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 struct bio_alloc_cache
{
29 struct bio_list 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
, SLAB_HWCACHE_ALIGN
, NULL
);
95 bslab
->slab_size
= size
;
97 if (!xa_err(xa_store(&bio_slabs
, size
, bslab
, GFP_KERNEL
)))
100 kmem_cache_destroy(bslab
->slab
);
107 static inline unsigned int bs_bio_slab_size(struct bio_set
*bs
)
109 return bs
->front_pad
+ sizeof(struct bio
) + bs
->back_pad
;
112 static struct kmem_cache
*bio_find_or_create_slab(struct bio_set
*bs
)
114 unsigned int size
= bs_bio_slab_size(bs
);
115 struct bio_slab
*bslab
;
117 mutex_lock(&bio_slab_lock
);
118 bslab
= xa_load(&bio_slabs
, size
);
122 bslab
= create_bio_slab(size
);
123 mutex_unlock(&bio_slab_lock
);
130 static void bio_put_slab(struct bio_set
*bs
)
132 struct bio_slab
*bslab
= NULL
;
133 unsigned int slab_size
= bs_bio_slab_size(bs
);
135 mutex_lock(&bio_slab_lock
);
137 bslab
= xa_load(&bio_slabs
, slab_size
);
138 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
141 WARN_ON_ONCE(bslab
->slab
!= bs
->bio_slab
);
143 WARN_ON(!bslab
->slab_ref
);
145 if (--bslab
->slab_ref
)
148 xa_erase(&bio_slabs
, slab_size
);
150 kmem_cache_destroy(bslab
->slab
);
154 mutex_unlock(&bio_slab_lock
);
157 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned short nr_vecs
)
159 BIO_BUG_ON(nr_vecs
> BIO_MAX_VECS
);
161 if (nr_vecs
== BIO_MAX_VECS
)
162 mempool_free(bv
, pool
);
163 else if (nr_vecs
> BIO_INLINE_VECS
)
164 kmem_cache_free(biovec_slab(nr_vecs
)->slab
, bv
);
168 * Make the first allocation restricted and don't dump info on allocation
169 * failures, since we'll fall back to the mempool in case of failure.
171 static inline gfp_t
bvec_alloc_gfp(gfp_t gfp
)
173 return (gfp
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
)) |
174 __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
177 struct bio_vec
*bvec_alloc(mempool_t
*pool
, unsigned short *nr_vecs
,
180 struct biovec_slab
*bvs
= biovec_slab(*nr_vecs
);
182 if (WARN_ON_ONCE(!bvs
))
186 * Upgrade the nr_vecs request to take full advantage of the allocation.
187 * We also rely on this in the bvec_free path.
189 *nr_vecs
= bvs
->nr_vecs
;
192 * Try a slab allocation first for all smaller allocations. If that
193 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
194 * The mempool is sized to handle up to BIO_MAX_VECS entries.
196 if (*nr_vecs
< BIO_MAX_VECS
) {
199 bvl
= kmem_cache_alloc(bvs
->slab
, bvec_alloc_gfp(gfp_mask
));
200 if (likely(bvl
) || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
202 *nr_vecs
= BIO_MAX_VECS
;
205 return mempool_alloc(pool
, gfp_mask
);
208 void bio_uninit(struct bio
*bio
)
210 #ifdef CONFIG_BLK_CGROUP
212 blkg_put(bio
->bi_blkg
);
216 if (bio_integrity(bio
))
217 bio_integrity_free(bio
);
219 bio_crypt_free_ctx(bio
);
221 EXPORT_SYMBOL(bio_uninit
);
223 static void bio_free(struct bio
*bio
)
225 struct bio_set
*bs
= bio
->bi_pool
;
231 bvec_free(&bs
->bvec_pool
, bio
->bi_io_vec
, bio
->bi_max_vecs
);
234 * If we have front padding, adjust the bio pointer before freeing
239 mempool_free(p
, &bs
->bio_pool
);
241 /* Bio was allocated by bio_kmalloc() */
247 * Users of this function have their own bio allocation. Subsequently,
248 * they must remember to pair any call to bio_init() with bio_uninit()
249 * when IO has completed, or when the bio is released.
251 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
252 unsigned short max_vecs
)
259 bio
->bi_write_hint
= 0;
261 bio
->bi_iter
.bi_sector
= 0;
262 bio
->bi_iter
.bi_size
= 0;
263 bio
->bi_iter
.bi_idx
= 0;
264 bio
->bi_iter
.bi_bvec_done
= 0;
265 bio
->bi_end_io
= NULL
;
266 bio
->bi_private
= NULL
;
267 #ifdef CONFIG_BLK_CGROUP
269 bio
->bi_issue
.value
= 0;
270 #ifdef CONFIG_BLK_CGROUP_IOCOST
271 bio
->bi_iocost_cost
= 0;
274 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
275 bio
->bi_crypt_context
= NULL
;
277 #ifdef CONFIG_BLK_DEV_INTEGRITY
278 bio
->bi_integrity
= NULL
;
282 atomic_set(&bio
->__bi_remaining
, 1);
283 atomic_set(&bio
->__bi_cnt
, 1);
285 bio
->bi_max_vecs
= max_vecs
;
286 bio
->bi_io_vec
= table
;
289 EXPORT_SYMBOL(bio_init
);
292 * bio_reset - reinitialize a bio
296 * After calling bio_reset(), @bio will be in the same state as a freshly
297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
298 * preserved are the ones that are initialized by bio_alloc_bioset(). See
299 * comment in struct bio.
