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>
23 #include <trace/events/block.h>
25 #include "blk-rq-qos.h"
28 * Test patch to inline a certain number of bi_io_vec's inside the bio
29 * itself, to shrink a bio data allocation from two mempool calls to one
31 #define BIO_INLINE_VECS 4
34 * if you change this list, also change bvec_alloc or things will
35 * break badly! cannot be bigger than what you can fit into an
38 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
39 static struct biovec_slab bvec_slabs
[BVEC_POOL_NR
] __read_mostly
= {
40 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES
, max
),
45 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
46 * IO code that does not need private memory pools.
48 struct bio_set fs_bio_set
;
49 EXPORT_SYMBOL(fs_bio_set
);
52 * Our slab pool management
55 struct kmem_cache
*slab
;
56 unsigned int slab_ref
;
57 unsigned int slab_size
;
60 static DEFINE_MUTEX(bio_slab_lock
);
61 static struct bio_slab
*bio_slabs
;
62 static unsigned int bio_slab_nr
, bio_slab_max
;
64 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
66 unsigned int sz
= sizeof(struct bio
) + extra_size
;
67 struct kmem_cache
*slab
= NULL
;
68 struct bio_slab
*bslab
, *new_bio_slabs
;
69 unsigned int new_bio_slab_max
;
70 unsigned int i
, entry
= -1;
72 mutex_lock(&bio_slab_lock
);
75 while (i
< bio_slab_nr
) {
76 bslab
= &bio_slabs
[i
];
78 if (!bslab
->slab
&& entry
== -1)
80 else if (bslab
->slab_size
== sz
) {
91 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
92 new_bio_slab_max
= bio_slab_max
<< 1;
93 new_bio_slabs
= krealloc(bio_slabs
,
94 new_bio_slab_max
* sizeof(struct bio_slab
),
98 bio_slab_max
= new_bio_slab_max
;
99 bio_slabs
= new_bio_slabs
;
102 entry
= bio_slab_nr
++;
104 bslab
= &bio_slabs
[entry
];
106 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
107 slab
= kmem_cache_create(bslab
->name
, sz
, ARCH_KMALLOC_MINALIGN
,
108 SLAB_HWCACHE_ALIGN
, NULL
);
114 bslab
->slab_size
= sz
;
116 mutex_unlock(&bio_slab_lock
);
120 static void bio_put_slab(struct bio_set
*bs
)
122 struct bio_slab
*bslab
= NULL
;
125 mutex_lock(&bio_slab_lock
);
127 for (i
= 0; i
< bio_slab_nr
; i
++) {
128 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
129 bslab
= &bio_slabs
[i
];
134 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
137 WARN_ON(!bslab
->slab_ref
);
139 if (--bslab
->slab_ref
)
142 kmem_cache_destroy(bslab
->slab
);
146 mutex_unlock(&bio_slab_lock
);
149 unsigned int bvec_nr_vecs(unsigned short idx
)
151 return bvec_slabs
[--idx
].nr_vecs
;
154 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
160 BIO_BUG_ON(idx
>= BVEC_POOL_NR
);
162 if (idx
== BVEC_POOL_MAX
) {
163 mempool_free(bv
, pool
);
165 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
167 kmem_cache_free(bvs
->slab
, bv
);
171 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
177 * see comment near bvec_array define!
195 case 129 ... BIO_MAX_PAGES
:
203 * idx now points to the pool we want to allocate from. only the
204 * 1-vec entry pool is mempool backed.
206 if (*idx
== BVEC_POOL_MAX
) {
208 bvl
= mempool_alloc(pool
, gfp_mask
);
210 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
211 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
);
214 * Make this allocation restricted and don't dump info on
215 * allocation failures, since we'll fallback to the mempool
216 * in case of failure.
218 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
221 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
222 * is set, retry with the 1-entry mempool
224 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
225 if (unlikely(!bvl
&& (gfp_mask
& __GFP_DIRECT_RECLAIM
))) {
226 *idx
= BVEC_POOL_MAX
;
235 void bio_uninit(struct bio
*bio
)
237 bio_disassociate_blkg(bio
);
239 if (bio_integrity(bio
))
240 bio_integrity_free(bio
);
242 bio_crypt_free_ctx(bio
);
244 EXPORT_SYMBOL(bio_uninit
);
246 static void bio_free(struct bio
*bio
)
248 struct bio_set
*bs
= bio
->bi_pool
;
254 bvec_free(&bs
->bvec_pool
, bio
->bi_io_vec
, BVEC_POOL_IDX(bio
));
257 * If we have front padding, adjust the bio pointer before freeing
262 mempool_free(p
, &bs
->bio_pool
);
264 /* Bio was allocated by bio_kmalloc() */
270 * Users of this function have their own bio allocation. Subsequently,
271 * they must remember to pair any call to bio_init() with bio_uninit()
272 * when IO has completed, or when the bio is released.
274 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
275 unsigned short max_vecs
)
277 memset(bio
, 0, sizeof(*bio
));
278 atomic_set(&bio
->__bi_remaining
, 1);
279 atomic_set(&bio
->__bi_cnt
, 1);
281 bio
->bi_io_vec
= table
;
282 bio
->bi_max_vecs
= max_vecs
;
284 EXPORT_SYMBOL(bio_init
);
287 * bio_reset - reinitialize a bio
291 * After calling bio_reset(), @bio will be in the same state as a freshly
292 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
293 * preserved are the ones that are initialized by bio_alloc_bioset(). See
294 * comment in struct bio.
