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>
22 #include <trace/events/block.h>
24 #include "blk-rq-qos.h"
27 * Test patch to inline a certain number of bi_io_vec's inside the bio
28 * itself, to shrink a bio data allocation from two mempool calls to one
30 #define BIO_INLINE_VECS 4
33 * if you change this list, also change bvec_alloc or things will
34 * break badly! cannot be bigger than what you can fit into an
37 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
38 static struct biovec_slab bvec_slabs
[BVEC_POOL_NR
] __read_mostly
= {
39 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES
, max
),
44 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
45 * IO code that does not need private memory pools.
47 struct bio_set fs_bio_set
;
48 EXPORT_SYMBOL(fs_bio_set
);
51 * Our slab pool management
54 struct kmem_cache
*slab
;
55 unsigned int slab_ref
;
56 unsigned int slab_size
;
59 static DEFINE_MUTEX(bio_slab_lock
);
60 static struct bio_slab
*bio_slabs
;
61 static unsigned int bio_slab_nr
, bio_slab_max
;
63 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
65 unsigned int sz
= sizeof(struct bio
) + extra_size
;
66 struct kmem_cache
*slab
= NULL
;
67 struct bio_slab
*bslab
, *new_bio_slabs
;
68 unsigned int new_bio_slab_max
;
69 unsigned int i
, entry
= -1;
71 mutex_lock(&bio_slab_lock
);
74 while (i
< bio_slab_nr
) {
75 bslab
= &bio_slabs
[i
];
77 if (!bslab
->slab
&& entry
== -1)
79 else if (bslab
->slab_size
== sz
) {
90 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
91 new_bio_slab_max
= bio_slab_max
<< 1;
92 new_bio_slabs
= krealloc(bio_slabs
,
93 new_bio_slab_max
* sizeof(struct bio_slab
),
97 bio_slab_max
= new_bio_slab_max
;
98 bio_slabs
= new_bio_slabs
;
101 entry
= bio_slab_nr
++;
103 bslab
= &bio_slabs
[entry
];
105 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
106 slab
= kmem_cache_create(bslab
->name
, sz
, ARCH_KMALLOC_MINALIGN
,
107 SLAB_HWCACHE_ALIGN
, NULL
);
113 bslab
->slab_size
= sz
;
115 mutex_unlock(&bio_slab_lock
);
119 static void bio_put_slab(struct bio_set
*bs
)
121 struct bio_slab
*bslab
= NULL
;
124 mutex_lock(&bio_slab_lock
);
126 for (i
= 0; i
< bio_slab_nr
; i
++) {
127 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
128 bslab
= &bio_slabs
[i
];
133 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
136 WARN_ON(!bslab
->slab_ref
);
138 if (--bslab
->slab_ref
)
141 kmem_cache_destroy(bslab
->slab
);
145 mutex_unlock(&bio_slab_lock
);
148 unsigned int bvec_nr_vecs(unsigned short idx
)
150 return bvec_slabs
[--idx
].nr_vecs
;
153 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
159 BIO_BUG_ON(idx
>= BVEC_POOL_NR
);
161 if (idx
== BVEC_POOL_MAX
) {
162 mempool_free(bv
, pool
);
164 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
166 kmem_cache_free(bvs
->slab
, bv
);
170 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
176 * see comment near bvec_array define!
194 case 129 ... BIO_MAX_PAGES
:
202 * idx now points to the pool we want to allocate from. only the
203 * 1-vec entry pool is mempool backed.
205 if (*idx
== BVEC_POOL_MAX
) {
207 bvl
= mempool_alloc(pool
, gfp_mask
);
209 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
210 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
);
213 * Make this allocation restricted and don't dump info on
214 * allocation failures, since we'll fallback to the mempool
215 * in case of failure.
217 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
220 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
221 * is set, retry with the 1-entry mempool
223 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
224 if (unlikely(!bvl
&& (gfp_mask
& __GFP_DIRECT_RECLAIM
))) {
225 *idx
= BVEC_POOL_MAX
;
234 void bio_uninit(struct bio
*bio
)
236 bio_disassociate_blkg(bio
);
238 if (bio_integrity(bio
))
239 bio_integrity_free(bio
);
241 EXPORT_SYMBOL(bio_uninit
);
243 static void bio_free(struct bio
*bio
)
245 struct bio_set
*bs
= bio
->bi_pool
;
251 bvec_free(&bs
->bvec_pool
, bio
->bi_io_vec
, BVEC_POOL_IDX(bio
));
254 * If we have front padding, adjust the bio pointer before freeing
259 mempool_free(p
, &bs
->bio_pool
);
261 /* Bio was allocated by bio_kmalloc() */
267 * Users of this function have their own bio allocation. Subsequently,
268 * they must remember to pair any call to bio_init() with bio_uninit()
269 * when IO has completed, or when the bio is released.
271 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
272 unsigned short max_vecs
)
274 memset(bio
, 0, sizeof(*bio
));
275 atomic_set(&bio
->__bi_remaining
, 1);
276 atomic_set(&bio
->__bi_cnt
, 1);
278 bio
->bi_io_vec
= table
;
279 bio
->bi_max_vecs
= max_vecs
;
281 EXPORT_SYMBOL(bio_init
);
284 * bio_reset - reinitialize a bio
288 * After calling bio_reset(), @bio will be in the same state as a freshly
289 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
290 * preserved are the ones that are initialized by bio_alloc_bioset(). See
291 * comment in struct bio.
293 void bio_reset(struct bio
*bio
)
295 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
299 memset(bio
, 0, BIO_RESET_BYTES
);
300 bio
->bi_flags
= flags
;
301 atomic_set(&bio
->__bi_remaining
, 1);
303 EXPORT_SYMBOL(bio_reset
);
305 static struct bio
*__bio_chain_endio(struct bio
*bio
)
307 struct bio
*parent
= bio
->bi_private
;
309 if (!parent
->bi_status
)
310 parent
->bi_status
= bio
->bi_status
;
315 static void bio_chain_endio(struct bio
*bio
)
317 bio_endio(__bio_chain_endio(bio
));
321 * bio_chain - chain bio completions
322 * @bio: the target bio
323 * @parent: the @bio's parent bio
325 * The caller won't have a bi_end_io called when @bio completes - instead,
326 * @parent's bi_end_io won't be called until both @parent and @bio have
327 * completed; the chained bio will also be freed when it completes.
329 * The caller must not set bi_private or bi_end_io in @bio.