301 void bio_reset(struct bio
*bio
)
304 memset(bio
, 0, BIO_RESET_BYTES
);
305 atomic_set(&bio
->__bi_remaining
, 1);
307 EXPORT_SYMBOL(bio_reset
);
309 static struct bio
*__bio_chain_endio(struct bio
*bio
)
311 struct bio
*parent
= bio
->bi_private
;
313 if (bio
->bi_status
&& !parent
->bi_status
)
314 parent
->bi_status
= bio
->bi_status
;
319 static void bio_chain_endio(struct bio
*bio
)
321 bio_endio(__bio_chain_endio(bio
));
325 * bio_chain - chain bio completions
326 * @bio: the target bio
327 * @parent: the parent bio of @bio
329 * The caller won't have a bi_end_io called when @bio completes - instead,
330 * @parent's bi_end_io won't be called until both @parent and @bio have
331 * completed; the chained bio will also be freed when it completes.
333 * The caller must not set bi_private or bi_end_io in @bio.
335 void bio_chain(struct bio
*bio
, struct bio
*parent
)
337 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
339 bio
->bi_private
= parent
;
340 bio
->bi_end_io
= bio_chain_endio
;
341 bio_inc_remaining(parent
);
343 EXPORT_SYMBOL(bio_chain
);
345 static void bio_alloc_rescue(struct work_struct
*work
)
347 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
351 spin_lock(&bs
->rescue_lock
);
352 bio
= bio_list_pop(&bs
->rescue_list
);
353 spin_unlock(&bs
->rescue_lock
);
358 submit_bio_noacct(bio
);
362 static void punt_bios_to_rescuer(struct bio_set
*bs
)
364 struct bio_list punt
, nopunt
;
367 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
370 * In order to guarantee forward progress we must punt only bios that
371 * were allocated from this bio_set; otherwise, if there was a bio on
372 * there for a stacking driver higher up in the stack, processing it
373 * could require allocating bios from this bio_set, and doing that from
374 * our own rescuer would be bad.
376 * Since bio lists are singly linked, pop them all instead of trying to
377 * remove from the middle of the list:
380 bio_list_init(&punt
);
381 bio_list_init(&nopunt
);
383 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
384 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
385 current
->bio_list
[0] = nopunt
;
387 bio_list_init(&nopunt
);
388 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
389 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
390 current
->bio_list
[1] = nopunt
;
392 spin_lock(&bs
->rescue_lock
);
393 bio_list_merge(&bs
->rescue_list
, &punt
);
394 spin_unlock(&bs
->rescue_lock
);
396 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
400 * bio_alloc_bioset - allocate a bio for I/O
401 * @gfp_mask: the GFP_* mask given to the slab allocator
402 * @nr_iovecs: number of iovecs to pre-allocate
403 * @bs: the bio_set to allocate from.
405 * Allocate a bio from the mempools in @bs.
407 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
408 * allocate a bio. This is due to the mempool guarantees. To make this work,
409 * callers must never allocate more than 1 bio at a time from the general pool.
410 * Callers that need to allocate more than 1 bio must always submit the
411 * previously allocated bio for IO before attempting to allocate a new one.
412 * Failure to do so can cause deadlocks under memory pressure.
414 * Note that when running under submit_bio_noacct() (i.e. any block driver),
415 * bios are not submitted until after you return - see the code in
416 * submit_bio_noacct() that converts recursion into iteration, to prevent
419 * This would normally mean allocating multiple bios under submit_bio_noacct()
420 * would be susceptible to deadlocks, but we have
421 * deadlock avoidance code that resubmits any blocked bios from a rescuer
424 * However, we do not guarantee forward progress for allocations from other
425 * mempools. Doing multiple allocations from the same mempool under
426 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
427 * for per bio allocations.
429 * Returns: Pointer to new bio on success, NULL on failure.
431 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned short nr_iovecs
,
434 gfp_t saved_gfp
= gfp_mask
;
438 /* should not use nobvec bioset for nr_iovecs > 0 */
439 if (WARN_ON_ONCE(!mempool_initialized(&bs
->bvec_pool
) && nr_iovecs
> 0))
443 * submit_bio_noacct() converts recursion to iteration; this means if
444 * we're running beneath it, any bios we allocate and submit will not be
445 * submitted (and thus freed) until after we return.
447 * This exposes us to a potential deadlock if we allocate multiple bios
448 * from the same bio_set() while running underneath submit_bio_noacct().
449 * If we were to allocate multiple bios (say a stacking block driver
450 * that was splitting bios), we would deadlock if we exhausted the
453 * We solve this, and guarantee forward progress, with a rescuer
454 * workqueue per bio_set. If we go to allocate and there are bios on
455 * current->bio_list, we first try the allocation without
456 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
457 * blocking to the rescuer workqueue before we retry with the original
460 if (current
->bio_list
&&
461 (!bio_list_empty(¤t
->bio_list
[0]) ||
462 !bio_list_empty(¤t
->bio_list
[1])) &&
463 bs
->rescue_workqueue
)
464 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
466 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
467 if (!p
&& gfp_mask
!= saved_gfp
) {
468 punt_bios_to_rescuer(bs
);
469 gfp_mask
= saved_gfp
;
470 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
475 bio
= p
+ bs
->front_pad
;
476 if (nr_iovecs
> BIO_INLINE_VECS
) {
477 struct bio_vec
*bvl
= NULL
;
479 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_iovecs
, gfp_mask
);
480 if (!bvl
&& gfp_mask
!= saved_gfp
) {
481 punt_bios_to_rescuer(bs
);
482 gfp_mask
= saved_gfp
;
483 bvl
= bvec_alloc(&bs
->bvec_pool
, &nr_iovecs
, gfp_mask
);
488 bio_init(bio
, bvl
, nr_iovecs
);
489 } else if (nr_iovecs
) {
490 bio_init(bio
, bio
->bi_inline_vecs
, BIO_INLINE_VECS
);
492 bio_init(bio
, NULL
, 0);
499 mempool_free(p
, &bs
->bio_pool
);
502 EXPORT_SYMBOL(bio_alloc_bioset
);
505 * bio_kmalloc - kmalloc a bio for I/O
506 * @gfp_mask: the GFP_* mask given to the slab allocator
507 * @nr_iovecs: number of iovecs to pre-allocate
509 * Use kmalloc to allocate and initialize a bio.