296 void bio_reset(struct bio
*bio
)
298 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
302 memset(bio
, 0, BIO_RESET_BYTES
);
303 bio
->bi_flags
= flags
;
304 atomic_set(&bio
->__bi_remaining
, 1);
306 EXPORT_SYMBOL(bio_reset
);
308 static struct bio
*__bio_chain_endio(struct bio
*bio
)
310 struct bio
*parent
= bio
->bi_private
;
312 if (!parent
->bi_status
)
313 parent
->bi_status
= bio
->bi_status
;
318 static void bio_chain_endio(struct bio
*bio
)
320 bio_endio(__bio_chain_endio(bio
));
324 * bio_chain - chain bio completions
325 * @bio: the target bio
326 * @parent: the @bio's parent bio
328 * The caller won't have a bi_end_io called when @bio completes - instead,
329 * @parent's bi_end_io won't be called until both @parent and @bio have
330 * completed; the chained bio will also be freed when it completes.
332 * The caller must not set bi_private or bi_end_io in @bio.
334 void bio_chain(struct bio
*bio
, struct bio
*parent
)
336 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
338 bio
->bi_private
= parent
;
339 bio
->bi_end_io
= bio_chain_endio
;
340 bio_inc_remaining(parent
);
342 EXPORT_SYMBOL(bio_chain
);
344 static void bio_alloc_rescue(struct work_struct
*work
)
346 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
350 spin_lock(&bs
->rescue_lock
);
351 bio
= bio_list_pop(&bs
->rescue_list
);
352 spin_unlock(&bs
->rescue_lock
);
357 generic_make_request(bio
);
361 static void punt_bios_to_rescuer(struct bio_set
*bs
)
363 struct bio_list punt
, nopunt
;
366 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
369 * In order to guarantee forward progress we must punt only bios that
370 * were allocated from this bio_set; otherwise, if there was a bio on
371 * there for a stacking driver higher up in the stack, processing it
372 * could require allocating bios from this bio_set, and doing that from
373 * our own rescuer would be bad.
375 * Since bio lists are singly linked, pop them all instead of trying to
376 * remove from the middle of the list:
379 bio_list_init(&punt
);
380 bio_list_init(&nopunt
);
382 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
383 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
384 current
->bio_list
[0] = nopunt
;
386 bio_list_init(&nopunt
);
387 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
388 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
389 current
->bio_list
[1] = nopunt
;
391 spin_lock(&bs
->rescue_lock
);
392 bio_list_merge(&bs
->rescue_list
, &punt
);
393 spin_unlock(&bs
->rescue_lock
);
395 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
399 * bio_alloc_bioset - allocate a bio for I/O
400 * @gfp_mask: the GFP_* mask given to the slab allocator
401 * @nr_iovecs: number of iovecs to pre-allocate
402 * @bs: the bio_set to allocate from.
405 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
406 * backed by the @bs's mempool.
408 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
409 * always be able to allocate a bio. This is due to the mempool guarantees.
410 * To make this work, callers must never allocate more than 1 bio at a time
411 * from this pool. Callers that need to allocate more than 1 bio must always
412 * submit the previously allocated bio for IO before attempting to allocate
413 * a new one. Failure to do so can cause deadlocks under memory pressure.
415 * Note that when running under generic_make_request() (i.e. any block
416 * driver), bios are not submitted until after you return - see the code in
417 * generic_make_request() that converts recursion into iteration, to prevent
420 * This would normally mean allocating multiple bios under
421 * generic_make_request() would be susceptible to deadlocks, but we have
422 * deadlock avoidance code that resubmits any blocked bios from a rescuer
425 * However, we do not guarantee forward progress for allocations from other
426 * mempools. Doing multiple allocations from the same mempool under
427 * generic_make_request() should be avoided - instead, use bio_set's front_pad
428 * for per bio allocations.
431 * Pointer to new bio on success, NULL on failure.
433 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned int nr_iovecs
,
436 gfp_t saved_gfp
= gfp_mask
;
438 unsigned inline_vecs
;
439 struct bio_vec
*bvl
= NULL
;
444 if (nr_iovecs
> UIO_MAXIOV
)
447 p
= kmalloc(sizeof(struct bio
) +
448 nr_iovecs
* sizeof(struct bio_vec
),
451 inline_vecs
= nr_iovecs
;
453 /* should not use nobvec bioset for nr_iovecs > 0 */
454 if (WARN_ON_ONCE(!mempool_initialized(&bs
->bvec_pool
) &&
458 * generic_make_request() converts recursion to iteration; this
459 * means if we're running beneath it, any bios we allocate and
460 * submit will not be submitted (and thus freed) until after we
463 * This exposes us to a potential deadlock if we allocate
464 * multiple bios from the same bio_set() while running
465 * underneath generic_make_request(). If we were to allocate
466 * multiple bios (say a stacking block driver that was splitting
467 * bios), we would deadlock if we exhausted the mempool's
470 * We solve this, and guarantee forward progress, with a rescuer
471 * workqueue per bio_set. If we go to allocate and there are
472 * bios on current->bio_list, we first try the allocation
473 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
474 * bios we would be blocking to the rescuer workqueue before
475 * we retry with the original gfp_flags.