331 void bio_chain(struct bio
*bio
, struct bio
*parent
)
333 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
335 bio
->bi_private
= parent
;
336 bio
->bi_end_io
= bio_chain_endio
;
337 bio_inc_remaining(parent
);
339 EXPORT_SYMBOL(bio_chain
);
341 static void bio_alloc_rescue(struct work_struct
*work
)
343 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
347 spin_lock(&bs
->rescue_lock
);
348 bio
= bio_list_pop(&bs
->rescue_list
);
349 spin_unlock(&bs
->rescue_lock
);
354 generic_make_request(bio
);
358 static void punt_bios_to_rescuer(struct bio_set
*bs
)
360 struct bio_list punt
, nopunt
;
363 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
366 * In order to guarantee forward progress we must punt only bios that
367 * were allocated from this bio_set; otherwise, if there was a bio on
368 * there for a stacking driver higher up in the stack, processing it
369 * could require allocating bios from this bio_set, and doing that from
370 * our own rescuer would be bad.
372 * Since bio lists are singly linked, pop them all instead of trying to
373 * remove from the middle of the list:
376 bio_list_init(&punt
);
377 bio_list_init(&nopunt
);
379 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
380 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
381 current
->bio_list
[0] = nopunt
;
383 bio_list_init(&nopunt
);
384 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
385 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
386 current
->bio_list
[1] = nopunt
;
388 spin_lock(&bs
->rescue_lock
);
389 bio_list_merge(&bs
->rescue_list
, &punt
);
390 spin_unlock(&bs
->rescue_lock
);
392 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
396 * bio_alloc_bioset - allocate a bio for I/O
397 * @gfp_mask: the GFP_* mask given to the slab allocator
398 * @nr_iovecs: number of iovecs to pre-allocate
399 * @bs: the bio_set to allocate from.
402 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
403 * backed by the @bs's mempool.
405 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
406 * always be able to allocate a bio. This is due to the mempool guarantees.
407 * To make this work, callers must never allocate more than 1 bio at a time
408 * from this pool. Callers that need to allocate more than 1 bio must always
409 * submit the previously allocated bio for IO before attempting to allocate
410 * a new one. Failure to do so can cause deadlocks under memory pressure.
412 * Note that when running under generic_make_request() (i.e. any block
413 * driver), bios are not submitted until after you return - see the code in
414 * generic_make_request() that converts recursion into iteration, to prevent
417 * This would normally mean allocating multiple bios under
418 * generic_make_request() would be susceptible to deadlocks, but we have
419 * deadlock avoidance code that resubmits any blocked bios from a rescuer
422 * However, we do not guarantee forward progress for allocations from other
423 * mempools. Doing multiple allocations from the same mempool under
424 * generic_make_request() should be avoided - instead, use bio_set's front_pad
425 * for per bio allocations.
428 * Pointer to new bio on success, NULL on failure.
430 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned int nr_iovecs
,
433 gfp_t saved_gfp
= gfp_mask
;
435 unsigned inline_vecs
;
436 struct bio_vec
*bvl
= NULL
;
441 if (nr_iovecs
> UIO_MAXIOV
)
444 p
= kmalloc(sizeof(struct bio
) +
445 nr_iovecs
* sizeof(struct bio_vec
),
448 inline_vecs
= nr_iovecs
;
450 /* should not use nobvec bioset for nr_iovecs > 0 */
451 if (WARN_ON_ONCE(!mempool_initialized(&bs
->bvec_pool
) &&
455 * generic_make_request() converts recursion to iteration; this
456 * means if we're running beneath it, any bios we allocate and
457 * submit will not be submitted (and thus freed) until after we
460 * This exposes us to a potential deadlock if we allocate
461 * multiple bios from the same bio_set() while running
462 * underneath generic_make_request(). If we were to allocate
463 * multiple bios (say a stacking block driver that was splitting
464 * bios), we would deadlock if we exhausted the mempool's
467 * We solve this, and guarantee forward progress, with a rescuer
468 * workqueue per bio_set. If we go to allocate and there are
469 * bios on current->bio_list, we first try the allocation
470 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
471 * bios we would be blocking to the rescuer workqueue before
472 * we retry with the original gfp_flags.
475 if (current
->bio_list
&&
476 (!bio_list_empty(¤t
->bio_list
[0]) ||
477 !bio_list_empty(¤t
->bio_list
[1])) &&
478 bs
->rescue_workqueue
)
479 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
481 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
482 if (!p
&& gfp_mask
!= saved_gfp
) {
483 punt_bios_to_rescuer(bs
);
484 gfp_mask
= saved_gfp
;
485 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
488 front_pad
= bs
->front_pad
;
489 inline_vecs
= BIO_INLINE_VECS
;
496 bio_init(bio
, NULL
, 0);
498 if (nr_iovecs
> inline_vecs
) {
499 unsigned long idx
= 0;
501 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, &bs
->bvec_pool
);
502 if (!bvl
&& gfp_mask
!= saved_gfp
) {
503 punt_bios_to_rescuer(bs
);
504 gfp_mask
= saved_gfp
;
505 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, &bs
->bvec_pool
);
511 bio
->bi_flags
|= idx
<< BVEC_POOL_OFFSET
;
512 } else if (nr_iovecs
) {
513 bvl
= bio
->bi_inline_vecs
;
517 bio
->bi_max_vecs
= nr_iovecs
;
518 bio
->bi_io_vec
= bvl
;
522 mempool_free(p
, &bs
->bio_pool
);
525 EXPORT_SYMBOL(bio_alloc_bioset
);
527 void zero_fill_bio_iter(struct bio
*bio
, struct bvec_iter start
)
531 struct bvec_iter iter
;
533 __bio_for_each_segment(bv
, bio
, iter
, start
) {
534 char *data
= bvec_kmap_irq(&bv
, &flags
);
535 memset(data
, 0, bv
.bv_len
);
536 flush_dcache_page(bv
.bv_page
);
537 bvec_kunmap_irq(data
, &flags
);
540 EXPORT_SYMBOL(zero_fill_bio_iter
);
543 * bio_truncate - truncate the bio to small size of @new_size
544 * @bio: the bio to be truncated
545 * @new_size: new size for truncating the bio
548 * Truncate the bio to new size of @new_size. If bio_op(bio) is
549 * REQ_OP_READ, zero the truncated part. This function should only
550 * be used for handling corner cases, such as bio eod.
552 void bio_truncate(struct bio
*bio
, unsigned new_size
)
555 struct bvec_iter iter
;
556 unsigned int done
= 0;
557 bool truncated
= false;
559 if (new_size
>= bio
->bi_iter
.bi_size
)
562 if (bio_op(bio
) != REQ_OP_READ
)
565 bio_for_each_segment(bv
, bio
, iter
) {
566 if (done
+ bv
.bv_len
> new_size
) {
570 offset
= new_size
- done
;
573 zero_user(bv
.bv_page
, offset
, bv
.bv_len
- offset
);
581 * Don't touch bvec table here and make it really immutable, since
582 * fs bio user has to retrieve all pages via bio_for_each_segment_all
583 * in its .end_bio() callback.
585 * It is enough to truncate bio by updating .bi_size since we can make
586 * correct bvec with the updated .bi_size for drivers.