511 * Returns: Pointer to new bio on success, NULL on failure.
513 struct bio
*bio_kmalloc(gfp_t gfp_mask
, unsigned short nr_iovecs
)
517 if (nr_iovecs
> UIO_MAXIOV
)
520 bio
= kmalloc(struct_size(bio
, bi_inline_vecs
, nr_iovecs
), gfp_mask
);
523 bio_init(bio
, nr_iovecs
? bio
->bi_inline_vecs
: NULL
, nr_iovecs
);
527 EXPORT_SYMBOL(bio_kmalloc
);
529 void zero_fill_bio(struct bio
*bio
)
532 struct bvec_iter iter
;
534 bio_for_each_segment(bv
, bio
, iter
)
537 EXPORT_SYMBOL(zero_fill_bio
);
540 * bio_truncate - truncate the bio to small size of @new_size
541 * @bio: the bio to be truncated
542 * @new_size: new size for truncating the bio
545 * Truncate the bio to new size of @new_size. If bio_op(bio) is
546 * REQ_OP_READ, zero the truncated part. This function should only
547 * be used for handling corner cases, such as bio eod.
549 void bio_truncate(struct bio
*bio
, unsigned new_size
)
552 struct bvec_iter iter
;
553 unsigned int done
= 0;
554 bool truncated
= false;
556 if (new_size
>= bio
->bi_iter
.bi_size
)
559 if (bio_op(bio
) != REQ_OP_READ
)
562 bio_for_each_segment(bv
, bio
, iter
) {
563 if (done
+ bv
.bv_len
> new_size
) {
567 offset
= new_size
- done
;
570 zero_user(bv
.bv_page
, bv
.bv_offset
+ offset
,
579 * Don't touch bvec table here and make it really immutable, since
580 * fs bio user has to retrieve all pages via bio_for_each_segment_all
581 * in its .end_bio() callback.
583 * It is enough to truncate bio by updating .bi_size since we can make
584 * correct bvec with the updated .bi_size for drivers.
586 bio
->bi_iter
.bi_size
= new_size
;
590 * guard_bio_eod - truncate a BIO to fit the block device
591 * @bio: bio to truncate
593 * This allows us to do IO even on the odd last sectors of a device, even if the
594 * block size is some multiple of the physical sector size.
596 * We'll just truncate the bio to the size of the device, and clear the end of
597 * the buffer head manually. Truly out-of-range accesses will turn into actual
598 * I/O errors, this only handles the "we need to be able to do I/O at the final
601 void guard_bio_eod(struct bio
*bio
)
603 sector_t maxsector
= bdev_nr_sectors(bio
->bi_bdev
);
609 * If the *whole* IO is past the end of the device,
610 * let it through, and the IO layer will turn it into
613 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
616 maxsector
-= bio
->bi_iter
.bi_sector
;
617 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
620 bio_truncate(bio
, maxsector
<< 9);
623 #define ALLOC_CACHE_MAX 512
624 #define ALLOC_CACHE_SLACK 64
626 static void bio_alloc_cache_prune(struct bio_alloc_cache
*cache
,
632 while ((bio
= bio_list_pop(&cache
->free_list
)) != NULL
) {
640 static int bio_cpu_dead(unsigned int cpu
, struct hlist_node
*node
)
644 bs
= hlist_entry_safe(node
, struct bio_set
, cpuhp_dead
);
646 struct bio_alloc_cache
*cache
= per_cpu_ptr(bs
->cache
, cpu
);
648 bio_alloc_cache_prune(cache
, -1U);
653 static void bio_alloc_cache_destroy(struct bio_set
*bs
)
660 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD
, &bs
->cpuhp_dead
);
661 for_each_possible_cpu(cpu
) {
662 struct bio_alloc_cache
*cache
;
664 cache
= per_cpu_ptr(bs
->cache
, cpu
);
665 bio_alloc_cache_prune(cache
, -1U);
667 free_percpu(bs
->cache
);
671 * bio_put - release a reference to a bio
672 * @bio: bio to release reference to
675 * Put a reference to a &struct bio, either one you have gotten with
676 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
678 void bio_put(struct bio
*bio
)
680 if (unlikely(bio_flagged(bio
, BIO_REFFED
))) {
681 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
682 if (!atomic_dec_and_test(&bio
->__bi_cnt
))
686 if (bio_flagged(bio
, BIO_PERCPU_CACHE
)) {
687 struct bio_alloc_cache
*cache
;
690 cache
= per_cpu_ptr(bio
->bi_pool
->cache
, get_cpu());
691 bio_list_add_head(&cache
->free_list
, bio
);
692 if (++cache
->nr
> ALLOC_CACHE_MAX
+ ALLOC_CACHE_SLACK
)
693 bio_alloc_cache_prune(cache
, ALLOC_CACHE_SLACK
);
699 EXPORT_SYMBOL(bio_put
);
702 * __bio_clone_fast - clone a bio that shares the original bio's biovec
703 * @bio: destination bio
704 * @bio_src: bio to clone
706 * Clone a &bio. Caller will own the returned bio, but not
707 * the actual data it points to. Reference count of returned
710 * Caller must ensure that @bio_src is not freed before @bio.