478 if (current
->bio_list
&&
479 (!bio_list_empty(¤t
->bio_list
[0]) ||
480 !bio_list_empty(¤t
->bio_list
[1])) &&
481 bs
->rescue_workqueue
)
482 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
484 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
485 if (!p
&& gfp_mask
!= saved_gfp
) {
486 punt_bios_to_rescuer(bs
);
487 gfp_mask
= saved_gfp
;
488 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
491 front_pad
= bs
->front_pad
;
492 inline_vecs
= BIO_INLINE_VECS
;
499 bio_init(bio
, NULL
, 0);
501 if (nr_iovecs
> inline_vecs
) {
502 unsigned long idx
= 0;
504 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, &bs
->bvec_pool
);
505 if (!bvl
&& gfp_mask
!= saved_gfp
) {
506 punt_bios_to_rescuer(bs
);
507 gfp_mask
= saved_gfp
;
508 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, &bs
->bvec_pool
);
514 bio
->bi_flags
|= idx
<< BVEC_POOL_OFFSET
;
515 } else if (nr_iovecs
) {
516 bvl
= bio
->bi_inline_vecs
;
520 bio
->bi_max_vecs
= nr_iovecs
;
521 bio
->bi_io_vec
= bvl
;
525 mempool_free(p
, &bs
->bio_pool
);
528 EXPORT_SYMBOL(bio_alloc_bioset
);
530 void zero_fill_bio_iter(struct bio
*bio
, struct bvec_iter start
)
534 struct bvec_iter iter
;
536 __bio_for_each_segment(bv
, bio
, iter
, start
) {
537 char *data
= bvec_kmap_irq(&bv
, &flags
);
538 memset(data
, 0, bv
.bv_len
);
539 flush_dcache_page(bv
.bv_page
);
540 bvec_kunmap_irq(data
, &flags
);
543 EXPORT_SYMBOL(zero_fill_bio_iter
);
546 * bio_truncate - truncate the bio to small size of @new_size
547 * @bio: the bio to be truncated
548 * @new_size: new size for truncating the bio
551 * Truncate the bio to new size of @new_size. If bio_op(bio) is
552 * REQ_OP_READ, zero the truncated part. This function should only
553 * be used for handling corner cases, such as bio eod.
555 void bio_truncate(struct bio
*bio
, unsigned new_size
)
558 struct bvec_iter iter
;
559 unsigned int done
= 0;
560 bool truncated
= false;
562 if (new_size
>= bio
->bi_iter
.bi_size
)
565 if (bio_op(bio
) != REQ_OP_READ
)
568 bio_for_each_segment(bv
, bio
, iter
) {
569 if (done
+ bv
.bv_len
> new_size
) {
573 offset
= new_size
- done
;
576 zero_user(bv
.bv_page
, offset
, bv
.bv_len
- offset
);
584 * Don't touch bvec table here and make it really immutable, since
585 * fs bio user has to retrieve all pages via bio_for_each_segment_all
586 * in its .end_bio() callback.
588 * It is enough to truncate bio by updating .bi_size since we can make
589 * correct bvec with the updated .bi_size for drivers.
591 bio
->bi_iter
.bi_size
= new_size
;
595 * guard_bio_eod - truncate a BIO to fit the block device
596 * @bio: bio to truncate
598 * This allows us to do IO even on the odd last sectors of a device, even if the
599 * block size is some multiple of the physical sector size.
601 * We'll just truncate the bio to the size of the device, and clear the end of
602 * the buffer head manually. Truly out-of-range accesses will turn into actual
603 * I/O errors, this only handles the "we need to be able to do I/O at the final
606 void guard_bio_eod(struct bio
*bio
)
609 struct hd_struct
*part
;
612 part
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
614 maxsector
= part_nr_sects_read(part
);
616 maxsector
= get_capacity(bio
->bi_disk
);
623 * If the *whole* IO is past the end of the device,
624 * let it through, and the IO layer will turn it into
627 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
630 maxsector
-= bio
->bi_iter
.bi_sector
;
631 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
634 bio_truncate(bio
, maxsector
<< 9);
638 * bio_put - release a reference to a bio
639 * @bio: bio to release reference to
642 * Put a reference to a &struct bio, either one you have gotten with
643 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
645 void bio_put(struct bio
*bio
)
647 if (!bio_flagged(bio
, BIO_REFFED
))
650 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
655 if (atomic_dec_and_test(&bio
->__bi_cnt
))
659 EXPORT_SYMBOL(bio_put
);
662 * __bio_clone_fast - clone a bio that shares the original bio's biovec
663 * @bio: destination bio
664 * @bio_src: bio to clone
666 * Clone a &bio. Caller will own the returned bio, but not
667 * the actual data it points to. Reference count of returned
670 * Caller must ensure that @bio_src is not freed before @bio.