588 bio
->bi_iter
.bi_size
= new_size
;
592 * guard_bio_eod - truncate a BIO to fit the block device
593 * @bio: bio to truncate
595 * This allows us to do IO even on the odd last sectors of a device, even if the
596 * block size is some multiple of the physical sector size.
598 * We'll just truncate the bio to the size of the device, and clear the end of
599 * the buffer head manually. Truly out-of-range accesses will turn into actual
600 * I/O errors, this only handles the "we need to be able to do I/O at the final
603 void guard_bio_eod(struct bio
*bio
)
606 struct hd_struct
*part
;
609 part
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
611 maxsector
= part_nr_sects_read(part
);
613 maxsector
= get_capacity(bio
->bi_disk
);
620 * If the *whole* IO is past the end of the device,
621 * let it through, and the IO layer will turn it into
624 if (unlikely(bio
->bi_iter
.bi_sector
>= maxsector
))
627 maxsector
-= bio
->bi_iter
.bi_sector
;
628 if (likely((bio
->bi_iter
.bi_size
>> 9) <= maxsector
))
631 bio_truncate(bio
, maxsector
<< 9);
635 * bio_put - release a reference to a bio
636 * @bio: bio to release reference to
639 * Put a reference to a &struct bio, either one you have gotten with
640 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
642 void bio_put(struct bio
*bio
)
644 if (!bio_flagged(bio
, BIO_REFFED
))
647 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
652 if (atomic_dec_and_test(&bio
->__bi_cnt
))
656 EXPORT_SYMBOL(bio_put
);
659 * __bio_clone_fast - clone a bio that shares the original bio's biovec
660 * @bio: destination bio
661 * @bio_src: bio to clone
663 * Clone a &bio. Caller will own the returned bio, but not
664 * the actual data it points to. Reference count of returned
667 * Caller must ensure that @bio_src is not freed before @bio.
669 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
671 BUG_ON(bio
->bi_pool
&& BVEC_POOL_IDX(bio
));
674 * most users will be overriding ->bi_disk with a new target,
675 * so we don't set nor calculate new physical/hw segment counts here
677 bio
->bi_disk
= bio_src
->bi_disk
;
678 bio
->bi_partno
= bio_src
->bi_partno
;
679 bio_set_flag(bio
, BIO_CLONED
);
680 if (bio_flagged(bio_src
, BIO_THROTTLED
))
681 bio_set_flag(bio
, BIO_THROTTLED
);
682 bio
->bi_opf
= bio_src
->bi_opf
;
683 bio
->bi_ioprio
= bio_src
->bi_ioprio
;
684 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
685 bio
->bi_iter
= bio_src
->bi_iter
;
686 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
688 bio_clone_blkg_association(bio
, bio_src
);
689 blkcg_bio_issue_init(bio
);
691 EXPORT_SYMBOL(__bio_clone_fast
);
694 * bio_clone_fast - clone a bio that shares the original bio's biovec
696 * @gfp_mask: allocation priority
697 * @bs: bio_set to allocate from
699 * Like __bio_clone_fast, only also allocates the returned bio
701 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
705 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
709 __bio_clone_fast(b
, bio
);
711 if (bio_integrity(bio
)) {
714 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
724 EXPORT_SYMBOL(bio_clone_fast
);
726 const char *bio_devname(struct bio
*bio
, char *buf
)
728 return disk_name(bio
->bi_disk
, bio
->bi_partno
, buf
);
730 EXPORT_SYMBOL(bio_devname
);
732 static inline bool page_is_mergeable(const struct bio_vec
*bv
,
733 struct page
*page
, unsigned int len
, unsigned int off
,
736 phys_addr_t vec_end_addr
= page_to_phys(bv
->bv_page
) +
737 bv
->bv_offset
+ bv
->bv_len
- 1;
738 phys_addr_t page_addr
= page_to_phys(page
);
740 if (vec_end_addr
+ 1 != page_addr
+ off
)
742 if (xen_domain() && !xen_biovec_phys_mergeable(bv
, page
))
745 *same_page
= ((vec_end_addr
& PAGE_MASK
) == page_addr
);
746 if (!*same_page
&& pfn_to_page(PFN_DOWN(vec_end_addr
)) + 1 != page
)
751 static bool bio_try_merge_pc_page(struct request_queue
*q
, struct bio
*bio
,
752 struct page
*page
, unsigned len
, unsigned offset
,
755 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
756 unsigned long mask
= queue_segment_boundary(q
);
757 phys_addr_t addr1
= page_to_phys(bv
->bv_page
) + bv
->bv_offset
;
758 phys_addr_t addr2
= page_to_phys(page
) + offset
+ len
- 1;
760 if ((addr1
| mask
) != (addr2
| mask
))
762 if (bv
->bv_len
+ len
> queue_max_segment_size(q
))
764 return __bio_try_merge_page(bio
, page
, len
, offset
, same_page
);
768 * __bio_add_pc_page - attempt to add page to passthrough bio
769 * @q: the target queue
770 * @bio: destination bio
772 * @len: vec entry length
773 * @offset: vec entry offset
774 * @same_page: return if the merge happen inside the same page
776 * Attempt to add a page to the bio_vec maplist. This can fail for a
777 * number of reasons, such as the bio being full or target block device
778 * limitations. The target block device must allow bio's up to PAGE_SIZE,
779 * so it is always possible to add a single page to an empty bio.
781 * This should only be used by passthrough bios.
783 static int __bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
,
784 struct page
*page
, unsigned int len
, unsigned int offset
,
787 struct bio_vec
*bvec
;
790 * cloned bio must not modify vec list
792 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
795 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > queue_max_hw_sectors(q
))
798 if (bio
->bi_vcnt
> 0) {
799 if (bio_try_merge_pc_page(q
, bio
, page
, len
, offset
, same_page
))
803 * If the queue doesn't support SG gaps and adding this segment
804 * would create a gap, disallow it.
806 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
807 if (bvec_gap_to_prev(q
, bvec
, offset
))
811 if (bio_full(bio
, len
))
814 if (bio
->bi_vcnt
>= queue_max_segments(q
))
817 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
818 bvec
->bv_page
= page
;
820 bvec
->bv_offset
= offset
;
822 bio
->bi_iter
.bi_size
+= len
;
826 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
,
827 struct page
*page
, unsigned int len
, unsigned int offset
)
829 bool same_page
= false;
830 return __bio_add_pc_page(q
, bio
, page
, len
, offset
, &same_page
);
832 EXPORT_SYMBOL(bio_add_pc_page
);
835 * __bio_try_merge_page - try appending data to an existing bvec.
836 * @bio: destination bio
837 * @page: start page to add
838 * @len: length of the data to add
839 * @off: offset of the data relative to @page
840 * @same_page: return if the segment has been merged inside the same page
842 * Try to add the data at @page + @off to the last bvec of @bio. This is a
843 * a useful optimisation for file systems with a block size smaller than the
846 * Warn if (@len, @off) crosses pages in case that @same_page is true.
