1 /*
2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
16 *
17 */
18 #include <linux/mm.h>
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
31 #include <linux/blk-cgroup.h>
33 #include <trace/events/block.h>
34 #include "blk.h"
35 #include "blk-rq-qos.h"
37 /*
38 * Test patch to inline a certain number of bi_io_vec's inside the bio
39 * itself, to shrink a bio data allocation from two mempool calls to one
40 */
41 #define BIO_INLINE_VECS 4
43 /*
44 * if you change this list, also change bvec_alloc or things will
45 * break badly! cannot be bigger than what you can fit into an
46 * unsigned short
47 */
48 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
49 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
50 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
51 };
52 #undef BV
54 /*
55 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
56 * IO code that does not need private memory pools.
57 */
58 struct bio_set fs_bio_set;
59 EXPORT_SYMBOL(fs_bio_set);
61 /*
62 * Our slab pool management
63 */
64 struct bio_slab {
65 struct kmem_cache *slab;
66 unsigned int slab_ref;
67 unsigned int slab_size;
68 char name[8];
69 };
70 static DEFINE_MUTEX(bio_slab_lock);
71 static struct bio_slab *bio_slabs;
72 static unsigned int bio_slab_nr, bio_slab_max;
74 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
75 {
76 unsigned int sz = sizeof(struct bio) + extra_size;
77 struct kmem_cache *slab = NULL;
78 struct bio_slab *bslab, *new_bio_slabs;
79 unsigned int new_bio_slab_max;
80 unsigned int i, entry = -1;
82 mutex_lock(&bio_slab_lock);
84 i = 0;
85 while (i < bio_slab_nr) {
86 bslab = &bio_slabs[i];
88 if (!bslab->slab && entry == -1)
89 entry = i;
90 else if (bslab->slab_size == sz) {
91 slab = bslab->slab;
92 bslab->slab_ref++;
93 break;
94 }
95 i++;
96 }
98 if (slab)
99 goto out_unlock;
101 if (bio_slab_nr == bio_slab_max && entry == -1) {
102 new_bio_slab_max = bio_slab_max << 1;
103 new_bio_slabs = krealloc(bio_slabs,
104 new_bio_slab_max * sizeof(struct bio_slab),
105 GFP_KERNEL);
106 if (!new_bio_slabs)
107 goto out_unlock;
108 bio_slab_max = new_bio_slab_max;
109 bio_slabs = new_bio_slabs;
110 }
111 if (entry == -1)
112 entry = bio_slab_nr++;
114 bslab = &bio_slabs[entry];
116 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
117 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
118 SLAB_HWCACHE_ALIGN, NULL);
119 if (!slab)
120 goto out_unlock;
122 bslab->slab = slab;
123 bslab->slab_ref = 1;
124 bslab->slab_size = sz;
125 out_unlock:
126 mutex_unlock(&bio_slab_lock);
127 return slab;
128 }
130 static void bio_put_slab(struct bio_set *bs)
131 {
132 struct bio_slab *bslab = NULL;
133 unsigned int i;
135 mutex_lock(&bio_slab_lock);
137 for (i = 0; i < bio_slab_nr; i++) {
138 if (bs->bio_slab == bio_slabs[i].slab) {
139 bslab = &bio_slabs[i];
140 break;
141 }
142 }
144 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 goto out;
147 WARN_ON(!bslab->slab_ref);
149 if (--bslab->slab_ref)
150 goto out;
152 kmem_cache_destroy(bslab->slab);
153 bslab->slab = NULL;
155 out:
156 mutex_unlock(&bio_slab_lock);
157 }
159 unsigned int bvec_nr_vecs(unsigned short idx)
160 {
161 return bvec_slabs[--idx].nr_vecs;
162 }
164 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
165 {
166 if (!idx)
167 return;
168 idx--;
170 BIO_BUG_ON(idx >= BVEC_POOL_NR);
172 if (idx == BVEC_POOL_MAX) {
173 mempool_free(bv, pool);
174 } else {
175 struct biovec_slab *bvs = bvec_slabs + idx;
177 kmem_cache_free(bvs->slab, bv);
178 }
179 }
181 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
182 mempool_t *pool)
183 {
184 struct bio_vec *bvl;
186 /*
187 * see comment near bvec_array define!