672 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
674 BUG_ON(bio
->bi_pool
&& BVEC_POOL_IDX(bio
));
677 * most users will be overriding ->bi_disk with a new target,
678 * so we don't set nor calculate new physical/hw segment counts here
680 bio
->bi_disk
= bio_src
->bi_disk
;
681 bio
->bi_partno
= bio_src
->bi_partno
;
682 bio_set_flag(bio
, BIO_CLONED
);
683 if (bio_flagged(bio_src
, BIO_THROTTLED
))
684 bio_set_flag(bio
, BIO_THROTTLED
);
685 bio
->bi_opf
= bio_src
->bi_opf
;
686 bio
->bi_ioprio
= bio_src
->bi_ioprio
;
687 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
688 bio
->bi_iter
= bio_src
->bi_iter
;
689 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
691 bio_clone_blkg_association(bio
, bio_src
);
692 blkcg_bio_issue_init(bio
);
694 EXPORT_SYMBOL(__bio_clone_fast
);
697 * bio_clone_fast - clone a bio that shares the original bio's biovec
699 * @gfp_mask: allocation priority
700 * @bs: bio_set to allocate from
702 * Like __bio_clone_fast, only also allocates the returned bio
704 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
708 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
712 __bio_clone_fast(b
, bio
);
714 bio_crypt_clone(b
, bio
, gfp_mask
);
716 if (bio_integrity(bio
)) {
719 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
729 EXPORT_SYMBOL(bio_clone_fast
);
731 const char *bio_devname(struct bio
*bio
, char *buf
)
733 return disk_name(bio
->bi_disk
, bio
->bi_partno
, buf
);
735 EXPORT_SYMBOL(bio_devname
);
737 static inline bool page_is_mergeable(const struct bio_vec
*bv
,
738 struct page
*page
, unsigned int len
, unsigned int off
,
741 phys_addr_t vec_end_addr
= page_to_phys(bv
->bv_page
) +
742 bv
->bv_offset
+ bv
->bv_len
- 1;
743 phys_addr_t page_addr
= page_to_phys(page
);
745 if (vec_end_addr
+ 1 != page_addr
+ off
)
747 if (xen_domain() && !xen_biovec_phys_mergeable(bv
, page
))
750 *same_page
= ((vec_end_addr
& PAGE_MASK
) == page_addr
);
751 if (!*same_page
&& pfn_to_page(PFN_DOWN(vec_end_addr
)) + 1 != page
)
757 * Try to merge a page into a segment, while obeying the hardware segment
758 * size limit. This is not for normal read/write bios, but for passthrough
759 * or Zone Append operations that we can't split.
761 static bool bio_try_merge_hw_seg(struct request_queue
*q
, struct bio
*bio
,
762 struct page
*page
, unsigned len
,
763 unsigned offset
, bool *same_page
)
765 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
766 unsigned long mask
= queue_segment_boundary(q
);
767 phys_addr_t addr1
= page_to_phys(bv
->bv_page
) + bv
->bv_offset
;
768 phys_addr_t addr2
= page_to_phys(page
) + offset
+ len
- 1;
770 if ((addr1
| mask
) != (addr2
| mask
))
772 if (bv
->bv_len
+ len
> queue_max_segment_size(q
))
774 return __bio_try_merge_page(bio
, page
, len
, offset
, same_page
);
778 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
779 * @q: the target queue
780 * @bio: destination bio
782 * @len: vec entry length
783 * @offset: vec entry offset
784 * @max_sectors: maximum number of sectors that can be added
785 * @same_page: return if the segment has been merged inside the same page
787 * Add a page to a bio while respecting the hardware max_sectors, max_segment
788 * and gap limitations.
790 int bio_add_hw_page(struct request_queue
*q
, struct bio
*bio
,
791 struct page
*page
, unsigned int len
, unsigned int offset
,
792 unsigned int max_sectors
, bool *same_page
)
794 struct bio_vec
*bvec
;
796 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
799 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
802 if (bio
->bi_vcnt
> 0) {
803 if (bio_try_merge_hw_seg(q
, bio
, page
, len
, offset
, same_page
))
807 * If the queue doesn't support SG gaps and adding this segment
808 * would create a gap, disallow it.
810 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
811 if (bvec_gap_to_prev(q
, bvec
, offset
))
815 if (bio_full(bio
, len
))
818 if (bio
->bi_vcnt
>= queue_max_segments(q
))
821 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
822 bvec
->bv_page
= page
;
824 bvec
->bv_offset
= offset
;
826 bio
->bi_iter
.bi_size
+= len
;
831 * bio_add_pc_page - attempt to add page to passthrough bio
832 * @q: the target queue
833 * @bio: destination bio
835 * @len: vec entry length
836 * @offset: vec entry offset
838 * Attempt to add a page to the bio_vec maplist. This can fail for a
839 * number of reasons, such as the bio being full or target block device
840 * limitations. The target block device must allow bio's up to PAGE_SIZE,
841 * so it is always possible to add a single page to an empty bio.
843 * This should only be used by passthrough bios.
845 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
,
846 struct page
*page
, unsigned int len
, unsigned int offset
)
848 bool same_page
= false;
849 return bio_add_hw_page(q
, bio
, page
, len
, offset
,
850 queue_max_hw_sectors(q
), &same_page
);
852 EXPORT_SYMBOL(bio_add_pc_page
);
855 * __bio_try_merge_page - try appending data to an existing bvec.
856 * @bio: destination bio
857 * @page: start page to add
858 * @len: length of the data to add
859 * @off: offset of the data relative to @page
860 * @same_page: return if the segment has been merged inside the same page
862 * Try to add the data at @page + @off to the last bvec of @bio. This is a
863 * a useful optimisation for file systems with a block size smaller than the
866 * Warn if (@len, @off) crosses pages in case that @same_page is true.
868 * Return %true on success or %false on failure.
870 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
871 unsigned int len
, unsigned int off
, bool *same_page
)
873 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
876 if (bio
->bi_vcnt
> 0) {
877 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
879 if (page_is_mergeable(bv
, page
, len
, off
, same_page
)) {
880 if (bio
->bi_iter
.bi_size
> UINT_MAX
- len
)
883 bio
->bi_iter
.bi_size
+= len
;
889 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
892 * __bio_add_page - add page(s) to a bio in a new segment
893 * @bio: destination bio
894 * @page: start page to add
895 * @len: length of the data to add, may cross pages
896 * @off: offset of the data relative to @page, may cross pages
898 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
899 * that @bio has space for another bvec.