712 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
714 WARN_ON_ONCE(bio
->bi_pool
&& bio
->bi_max_vecs
);
717 * most users will be overriding ->bi_bdev with a new target,
718 * so we don't set nor calculate new physical/hw segment counts here
720 bio
->bi_bdev
= bio_src
->bi_bdev
;
721 bio_set_flag(bio
, BIO_CLONED
);
722 if (bio_flagged(bio_src
, BIO_THROTTLED
))
723 bio_set_flag(bio
, BIO_THROTTLED
);
724 if (bio_flagged(bio_src
, BIO_REMAPPED
))
725 bio_set_flag(bio
, BIO_REMAPPED
);
726 bio
->bi_opf
= bio_src
->bi_opf
;
727 bio
->bi_ioprio
= bio_src
->bi_ioprio
;
728 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
729 bio
->bi_iter
= bio_src
->bi_iter
;
730 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
732 bio_clone_blkg_association(bio
, bio_src
);
733 blkcg_bio_issue_init(bio
);
735 EXPORT_SYMBOL(__bio_clone_fast
);
738 * bio_clone_fast - clone a bio that shares the original bio's biovec
740 * @gfp_mask: allocation priority
741 * @bs: bio_set to allocate from
743 * Like __bio_clone_fast, only also allocates the returned bio
745 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
749 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
753 __bio_clone_fast(b
, bio
);
755 if (bio_crypt_clone(b
, bio
, gfp_mask
) < 0)
758 if (bio_integrity(bio
) &&
759 bio_integrity_clone(b
, bio
, gfp_mask
) < 0)
768 EXPORT_SYMBOL(bio_clone_fast
);
770 const char *bio_devname(struct bio
*bio
, char *buf
)
772 return bdevname(bio
->bi_bdev
, buf
);
774 EXPORT_SYMBOL(bio_devname
);
776 static inline bool page_is_mergeable(const struct bio_vec
*bv
,
777 struct page
*page
, unsigned int len
, unsigned int off
,
780 size_t bv_end
= bv
->bv_offset
+ bv
->bv_len
;
781 phys_addr_t vec_end_addr
= page_to_phys(bv
->bv_page
) + bv_end
- 1;
782 phys_addr_t page_addr
= page_to_phys(page
);
784 if (vec_end_addr
+ 1 != page_addr
+ off
)
786 if (xen_domain() && !xen_biovec_phys_mergeable(bv
, page
))
789 *same_page
= ((vec_end_addr
& PAGE_MASK
) == page_addr
);
792 return (bv
->bv_page
+ bv_end
/ PAGE_SIZE
) == (page
+ off
/ PAGE_SIZE
);
796 * Try to merge a page into a segment, while obeying the hardware segment
797 * size limit. This is not for normal read/write bios, but for passthrough
798 * or Zone Append operations that we can't split.
800 static bool bio_try_merge_hw_seg(struct request_queue
*q
, struct bio
*bio
,
801 struct page
*page
, unsigned len
,
802 unsigned offset
, bool *same_page
)
804 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
805 unsigned long mask
= queue_segment_boundary(q
);
806 phys_addr_t addr1
= page_to_phys(bv
->bv_page
) + bv
->bv_offset
;
807 phys_addr_t addr2
= page_to_phys(page
) + offset
+ len
- 1;
809 if ((addr1
| mask
) != (addr2
| mask
))
811 if (bv
->bv_len
+ len
> queue_max_segment_size(q
))
813 return __bio_try_merge_page(bio
, page
, len
, offset
, same_page
);
817 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
818 * @q: the target queue
819 * @bio: destination bio
821 * @len: vec entry length
822 * @offset: vec entry offset
823 * @max_sectors: maximum number of sectors that can be added
824 * @same_page: return if the segment has been merged inside the same page
826 * Add a page to a bio while respecting the hardware max_sectors, max_segment
827 * and gap limitations.
829 int bio_add_hw_page(struct request_queue
*q
, struct bio
*bio
,
830 struct page
*page
, unsigned int len
, unsigned int offset
,
831 unsigned int max_sectors
, bool *same_page
)
833 struct bio_vec
*bvec
;
835 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
838 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
841 if (bio
->bi_vcnt
> 0) {
842 if (bio_try_merge_hw_seg(q
, bio
, page
, len
, offset
, same_page
))
846 * If the queue doesn't support SG gaps and adding this segment
847 * would create a gap, disallow it.
849 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
850 if (bvec_gap_to_prev(q
, bvec
, offset
))
854 if (bio_full(bio
, len
))
857 if (bio
->bi_vcnt
>= queue_max_segments(q
))
860 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
861 bvec
->bv_page
= page
;
863 bvec
->bv_offset
= offset
;
865 bio
->bi_iter
.bi_size
+= len
;
870 * bio_add_pc_page - attempt to add page to passthrough bio
871 * @q: the target queue
872 * @bio: destination bio
874 * @len: vec entry length
875 * @offset: vec entry offset
877 * Attempt to add a page to the bio_vec maplist. This can fail for a
878 * number of reasons, such as the bio being full or target block device
879 * limitations. The target block device must allow bio's up to PAGE_SIZE,
880 * so it is always possible to add a single page to an empty bio.
882 * This should only be used by passthrough bios.
884 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
,
885 struct page
*page
, unsigned int len
, unsigned int offset
)
887 bool same_page
= false;
888 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
889 queue_max_hw_sectors(q
), &same_page
);
891 EXPORT_SYMBOL(bio_add_pc_page
);
894 * bio_add_zone_append_page - attempt to add page to zone-append bio
895 * @bio: destination bio
897 * @len: vec entry length
898 * @offset: vec entry offset
900 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
901 * for a zone-append request. This can fail for a number of reasons, such as the
902 * bio being full or the target block device is not a zoned block device or
903 * other limitations of the target block device. The target block device must
904 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
907 * Returns: number of bytes added to the bio, or 0 in case of a failure.