188 */
189 switch (nr) {
190 case 1:
191 *idx = 0;
192 break;
193 case 2 ... 4:
194 *idx = 1;
195 break;
196 case 5 ... 16:
197 *idx = 2;
198 break;
199 case 17 ... 64:
200 *idx = 3;
201 break;
202 case 65 ... 128:
203 *idx = 4;
204 break;
205 case 129 ... BIO_MAX_PAGES:
206 *idx = 5;
207 break;
208 default:
209 return NULL;
210 }
212 /*
213 * idx now points to the pool we want to allocate from. only the
214 * 1-vec entry pool is mempool backed.
215 */
216 if (*idx == BVEC_POOL_MAX) {
217 fallback:
218 bvl = mempool_alloc(pool, gfp_mask);
219 } else {
220 struct biovec_slab *bvs = bvec_slabs + *idx;
221 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
223 /*
224 * Make this allocation restricted and don't dump info on
225 * allocation failures, since we'll fallback to the mempool
226 * in case of failure.
227 */
228 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
230 /*
231 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
232 * is set, retry with the 1-entry mempool
233 */
234 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
235 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
236 *idx = BVEC_POOL_MAX;
237 goto fallback;
238 }
239 }
241 (*idx)++;
242 return bvl;
243 }
245 void bio_uninit(struct bio *bio)
246 {
247 bio_disassociate_task(bio);
248 }
249 EXPORT_SYMBOL(bio_uninit);
251 static void bio_free(struct bio *bio)
252 {
253 struct bio_set *bs = bio->bi_pool;
254 void *p;
256 bio_uninit(bio);
258 if (bs) {
259 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
261 /*
262 * If we have front padding, adjust the bio pointer before freeing
263 */
264 p = bio;
265 p -= bs->front_pad;
267 mempool_free(p, &bs->bio_pool);
268 } else {
269 /* Bio was allocated by bio_kmalloc() */
270 kfree(bio);
271 }
272 }
274 /*
275 * Users of this function have their own bio allocation. Subsequently,
276 * they must remember to pair any call to bio_init() with bio_uninit()
277 * when IO has completed, or when the bio is released.
278 */
279 void bio_init(struct bio *bio, struct bio_vec *table,
280 unsigned short max_vecs)
281 {
282 memset(bio, 0, sizeof(*bio));
283 atomic_set(&bio->__bi_remaining, 1);
284 atomic_set(&bio->__bi_cnt, 1);
286 bio->bi_io_vec = table;
287 bio->bi_max_vecs = max_vecs;
288 }
289 EXPORT_SYMBOL(bio_init);
291 /**
292 * bio_reset - reinitialize a bio
293 * @bio: bio to reset
294 *
295 * Description:
296 * After calling bio_reset(), @bio will be in the same state as a freshly
297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
298 * preserved are the ones that are initialized by bio_alloc_bioset(). See
299 * comment in struct bio.
300 */
301 void bio_reset(struct bio *bio)
302 {
303 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
305 bio_uninit(bio);
307 memset(bio, 0, BIO_RESET_BYTES);
308 bio->bi_flags = flags;
309 atomic_set(&bio->__bi_remaining, 1);
310 }
311 EXPORT_SYMBOL(bio_reset);
313 static struct bio *__bio_chain_endio(struct bio *bio)
314 {
315 struct bio *parent = bio->bi_private;
317 if (!parent->bi_status)
318 parent->bi_status = bio->bi_status;
319 bio_put(bio);
320 return parent;
321 }
323 static void bio_chain_endio(struct bio *bio)
324 {
325 bio_endio(__bio_chain_endio(bio));
326 }
328 /**
329 * bio_chain - chain bio completions
330 * @bio: the target bio
331 * @parent: the @bio's parent bio
332 *
333 * The caller won't have a bi_end_io called when @bio completes - instead,
334 * @parent's bi_end_io won't be called until both @parent and @bio have
335 * completed; the chained bio will also be freed when it completes.
336 *
337 * The caller must not set bi_private or bi_end_io in @bio.