901 void __bio_add_page(struct bio
*bio
, struct page
*page
,
902 unsigned int len
, unsigned int off
)
904 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
906 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
907 WARN_ON_ONCE(bio_full(bio
, len
));
913 bio
->bi_iter
.bi_size
+= len
;
916 if (!bio_flagged(bio
, BIO_WORKINGSET
) && unlikely(PageWorkingset(page
)))
917 bio_set_flag(bio
, BIO_WORKINGSET
);
919 EXPORT_SYMBOL_GPL(__bio_add_page
);
922 * bio_add_page - attempt to add page(s) to bio
923 * @bio: destination bio
924 * @page: start page to add
925 * @len: vec entry length, may cross pages
926 * @offset: vec entry offset relative to @page, may cross pages
928 * Attempt to add page(s) to the bio_vec maplist. This will only fail
929 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
931 int bio_add_page(struct bio
*bio
, struct page
*page
,
932 unsigned int len
, unsigned int offset
)
934 bool same_page
= false;
936 if (!__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
937 if (bio_full(bio
, len
))
939 __bio_add_page(bio
, page
, len
, offset
);
943 EXPORT_SYMBOL(bio_add_page
);
945 void bio_release_pages(struct bio
*bio
, bool mark_dirty
)
947 struct bvec_iter_all iter_all
;
948 struct bio_vec
*bvec
;
950 if (bio_flagged(bio
, BIO_NO_PAGE_REF
))
953 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
954 if (mark_dirty
&& !PageCompound(bvec
->bv_page
))
955 set_page_dirty_lock(bvec
->bv_page
);
956 put_page(bvec
->bv_page
);
959 EXPORT_SYMBOL_GPL(bio_release_pages
);
961 static int __bio_iov_bvec_add_pages(struct bio
*bio
, struct iov_iter
*iter
)
963 const struct bio_vec
*bv
= iter
->bvec
;
967 if (WARN_ON_ONCE(iter
->iov_offset
> bv
->bv_len
))
970 len
= min_t(size_t, bv
->bv_len
- iter
->iov_offset
, iter
->count
);
971 size
= bio_add_page(bio
, bv
->bv_page
, len
,
972 bv
->bv_offset
+ iter
->iov_offset
);
973 if (unlikely(size
!= len
))
975 iov_iter_advance(iter
, size
);
979 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
982 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
983 * @bio: bio to add pages to
984 * @iter: iov iterator describing the region to be mapped
986 * Pins pages from *iter and appends them to @bio's bvec array. The
987 * pages will have to be released using put_page() when done.
988 * For multi-segment *iter, this function only adds pages from the
989 * the next non-empty segment of the iov iterator.
991 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
993 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
994 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
995 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
996 struct page
**pages
= (struct page
**)bv
;
997 bool same_page
= false;
1003 * Move page array up in the allocated memory for the bio vecs as far as
1004 * possible so that we can start filling biovecs from the beginning
1005 * without overwriting the temporary page array.
1007 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1008 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1010 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1011 if (unlikely(size
<= 0))
1012 return size
? size
: -EFAULT
;
1014 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1015 struct page
*page
= pages
[i
];
1017 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1019 if (__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1023 if (WARN_ON_ONCE(bio_full(bio
, len
)))
1025 __bio_add_page(bio
, page
, len
, offset
);
1030 iov_iter_advance(iter
, size
);
1034 static int __bio_iov_append_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1036 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1037 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
1038 struct request_queue
*q
= bio
->bi_disk
->queue
;
1039 unsigned int max_append_sectors
= queue_max_zone_append_sectors(q
);
1040 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
1041 struct page
**pages
= (struct page
**)bv
;
1046 if (WARN_ON_ONCE(!max_append_sectors
))
1050 * Move page array up in the allocated memory for the bio vecs as far as
1051 * possible so that we can start filling biovecs from the beginning
1052 * without overwriting the temporary page array.
1054 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
1055 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
1057 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
1058 if (unlikely(size
<= 0))
1059 return size
? size
: -EFAULT
;
1061 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
1062 struct page
*page
= pages
[i
];
1063 bool same_page
= false;
1065 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
1066 if (bio_add_hw_page(q
, bio
, page
, len
, offset
,
1067 max_append_sectors
, &same_page
) != len
)
1074 iov_iter_advance(iter
, size
);
1079 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1080 * @bio: bio to add pages to
1081 * @iter: iov iterator describing the region to be added
1083 * This takes either an iterator pointing to user memory, or one pointing to
1084 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1085 * map them into the kernel. On IO completion, the caller should put those
1086 * pages. If we're adding kernel pages, and the caller told us it's safe to
1087 * do so, we just have to add the pages to the bio directly. We don't grab an
1088 * extra reference to those pages (the user should already have that), and we
1089 * don't put the page on IO completion. The caller needs to check if the bio is
1090 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
1093 * The function tries, but does not guarantee, to pin as many pages as
1094 * fit into the bio, or are requested in *iter, whatever is smaller. If
1095 * MM encounters an error pinning the requested pages, it stops. Error
1096 * is returned only if 0 pages could be pinned.
1098 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1100 const bool is_bvec
= iov_iter_is_bvec(iter
);
1103 if (WARN_ON_ONCE(bio
->bi_vcnt
))
1107 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
) {
1108 if (WARN_ON_ONCE(is_bvec
))
1110 ret
= __bio_iov_append_get_pages(bio
, iter
);
1113 ret
= __bio_iov_bvec_add_pages(bio
, iter
);
1115 ret
= __bio_iov_iter_get_pages(bio
, iter
);
1117 } while (!ret
&& iov_iter_count(iter
) && !bio_full(bio
, 0));
1120 bio_set_flag(bio
, BIO_NO_PAGE_REF
);
1121 return bio
->bi_vcnt
? 0 : ret
;
1123 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
1125 static void submit_bio_wait_endio(struct bio
*bio
)
1127 complete(bio
->bi_private
);
1131 * submit_bio_wait - submit a bio, and wait until it completes
1132 * @bio: The &struct bio which describes the I/O
1134 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1135 * bio_endio() on failure.