909 int bio_add_zone_append_page(struct bio
*bio
, struct page
*page
,
910 unsigned int len
, unsigned int offset
)
912 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
913 bool same_page
= false;
915 if (WARN_ON_ONCE(bio_op(bio
) != REQ_OP_ZONE_APPEND
))
918 if (WARN_ON_ONCE(!blk_queue_is_zoned(q
)))
921 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
922 queue_max_zone_append_sectors(q
), &same_page
);
924 EXPORT_SYMBOL_GPL(bio_add_zone_append_page
);
927 * __bio_try_merge_page - try appending data to an existing bvec.
928 * @bio: destination bio
929 * @page: start page to add
930 * @len: length of the data to add
931 * @off: offset of the data relative to @page
932 * @same_page: return if the segment has been merged inside the same page
934 * Try to add the data at @page + @off to the last bvec of @bio. This is a
935 * useful optimisation for file systems with a block size smaller than the
938 * Warn if (@len, @off) crosses pages in case that @same_page is true.
940 * Return %true on success or %false on failure.
942 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
943 unsigned int len
, unsigned int off
, bool *same_page
)
945 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
948 if (bio
->bi_vcnt
> 0) {
949 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
951 if (page_is_mergeable(bv
, page
, len
, off
, same_page
)) {
952 if (bio
->bi_iter
.bi_size
> UINT_MAX
- len
) {
957 bio
->bi_iter
.bi_size
+= len
;
963 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
966 * __bio_add_page - add page(s) to a bio in a new segment
967 * @bio: destination bio
968 * @page: start page to add
969 * @len: length of the data to add, may cross pages
970 * @off: offset of the data relative to @page, may cross pages
972 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
973 * that @bio has space for another bvec.
975 void __bio_add_page(struct bio
*bio
, struct page
*page
,
976 unsigned int len
, unsigned int off
)
978 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
980 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
981 WARN_ON_ONCE(bio_full(bio
, len
));
987 bio
->bi_iter
.bi_size
+= len
;
990 if (!bio_flagged(bio
, BIO_WORKINGSET
) && unlikely(PageWorkingset(page
)))
991 bio_set_flag(bio
, BIO_WORKINGSET
);
993 EXPORT_SYMBOL_GPL(__bio_add_page
);
996 * bio_add_page - attempt to add page(s) to bio
997 * @bio: destination bio
998 * @page: start page to add
999 * @len: vec entry length, may cross pages
1000 * @offset: vec entry offset relative to @page, may cross pages
1002 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1003 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1005 int bio_add_page(struct bio
*bio
, struct page
*page
,
1006 unsigned int len
, unsigned int offset
)
1008 bool same_page
= false;
1010 if (!__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1011 if (bio_full(bio
, len
))
1013 __bio_add_page(bio
, page
, len
, offset
);
1017 EXPORT_SYMBOL(bio_add_page
);
1019 void bio_release_pages(struct bio
*bio
, bool mark_dirty
)
1021 struct bvec_iter_all iter_all
;
1022 struct bio_vec
*bvec
;
1024 if (bio_flagged(bio
, BIO_NO_PAGE_REF
))
1027 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1028 if (mark_dirty
&& !PageCompound(bvec
->bv_page
))
1029 set_page_dirty_lock(bvec
->bv_page
);
1030 put_page(bvec
->bv_page
);
1033 EXPORT_SYMBOL_GPL(bio_release_pages
);
1035 static void __bio_iov_bvec_set(struct bio
*bio
, struct iov_iter
*iter
)
1037 WARN_ON_ONCE(bio
->bi_max_vecs
);
1039 bio
->bi_vcnt
= iter
->nr_segs
;
1040 bio
->bi_io_vec
= (struct bio_vec
*)iter
->bvec
;
1041 bio
->bi_iter
.bi_bvec_done
= iter
->iov_offset
;
1042 bio
->bi_iter
.bi_size
= iter
->count
;
1043 bio_set_flag(bio
, BIO_NO_PAGE_REF
);
1044 bio_set_flag(bio
, BIO_CLONED
);
1047 static int bio_iov_bvec_set(struct bio
*bio
, struct iov_iter
*iter
)
1049 __bio_iov_bvec_set(bio
, iter
);
1050 iov_iter_advance(iter
, iter
->count
);
1054 static int bio_iov_bvec_set_append(struct bio
*bio
, struct iov_iter
*iter
)
1056 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
1057 struct iov_iter i
= *iter
;
1059 iov_iter_truncate(&i
, queue_max_zone_append_sectors(q
) << 9);
1060 __bio_iov_bvec_set(bio
, &i
);
1061 iov_iter_advance(iter
, i
.count
);
1065 static void bio_put_pages(struct page
**pages
, size_t size
, size_t off
)
1067 size_t i
, nr
= DIV_ROUND_UP(size
+ (off
& ~PAGE_MASK
), PAGE_SIZE
);
1069 for (i
= 0; i
< nr
; i
++)
1073 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1076 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1077 * @bio: bio to add pages to
1078 * @iter: iov iterator describing the region to be mapped
1080 * Pins pages from *iter and appends them to @bio's bvec array. The
1081 * pages will have to be released using put_page() when done.
1082 * For multi-segment *iter, this function only adds pages from the
1083 * next non-empty segment of the iov iterator.
1085 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1087 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1088 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1089 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1090 struct page
**pages
= (struct page
**)bv
;
1091 bool same_page
= false;
1097 * Move page array up in the allocated memory for the bio vecs as far as
1098 * possible so that we can start filling biovecs from the beginning
1099 * without overwriting the temporary page array.
