * Note that when running under submit_bio_noacct() (i.e. any block driver),
451 * bios are not submitted until after you return - see the code in
452 * submit_bio_noacct() that converts recursion into iteration, to prevent
455 * This would normally mean allocating multiple bios under submit_bio_noacct()
456 * would be susceptible to deadlocks, but we have
457 * deadlock avoidance code that resubmits any blocked bios from a rescuer
460 * However, we do not guarantee forward progress for allocations from other
461 * mempools. Doing multiple allocations from the same mempool under
462 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
463 * for per bio allocations.
465 * If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done from process
466 * context, not hard/soft IRQ.
468 * Returns: Pointer to new bio on success, NULL on failure.
470 struct bio
*bio_alloc_bioset(struct block_device
*bdev
, unsigned short nr_vecs
,
471 unsigned int opf
, gfp_t gfp_mask
,
474 gfp_t saved_gfp
= gfp_mask
;
478 /* should not use nobvec bioset for nr_vecs > 0 */
479 if (WARN_ON_ONCE(!mempool_initialized(&bs
->bvec_pool
) && nr_vecs
> 0))
482 if (opf
& REQ_ALLOC_CACHE
) {
483 if (bs
->cache
&& nr_vecs
<= BIO_INLINE_VECS
) {
484 bio
= bio_alloc_percpu_cache(bdev
, nr_vecs
, opf
,
489 * No cached bio available, bio returned below marked with
490 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
493 opf
&= ~REQ_ALLOC_CACHE
;
498 * submit_bio_noacct() converts recursion to iteration; this means if
499 * we're running beneath it, any bios we allocate and submit will not be
500 * submitted (and thus freed) until after we return.
502 * This exposes us to a potential deadlock if we allocate multiple bios
503 * from the same bio_set() while running underneath submit_bio_noacct().
504 * If we were to allocate multiple bios (say a stacking block driver
505 * that was splitting bios), we would deadlock if we exhausted the
508 * We solve this, and guarantee forward progress, with a rescuer
509 * workqueue per bio_set. If we go to allocate and there are bios on
510 * current->bio_list, we first try the allocation without
511 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
512 * blocking to the rescuer workqueue before we retry with the original
515 if (current
->bio_list
&&
516 (!bio_list_empty(¤t
->bio_list
[0]) ||
517 !bio_list_empty(¤t
->bio_list
[1])) &&
518 bs
->rescue_workqueue
)
519 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
521 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
522 if (!p
&& gfp_mask
!= saved_gfp
) {
523 punt_bios_to_rescuer(bs
);
524 gfp_mask
= saved_gfp
;
525 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
530 bio
= p
+ bs
->front_pad
;
531 if (nr_vecs
> BIO_INLINE_VECS
) {
532 struct bio_vec
*bvl
= NULL
;
534 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_vecs
, gfp_mask
);
535 if (!bvl
&& gfp_mask
!= saved_gfp
) {
536 punt_bios_to_rescuer(bs
);
537 gfp_mask
= saved_gfp
;
538 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_vecs
, gfp_mask
);
543 bio_init(bio
, bdev
, bvl
, nr_vecs
, opf
);
544 } else if (nr_vecs
) {
545 bio_init(bio
, bdev
, bio
->bi_inline_vecs
, BIO_INLINE_VECS
, opf
);
547 bio_init(bio
, bdev
, NULL
, 0, opf
);
554 mempool_free(p
, &bs
->bio_pool
);
557 EXPORT_SYMBOL(bio_alloc_bioset
);
560 * bio_kmalloc - kmalloc a bio
561 * @nr_vecs: number of bio_vecs to allocate
562 * @gfp_mask: the GFP_* mask given to the slab allocator
564 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
565 * using bio_init() before use. To free a bio returned from this function use
566 * kfree() after calling bio_uninit(). A bio returned from this function can
567 * be reused by calling bio_uninit() before calling bio_init() again.
569 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
570 * function are not backed by a mempool can can fail. Do not use this function
571 * for allocations in the file system I/O path.
573 * Returns: Pointer to new bio on success, NULL on failure.
575 struct bio
*bio_kmalloc(unsigned short nr_vecs
, gfp_t gfp_mask
)
579 if (nr_vecs
> UIO_MAXIOV
)
581 return kmalloc(struct_size(bio
, bi_inline_vecs
, nr_vecs
), gfp_mask
);
583 EXPORT_SYMBOL(bio_kmalloc
);
585 void zero_fill_bio(struct bio
*bio
)
588 struct bvec_iter iter
;
590 bio_for_each_segment(bv
, bio
, iter
)
593 EXPORT_SYMBOL(zero_fill_bio
);
596 * bio_truncate - truncate the bio to small size of @new_size
597 * @bio: the bio to be truncated
598 * @new_size: new size for truncating the bio
601 * Truncate the bio to new size of @new_size. If bio_op(bio) is
602 * REQ_OP_READ, zero the truncated part. This function should only
603 * be used for handling corner cases, such as bio eod.