1137 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1138 * result in bio reference to be consumed. The caller must drop the reference
1141 int submit_bio_wait(struct bio
*bio
)
1143 DECLARE_COMPLETION_ONSTACK_MAP(done
, bio
->bi_disk
->lockdep_map
);
1144 unsigned long hang_check
;
1146 bio
->bi_private
= &done
;
1147 bio
->bi_end_io
= submit_bio_wait_endio
;
1148 bio
->bi_opf
|= REQ_SYNC
;
1151 /* Prevent hang_check timer from firing at us during very long I/O */
1152 hang_check
= sysctl_hung_task_timeout_secs
;
1154 while (!wait_for_completion_io_timeout(&done
,
1155 hang_check
* (HZ
/2)))
1158 wait_for_completion_io(&done
);
1160 return blk_status_to_errno(bio
->bi_status
);
1162 EXPORT_SYMBOL(submit_bio_wait
);
1165 * bio_advance - increment/complete a bio by some number of bytes
1166 * @bio: bio to advance
1167 * @bytes: number of bytes to complete
1169 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1170 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1171 * be updated on the last bvec as well.
1173 * @bio will then represent the remaining, uncompleted portion of the io.
1175 void bio_advance(struct bio
*bio
, unsigned bytes
)
1177 if (bio_integrity(bio
))
1178 bio_integrity_advance(bio
, bytes
);
1180 bio_crypt_advance(bio
, bytes
);
1181 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
1183 EXPORT_SYMBOL(bio_advance
);
1185 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
1186 struct bio
*src
, struct bvec_iter
*src_iter
)
1188 struct bio_vec src_bv
, dst_bv
;
1189 void *src_p
, *dst_p
;
1192 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
1193 src_bv
= bio_iter_iovec(src
, *src_iter
);
1194 dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
1196 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1198 src_p
= kmap_atomic(src_bv
.bv_page
);
1199 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1201 memcpy(dst_p
+ dst_bv
.bv_offset
,
1202 src_p
+ src_bv
.bv_offset
,
1205 kunmap_atomic(dst_p
);
1206 kunmap_atomic(src_p
);
1208 flush_dcache_page(dst_bv
.bv_page
);
1210 bio_advance_iter(src
, src_iter
, bytes
);
1211 bio_advance_iter(dst
, dst_iter
, bytes
);
1214 EXPORT_SYMBOL(bio_copy_data_iter
);
1217 * bio_copy_data - copy contents of data buffers from one bio to another
1219 * @dst: destination bio
1221 * Stops when it reaches the end of either @src or @dst - that is, copies
1222 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1224 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1226 struct bvec_iter src_iter
= src
->bi_iter
;
1227 struct bvec_iter dst_iter
= dst
->bi_iter
;
1229 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1231 EXPORT_SYMBOL(bio_copy_data
);
1234 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1236 * @src: source bio list
1237 * @dst: destination bio list
1239 * Stops when it reaches the end of either the @src list or @dst list - that is,
1240 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1243 void bio_list_copy_data(struct bio
*dst
, struct bio
*src
)
1245 struct bvec_iter src_iter
= src
->bi_iter
;
1246 struct bvec_iter dst_iter
= dst
->bi_iter
;
1249 if (!src_iter
.bi_size
) {
1254 src_iter
= src
->bi_iter
;
1257 if (!dst_iter
.bi_size
) {
1262 dst_iter
= dst
->bi_iter
;
1265 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1268 EXPORT_SYMBOL(bio_list_copy_data
);
1270 void bio_free_pages(struct bio
*bio
)
1272 struct bio_vec
*bvec
;
1273 struct bvec_iter_all iter_all
;
1275 bio_for_each_segment_all(bvec
, bio
, iter_all
)
1276 __free_page(bvec
->bv_page
);
1278 EXPORT_SYMBOL(bio_free_pages
);
1281 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1282 * for performing direct-IO in BIOs.
1284 * The problem is that we cannot run set_page_dirty() from interrupt context
1285 * because the required locks are not interrupt-safe. So what we can do is to
1286 * mark the pages dirty _before_ performing IO. And in interrupt context,
1287 * check that the pages are still dirty. If so, fine. If not, redirty them
1288 * in process context.
1290 * We special-case compound pages here: normally this means reads into hugetlb
1291 * pages. The logic in here doesn't really work right for compound pages
1292 * because the VM does not uniformly chase down the head page in all cases.
1293 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1294 * handle them at all. So we skip compound pages here at an early stage.
1296 * Note that this code is very hard to test under normal circumstances because
1297 * direct-io pins the pages with get_user_pages(). This makes
1298 * is_page_cache_freeable return false, and the VM will not clean the pages.
1299 * But other code (eg, flusher threads) could clean the pages if they are mapped
1302 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1303 * deferred bio dirtying paths.
1307 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1309 void bio_set_pages_dirty(struct bio
*bio
)
1311 struct bio_vec
*bvec
;
1312 struct bvec_iter_all iter_all
;
1314 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1315 if (!PageCompound(bvec
->bv_page
))
1316 set_page_dirty_lock(bvec
->bv_page
);
1321 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1322 * If they are, then fine. If, however, some pages are clean then they must
1323 * have been written out during the direct-IO read. So we take another ref on
1324 * the BIO and re-dirty the pages in process context.
1326 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1327 * here on. It will run one put_page() against each page and will run one
1328 * bio_put() against the BIO.