1101 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1102 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1104 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1105 if (unlikely(size
<= 0))
1106 return size
? size
: -EFAULT
;
1108 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1109 struct page
*page
= pages
[i
];
1111 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1113 if (__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1117 if (WARN_ON_ONCE(bio_full(bio
, len
))) {
1118 bio_put_pages(pages
+ i
, left
, offset
);
1121 __bio_add_page(bio
, page
, len
, offset
);
1126 iov_iter_advance(iter
, size
);
1130 static int __bio_iov_append_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1132 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1133 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1134 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
1135 unsigned int max_append_sectors
= queue_max_zone_append_sectors(q
);
1136 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1137 struct page
**pages
= (struct page
**)bv
;
1143 if (WARN_ON_ONCE(!max_append_sectors
))
1147 * Move page array up in the allocated memory for the bio vecs as far as
1148 * possible so that we can start filling biovecs from the beginning
1149 * without overwriting the temporary page array.
1151 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1152 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1154 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1155 if (unlikely(size
<= 0))
1156 return size
? size
: -EFAULT
;
1158 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1159 struct page
*page
= pages
[i
];
1160 bool same_page
= false;
1162 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1163 if (bio_add_hw_page(q
, bio
, page
, len
, offset
,
1164 max_append_sectors
, &same_page
) != len
) {
1165 bio_put_pages(pages
+ i
, left
, offset
);
1174 iov_iter_advance(iter
, size
- left
);
1179 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1180 * @bio: bio to add pages to
1181 * @iter: iov iterator describing the region to be added
1183 * This takes either an iterator pointing to user memory, or one pointing to
1184 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1185 * map them into the kernel. On IO completion, the caller should put those
1186 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1187 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1188 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1189 * completed by a call to ->ki_complete() or returns with an error other than
1190 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1191 * on IO completion. If it isn't, then pages should be released.
1193 * The function tries, but does not guarantee, to pin as many pages as
1194 * fit into the bio, or are requested in @iter, whatever is smaller. If
1195 * MM encounters an error pinning the requested pages, it stops. Error
1196 * is returned only if 0 pages could be pinned.
1198 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1199 * responsible for setting BIO_WORKINGSET if necessary.
1201 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1205 if (iov_iter_is_bvec(iter
)) {
1206 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
)
1207 return bio_iov_bvec_set_append(bio
, iter
);
1208 return bio_iov_bvec_set(bio
, iter
);
1212 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
)
1213 ret
= __bio_iov_append_get_pages(bio
, iter
);
1215 ret
= __bio_iov_iter_get_pages(bio
, iter
);
1216 } while (!ret
&& iov_iter_count(iter
) && !bio_full(bio
, 0));
1218 /* don't account direct I/O as memory stall */
1219 bio_clear_flag(bio
, BIO_WORKINGSET
);
1220 return bio
->bi_vcnt
? 0 : ret
;
1222 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
1224 static void submit_bio_wait_endio(struct bio
*bio
)
1226 complete(bio
->bi_private
);
1230 * submit_bio_wait - submit a bio, and wait until it completes
1231 * @bio: The &struct bio which describes the I/O
1233 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1234 * bio_endio() on failure.
1236 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1237 * result in bio reference to be consumed. The caller must drop the reference
1240 int submit_bio_wait(struct bio
*bio
)
1242 DECLARE_COMPLETION_ONSTACK_MAP(done
,
1243 bio
->bi_bdev
->bd_disk
->lockdep_map
);
1244 unsigned long hang_check
;
1246 bio
->bi_private
= &done
;
1247 bio
->bi_end_io
= submit_bio_wait_endio
;
1248 bio
->bi_opf
|= REQ_SYNC
;
1251 /* Prevent hang_check timer from firing at us during very long I/O */
1252 hang_check
= sysctl_hung_task_timeout_secs
;
1254 while (!wait_for_completion_io_timeout(&done
,
1255 hang_check
* (HZ
/2)))
1258 wait_for_completion_io(&done
);
1260 return blk_status_to_errno(bio
->bi_status
);
1262 EXPORT_SYMBOL(submit_bio_wait
);
1265 * bio_advance - increment/complete a bio by some number of bytes
1266 * @bio: bio to advance
1267 * @bytes: number of bytes to complete
1269 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1270 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1271 * be updated on the last bvec as well.
1273 * @bio will then represent the remaining, uncompleted portion of the io.
1275 void bio_advance(struct bio
*bio
, unsigned bytes
)
1277 if (bio_integrity(bio
))
1278 bio_integrity_advance(bio
, bytes
);
1280 bio_crypt_advance(bio
, bytes
);
1281 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
1283 EXPORT_SYMBOL(bio_advance
);
1285 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
1286 struct bio
*src
, struct bvec_iter
*src_iter
)
1288 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
1289 struct bio_vec src_bv
= bio_iter_iovec(src
, *src_iter
);
1290 struct bio_vec dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
1291 unsigned int bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1294 src_buf
= bvec_kmap_local(&src_bv
);
1295 memcpy_to_bvec(&dst_bv
, src_buf
);
1296 kunmap_local(src_buf
);
1298 bio_advance_iter_single(src
, src_iter
, bytes
);
1299 bio_advance_iter_single(dst
, dst_iter
, bytes
);
1302 EXPORT_SYMBOL(bio_copy_data_iter
);
1305 * bio_copy_data - copy contents of data buffers from one bio to another
1307 * @dst: destination bio
1309 * Stops when it reaches the end of either @src or @dst - that is, copies
1310 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1312 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1314 struct bvec_iter src_iter
= src
->bi_iter
;
1315 struct bvec_iter dst_iter
= dst
->bi_iter
;
1317 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1319 EXPORT_SYMBOL(bio_copy_data
);
1321 void bio_free_pages(struct bio
*bio
)
1323 struct bio_vec
*bvec
;
1324 struct bvec_iter_all iter_all
;
1326 bio_for_each_segment_all(bvec
, bio
, iter_all
)
1327 __free_page(bvec
->bv_page
);
1329 EXPORT_SYMBOL(bio_free_pages
);
1332 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1333 * for performing direct-IO in BIOs.