848 * Return %true on success or %false on failure.
850 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
851 unsigned int len
, unsigned int off
, bool *same_page
)
853 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
856 if (bio
->bi_vcnt
> 0) {
857 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
859 if (page_is_mergeable(bv
, page
, len
, off
, same_page
)) {
860 if (bio
->bi_iter
.bi_size
> UINT_MAX
- len
)
863 bio
->bi_iter
.bi_size
+= len
;
869 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
872 * __bio_add_page - add page(s) to a bio in a new segment
873 * @bio: destination bio
874 * @page: start page to add
875 * @len: length of the data to add, may cross pages
876 * @off: offset of the data relative to @page, may cross pages
878 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
879 * that @bio has space for another bvec.
881 void __bio_add_page(struct bio
*bio
, struct page
*page
,
882 unsigned int len
, unsigned int off
)
884 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
886 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
887 WARN_ON_ONCE(bio_full(bio
, len
));
893 bio
->bi_iter
.bi_size
+= len
;
896 if (!bio_flagged(bio
, BIO_WORKINGSET
) && unlikely(PageWorkingset(page
)))
897 bio_set_flag(bio
, BIO_WORKINGSET
);
899 EXPORT_SYMBOL_GPL(__bio_add_page
);
902 * bio_add_page - attempt to add page(s) to bio
903 * @bio: destination bio
904 * @page: start page to add
905 * @len: vec entry length, may cross pages
906 * @offset: vec entry offset relative to @page, may cross pages
908 * Attempt to add page(s) to the bio_vec maplist. This will only fail
909 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
911 int bio_add_page(struct bio
*bio
, struct page
*page
,
912 unsigned int len
, unsigned int offset
)
914 bool same_page
= false;
916 if (!__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
917 if (bio_full(bio
, len
))
919 __bio_add_page(bio
, page
, len
, offset
);
923 EXPORT_SYMBOL(bio_add_page
);
925 void bio_release_pages(struct bio
*bio
, bool mark_dirty
)
927 struct bvec_iter_all iter_all
;
928 struct bio_vec
*bvec
;
930 if (bio_flagged(bio
, BIO_NO_PAGE_REF
))
933 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
934 if (mark_dirty
&& !PageCompound(bvec
->bv_page
))
935 set_page_dirty_lock(bvec
->bv_page
);
936 put_page(bvec
->bv_page
);
940 static int __bio_iov_bvec_add_pages(struct bio
*bio
, struct iov_iter
*iter
)
942 const struct bio_vec
*bv
= iter
->bvec
;
946 if (WARN_ON_ONCE(iter
->iov_offset
> bv
->bv_len
))
949 len
= min_t(size_t, bv
->bv_len
- iter
->iov_offset
, iter
->count
);
950 size
= bio_add_page(bio
, bv
->bv_page
, len
,
951 bv
->bv_offset
+ iter
->iov_offset
);
952 if (unlikely(size
!= len
))
954 iov_iter_advance(iter
, size
);
958 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
961 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
962 * @bio: bio to add pages to
963 * @iter: iov iterator describing the region to be mapped
965 * Pins pages from *iter and appends them to @bio's bvec array. The
966 * pages will have to be released using put_page() when done.
967 * For multi-segment *iter, this function only adds pages from the
968 * the next non-empty segment of the iov iterator.
970 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
972 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
973 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
974 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
975 struct page
**pages
= (struct page
**)bv
;
976 bool same_page
= false;
982 * Move page array up in the allocated memory for the bio vecs as far as
983 * possible so that we can start filling biovecs from the beginning
984 * without overwriting the temporary page array.
986 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
987 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
989 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
990 if (unlikely(size
<= 0))
991 return size
? size
: -EFAULT
;
993 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
994 struct page
*page
= pages
[i
];
996 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
998 if (__bio_try_merge_page(bio
, page
, len
, offset
, &same_page
)) {
1002 if (WARN_ON_ONCE(bio_full(bio
, len
)))
1004 __bio_add_page(bio
, page
, len
, offset
);
1009 iov_iter_advance(iter
, size
);
1014 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1015 * @bio: bio to add pages to
1016 * @iter: iov iterator describing the region to be added
1018 * This takes either an iterator pointing to user memory, or one pointing to
1019 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1020 * map them into the kernel. On IO completion, the caller should put those
1021 * pages. If we're adding kernel pages, and the caller told us it's safe to
1022 * do so, we just have to add the pages to the bio directly. We don't grab an
1023 * extra reference to those pages (the user should already have that), and we
1024 * don't put the page on IO completion. The caller needs to check if the bio is
1025 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
1028 * The function tries, but does not guarantee, to pin as many pages as
1029 * fit into the bio, or are requested in *iter, whatever is smaller. If
1030 * MM encounters an error pinning the requested pages, it stops. Error
1031 * is returned only if 0 pages could be pinned.
1033 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
1035 const bool is_bvec
= iov_iter_is_bvec(iter
);
1038 if (WARN_ON_ONCE(bio
->bi_vcnt
))
1043 ret
= __bio_iov_bvec_add_pages(bio
, iter
);
1045 ret
= __bio_iov_iter_get_pages(bio
, iter
);
1046 } while (!ret
&& iov_iter_count(iter
) && !bio_full(bio
, 0));
1049 bio_set_flag(bio
, BIO_NO_PAGE_REF
);
1050 return bio
->bi_vcnt
? 0 : ret
;
1053 static void submit_bio_wait_endio(struct bio
*bio
)
1055 complete(bio
->bi_private
);
1059 * submit_bio_wait - submit a bio, and wait until it completes
1060 * @bio: The &struct bio which describes the I/O
1062 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1063 * bio_endio() on failure.
1065 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1066 * result in bio reference to be consumed. The caller must drop the reference
1069 int submit_bio_wait(struct bio
*bio
)
1071 DECLARE_COMPLETION_ONSTACK_MAP(done
, bio
->bi_disk
->lockdep_map
);
1072 unsigned long hang_check
;
1074 bio
->bi_private
= &done
;
1075 bio
->bi_end_io
= submit_bio_wait_endio
;
1076 bio
->bi_opf
|= REQ_SYNC
;
1079 /* Prevent hang_check timer from firing at us during very long I/O */
1080 hang_check
= sysctl_hung_task_timeout_secs
;
1082 while (!wait_for_completion_io_timeout(&done
,
1083 hang_check
* (HZ
/2)))
1086 wait_for_completion_io(&done
);
1088 return blk_status_to_errno(bio
->bi_status
);
1090 EXPORT_SYMBOL(submit_bio_wait
);
1093 * bio_advance - increment/complete a bio by some number of bytes
1094 * @bio: bio to advance
1095 * @bytes: number of bytes to complete
1097 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1098 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1099 * be updated on the last bvec as well.