338 */
339 void bio_chain(struct bio *bio, struct bio *parent)
340 {
341 BUG_ON(bio->bi_private || bio->bi_end_io);
343 bio->bi_private = parent;
344 bio->bi_end_io = bio_chain_endio;
345 bio_inc_remaining(parent);
346 }
347 EXPORT_SYMBOL(bio_chain);
349 static void bio_alloc_rescue(struct work_struct *work)
350 {
351 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
352 struct bio *bio;
354 while (1) {
355 spin_lock(&bs->rescue_lock);
356 bio = bio_list_pop(&bs->rescue_list);
357 spin_unlock(&bs->rescue_lock);
359 if (!bio)
360 break;
362 generic_make_request(bio);
363 }
364 }
366 static void punt_bios_to_rescuer(struct bio_set *bs)
367 {
368 struct bio_list punt, nopunt;
369 struct bio *bio;
371 if (WARN_ON_ONCE(!bs->rescue_workqueue))
372 return;
373 /*
374 * In order to guarantee forward progress we must punt only bios that
375 * were allocated from this bio_set; otherwise, if there was a bio on
376 * there for a stacking driver higher up in the stack, processing it
377 * could require allocating bios from this bio_set, and doing that from
378 * our own rescuer would be bad.
379 *
380 * Since bio lists are singly linked, pop them all instead of trying to
381 * remove from the middle of the list:
382 */
384 bio_list_init(&punt);
385 bio_list_init(&nopunt);
387 while ((bio = bio_list_pop(¤t->bio_list[0])))
388 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
389 current->bio_list[0] = nopunt;
391 bio_list_init(&nopunt);
392 while ((bio = bio_list_pop(¤t->bio_list[1])))
393 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
394 current->bio_list[1] = nopunt;
396 spin_lock(&bs->rescue_lock);
397 bio_list_merge(&bs->rescue_list, &punt);
398 spin_unlock(&bs->rescue_lock);
400 queue_work(bs->rescue_workqueue, &bs->rescue_work);
401 }
403 /**
404 * bio_alloc_bioset - allocate a bio for I/O
405 * @gfp_mask: the GFP_* mask given to the slab allocator
406 * @nr_iovecs: number of iovecs to pre-allocate
407 * @bs: the bio_set to allocate from.
408 *
409 * Description:
410 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
411 * backed by the @bs's mempool.
412 *
413 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
414 * always be able to allocate a bio. This is due to the mempool guarantees.
415 * To make this work, callers must never allocate more than 1 bio at a time
416 * from this pool. Callers that need to allocate more than 1 bio must always
417 * submit the previously allocated bio for IO before attempting to allocate
418 * a new one. Failure to do so can cause deadlocks under memory pressure.
419 *
420 * Note that when running under generic_make_request() (i.e. any block
421 * driver), bios are not submitted until after you return - see the code in
422 * generic_make_request() that converts recursion into iteration, to prevent
423 * stack overflows.
424 *
425 * This would normally mean allocating multiple bios under
426 * generic_make_request() would be susceptible to deadlocks, but we have
427 * deadlock avoidance code that resubmits any blocked bios from a rescuer
428 * thread.
429 *
430 * However, we do not guarantee forward progress for allocations from other
431 * mempools. Doing multiple allocations from the same mempool under
432 * generic_make_request() should be avoided - instead, use bio_set's front_pad
433 * for per bio allocations.
434 *
435 * RETURNS:
436 * Pointer to new bio on success, NULL on failure.
437 */
438 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
439 struct bio_set *bs)
440 {
441 gfp_t saved_gfp = gfp_mask;
442 unsigned front_pad;
443 unsigned inline_vecs;
444 struct bio_vec *bvl = NULL;
445 struct bio *bio;
446 void *p;
448 if (!bs) {
449 if (nr_iovecs > UIO_MAXIOV)
450 return NULL;
452 p = kmalloc(sizeof(struct bio) +
453 nr_iovecs * sizeof(struct bio_vec),
454 gfp_mask);
455 front_pad = 0;
456 inline_vecs = nr_iovecs;
457 } else {
458 /* should not use nobvec bioset for nr_iovecs > 0 */
459 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
460 nr_iovecs > 0))
461 return NULL;
462 /*
463 * generic_make_request() converts recursion to iteration; this
464 * means if we're running beneath it, any bios we allocate and
465 * submit will not be submitted (and thus freed) until after we
466 * return.