1335 * The problem is that we cannot run set_page_dirty() from interrupt context
1336 * because the required locks are not interrupt-safe. So what we can do is to
1337 * mark the pages dirty _before_ performing IO. And in interrupt context,
1338 * check that the pages are still dirty. If so, fine. If not, redirty them
1339 * in process context.
1341 * We special-case compound pages here: normally this means reads into hugetlb
1342 * pages. The logic in here doesn't really work right for compound pages
1343 * because the VM does not uniformly chase down the head page in all cases.
1344 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1345 * handle them at all. So we skip compound pages here at an early stage.
1347 * Note that this code is very hard to test under normal circumstances because
1348 * direct-io pins the pages with get_user_pages(). This makes
1349 * is_page_cache_freeable return false, and the VM will not clean the pages.
1350 * But other code (eg, flusher threads) could clean the pages if they are mapped
1353 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1354 * deferred bio dirtying paths.
1358 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1360 void bio_set_pages_dirty(struct bio
*bio
)
1362 struct bio_vec
*bvec
;
1363 struct bvec_iter_all iter_all
;
1365 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1366 if (!PageCompound(bvec
->bv_page
))
1367 set_page_dirty_lock(bvec
->bv_page
);
1372 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1373 * If they are, then fine. If, however, some pages are clean then they must
1374 * have been written out during the direct-IO read. So we take another ref on
1375 * the BIO and re-dirty the pages in process context.
1377 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1378 * here on. It will run one put_page() against each page and will run one
1379 * bio_put() against the BIO.
1382 static void bio_dirty_fn(struct work_struct
*work
);
1384 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1385 static DEFINE_SPINLOCK(bio_dirty_lock
);
1386 static struct bio
*bio_dirty_list
;
1389 * This runs in process context
1391 static void bio_dirty_fn(struct work_struct
*work
)
1393 struct bio
*bio
, *next
;
1395 spin_lock_irq(&bio_dirty_lock
);
1396 next
= bio_dirty_list
;
1397 bio_dirty_list
= NULL
;
1398 spin_unlock_irq(&bio_dirty_lock
);
1400 while ((bio
= next
) != NULL
) {
1401 next
= bio
->bi_private
;
1403 bio_release_pages(bio
, true);
1408 void bio_check_pages_dirty(struct bio
*bio
)
1410 struct bio_vec
*bvec
;
1411 unsigned long flags
;
1412 struct bvec_iter_all iter_all
;
1414 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1415 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1419 bio_release_pages(bio
, false);
1423 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1424 bio
->bi_private
= bio_dirty_list
;
1425 bio_dirty_list
= bio
;
1426 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1427 schedule_work(&bio_dirty_work
);
1430 static inline bool bio_remaining_done(struct bio
*bio
)
1433 * If we're not chaining, then ->__bi_remaining is always 1 and
1434 * we always end io on the first invocation.
1436 if (!bio_flagged(bio
, BIO_CHAIN
))
1439 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1441 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1442 bio_clear_flag(bio
, BIO_CHAIN
);
1450 * bio_endio - end I/O on a bio
1454 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1455 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1456 * bio unless they own it and thus know that it has an end_io function.
1458 * bio_endio() can be called several times on a bio that has been chained
1459 * using bio_chain(). The ->bi_end_io() function will only be called the
1462 void bio_endio(struct bio
*bio
)
1465 if (!bio_remaining_done(bio
))
1467 if (!bio_integrity_endio(bio
))
1470 if (bio
->bi_bdev
&& bio_flagged(bio
, BIO_TRACKED
))
1471 rq_qos_done_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
);
1473 if (bio
->bi_bdev
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1474 trace_block_bio_complete(bio
->bi_bdev
->bd_disk
->queue
, bio
);
1475 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1479 * Need to have a real endio function for chained bios, otherwise
1480 * various corner cases will break (like stacking block devices that
1481 * save/restore bi_end_io) - however, we want to avoid unbounded
1482 * recursion and blowing the stack. Tail call optimization would
1483 * handle this, but compiling with frame pointers also disables
1484 * gcc's sibling call optimization.
1486 if (bio
->bi_end_io
== bio_chain_endio
) {
1487 bio
= __bio_chain_endio(bio
);
1491 blk_throtl_bio_endio(bio
);
1492 /* release cgroup info */
1495 bio
->bi_end_io(bio
);
1497 EXPORT_SYMBOL(bio_endio
);
1500 * bio_split - split a bio
1501 * @bio: bio to split
1502 * @sectors: number of sectors to split from the front of @bio
1504 * @bs: bio set to allocate from
1506 * Allocates and returns a new bio which represents @sectors from the start of
1507 * @bio, and updates @bio to represent the remaining sectors.
1509 * Unless this is a discard request the newly allocated bio will point
1510 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1511 * neither @bio nor @bs are freed before the split bio.
1513 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1514 gfp_t gfp
, struct bio_set
*bs
)
1518 BUG_ON(sectors
<= 0);
1519 BUG_ON(sectors
>= bio_sectors(bio
));
1521 /* Zone append commands cannot be split */
1522 if (WARN_ON_ONCE(bio_op(bio
) == REQ_OP_ZONE_APPEND
))
1525 split
= bio_clone_fast(bio
, gfp
, bs
);
1529 split
->bi_iter
.bi_size
= sectors
<< 9;
1531 if (bio_integrity(split
))
1532 bio_integrity_trim(split
);
1534 bio_advance(bio
, split
->bi_iter
.bi_size
);
1536 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1537 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1541 EXPORT_SYMBOL(bio_split
);
1544 * bio_trim - trim a bio
1546 * @offset: number of sectors to trim from the front of @bio
1547 * @size: size we want to trim @bio to, in sectors
1549 * This function is typically used for bios that are cloned and submitted
1550 * to the underlying device in parts.