605 static void bio_truncate(struct bio
*bio
, unsigned new_size
)
608 struct bvec_iter iter
;
609 unsigned int done
= 0;
610 bool truncated
= false;
612 if (new_size
>= bio
->bi_iter
.bi_size
)
615 if (bio_op(bio
) != REQ_OP_READ
)
618 bio_for_each_segment(bv
, bio
, iter
) {
619 if (done
+ bv
.bv_len
> new_size
) {
623 offset
= new_size
- done
;
626 zero_user(bv
.bv_page
, bv
.bv_offset
+ offset
,
635 * Don't touch bvec table here and make it really immutable, since
636 * fs bio user has to retrieve all pages via bio_for_each_segment_all
637 * in its .end_bio() callback.
639 * It is enough to truncate bio by updating .bi_size since we can make
640 * correct bvec with the updated .bi_size for drivers.
642 bio
->bi_iter
.bi_size
= new_size
;
646 * guard_bio_eod - truncate a BIO to fit the block device
647 * @bio: bio to truncate
649 * This allows us to do IO even on the odd last sectors of a device, even if the
650 * block size is some multiple of the physical sector size.
652 * We'll just truncate the bio to the size of the device, and clear the end of
653 * the buffer head manually. Truly out-of-range accesses will turn into actual
654 * I/O errors, this only handles the "we need to be able to do I/O at the final
657 void guard_bio_eod(struct bio
*bio
)
659 sector_t maxsector
= bdev_nr_sectors(bio
->bi_bdev
);
665 * If the *whole* IO is past the end of the device,
666 * let it through, and the IO layer will turn it into
669 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
672 maxsector
-= bio
->bi_iter
.bi_sector
;
673 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
676 bio_truncate(bio
, maxsector
<< 9);
679 #define ALLOC_CACHE_MAX 512
680 #define ALLOC_CACHE_SLACK 64
682 static void bio_alloc_cache_prune(struct bio_alloc_cache
*cache
,
688 while ((bio
= cache
->free_list
) != NULL
) {
689 cache
->free_list
= bio
->bi_next
;
697 static int bio_cpu_dead(unsigned int cpu
, struct hlist_node
*node
)
701 bs
= hlist_entry_safe(node
, struct bio_set
, cpuhp_dead
);
703 struct bio_alloc_cache
*cache
= per_cpu_ptr(bs
->cache
, cpu
);
705 bio_alloc_cache_prune(cache
, -1U);
710 static void bio_alloc_cache_destroy(struct bio_set
*bs
)
717 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD
, &bs
->cpuhp_dead
);
718 for_each_possible_cpu(cpu
) {
719 struct bio_alloc_cache
*cache
;
721 cache
= per_cpu_ptr(bs
->cache
, cpu
);
722 bio_alloc_cache_prune(cache
, -1U);
724 free_percpu(bs
->cache
);
729 * bio_put - release a reference to a bio
730 * @bio: bio to release reference to
733 * Put a reference to a &struct bio, either one you have gotten with
734 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
736 void bio_put(struct bio
*bio
)
738 if (unlikely(bio_flagged(bio
, BIO_REFFED
))) {
739 BUG_ON(!atomic_read(&bio
->__bi_cnt
));
740 if (!atomic_dec_and_test(&bio
->__bi_cnt
))
744 if (bio
->bi_opf
& REQ_ALLOC_CACHE
) {
745 struct bio_alloc_cache
*cache
;
748 cache
= per_cpu_ptr(bio
->bi_pool
->cache
, get_cpu());
749 bio
->bi_next
= cache
->free_list
;
750 cache
->free_list
= bio
;
751 if (++cache
->nr
> ALLOC_CACHE_MAX
+ ALLOC_CACHE_SLACK
)
752 bio_alloc_cache_prune(cache
, ALLOC_CACHE_SLACK
);
758 EXPORT_SYMBOL(bio_put
);
760 static int __bio_clone(struct bio
*bio
, struct bio
*bio_src
, gfp_t gfp
)
762 bio_set_flag(bio
, BIO_CLONED
);
763 if (bio_flagged(bio_src
, BIO_THROTTLED
))
764 bio_set_flag(bio
, BIO_THROTTLED
);
765 bio
->bi_ioprio
= bio_src
->bi_ioprio
;
766 bio
->bi_iter
= bio_src
->bi_iter
;
769 if (bio
->bi_bdev
== bio_src
->bi_bdev
&&
770 bio_flagged(bio_src
, BIO_REMAPPED
))
771 bio_set_flag(bio
, BIO_REMAPPED
);
772 bio_clone_blkg_association(bio
, bio_src
);
775 if (bio_crypt_clone(bio
, bio_src
, gfp
) < 0)
777 if (bio_integrity(bio_src
) &&
778 bio_integrity_clone(bio
, bio_src
, gfp
) < 0)
784 * bio_alloc_clone - clone a bio that shares the original bio's biovec
785 * @bdev: block_device to clone onto
786 * @bio_src: bio to clone from
787 * @gfp: allocation priority
788 * @bs: bio_set to allocate from
790 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
791 * bio, but not the actual data it points to.