467 *
468 * This exposes us to a potential deadlock if we allocate
469 * multiple bios from the same bio_set() while running
470 * underneath generic_make_request(). If we were to allocate
471 * multiple bios (say a stacking block driver that was splitting
472 * bios), we would deadlock if we exhausted the mempool's
473 * reserve.
474 *
475 * We solve this, and guarantee forward progress, with a rescuer
476 * workqueue per bio_set. If we go to allocate and there are
477 * bios on current->bio_list, we first try the allocation
478 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
479 * bios we would be blocking to the rescuer workqueue before
480 * we retry with the original gfp_flags.
481 */
483 if (current->bio_list &&
484 (!bio_list_empty(¤t->bio_list[0]) ||
485 !bio_list_empty(¤t->bio_list[1])) &&
486 bs->rescue_workqueue)
487 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
489 p = mempool_alloc(&bs->bio_pool, gfp_mask);
490 if (!p && gfp_mask != saved_gfp) {
491 punt_bios_to_rescuer(bs);
492 gfp_mask = saved_gfp;
493 p = mempool_alloc(&bs->bio_pool, gfp_mask);
494 }
496 front_pad = bs->front_pad;
497 inline_vecs = BIO_INLINE_VECS;
498 }
500 if (unlikely(!p))
501 return NULL;
503 bio = p + front_pad;
504 bio_init(bio, NULL, 0);
506 if (nr_iovecs > inline_vecs) {
507 unsigned long idx = 0;
509 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
510 if (!bvl && gfp_mask != saved_gfp) {
511 punt_bios_to_rescuer(bs);
512 gfp_mask = saved_gfp;
513 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
514 }
516 if (unlikely(!bvl))
517 goto err_free;
519 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
520 } else if (nr_iovecs) {
521 bvl = bio->bi_inline_vecs;
522 }
524 bio->bi_pool = bs;
525 bio->bi_max_vecs = nr_iovecs;
526 bio->bi_io_vec = bvl;
527 return bio;
529 err_free:
530 mempool_free(p, &bs->bio_pool);
531 return NULL;
532 }
533 EXPORT_SYMBOL(bio_alloc_bioset);
535 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
536 {
537 unsigned long flags;
538 struct bio_vec bv;
539 struct bvec_iter iter;
541 __bio_for_each_segment(bv, bio, iter, start) {
542 char *data = bvec_kmap_irq(&bv, &flags);
543 memset(data, 0, bv.bv_len);
544 flush_dcache_page(bv.bv_page);
545 bvec_kunmap_irq(data, &flags);
546 }
547 }
548 EXPORT_SYMBOL(zero_fill_bio_iter);
550 /**
551 * bio_put - release a reference to a bio
552 * @bio: bio to release reference to
553 *
554 * Description:
555 * Put a reference to a &struct bio, either one you have gotten with
556 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
557 **/
558 void bio_put(struct bio *bio)
559 {
560 if (!bio_flagged(bio, BIO_REFFED))
561 bio_free(bio);
562 else {
563 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
565 /*
566 * last put frees it
567 */
568 if (atomic_dec_and_test(&bio->__bi_cnt))
569 bio_free(bio);
570 }
571 }
572 EXPORT_SYMBOL(bio_put);
574 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
575 {
576 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
577 blk_recount_segments(q, bio);
579 return bio->bi_phys_segments;
580 }
581 EXPORT_SYMBOL(bio_phys_segments);
583 /**
584 * __bio_clone_fast - clone a bio that shares the original bio's biovec
585 * @bio: destination bio
586 * @bio_src: bio to clone
587 *
588 * Clone a &bio. Caller will own the returned bio, but not
589 * the actual data it points to. Reference count of returned
590 * bio will be one.
591 *
592 * Caller must ensure that @bio_src is not freed before @bio.