1331 static void bio_dirty_fn(struct work_struct
*work
);
1333 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1334 static DEFINE_SPINLOCK(bio_dirty_lock
);
1335 static struct bio
*bio_dirty_list
;
1338 * This runs in process context
1340 static void bio_dirty_fn(struct work_struct
*work
)
1342 struct bio
*bio
, *next
;
1344 spin_lock_irq(&bio_dirty_lock
);
1345 next
= bio_dirty_list
;
1346 bio_dirty_list
= NULL
;
1347 spin_unlock_irq(&bio_dirty_lock
);
1349 while ((bio
= next
) != NULL
) {
1350 next
= bio
->bi_private
;
1352 bio_release_pages(bio
, true);
1357 void bio_check_pages_dirty(struct bio
*bio
)
1359 struct bio_vec
*bvec
;
1360 unsigned long flags
;
1361 struct bvec_iter_all iter_all
;
1363 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1364 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1368 bio_release_pages(bio
, false);
1372 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1373 bio
->bi_private
= bio_dirty_list
;
1374 bio_dirty_list
= bio
;
1375 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1376 schedule_work(&bio_dirty_work
);
1379 static inline bool bio_remaining_done(struct bio
*bio
)
1382 * If we're not chaining, then ->__bi_remaining is always 1 and
1383 * we always end io on the first invocation.
1385 if (!bio_flagged(bio
, BIO_CHAIN
))
1388 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1390 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1391 bio_clear_flag(bio
, BIO_CHAIN
);
1399 * bio_endio - end I/O on a bio
1403 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1404 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1405 * bio unless they own it and thus know that it has an end_io function.
1407 * bio_endio() can be called several times on a bio that has been chained
1408 * using bio_chain(). The ->bi_end_io() function will only be called the
1409 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1410 * generated if BIO_TRACE_COMPLETION is set.
1412 void bio_endio(struct bio
*bio
)
1415 if (!bio_remaining_done(bio
))
1417 if (!bio_integrity_endio(bio
))
1421 rq_qos_done_bio(bio
->bi_disk
->queue
, bio
);
1424 * Need to have a real endio function for chained bios, otherwise
1425 * various corner cases will break (like stacking block devices that
1426 * save/restore bi_end_io) - however, we want to avoid unbounded
1427 * recursion and blowing the stack. Tail call optimization would
1428 * handle this, but compiling with frame pointers also disables
1429 * gcc's sibling call optimization.
1431 if (bio
->bi_end_io
== bio_chain_endio
) {
1432 bio
= __bio_chain_endio(bio
);
1436 if (bio
->bi_disk
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1437 trace_block_bio_complete(bio
->bi_disk
->queue
, bio
);
1438 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1441 blk_throtl_bio_endio(bio
);
1442 /* release cgroup info */
1445 bio
->bi_end_io(bio
);
1447 EXPORT_SYMBOL(bio_endio
);
1450 * bio_split - split a bio
1451 * @bio: bio to split
1452 * @sectors: number of sectors to split from the front of @bio
1454 * @bs: bio set to allocate from
1456 * Allocates and returns a new bio which represents @sectors from the start of
1457 * @bio, and updates @bio to represent the remaining sectors.
1459 * Unless this is a discard request the newly allocated bio will point
1460 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1461 * neither @bio nor @bs are freed before the split bio.
1463 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1464 gfp_t gfp
, struct bio_set
*bs
)
1468 BUG_ON(sectors
<= 0);
1469 BUG_ON(sectors
>= bio_sectors(bio
));
1471 /* Zone append commands cannot be split */
1472 if (WARN_ON_ONCE(bio_op(bio
) == REQ_OP_ZONE_APPEND
))
1475 split
= bio_clone_fast(bio
, gfp
, bs
);
1479 split
->bi_iter
.bi_size
= sectors
<< 9;
1481 if (bio_integrity(split
))
1482 bio_integrity_trim(split
);
1484 bio_advance(bio
, split
->bi_iter
.bi_size
);
1486 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1487 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1491 EXPORT_SYMBOL(bio_split
);
1494 * bio_trim - trim a bio
1496 * @offset: number of sectors to trim from the front of @bio
1497 * @size: size we want to trim @bio to, in sectors
1499 void bio_trim(struct bio
*bio
, int offset
, int size
)
1501 /* 'bio' is a cloned bio which we need to trim to match
1502 * the given offset and size.
1506 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1509 bio_advance(bio
, offset
<< 9);
1510 bio
->bi_iter
.bi_size
= size
;
1512 if (bio_integrity(bio
))
1513 bio_integrity_trim(bio
);
1516 EXPORT_SYMBOL_GPL(bio_trim
);
1519 * create memory pools for biovec's in a bio_set.
1520 * use the global biovec slabs created for general use.
1522 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1524 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
1526 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1530 * bioset_exit - exit a bioset initialized with bioset_init()
1532 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1535 void bioset_exit(struct bio_set
*bs
)
1537 if (bs
->rescue_workqueue
)
1538 destroy_workqueue(bs
->rescue_workqueue
);
1539 bs
->rescue_workqueue
= NULL
;
1541 mempool_exit(&bs
->bio_pool
);
1542 mempool_exit(&bs
->bvec_pool
);
1544 bioset_integrity_free(bs
);
1547 bs
->bio_slab
= NULL
;
1549 EXPORT_SYMBOL(bioset_exit
);
1552 * bioset_init - Initialize a bio_set
1553 * @bs: pool to initialize
1554 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1555 * @front_pad: Number of bytes to allocate in front of the returned bio
1556 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1557 * and %BIOSET_NEED_RESCUER
1560 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1561 * to ask for a number of bytes to be allocated in front of the bio.