793 * The caller must ensure that the return bio is not freed before @bio_src.
795 struct bio
*bio_alloc_clone(struct block_device
*bdev
, struct bio
*bio_src
,
796 gfp_t gfp
, struct bio_set
*bs
)
800 bio
= bio_alloc_bioset(bdev
, 0, bio_src
->bi_opf
, gfp
, bs
);
804 if (__bio_clone(bio
, bio_src
, gfp
) < 0) {
808 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
812 EXPORT_SYMBOL(bio_alloc_clone
);
815 * bio_init_clone - clone a bio that shares the original bio's biovec
816 * @bdev: block_device to clone onto
817 * @bio: bio to clone into
818 * @bio_src: bio to clone from
819 * @gfp: allocation priority
821 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
822 * The caller owns the returned bio, but not the actual data it points to.
824 * The caller must ensure that @bio_src is not freed before @bio.
826 int bio_init_clone(struct block_device
*bdev
, struct bio
*bio
,
827 struct bio
*bio_src
, gfp_t gfp
)
831 bio_init(bio
, bdev
, bio_src
->bi_io_vec
, 0, bio_src
->bi_opf
);
832 ret
= __bio_clone(bio
, bio_src
, gfp
);
837 EXPORT_SYMBOL(bio_init_clone
);
840 * bio_full - check if the bio is full
842 * @len: length of one segment to be added
844 * Return true if @bio is full and one segment with @len bytes can't be
845 * added to the bio, otherwise return false
847 static inline bool bio_full(struct bio
*bio
, unsigned len
)
849 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
851 if (bio
->bi_iter
.bi_size
> UINT_MAX
- len
)
856 static inline bool page_is_mergeable(const struct bio_vec
*bv
,
857 struct page
*page
, unsigned int len
, unsigned int off
,
860 size_t bv_end
= bv
->bv_offset
+ bv
->bv_len
;
861 phys_addr_t vec_end_addr
= page_to_phys(bv
->bv_page
) + bv_end
- 1;
862 phys_addr_t page_addr
= page_to_phys(page
);
864 if (vec_end_addr
+ 1 != page_addr
+ off
)
866 if (xen_domain() && !xen_biovec_phys_mergeable(bv
, page
))
869 *same_page
= ((vec_end_addr
& PAGE_MASK
) == page_addr
);
872 return (bv
->bv_page
+ bv_end
/ PAGE_SIZE
) == (page
+ off
/ PAGE_SIZE
);
876 * __bio_try_merge_page - try appending data to an existing bvec.
877 * @bio: destination bio
878 * @page: start page to add
879 * @len: length of the data to add
880 * @off: offset of the data relative to @page
881 * @same_page: return if the segment has been merged inside the same page
883 * Try to add the data at @page + @off to the last bvec of @bio. This is a
884 * useful optimisation for file systems with a block size smaller than the
887 * Warn if (@len, @off) crosses pages in case that @same_page is true.
889 * Return %true on success or %false on failure.
891 static bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
892 unsigned int len
, unsigned int off
, bool *same_page
)
894 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
897 if (bio
->bi_vcnt
> 0) {
898 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
900 if (page_is_mergeable(bv
, page
, len
, off
, same_page
)) {
901 if (bio
->bi_iter
.bi_size
> UINT_MAX
- len
) {
906 bio
->bi_iter
.bi_size
+= len
;
914 * Try to merge a page into a segment, while obeying the hardware segment
915 * size limit. This is not for normal read/write bios, but for passthrough
916 * or Zone Append operations that we can't split.