1101 * @bio will then represent the remaining, uncompleted portion of the io.
1103 void bio_advance(struct bio
*bio
, unsigned bytes
)
1105 if (bio_integrity(bio
))
1106 bio_integrity_advance(bio
, bytes
);
1108 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
1110 EXPORT_SYMBOL(bio_advance
);
1112 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
1113 struct bio
*src
, struct bvec_iter
*src_iter
)
1115 struct bio_vec src_bv
, dst_bv
;
1116 void *src_p
, *dst_p
;
1119 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
1120 src_bv
= bio_iter_iovec(src
, *src_iter
);
1121 dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
1123 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1125 src_p
= kmap_atomic(src_bv
.bv_page
);
1126 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1128 memcpy(dst_p
+ dst_bv
.bv_offset
,
1129 src_p
+ src_bv
.bv_offset
,
1132 kunmap_atomic(dst_p
);
1133 kunmap_atomic(src_p
);
1135 flush_dcache_page(dst_bv
.bv_page
);
1137 bio_advance_iter(src
, src_iter
, bytes
);
1138 bio_advance_iter(dst
, dst_iter
, bytes
);
1141 EXPORT_SYMBOL(bio_copy_data_iter
);
1144 * bio_copy_data - copy contents of data buffers from one bio to another
1146 * @dst: destination bio
1148 * Stops when it reaches the end of either @src or @dst - that is, copies
1149 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1151 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1153 struct bvec_iter src_iter
= src
->bi_iter
;
1154 struct bvec_iter dst_iter
= dst
->bi_iter
;
1156 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1158 EXPORT_SYMBOL(bio_copy_data
);
1161 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1163 * @src: source bio list
1164 * @dst: destination bio list
1166 * Stops when it reaches the end of either the @src list or @dst list - that is,
1167 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1170 void bio_list_copy_data(struct bio
*dst
, struct bio
*src
)
1172 struct bvec_iter src_iter
= src
->bi_iter
;
1173 struct bvec_iter dst_iter
= dst
->bi_iter
;
1176 if (!src_iter
.bi_size
) {
1181 src_iter
= src
->bi_iter
;
1184 if (!dst_iter
.bi_size
) {
1189 dst_iter
= dst
->bi_iter
;
1192 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1195 EXPORT_SYMBOL(bio_list_copy_data
);
1197 struct bio_map_data
{
1199 struct iov_iter iter
;
1203 static struct bio_map_data
*bio_alloc_map_data(struct iov_iter
*data
,
1206 struct bio_map_data
*bmd
;
1207 if (data
->nr_segs
> UIO_MAXIOV
)
1210 bmd
= kmalloc(struct_size(bmd
, iov
, data
->nr_segs
), gfp_mask
);
1213 memcpy(bmd
->iov
, data
->iov
, sizeof(struct iovec
) * data
->nr_segs
);
1215 bmd
->iter
.iov
= bmd
->iov
;
1220 * bio_copy_from_iter - copy all pages from iov_iter to bio
1221 * @bio: The &struct bio which describes the I/O as destination
1222 * @iter: iov_iter as source
1224 * Copy all pages from iov_iter to bio.
1225 * Returns 0 on success, or error on failure.
1227 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter
*iter
)
1229 struct bio_vec
*bvec
;
1230 struct bvec_iter_all iter_all
;
1232 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1235 ret
= copy_page_from_iter(bvec
->bv_page
,
1240 if (!iov_iter_count(iter
))
1243 if (ret
< bvec
->bv_len
)
1251 * bio_copy_to_iter - copy all pages from bio to iov_iter
1252 * @bio: The &struct bio which describes the I/O as source
1253 * @iter: iov_iter as destination
1255 * Copy all pages from bio to iov_iter.
1256 * Returns 0 on success, or error on failure.
1258 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1260 struct bio_vec
*bvec
;
1261 struct bvec_iter_all iter_all
;
1263 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1266 ret
= copy_page_to_iter(bvec
->bv_page
,
1271 if (!iov_iter_count(&iter
))
1274 if (ret
< bvec
->bv_len
)
1281 void bio_free_pages(struct bio
*bio
)
1283 struct bio_vec
*bvec
;
1284 struct bvec_iter_all iter_all
;
1286 bio_for_each_segment_all(bvec
, bio
, iter_all
)
1287 __free_page(bvec
->bv_page
);
1289 EXPORT_SYMBOL(bio_free_pages
);
1292 * bio_uncopy_user - finish previously mapped bio
1293 * @bio: bio being terminated
1295 * Free pages allocated from bio_copy_user_iov() and write back data
1296 * to user space in case of a read.
1298 int bio_uncopy_user(struct bio
*bio
)
1300 struct bio_map_data
*bmd
= bio
->bi_private
;
1303 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1305 * if we're in a workqueue, the request is orphaned, so
1306 * don't copy into a random user address space, just free
1307 * and return -EINTR so user space doesn't expect any data.
1311 else if (bio_data_dir(bio
) == READ
)
1312 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1313 if (bmd
->is_our_pages
)
1314 bio_free_pages(bio
);
1322 * bio_copy_user_iov - copy user data to bio
1323 * @q: destination block queue
1324 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1325 * @iter: iovec iterator
1326 * @gfp_mask: memory allocation flags
1328 * Prepares and returns a bio for indirect user io, bouncing data
1329 * to/from kernel pages as necessary. Must be paired with
1330 * call bio_uncopy_user() on io completion.
1332 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1333 struct rq_map_data
*map_data
,
1334 struct iov_iter
*iter
,
1337 struct bio_map_data
*bmd
;
1342 unsigned int len
= iter
->count
;
1343 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
1345 bmd
= bio_alloc_map_data(iter
, gfp_mask
);
1347 return ERR_PTR(-ENOMEM
);
1350 * We need to do a deep copy of the iov_iter including the iovecs.