1552 void bio_trim(struct bio
*bio
, sector_t offset
, sector_t size
)
1554 if (WARN_ON_ONCE(offset
> BIO_MAX_SECTORS
|| size
> BIO_MAX_SECTORS
||
1555 offset
+ size
> bio_sectors(bio
)))
1559 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1562 bio_advance(bio
, offset
<< 9);
1563 bio
->bi_iter
.bi_size
= size
;
1565 if (bio_integrity(bio
))
1566 bio_integrity_trim(bio
);
1568 EXPORT_SYMBOL_GPL(bio_trim
);
1571 * create memory pools for biovec's in a bio_set.
1572 * use the global biovec slabs created for general use.
1574 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1576 struct biovec_slab
*bp
= bvec_slabs
+ ARRAY_SIZE(bvec_slabs
) - 1;
1578 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1582 * bioset_exit - exit a bioset initialized with bioset_init()
1584 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1587 void bioset_exit(struct bio_set
*bs
)
1589 bio_alloc_cache_destroy(bs
);
1590 if (bs
->rescue_workqueue
)
1591 destroy_workqueue(bs
->rescue_workqueue
);
1592 bs
->rescue_workqueue
= NULL
;
1594 mempool_exit(&bs
->bio_pool
);
1595 mempool_exit(&bs
->bvec_pool
);
1597 bioset_integrity_free(bs
);
1600 bs
->bio_slab
= NULL
;
1602 EXPORT_SYMBOL(bioset_exit
);
1605 * bioset_init - Initialize a bio_set
1606 * @bs: pool to initialize
1607 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1608 * @front_pad: Number of bytes to allocate in front of the returned bio
1609 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1610 * and %BIOSET_NEED_RESCUER
1613 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1614 * to ask for a number of bytes to be allocated in front of the bio.
1615 * Front pad allocation is useful for embedding the bio inside
1616 * another structure, to avoid allocating extra data to go with the bio.
1617 * Note that the bio must be embedded at the END of that structure always,
1618 * or things will break badly.
1619 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1620 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1621 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1622 * dispatch queued requests when the mempool runs out of space.
1625 int bioset_init(struct bio_set
*bs
,
1626 unsigned int pool_size
,
1627 unsigned int front_pad
,
1630 bs
->front_pad
= front_pad
;
1631 if (flags
& BIOSET_NEED_BVECS
)
1632 bs
->back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1636 spin_lock_init(&bs
->rescue_lock
);
1637 bio_list_init(&bs
->rescue_list
);
1638 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1640 bs
->bio_slab
= bio_find_or_create_slab(bs
);
1644 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
1647 if ((flags
& BIOSET_NEED_BVECS
) &&
1648 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
1651 if (flags
& BIOSET_NEED_RESCUER
) {
1652 bs
->rescue_workqueue
= alloc_workqueue("bioset",
1654 if (!bs
->rescue_workqueue
)
1657 if (flags
& BIOSET_PERCPU_CACHE
) {
1658 bs
->cache
= alloc_percpu(struct bio_alloc_cache
);
1661 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD
, &bs
->cpuhp_dead
);
1669 EXPORT_SYMBOL(bioset_init
);
1672 * Initialize and setup a new bio_set, based on the settings from
1675 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
1680 if (src
->bvec_pool
.min_nr
)
1681 flags
|= BIOSET_NEED_BVECS
;
1682 if (src
->rescue_workqueue
)
1683 flags
|= BIOSET_NEED_RESCUER
;
1685 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
1687 EXPORT_SYMBOL(bioset_init_from_src
);
1690 * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb
1691 * @kiocb: kiocb describing the IO
1692 * @nr_vecs: number of iovecs to pre-allocate
1693 * @bs: bio_set to allocate from
1696 * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only
1697 * used to check if we should dip into the per-cpu bio_set allocation
1698 * cache. The allocation uses GFP_KERNEL internally. On return, the
1699 * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio
1700 * MUST be done from process context, not hard/soft IRQ.
1703 struct bio
*bio_alloc_kiocb(struct kiocb
*kiocb
, unsigned short nr_vecs
,
1706 struct bio_alloc_cache
*cache
;
1709 if (!(kiocb
->ki_flags
& IOCB_ALLOC_CACHE
) || nr_vecs
> BIO_INLINE_VECS
)
1710 return bio_alloc_bioset(GFP_KERNEL
, nr_vecs
, bs
);
1712 cache
= per_cpu_ptr(bs
->cache
, get_cpu());
1713 bio
= bio_list_pop(&cache
->free_list
);
1717 bio_init(bio
, nr_vecs
? bio
->bi_inline_vecs
: NULL
, nr_vecs
);
1719 bio_set_flag(bio
, BIO_PERCPU_CACHE
);
1723 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_vecs
, bs
);
1724 bio_set_flag(bio
, BIO_PERCPU_CACHE
);
1727 EXPORT_SYMBOL_GPL(bio_alloc_kiocb
);
1729 static int __init
init_bio(void)
1733 bio_integrity_init();
1735 for (i
= 0; i
< ARRAY_SIZE(bvec_slabs
); i
++) {
1736 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1738 bvs
->slab
= kmem_cache_create(bvs
->name
,
1739 bvs
->nr_vecs
* sizeof(struct bio_vec
), 0,
1740 SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1743 cpuhp_setup_state_multi(CPUHP_BIO_DEAD
, "block/bio:dead", NULL
,
1746 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
1747 panic("bio: can't allocate bios\n");
1749 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
1750 panic("bio: can't create integrity pool\n");
1754 subsys_initcall(init_bio
);