918 static bool bio_try_merge_hw_seg(struct request_queue
*q
, struct bio
*bio
,
919 struct page
*page
, unsigned len
,
920 unsigned offset
, bool *same_page
)
922 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
923 unsigned long mask
= queue_segment_boundary(q
);
924 phys_addr_t addr1
= page_to_phys(bv
->bv_page
) + bv
->bv_offset
;
925 phys_addr_t addr2
= page_to_phys(page
) + offset
+ len
- 1;
927 if ((addr1
| mask
) != (addr2
| mask
))
929 if (bv
->bv_len
+ len
> queue_max_segment_size(q
))
931 return __bio_try_merge_page(bio
, page
, len
, offset
, same_page
);
935 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
936 * @q: the target queue
937 * @bio: destination bio
939 * @len: vec entry length
940 * @offset: vec entry offset
941 * @max_sectors: maximum number of sectors that can be added
942 * @same_page: return if the segment has been merged inside the same page
944 * Add a page to a bio while respecting the hardware max_sectors, max_segment
945 * and gap limitations.
947 int bio_add_hw_page(struct request_queue
*q
, struct bio
*bio
,
948 struct page
*page
, unsigned int len
, unsigned int offset
,
949 unsigned int max_sectors
, bool *same_page
)
951 struct bio_vec
*bvec
;
953 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
956 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
959 if (bio
->bi_vcnt
> 0) {
960 if (bio_try_merge_hw_seg(q
, bio
, page
, len
, offset
, same_page
))
964 * If the queue doesn't support SG gaps and adding this segment
965 * would create a gap, disallow it.
967 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
968 if (bvec_gap_to_prev(q
, bvec
, offset
))
972 if (bio_full(bio
, len
))
975 if (bio
->bi_vcnt
>= queue_max_segments(q
))
978 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
979 bvec
->bv_page
= page
;
981 bvec
->bv_offset
= offset
;
983 bio
->bi_iter
.bi_size
+= len
;
988 * bio_add_pc_page - attempt to add page to passthrough bio
989 * @q: the target queue
990 * @bio: destination bio
992 * @len: vec entry length
993 * @offset: vec entry offset
995 * Attempt to add a page to the bio_vec maplist. This can fail for a
996 * number of reasons, such as the bio being full or target block device
997 * limitations. The target block device must allow bio's up to PAGE_SIZE,
998 * so it is always possible to add a single page to an empty bio.
1000 * This should only be used by passthrough bios.
1002 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
,
1003 struct page
*page
, unsigned int len
, unsigned int offset
)
1005 bool same_page
= false;
1006 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
1007 queue_max_hw_sectors(q
), &same_page
);
1009 EXPORT_SYMBOL(bio_add_pc_page
);
1012 * bio_add_zone_append_page - attempt to add page to zone-append bio
1013 * @bio: destination bio
1014 * @page: page to add
1015 * @len: vec entry length
1016 * @offset: vec entry offset
1018 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1019 * for a zone-append request. This can fail for a number of reasons, such as the
1020 * bio being full or the target block device is not a zoned block device or
1021 * other limitations of the target block device. The target block device must
1022 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1025 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1027 int bio_add_zone_append_page(struct bio
*bio
, struct page
*page
,
1028 unsigned int len
, unsigned int offset
)
1030 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1031 bool same_page
= false;
1033 if (WARN_ON_ONCE(bio_op(bio
) != REQ_OP_ZONE_APPEND
))
1036 if (WARN_ON_ONCE(!blk_queue_is_zoned(q
)))
1039 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
1040 queue_max_zone_append_sectors(q
), &same_page
);
1042 EXPORT_SYMBOL_GPL(bio_add_zone_append_page
);
1045 * __bio_add_page - add page(s) to a bio in a new segment
1046 * @bio: destination bio
1047 * @page: start page to add
1048 * @len: length of the data to add, may cross pages
1049 * @off: offset of the data relative to @page, may cross pages
1051 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1052 * that @bio has space for another bvec.
1054 void __bio_add_page(struct bio
*bio
, struct page
*page
,
1055 unsigned int len
, unsigned int off
)
1057 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
1059 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
1060 WARN_ON_ONCE(bio_full(bio
, len
));
1063 bv
->bv_offset
= off
;
1066 bio
->bi_iter
.bi_size
+= len
;
1069 if (!bio_flagged(bio
, BIO_WORKINGSET
) && unlikely(PageWorkingset(page
)))
1070 bio_set_flag(bio
, BIO_WORKINGSET
);
1072 EXPORT_SYMBOL_GPL(__bio_add_page
);
1075 * bio_add_page - attempt to add page(s) to bio
1076 * @bio: destination bio
1077 * @page: start page to add
1078 * @len: vec entry length, may cross pages
1079 * @offset: vec entry offset relative to @page, may cross pages
1081 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1082 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1084 int bio_add_page(struct bio
*bio
, struct page
*page
,
1085 unsigned int len
, unsigned int offset
)
1087 bool same_page
= false;
1089 if (!__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1090 if (bio_full(bio
, len
))
1092 __bio_add_page(bio
, page
, len
, offset
);
1096 EXPORT_SYMBOL(bio_add_page
);
1099 * bio_add_folio - Attempt to add part of a folio to a bio.