1351 * The caller provided iov might point to an on-stack or otherwise
1354 bmd
->is_our_pages
= map_data
? 0 : 1;
1356 nr_pages
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
);
1357 if (nr_pages
> BIO_MAX_PAGES
)
1358 nr_pages
= BIO_MAX_PAGES
;
1361 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1368 nr_pages
= 1 << map_data
->page_order
;
1369 i
= map_data
->offset
/ PAGE_SIZE
;
1372 unsigned int bytes
= PAGE_SIZE
;
1380 if (i
== map_data
->nr_entries
* nr_pages
) {
1385 page
= map_data
->pages
[i
/ nr_pages
];
1386 page
+= (i
% nr_pages
);
1390 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1397 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
) {
1411 map_data
->offset
+= bio
->bi_iter
.bi_size
;
1416 if ((iov_iter_rw(iter
) == WRITE
&& (!map_data
|| !map_data
->null_mapped
)) ||
1417 (map_data
&& map_data
->from_user
)) {
1418 ret
= bio_copy_from_iter(bio
, iter
);
1422 if (bmd
->is_our_pages
)
1424 iov_iter_advance(iter
, bio
->bi_iter
.bi_size
);
1427 bio
->bi_private
= bmd
;
1428 if (map_data
&& map_data
->null_mapped
)
1429 bio_set_flag(bio
, BIO_NULL_MAPPED
);
1433 bio_free_pages(bio
);
1437 return ERR_PTR(ret
);
1441 * bio_map_user_iov - map user iovec into bio
1442 * @q: the struct request_queue for the bio
1443 * @iter: iovec iterator
1444 * @gfp_mask: memory allocation flags
1446 * Map the user space address into a bio suitable for io to a block
1447 * device. Returns an error pointer in case of error.
1449 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1450 struct iov_iter
*iter
,
1457 if (!iov_iter_count(iter
))
1458 return ERR_PTR(-EINVAL
);
1460 bio
= bio_kmalloc(gfp_mask
, iov_iter_npages(iter
, BIO_MAX_PAGES
));
1462 return ERR_PTR(-ENOMEM
);
1464 while (iov_iter_count(iter
)) {
1465 struct page
**pages
;
1467 size_t offs
, added
= 0;
1470 bytes
= iov_iter_get_pages_alloc(iter
, &pages
, LONG_MAX
, &offs
);
1471 if (unlikely(bytes
<= 0)) {
1472 ret
= bytes
? bytes
: -EFAULT
;
1476 npages
= DIV_ROUND_UP(offs
+ bytes
, PAGE_SIZE
);
1478 if (unlikely(offs
& queue_dma_alignment(q
))) {
1482 for (j
= 0; j
< npages
; j
++) {
1483 struct page
*page
= pages
[j
];
1484 unsigned int n
= PAGE_SIZE
- offs
;
1485 bool same_page
= false;
1490 if (!__bio_add_pc_page(q
, bio
, page
, n
, offs
,
1501 iov_iter_advance(iter
, added
);
1504 * release the pages we didn't map into the bio, if any
1507 put_page(pages
[j
++]);
1509 /* couldn't stuff something into bio? */
1514 bio_set_flag(bio
, BIO_USER_MAPPED
);
1517 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1518 * it would normally disappear when its bi_end_io is run.
1519 * however, we need it for the unmap, so grab an extra
1526 bio_release_pages(bio
, false);
1528 return ERR_PTR(ret
);
1532 * bio_unmap_user - unmap a bio
1533 * @bio: the bio being unmapped
1535 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1538 * bio_unmap_user() may sleep.
1540 void bio_unmap_user(struct bio
*bio
)
1542 bio_release_pages(bio
, bio_data_dir(bio
) == READ
);
1547 static void bio_invalidate_vmalloc_pages(struct bio
*bio
)
1549 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
1550 if (bio
->bi_private
&& !op_is_write(bio_op(bio
))) {
1551 unsigned long i
, len
= 0;
1553 for (i
= 0; i
< bio
->bi_vcnt
; i
++)
1554 len
+= bio
->bi_io_vec
[i
].bv_len
;
1555 invalidate_kernel_vmap_range(bio
->bi_private
, len
);
1560 static void bio_map_kern_endio(struct bio
*bio
)
1562 bio_invalidate_vmalloc_pages(bio
);
1567 * bio_map_kern - map kernel address into bio
1568 * @q: the struct request_queue for the bio
1569 * @data: pointer to buffer to map
1570 * @len: length in bytes
1571 * @gfp_mask: allocation flags for bio allocation
1573 * Map the kernel address into a bio suitable for io to a block
1574 * device. Returns an error pointer in case of error.
1576 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1579 unsigned long kaddr
= (unsigned long)data
;
1580 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1581 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1582 const int nr_pages
= end
- start
;
1583 bool is_vmalloc
= is_vmalloc_addr(data
);
1588 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1590 return ERR_PTR(-ENOMEM
);
1593 flush_kernel_vmap_range(data
, len
);
1594 bio
->bi_private
= data
;
1597 offset
= offset_in_page(kaddr
);
1598 for (i
= 0; i
< nr_pages
; i
++) {
1599 unsigned int bytes
= PAGE_SIZE
- offset
;
1608 page
= virt_to_page(data
);
1610 page
= vmalloc_to_page(data
);
1611 if (bio_add_pc_page(q
, bio
, page
, bytes
,
1613 /* we don't support partial mappings */
1615 return ERR_PTR(-EINVAL
);
1623 bio
->bi_end_io
= bio_map_kern_endio
;
1627 static void bio_copy_kern_endio(struct bio
*bio
)
1629 bio_free_pages(bio
);
1633 static void bio_copy_kern_endio_read(struct bio
*bio
)
1635 char *p
= bio
->bi_private
;
1636 struct bio_vec
*bvec
;
1637 struct bvec_iter_all iter_all
;
1639 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1640 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1644 bio_copy_kern_endio(bio
);
1648 * bio_copy_kern - copy kernel address into bio
1649 * @q: the struct request_queue for the bio
1650 * @data: pointer to buffer to copy
1651 * @len: length in bytes
1652 * @gfp_mask: allocation flags for bio and page allocation
1653 * @reading: data direction is READ
1655 * copy the kernel address into a bio suitable for io to a block
1656 * device. Returns an error pointer in case of error.
1658 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1659 gfp_t gfp_mask
, int reading
)
1661 unsigned long kaddr
= (unsigned long)data
;
1662 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1663 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1672 return ERR_PTR(-EINVAL
);
1674 nr_pages
= end
- start
;
1675 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1677 return ERR_PTR(-ENOMEM
);
1681 unsigned int bytes
= PAGE_SIZE
;
1686 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1691 memcpy(page_address(page
), p
, bytes
);
1693 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1701 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1702 bio
->bi_private
= data
;
1704 bio
->bi_end_io
= bio_copy_kern_endio
;
1710 bio_free_pages(bio
);
1712 return ERR_PTR(-ENOMEM
);
1716 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1717 * for performing direct-IO in BIOs.
1719 * The problem is that we cannot run set_page_dirty() from interrupt context
1720 * because the required locks are not interrupt-safe. So what we can do is to
1721 * mark the pages dirty _before_ performing IO. And in interrupt context,
1722 * check that the pages are still dirty. If so, fine. If not, redirty them
1723 * in process context.
1725 * We special-case compound pages here: normally this means reads into hugetlb
1726 * pages. The logic in here doesn't really work right for compound pages
1727 * because the VM does not uniformly chase down the head page in all cases.
1728 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1729 * handle them at all. So we skip compound pages here at an early stage.