593 */
594 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
595 {
596 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
598 /*
599 * most users will be overriding ->bi_disk with a new target,
600 * so we don't set nor calculate new physical/hw segment counts here
601 */
602 bio->bi_disk = bio_src->bi_disk;
603 bio->bi_partno = bio_src->bi_partno;
604 bio_set_flag(bio, BIO_CLONED);
605 if (bio_flagged(bio_src, BIO_THROTTLED))
606 bio_set_flag(bio, BIO_THROTTLED);
607 bio->bi_opf = bio_src->bi_opf;
608 bio->bi_write_hint = bio_src->bi_write_hint;
609 bio->bi_iter = bio_src->bi_iter;
610 bio->bi_io_vec = bio_src->bi_io_vec;
612 bio_clone_blkcg_association(bio, bio_src);
613 }
614 EXPORT_SYMBOL(__bio_clone_fast);
616 /**
617 * bio_clone_fast - clone a bio that shares the original bio's biovec
618 * @bio: bio to clone
619 * @gfp_mask: allocation priority
620 * @bs: bio_set to allocate from
621 *
622 * Like __bio_clone_fast, only also allocates the returned bio
623 */
624 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
625 {
626 struct bio *b;
628 b = bio_alloc_bioset(gfp_mask, 0, bs);
629 if (!b)
630 return NULL;
632 __bio_clone_fast(b, bio);
634 if (bio_integrity(bio)) {
635 int ret;
637 ret = bio_integrity_clone(b, bio, gfp_mask);
639 if (ret < 0) {
640 bio_put(b);
641 return NULL;
642 }
643 }
645 return b;
646 }
647 EXPORT_SYMBOL(bio_clone_fast);
649 /**
650 * bio_add_pc_page - attempt to add page to bio
651 * @q: the target queue
652 * @bio: destination bio
653 * @page: page to add
654 * @len: vec entry length
655 * @offset: vec entry offset
656 *
657 * Attempt to add a page to the bio_vec maplist. This can fail for a
658 * number of reasons, such as the bio being full or target block device
659 * limitations. The target block device must allow bio's up to PAGE_SIZE,
660 * so it is always possible to add a single page to an empty bio.
661 *
662 * This should only be used by REQ_PC bios.
663 */
664 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
665 *page, unsigned int len, unsigned int offset)
666 {
667 int retried_segments = 0;
668 struct bio_vec *bvec;
670 /*
671 * cloned bio must not modify vec list
672 */
673 if (unlikely(bio_flagged(bio, BIO_CLONED)))
674 return 0;
676 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
677 return 0;
679 /*
680 * For filesystems with a blocksize smaller than the pagesize
681 * we will often be called with the same page as last time and
682 * a consecutive offset. Optimize this special case.
683 */
684 if (bio->bi_vcnt > 0) {
685 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
687 if (page == prev->bv_page &&
688 offset == prev->bv_offset + prev->bv_len) {
689 prev->bv_len += len;
690 bio->bi_iter.bi_size += len;
691 goto done;
692 }
694 /*
695 * If the queue doesn't support SG gaps and adding this
696 * offset would create a gap, disallow it.
697 */
698 if (bvec_gap_to_prev(q, prev, offset))
699 return 0;
700 }
702 if (bio_full(bio))
703 return 0;
705 /*
706 * setup the new entry, we might clear it again later if we
707 * cannot add the page
708 */
709 bvec = &bio->bi_io_vec[bio->bi_vcnt];
710 bvec->bv_page = page;
711 bvec->bv_len = len;
712 bvec->bv_offset = offset;
713 bio->bi_vcnt++;
714 bio->bi_phys_segments++;
715 bio->bi_iter.bi_size += len;
717 /*
718 * Perform a recount if the number of segments is greater
719 * than queue_max_segments(q).
720 */
722 while (bio->bi_phys_segments > queue_max_segments(q)) {
724 if (retried_segments)
725 goto failed;
727 retried_segments = 1;
728 blk_recount_segments(q, bio);
729 }
731 /* If we may be able to merge these biovecs, force a recount */
732 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
733 bio_clear_flag(bio, BIO_SEG_VALID);
735 done:
736 return len;
738 failed:
739 bvec->bv_page = NULL;
740 bvec->bv_len = 0;
741 bvec->bv_offset = 0;
742 bio->bi_vcnt--;
743 bio->bi_iter.bi_size -= len;
744 blk_recount_segments(q, bio);
745 return 0;
746 }
747 EXPORT_SYMBOL(bio_add_pc_page);
749 /**
750 * __bio_try_merge_page - try appending data to an existing bvec.