1100 * @bio: BIO to add to.
1101 * @folio: Folio to add.
1102 * @len: How many bytes from the folio to add.
1103 * @off: First byte in this folio to add.
1105 * Filesystems that use folios can call this function instead of calling
1106 * bio_add_page() for each page in the folio. If @off is bigger than
1107 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1108 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1110 * Return: Whether the addition was successful.
1112 bool bio_add_folio(struct bio
*bio
, struct folio
*folio
, size_t len
,
1115 if (len
> UINT_MAX
|| off
> UINT_MAX
)
1117 return bio_add_page(bio
, &folio
->page
, len
, off
) > 0;
1120 void __bio_release_pages(struct bio
*bio
, bool mark_dirty
)
1122 struct bvec_iter_all iter_all
;
1123 struct bio_vec
*bvec
;
1125 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1126 if (mark_dirty
&& !PageCompound(bvec
->bv_page
))
1127 set_page_dirty_lock(bvec
->bv_page
);
1128 put_page(bvec
->bv_page
);
1131 EXPORT_SYMBOL_GPL(__bio_release_pages
);
1133 void bio_iov_bvec_set(struct bio
*bio
, struct iov_iter
*iter
)
1135 size_t size
= iov_iter_count(iter
);
1137 WARN_ON_ONCE(bio
->bi_max_vecs
);
1139 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
) {
1140 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1141 size_t max_sectors
= queue_max_zone_append_sectors(q
);
1143 size
= min(size
, max_sectors
<< SECTOR_SHIFT
);
1146 bio
->bi_vcnt
= iter
->nr_segs
;
1147 bio
->bi_io_vec
= (struct bio_vec
*)iter
->bvec
;
1148 bio
->bi_iter
.bi_bvec_done
= iter
->iov_offset
;
1149 bio
->bi_iter
.bi_size
= size
;
1150 bio_set_flag(bio
, BIO_NO_PAGE_REF
);
1151 bio_set_flag(bio
, BIO_CLONED
);
1154 static void bio_put_pages(struct page
**pages
, size_t size
, size_t off
)
1156 size_t i
, nr
= DIV_ROUND_UP(size
+ (off
& ~PAGE_MASK
), PAGE_SIZE
);
1158 for (i
= 0; i
< nr
; i
++)
1162 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1165 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1166 * @bio: bio to add pages to
1167 * @iter: iov iterator describing the region to be mapped
1169 * Pins pages from *iter and appends them to @bio's bvec array. The
1170 * pages will have to be released using put_page() when done.
1171 * For multi-segment *iter, this function only adds pages from the
1172 * next non-empty segment of the iov iterator.
1174 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1176 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1177 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1178 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1179 struct page
**pages
= (struct page
**)bv
;
1180 bool same_page
= false;
1186 * Move page array up in the allocated memory for the bio vecs as far as
1187 * possible so that we can start filling biovecs from the beginning
1188 * without overwriting the temporary page array.
1190 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1191 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1193 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1194 if (unlikely(size
<= 0))
1195 return size
? size
: -EFAULT
;
1197 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1198 struct page
*page
= pages
[i
];
1200 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1202 if (__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1206 if (WARN_ON_ONCE(bio_full(bio
, len
))) {
1207 bio_put_pages(pages
+ i
, left
, offset
);
1210 __bio_add_page(bio
, page
, len
, offset
);
1215 iov_iter_advance(iter
, size
);
1219 static int __bio_iov_append_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1221 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1222 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1223 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
1224 unsigned int max_append_sectors
= queue_max_zone_append_sectors(q
);
1225 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1226 struct page
**pages
= (struct page
**)bv
;
1232 if (WARN_ON_ONCE(!max_append_sectors
))
1236 * Move page array up in the allocated memory for the bio vecs as far as
1237 * possible so that we can start filling biovecs from the beginning
1238 * without overwriting the temporary page array.
1240 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1241 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1243 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1244 if (unlikely(size
<= 0))
1245 return size
? size
: -EFAULT
;
1247 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1248 struct page
*page
= pages
[i
];
1249 bool same_page
= false;
1251 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1252 if (bio_add_hw_page(q
, bio
, page
, len
, offset
,
1253 max_append_sectors
, &same_page
) != len
) {
1254 bio_put_pages(pages
+ i
, left
, offset
);
1263 iov_iter_advance(iter
, size
- left
);
1268 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1269 * @bio: bio to add pages to
1270 * @iter: iov iterator describing the region to be added
1272 * This takes either an iterator pointing to user memory, or one pointing to
1273 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1274 * map them into the kernel. On IO completion, the caller should put those
1275 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1276 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1277 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1278 * completed by a call to ->ki_complete() or returns with an error other than
1279 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1280 * on IO completion. If it isn't, then pages should be released.