1731 * Note that this code is very hard to test under normal circumstances because
1732 * direct-io pins the pages with get_user_pages(). This makes
1733 * is_page_cache_freeable return false, and the VM will not clean the pages.
1734 * But other code (eg, flusher threads) could clean the pages if they are mapped
1737 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1738 * deferred bio dirtying paths.
1742 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1744 void bio_set_pages_dirty(struct bio
*bio
)
1746 struct bio_vec
*bvec
;
1747 struct bvec_iter_all iter_all
;
1749 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1750 if (!PageCompound(bvec
->bv_page
))
1751 set_page_dirty_lock(bvec
->bv_page
);
1756 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1757 * If they are, then fine. If, however, some pages are clean then they must
1758 * have been written out during the direct-IO read. So we take another ref on
1759 * the BIO and re-dirty the pages in process context.
1761 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1762 * here on. It will run one put_page() against each page and will run one
1763 * bio_put() against the BIO.
1766 static void bio_dirty_fn(struct work_struct
*work
);
1768 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1769 static DEFINE_SPINLOCK(bio_dirty_lock
);
1770 static struct bio
*bio_dirty_list
;
1773 * This runs in process context
1775 static void bio_dirty_fn(struct work_struct
*work
)
1777 struct bio
*bio
, *next
;
1779 spin_lock_irq(&bio_dirty_lock
);
1780 next
= bio_dirty_list
;
1781 bio_dirty_list
= NULL
;
1782 spin_unlock_irq(&bio_dirty_lock
);
1784 while ((bio
= next
) != NULL
) {
1785 next
= bio
->bi_private
;
1787 bio_release_pages(bio
, true);
1792 void bio_check_pages_dirty(struct bio
*bio
)
1794 struct bio_vec
*bvec
;
1795 unsigned long flags
;
1796 struct bvec_iter_all iter_all
;
1798 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
1799 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1803 bio_release_pages(bio
, false);
1807 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1808 bio
->bi_private
= bio_dirty_list
;
1809 bio_dirty_list
= bio
;
1810 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1811 schedule_work(&bio_dirty_work
);
1814 void update_io_ticks(struct hd_struct
*part
, unsigned long now
, bool end
)
1816 unsigned long stamp
;
1818 stamp
= READ_ONCE(part
->stamp
);
1819 if (unlikely(stamp
!= now
)) {
1820 if (likely(cmpxchg(&part
->stamp
, stamp
, now
) == stamp
)) {
1821 __part_stat_add(part
, io_ticks
, end
? now
- stamp
: 1);
1825 part
= &part_to_disk(part
)->part0
;
1830 void generic_start_io_acct(struct request_queue
*q
, int op
,
1831 unsigned long sectors
, struct hd_struct
*part
)
1833 const int sgrp
= op_stat_group(op
);
1837 update_io_ticks(part
, jiffies
, false);
1838 part_stat_inc(part
, ios
[sgrp
]);
1839 part_stat_add(part
, sectors
[sgrp
], sectors
);
1840 part_inc_in_flight(q
, part
, op_is_write(op
));
1844 EXPORT_SYMBOL(generic_start_io_acct
);
1846 void generic_end_io_acct(struct request_queue
*q
, int req_op
,
1847 struct hd_struct
*part
, unsigned long start_time
)
1849 unsigned long now
= jiffies
;
1850 unsigned long duration
= now
- start_time
;
1851 const int sgrp
= op_stat_group(req_op
);
1855 update_io_ticks(part
, now
, true);
1856 part_stat_add(part
, nsecs
[sgrp
], jiffies_to_nsecs(duration
));
1857 part_dec_in_flight(q
, part
, op_is_write(req_op
));
1861 EXPORT_SYMBOL(generic_end_io_acct
);
1863 static inline bool bio_remaining_done(struct bio
*bio
)
1866 * If we're not chaining, then ->__bi_remaining is always 1 and
1867 * we always end io on the first invocation.
1869 if (!bio_flagged(bio
, BIO_CHAIN
))
1872 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1874 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1875 bio_clear_flag(bio
, BIO_CHAIN
);
1883 * bio_endio - end I/O on a bio
1887 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1888 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1889 * bio unless they own it and thus know that it has an end_io function.
1891 * bio_endio() can be called several times on a bio that has been chained
1892 * using bio_chain(). The ->bi_end_io() function will only be called the
1893 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1894 * generated if BIO_TRACE_COMPLETION is set.
1896 void bio_endio(struct bio
*bio
)
1899 if (!bio_remaining_done(bio
))
1901 if (!bio_integrity_endio(bio
))
1905 rq_qos_done_bio(bio
->bi_disk
->queue
, bio
);
1908 * Need to have a real endio function for chained bios, otherwise
1909 * various corner cases will break (like stacking block devices that
1910 * save/restore bi_end_io) - however, we want to avoid unbounded
1911 * recursion and blowing the stack. Tail call optimization would
1912 * handle this, but compiling with frame pointers also disables
1913 * gcc's sibling call optimization.
1915 if (bio
->bi_end_io
== bio_chain_endio
) {
1916 bio
= __bio_chain_endio(bio
);
1920 if (bio
->bi_disk
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1921 trace_block_bio_complete(bio
->bi_disk
->queue
, bio
,
1922 blk_status_to_errno(bio
->bi_status
));
1923 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1926 blk_throtl_bio_endio(bio
);
1927 /* release cgroup info */
1930 bio
->bi_end_io(bio
);
1932 EXPORT_SYMBOL(bio_endio
);
1935 * bio_split - split a bio
1936 * @bio: bio to split
1937 * @sectors: number of sectors to split from the front of @bio
1939 * @bs: bio set to allocate from
1941 * Allocates and returns a new bio which represents @sectors from the start of
1942 * @bio, and updates @bio to represent the remaining sectors.
1944 * Unless this is a discard request the newly allocated bio will point
1945 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1946 * neither @bio nor @bs are freed before the split bio.
1948 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1949 gfp_t gfp
, struct bio_set
*bs
)
1953 BUG_ON(sectors
<= 0);
1954 BUG_ON(sectors
>= bio_sectors(bio
));
1956 split
= bio_clone_fast(bio
, gfp
, bs
);
1960 split
->bi_iter
.bi_size
= sectors
<< 9;
1962 if (bio_integrity(split
))
1963 bio_integrity_trim(split
);
1965 bio_advance(bio
, split
->bi_iter
.bi_size
);
1967 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1968 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1972 EXPORT_SYMBOL(bio_split
);
1975 * bio_trim - trim a bio
1977 * @offset: number of sectors to trim from the front of @bio
1978 * @size: size we want to trim @bio to, in sectors
1980 void bio_trim(struct bio
*bio
, int offset
, int size
)
1982 /* 'bio' is a cloned bio which we need to trim to match
1983 * the given offset and size.