751 * @bio: destination bio
752 * @page: page to add
753 * @len: length of the data to add
754 * @off: offset of the data in @page
755 *
756 * Try to add the data at @page + @off to the last bvec of @bio. This is a
757 * a useful optimisation for file systems with a block size smaller than the
758 * page size.
759 *
760 * Return %true on success or %false on failure.
761 */
762 bool __bio_try_merge_page(struct bio *bio, struct page *page,
763 unsigned int len, unsigned int off)
764 {
765 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
766 return false;
768 if (bio->bi_vcnt > 0) {
769 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
771 if (page == bv->bv_page && off == bv->bv_offset + bv->bv_len) {
772 bv->bv_len += len;
773 bio->bi_iter.bi_size += len;
774 return true;
775 }
776 }
777 return false;
778 }
779 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
781 /**
782 * __bio_add_page - add page to a bio in a new segment
783 * @bio: destination bio
784 * @page: page to add
785 * @len: length of the data to add
786 * @off: offset of the data in @page
787 *
788 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
789 * that @bio has space for another bvec.
790 */
791 void __bio_add_page(struct bio *bio, struct page *page,
792 unsigned int len, unsigned int off)
793 {
794 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
796 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
797 WARN_ON_ONCE(bio_full(bio));
799 bv->bv_page = page;
800 bv->bv_offset = off;
801 bv->bv_len = len;
803 bio->bi_iter.bi_size += len;
804 bio->bi_vcnt++;
805 }
806 EXPORT_SYMBOL_GPL(__bio_add_page);
808 /**
809 * bio_add_page - attempt to add page to bio
810 * @bio: destination bio
811 * @page: page to add
812 * @len: vec entry length
813 * @offset: vec entry offset
814 *
815 * Attempt to add a page to the bio_vec maplist. This will only fail
816 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
817 */
818 int bio_add_page(struct bio *bio, struct page *page,
819 unsigned int len, unsigned int offset)
820 {
821 if (!__bio_try_merge_page(bio, page, len, offset)) {
822 if (bio_full(bio))
823 return 0;
824 __bio_add_page(bio, page, len, offset);
825 }
826 return len;
827 }
828 EXPORT_SYMBOL(bio_add_page);
830 /**
831 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
832 * @bio: bio to add pages to
833 * @iter: iov iterator describing the region to be mapped
834 *
835 * Pins pages from *iter and appends them to @bio's bvec array. The
836 * pages will have to be released using put_page() when done.
837 * For multi-segment *iter, this function only adds pages from the
838 * the next non-empty segment of the iov iterator.
839 */
840 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
841 {
842 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt, idx;
843 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
844 struct page **pages = (struct page **)bv;
845 size_t offset;
846 ssize_t size;
848 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
849 if (unlikely(size <= 0))
850 return size ? size : -EFAULT;
851 idx = nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE;
853 /*
854 * Deep magic below: We need to walk the pinned pages backwards
855 * because we are abusing the space allocated for the bio_vecs
856 * for the page array. Because the bio_vecs are larger than the
857 * page pointers by definition this will always work. But it also
858 * means we can't use bio_add_page, so any changes to it's semantics
859 * need to be reflected here as well.
860 */
861 bio->bi_iter.bi_size += size;
862 bio->bi_vcnt += nr_pages;
864 while (idx--) {
865 bv[idx].bv_page = pages[idx];
866 bv[idx].bv_len = PAGE_SIZE;
867 bv[idx].bv_offset = 0;
868 }
870 bv[0].bv_offset += offset;
871 bv[0].bv_len -= offset;
872 bv[nr_pages - 1].bv_len -= nr_pages * PAGE_SIZE - offset - size;
874 iov_iter_advance(iter, size);
875 return 0;
876 }
878 /**
879 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
880 * @bio: bio to add pages to
881 * @iter: iov iterator describing the region to be mapped
882 *
883 * Pins pages from *iter and appends them to @bio's bvec array. The
884 * pages will have to be released using put_page() when done.
885 * The function tries, but does not guarantee, to pin as many pages as
886 * fit into the bio, or are requested in *iter, whatever is smaller.