1282 * The function tries, but does not guarantee, to pin as many pages as
1283 * fit into the bio, or are requested in @iter, whatever is smaller. If
1284 * MM encounters an error pinning the requested pages, it stops. Error
1285 * is returned only if 0 pages could be pinned.
1287 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1288 * responsible for setting BIO_WORKINGSET if necessary.
1290 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1294 if (iov_iter_is_bvec(iter
)) {
1295 bio_iov_bvec_set(bio
, iter
);
1296 iov_iter_advance(iter
, bio
->bi_iter
.bi_size
);
1301 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
)
1302 ret
= __bio_iov_append_get_pages(bio
, iter
);
1304 ret
= __bio_iov_iter_get_pages(bio
, iter
);
1305 } while (!ret
&& iov_iter_count(iter
) && !bio_full(bio
, 0));
1307 /* don't account direct I/O as memory stall */
1308 bio_clear_flag(bio
, BIO_WORKINGSET
);
1309 return bio
->bi_vcnt
? 0 : ret
;
1311 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
1313 static void submit_bio_wait_endio(struct bio
*bio
)
1315 complete(bio
->bi_private
);
1319 * submit_bio_wait - submit a bio, and wait until it completes
1320 * @bio: The &struct bio which describes the I/O
1322 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1323 * bio_endio() on failure.
1325 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1326 * result in bio reference to be consumed. The caller must drop the reference
1329 int submit_bio_wait(struct bio
*bio
)
1331 DECLARE_COMPLETION_ONSTACK_MAP(done
,
1332 bio
->bi_bdev
->bd_disk
->lockdep_map
);
1333 unsigned long hang_check
;
1335 bio
->bi_private
= &done
;
1336 bio
->bi_end_io
= submit_bio_wait_endio
;
1337 bio
->bi_opf
|= REQ_SYNC
;
1340 /* Prevent hang_check timer from firing at us during very long I/O */
1341 hang_check
= sysctl_hung_task_timeout_secs
;
1343 while (!wait_for_completion_io_timeout(&done
,
1344 hang_check
* (HZ
/2)))
1347 wait_for_completion_io(&done
);
1349 return blk_status_to_errno(bio
->bi_status
);
1351 EXPORT_SYMBOL(submit_bio_wait
);
1353 void __bio_advance(struct bio
*bio
, unsigned bytes
)
1355 if (bio_integrity(bio
))
1356 bio_integrity_advance(bio
, bytes
);
1358 bio_crypt_advance(bio
, bytes
);
1359 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
1361 EXPORT_SYMBOL(__bio_advance
);
1363 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
1364 struct bio
*src
, struct bvec_iter
*src_iter
)
1366 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
1367 struct bio_vec src_bv
= bio_iter_iovec(src
, *src_iter
);
1368 struct bio_vec dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
1369 unsigned int bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1370 void *src_buf
= bvec_kmap_local(&src_bv
);
1371 void *dst_buf
= bvec_kmap_local(&dst_bv
);
1373 memcpy(dst_buf
, src_buf
, bytes
);
1375 kunmap_local(dst_buf
);
1376 kunmap_local(src_buf
);
1378 bio_advance_iter_single(src
, src_iter
, bytes
);
1379 bio_advance_iter_single(dst
, dst_iter
, bytes
);
1382 EXPORT_SYMBOL(bio_copy_data_iter
);
1385 * bio_copy_data - copy contents of data buffers from one bio to another
1387 * @dst: destination bio
1389 * Stops when it reaches the end of either @src or @dst - that is, copies
1390 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1392 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1394 struct bvec_iter src_iter
= src
->bi_iter
;
1395 struct bvec_iter dst_iter
= dst
->bi_iter
;
1397 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1399 EXPORT_SYMBOL(bio_copy_data
);
1401 void bio_free_pages(struct bio
*bio
)
1403 struct bio_vec
*bvec
;
1404 struct bvec_iter_all iter_all
;
1406 bio_for_each_segment_all(bvec
, bio
, iter_all
)
1407 __free_page(bvec
->bv_page
);
1409 EXPORT_SYMBOL(bio_free_pages
);
1412 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1413 * for performing direct-IO in BIOs.
1415 * The problem is that we cannot run set_page_dirty() from interrupt context
1416 * because the required locks are not interrupt-safe. So what we can do is to
1417 * mark the pages dirty _before_ performing IO. And in interrupt context,
1418 * check that the pages are still dirty. If so, fine. If not, redirty them
1419 * in process context.