1987 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1990 bio_advance(bio
, offset
<< 9);
1991 bio
->bi_iter
.bi_size
= size
;
1993 if (bio_integrity(bio
))
1994 bio_integrity_trim(bio
);
1997 EXPORT_SYMBOL_GPL(bio_trim
);
2000 * create memory pools for biovec's in a bio_set.
2001 * use the global biovec slabs created for general use.
2003 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
2005 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
2007 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
2011 * bioset_exit - exit a bioset initialized with bioset_init()
2013 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
2016 void bioset_exit(struct bio_set
*bs
)
2018 if (bs
->rescue_workqueue
)
2019 destroy_workqueue(bs
->rescue_workqueue
);
2020 bs
->rescue_workqueue
= NULL
;
2022 mempool_exit(&bs
->bio_pool
);
2023 mempool_exit(&bs
->bvec_pool
);
2025 bioset_integrity_free(bs
);
2028 bs
->bio_slab
= NULL
;
2030 EXPORT_SYMBOL(bioset_exit
);
2033 * bioset_init - Initialize a bio_set
2034 * @bs: pool to initialize
2035 * @pool_size: Number of bio and bio_vecs to cache in the mempool
2036 * @front_pad: Number of bytes to allocate in front of the returned bio
2037 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
2038 * and %BIOSET_NEED_RESCUER
2041 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
2042 * to ask for a number of bytes to be allocated in front of the bio.
2043 * Front pad allocation is useful for embedding the bio inside
2044 * another structure, to avoid allocating extra data to go with the bio.
2045 * Note that the bio must be embedded at the END of that structure always,
2046 * or things will break badly.
2047 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
2048 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
2049 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
2050 * dispatch queued requests when the mempool runs out of space.
2053 int bioset_init(struct bio_set
*bs
,
2054 unsigned int pool_size
,
2055 unsigned int front_pad
,
2058 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
2060 bs
->front_pad
= front_pad
;
2062 spin_lock_init(&bs
->rescue_lock
);
2063 bio_list_init(&bs
->rescue_list
);
2064 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
2066 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
2070 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
2073 if ((flags
& BIOSET_NEED_BVECS
) &&
2074 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
2077 if (!(flags
& BIOSET_NEED_RESCUER
))
2080 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
2081 if (!bs
->rescue_workqueue
)
2089 EXPORT_SYMBOL(bioset_init
);
2092 * Initialize and setup a new bio_set, based on the settings from
2095 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
2100 if (src
->bvec_pool
.min_nr
)
2101 flags
|= BIOSET_NEED_BVECS
;
2102 if (src
->rescue_workqueue
)
2103 flags
|= BIOSET_NEED_RESCUER
;
2105 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
2107 EXPORT_SYMBOL(bioset_init_from_src
);
2109 #ifdef CONFIG_BLK_CGROUP
2112 * bio_disassociate_blkg - puts back the blkg reference if associated
2115 * Helper to disassociate the blkg from @bio if a blkg is associated.
2117 void bio_disassociate_blkg(struct bio
*bio
)
2120 blkg_put(bio
->bi_blkg
);
2121 bio
->bi_blkg
= NULL
;
2124 EXPORT_SYMBOL_GPL(bio_disassociate_blkg
);
2127 * __bio_associate_blkg - associate a bio with the a blkg
2129 * @blkg: the blkg to associate
2131 * This tries to associate @bio with the specified @blkg. Association failure
2132 * is handled by walking up the blkg tree. Therefore, the blkg associated can
2133 * be anything between @blkg and the root_blkg. This situation only happens
2134 * when a cgroup is dying and then the remaining bios will spill to the closest
2137 * A reference will be taken on the @blkg and will be released when @bio is
2140 static void __bio_associate_blkg(struct bio
*bio
, struct blkcg_gq
*blkg
)
2142 bio_disassociate_blkg(bio
);
2144 bio
->bi_blkg
= blkg_tryget_closest(blkg
);
2148 * bio_associate_blkg_from_css - associate a bio with a specified css
2152 * Associate @bio with the blkg found by combining the css's blkg and the
2153 * request_queue of the @bio. This falls back to the queue's root_blkg if
2154 * the association fails with the css.
2156 void bio_associate_blkg_from_css(struct bio
*bio
,
2157 struct cgroup_subsys_state
*css
)
2159 struct request_queue
*q
= bio
->bi_disk
->queue
;
2160 struct blkcg_gq
*blkg
;
2164 if (!css
|| !css
->parent
)
2165 blkg
= q
->root_blkg
;
2167 blkg
= blkg_lookup_create(css_to_blkcg(css
), q
);
2169 __bio_associate_blkg(bio
, blkg
);
2173 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css
);
2177 * bio_associate_blkg_from_page - associate a bio with the page's blkg
2179 * @page: the page to lookup the blkcg from
2181 * Associate @bio with the blkg from @page's owning memcg and the respective
2182 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's
2185 void bio_associate_blkg_from_page(struct bio
*bio
, struct page
*page
)
2187 struct cgroup_subsys_state
*css
;
2189 if (!page
->mem_cgroup
)
2194 css
= cgroup_e_css(page
->mem_cgroup
->css
.cgroup
, &io_cgrp_subsys
);
2195 bio_associate_blkg_from_css(bio
, css
);
2199 #endif /* CONFIG_MEMCG */
2202 * bio_associate_blkg - associate a bio with a blkg
2205 * Associate @bio with the blkg found from the bio's css and request_queue.
2206 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is
2207 * already associated, the css is reused and association redone as the
2208 * request_queue may have changed.
2210 void bio_associate_blkg(struct bio
*bio
)
2212 struct cgroup_subsys_state
*css
;
2217 css
= &bio_blkcg(bio
)->css
;
2221 bio_associate_blkg_from_css(bio
, css
);
2225 EXPORT_SYMBOL_GPL(bio_associate_blkg
);
2228 * bio_clone_blkg_association - clone blkg association from src to dst bio
2229 * @dst: destination bio
2232 void bio_clone_blkg_association(struct bio
*dst
, struct bio
*src
)
2237 __bio_associate_blkg(dst
, src
->bi_blkg
);
2241 EXPORT_SYMBOL_GPL(bio_clone_blkg_association
);
2242 #endif /* CONFIG_BLK_CGROUP */
2244 static void __init
biovec_init_slabs(void)
2248 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
2250 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2252 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2257 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2258 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2259 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2263 static int __init
init_bio(void)
2267 bio_slabs
= kcalloc(bio_slab_max
, sizeof(struct bio_slab
),
2270 BUILD_BUG_ON(BIO_FLAG_LAST
> BVEC_POOL_OFFSET
);
2273 panic("bio: can't allocate bios\n");
2275 bio_integrity_init();
2276 biovec_init_slabs();
2278 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
2279 panic("bio: can't allocate bios\n");
2281 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
2282 panic("bio: can't create integrity pool\n");
2286 subsys_initcall(init_bio
);