1562 * Front pad allocation is useful for embedding the bio inside
1563 * another structure, to avoid allocating extra data to go with the bio.
1564 * Note that the bio must be embedded at the END of that structure always,
1565 * or things will break badly.
1566 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1567 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1568 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1569 * dispatch queued requests when the mempool runs out of space.
1572 int bioset_init(struct bio_set
*bs
,
1573 unsigned int pool_size
,
1574 unsigned int front_pad
,
1577 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1579 bs
->front_pad
= front_pad
;
1581 spin_lock_init(&bs
->rescue_lock
);
1582 bio_list_init(&bs
->rescue_list
);
1583 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1585 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1589 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
1592 if ((flags
& BIOSET_NEED_BVECS
) &&
1593 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
1596 if (!(flags
& BIOSET_NEED_RESCUER
))
1599 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1600 if (!bs
->rescue_workqueue
)
1608 EXPORT_SYMBOL(bioset_init
);
1611 * Initialize and setup a new bio_set, based on the settings from
1614 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
1619 if (src
->bvec_pool
.min_nr
)
1620 flags
|= BIOSET_NEED_BVECS
;
1621 if (src
->rescue_workqueue
)
1622 flags
|= BIOSET_NEED_RESCUER
;
1624 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
1626 EXPORT_SYMBOL(bioset_init_from_src
);
1628 #ifdef CONFIG_BLK_CGROUP
1631 * bio_disassociate_blkg - puts back the blkg reference if associated
1634 * Helper to disassociate the blkg from @bio if a blkg is associated.
1636 void bio_disassociate_blkg(struct bio
*bio
)
1639 blkg_put(bio
->bi_blkg
);
1640 bio
->bi_blkg
= NULL
;
1643 EXPORT_SYMBOL_GPL(bio_disassociate_blkg
);
1646 * __bio_associate_blkg - associate a bio with the a blkg
1648 * @blkg: the blkg to associate
1650 * This tries to associate @bio with the specified @blkg. Association failure
1651 * is handled by walking up the blkg tree. Therefore, the blkg associated can
1652 * be anything between @blkg and the root_blkg. This situation only happens
1653 * when a cgroup is dying and then the remaining bios will spill to the closest
1656 * A reference will be taken on the @blkg and will be released when @bio is
1659 static void __bio_associate_blkg(struct bio
*bio
, struct blkcg_gq
*blkg
)
1661 bio_disassociate_blkg(bio
);
1663 bio
->bi_blkg
= blkg_tryget_closest(blkg
);
1667 * bio_associate_blkg_from_css - associate a bio with a specified css
1671 * Associate @bio with the blkg found by combining the css's blkg and the
1672 * request_queue of the @bio. This falls back to the queue's root_blkg if
1673 * the association fails with the css.
1675 void bio_associate_blkg_from_css(struct bio
*bio
,
1676 struct cgroup_subsys_state
*css
)
1678 struct request_queue
*q
= bio
->bi_disk
->queue
;
1679 struct blkcg_gq
*blkg
;
1683 if (!css
|| !css
->parent
)
1684 blkg
= q
->root_blkg
;
1686 blkg
= blkg_lookup_create(css_to_blkcg(css
), q
);
1688 __bio_associate_blkg(bio
, blkg
);
1692 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css
);
1696 * bio_associate_blkg_from_page - associate a bio with the page's blkg
1698 * @page: the page to lookup the blkcg from
1700 * Associate @bio with the blkg from @page's owning memcg and the respective
1701 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's
1704 void bio_associate_blkg_from_page(struct bio
*bio
, struct page
*page
)
1706 struct cgroup_subsys_state
*css
;
1708 if (!page
->mem_cgroup
)
1713 css
= cgroup_e_css(page
->mem_cgroup
->css
.cgroup
, &io_cgrp_subsys
);
1714 bio_associate_blkg_from_css(bio
, css
);
1718 #endif /* CONFIG_MEMCG */
1721 * bio_associate_blkg - associate a bio with a blkg
1724 * Associate @bio with the blkg found from the bio's css and request_queue.
1725 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is
1726 * already associated, the css is reused and association redone as the
1727 * request_queue may have changed.
1729 void bio_associate_blkg(struct bio
*bio
)
1731 struct cgroup_subsys_state
*css
;
1736 css
= &bio_blkcg(bio
)->css
;
1740 bio_associate_blkg_from_css(bio
, css
);
1744 EXPORT_SYMBOL_GPL(bio_associate_blkg
);
1747 * bio_clone_blkg_association - clone blkg association from src to dst bio
1748 * @dst: destination bio
1751 void bio_clone_blkg_association(struct bio
*dst
, struct bio
*src
)
1756 __bio_associate_blkg(dst
, src
->bi_blkg
);
1760 EXPORT_SYMBOL_GPL(bio_clone_blkg_association
);
1761 #endif /* CONFIG_BLK_CGROUP */
1763 static void __init
biovec_init_slabs(void)
1767 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
1769 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1771 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
1776 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1777 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1778 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1782 static int __init
init_bio(void)
1786 bio_slabs
= kcalloc(bio_slab_max
, sizeof(struct bio_slab
),
1789 BUILD_BUG_ON(BIO_FLAG_LAST
> BVEC_POOL_OFFSET
);
1792 panic("bio: can't allocate bios\n");
1794 bio_integrity_init();
1795 biovec_init_slabs();
1797 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
1798 panic("bio: can't allocate bios\n");
1800 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
1801 panic("bio: can't create integrity pool\n");
1805 subsys_initcall(init_bio
);