1421 * We special-case compound pages here: normally this means reads into hugetlb
1422 * pages. The logic in here doesn't really work right for compound pages
1423 * because the VM does not uniformly chase down the head page in all cases.
1424 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1425 * handle them at all. So we skip compound pages here at an early stage.
1427 * Note that this code is very hard to test under normal circumstances because
1428 * direct-io pins the pages with get_user_pages(). This makes
1429 * is_page_cache_freeable return false, and the VM will not clean the pages.
1430 * But other code (eg, flusher threads) could clean the pages if they are mapped
1433 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1434 * deferred bio dirtying paths.
1438 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1440 void bio_set_pages_dirty(struct bio
*bio
)
1442 struct bio_vec
*bvec
;
1443 struct bvec_iter_all iter_all
;
1445 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1446 if (!PageCompound(bvec
->bv_page
))
1447 set_page_dirty_lock(bvec
->bv_page
);
1452 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1453 * If they are, then fine. If, however, some pages are clean then they must
1454 * have been written out during the direct-IO read. So we take another ref on
1455 * the BIO and re-dirty the pages in process context.
1457 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1458 * here on. It will run one put_page() against each page and will run one
1459 * bio_put() against the BIO.
1462 static void bio_dirty_fn(struct work_struct
*work
);
1464 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1465 static DEFINE_SPINLOCK(bio_dirty_lock
);
1466 static struct bio
*bio_dirty_list
;
1469 * This runs in process context
1471 static void bio_dirty_fn(struct work_struct
*work
)
1473 struct bio
*bio
, *next
;
1475 spin_lock_irq(&bio_dirty_lock
);
1476 next
= bio_dirty_list
;
1477 bio_dirty_list
= NULL
;
1478 spin_unlock_irq(&bio_dirty_lock
);
1480 while ((bio
= next
) != NULL
) {
1481 next
= bio
->bi_private
;
1483 bio_release_pages(bio
, true);
1488 void bio_check_pages_dirty(struct bio
*bio
)
1490 struct bio_vec
*bvec
;
1491 unsigned long flags
;
1492 struct bvec_iter_all iter_all
;
1494 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1495 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1499 bio_release_pages(bio
, false);
1503 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1504 bio
->bi_private
= bio_dirty_list
;
1505 bio_dirty_list
= bio
;
1506 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1507 schedule_work(&bio_dirty_work
);
1510 static inline bool bio_remaining_done(struct bio
*bio
)
1513 * If we're not chaining, then ->__bi_remaining is always 1 and
1514 * we always end io on the first invocation.
1516 if (!bio_flagged(bio
, BIO_CHAIN
))
1519 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1521 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1522 bio_clear_flag(bio
, BIO_CHAIN
);
1530 * bio_endio - end I/O on a bio
1534 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1535 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1536 * bio unless they own it and thus know that it has an end_io function.
1538 * bio_endio() can be called several times on a bio that has been chained
1539 * using bio_chain(). The ->bi_end_io() function will only be called the
1542 void bio_endio(struct bio
*bio
)
1545 if (!bio_remaining_done(bio
))
1547 if (!bio_integrity_endio(bio
))
1550 rq_qos_done_bio(bio
);
1552 if (bio
->bi_bdev
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1553 trace_block_bio_complete(bdev_get_queue(bio
->bi_bdev
), bio
);
1554 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1558 * Need to have a real endio function for chained bios, otherwise
1559 * various corner cases will break (like stacking block devices that
1560 * save/restore bi_end_io) - however, we want to avoid unbounded
1561 * recursion and blowing the stack. Tail call optimization would
1562 * handle this, but compiling with frame pointers also disables
1563 * gcc's sibling call optimization.
1565 if (bio
->bi_end_io
== bio_chain_endio
) {
1566 bio
= __bio_chain_endio(bio
);
1570 blk_throtl_bio_endio(bio
);
1571 /* release cgroup info */
1574 bio
->bi_end_io(bio
);
1576 EXPORT_SYMBOL(bio_endio
);
1579 * bio_split - split a bio
1580 * @bio: bio to split
1581 * @sectors: number of sectors to split from the front of @bio
1583 * @bs: bio set to allocate from
1585 * Allocates and returns a new bio which represents @sectors from the start of
1586 * @bio, and updates @bio to represent the remaining sectors.
1588 * Unless this is a discard request the newly allocated bio will point
1589 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1590 * neither @bio nor @bs are freed before the split bio.
1592 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1593 gfp_t gfp
, struct bio_set
*bs
)
1597 BUG_ON(sectors
<= 0);
1598 BUG_ON(sectors
>= bio_sectors(bio
));
1600 /* Zone append commands cannot be split */
1601 if (WARN_ON_ONCE(bio_op(bio
) == REQ_OP_ZONE_APPEND
))
1604 split
= bio_alloc_clone(bio
->bi_bdev
, bio
, gfp
, bs
);
1608 split
->bi_iter
.bi_size
= sectors
<< 9;
1610 if (bio_integrity(split
))
1611 bio_integrity_trim(split
);
1613 bio_advance(bio
, split
->bi_iter
.bi_size
);
1615 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1616 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1620 EXPORT_SYMBOL(bio_split
);
1623 * bio_trim - trim a bio
1625 * @offset: number of sectors to trim from the front of @bio
1626 * @size: size we want to trim @bio to, in sectors
1628 * This function is typically used for bios that are cloned and submitted
1629 * to the underlying device in parts.
1631 void bio_trim(struct bio
*bio
, sector_t offset
, sector_t size
)
1633 if (WARN_ON_ONCE(offset
> BIO_MAX_SECTORS
|| size
> BIO_MAX_SECTORS
||
1634 offset
+ size
> bio_sectors(bio
)))
1638 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1641 bio_advance(bio
, offset
<< 9);
1642 bio
->bi_iter
.bi_size
= size
;
1644 if (bio_integrity(bio
))
1645 bio_integrity_trim(bio
);
1647 EXPORT_SYMBOL_GPL(bio_trim
);
1650 * create memory pools for biovec's in a bio_set.
1651 * use the global biovec slabs created for general use.
1653 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1655 struct biovec_slab
*bp
= bvec_slabs
+ ARRAY_SIZE(bvec_slabs
) - 1;
1657 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1661 * bioset_exit - exit a bioset initialized with bioset_init()
1663 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1666 void bioset_exit(struct bio_set
*bs
)
1668 bio_alloc_cache_destroy(bs
);
1669 if (bs
->rescue_workqueue
)
1670 destroy_workqueue(bs
->rescue_workqueue
);
1671 bs
->rescue_workqueue
= NULL
;
1673 mempool_exit(&bs
->bio_pool
);
1674 mempool_exit(&bs
->bvec_pool
);
1676 bioset_integrity_free(bs
);
1679 bs
->bio_slab
= NULL
;
1681 EXPORT_SYMBOL(bioset_exit
);
1684 * bioset_init - Initialize a bio_set
1685 * @bs: pool to initialize
1686 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1687 * @front_pad: Number of bytes to allocate in front of the returned bio
1688 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1689 * and %BIOSET_NEED_RESCUER
1692 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1693 * to ask for a number of bytes to be allocated in front of the bio.
1694 * Front pad allocation is useful for embedding the bio inside
1695 * another structure, to avoid allocating extra data to go with the bio.
1696 * Note that the bio must be embedded at the END of that structure always,
1697 * or things will break badly.
1698 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1699 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1700 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1701 * to dispatch queued requests when the mempool runs out of space.
1704 int bioset_init(struct bio_set
*bs
,
1705 unsigned int pool_size
,
1706 unsigned int front_pad
,
1709 bs
->front_pad
= front_pad
;
1710 if (flags
& BIOSET_NEED_BVECS
)
1711 bs
->back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1715 spin_lock_init(&bs
->rescue_lock
);
1716 bio_list_init(&bs
->rescue_list
);
1717 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1719 bs
->bio_slab
= bio_find_or_create_slab(bs
);
1723 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
1726 if ((flags
& BIOSET_NEED_BVECS
) &&
1727 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
1730 if (flags
& BIOSET_NEED_RESCUER
) {
1731 bs
->rescue_workqueue
= alloc_workqueue("bioset",
1733 if (!bs
->rescue_workqueue
)
1736 if (flags
& BIOSET_PERCPU_CACHE
) {
1737 bs
->cache
= alloc_percpu(struct bio_alloc_cache
);
1740 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD
, &bs
->cpuhp_dead
);
1748 EXPORT_SYMBOL(bioset_init
);
1750 static int __init
init_bio(void)
1754 bio_integrity_init();
1756 for (i
= 0; i
< ARRAY_SIZE(bvec_slabs
); i
++) {
1757 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1759 bvs
->slab
= kmem_cache_create(bvs
->name
,
1760 bvs
->nr_vecs
* sizeof(struct bio_vec
), 0,
1761 SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1764 cpuhp_setup_state_multi(CPUHP_BIO_DEAD
, "block/bio:dead", NULL
,
1767 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
1768 panic("bio: can't allocate bios\n");
1770 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
1771 panic("bio: can't create integrity pool\n");
1775 subsys_initcall(init_bio
);