887 * If MM encounters an error pinning the requested pages, it stops.
888 * Error is returned only if 0 pages could be pinned.
889 */
890 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
891 {
892 unsigned short orig_vcnt = bio->bi_vcnt;
894 do {
895 int ret = __bio_iov_iter_get_pages(bio, iter);
897 if (unlikely(ret))
898 return bio->bi_vcnt > orig_vcnt ? 0 : ret;
900 } while (iov_iter_count(iter) && !bio_full(bio));
902 return 0;
903 }
904 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
906 static void submit_bio_wait_endio(struct bio *bio)
907 {
908 complete(bio->bi_private);
909 }
911 /**
912 * submit_bio_wait - submit a bio, and wait until it completes
913 * @bio: The &struct bio which describes the I/O
914 *
915 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
916 * bio_endio() on failure.
917 *
918 * WARNING: Unlike to how submit_bio() is usually used, this function does not
919 * result in bio reference to be consumed. The caller must drop the reference
920 * on his own.
921 */
922 int submit_bio_wait(struct bio *bio)
923 {
924 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
926 bio->bi_private = &done;
927 bio->bi_end_io = submit_bio_wait_endio;
928 bio->bi_opf |= REQ_SYNC;
929 submit_bio(bio);
930 wait_for_completion_io(&done);
932 return blk_status_to_errno(bio->bi_status);
933 }
934 EXPORT_SYMBOL(submit_bio_wait);
936 /**
937 * bio_advance - increment/complete a bio by some number of bytes
938 * @bio: bio to advance
939 * @bytes: number of bytes to complete
940 *
941 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
942 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
943 * be updated on the last bvec as well.
944 *
945 * @bio will then represent the remaining, uncompleted portion of the io.
946 */
947 void bio_advance(struct bio *bio, unsigned bytes)
948 {
949 if (bio_integrity(bio))
950 bio_integrity_advance(bio, bytes);
952 bio_advance_iter(bio, &bio->bi_iter, bytes);
953 }
954 EXPORT_SYMBOL(bio_advance);
956 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
957 struct bio *src, struct bvec_iter *src_iter)
958 {
959 struct bio_vec src_bv, dst_bv;
960 void *src_p, *dst_p;
961 unsigned bytes;
963 while (src_iter->bi_size && dst_iter->bi_size) {
964 src_bv = bio_iter_iovec(src, *src_iter);
965 dst_bv = bio_iter_iovec(dst, *dst_iter);
967 bytes = min(src_bv.bv_len, dst_bv.bv_len);
969 src_p = kmap_atomic(src_bv.bv_page);
970 dst_p = kmap_atomic(dst_bv.bv_page);
972 memcpy(dst_p + dst_bv.bv_offset,
973 src_p + src_bv.bv_offset,
974 bytes);
976 kunmap_atomic(dst_p);
977 kunmap_atomic(src_p);
979 flush_dcache_page(dst_bv.bv_page);
981 bio_advance_iter(src, src_iter, bytes);
982 bio_advance_iter(dst, dst_iter, bytes);
983 }
984 }
985 EXPORT_SYMBOL(bio_copy_data_iter);
987 /**
988 * bio_copy_data - copy contents of data buffers from one bio to another
989 * @src: source bio
990 * @dst: destination bio
991 *
992 * Stops when it reaches the end of either @src or @dst - that is, copies
993 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
994 */
995 void bio_copy_data(struct bio *dst, struct bio *src)
996 {
997 struct bvec_iter src_iter = src->bi_iter;
998 struct bvec_iter dst_iter = dst->bi_iter;
1000 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1001 }
1002 EXPORT_SYMBOL(bio_copy_data);
1004 /**
1005 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1006 * another
1007 * @src: source bio list
1008 * @dst: destination bio list
1009 *
1010 * Stops when it reaches the end of either the @src list or @dst list - that is,
1011 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1012 * bios).
1013 */
1014 void bio_list_copy_data(struct bio *dst, struct bio *src)
1015 {
1016 struct bvec_iter src_iter = src->bi_iter;
1017 struct bvec_iter dst_iter = dst->bi_iter;
1019 while (1) {
1020 if (!src_iter.bi_size) {
1021 src = src->bi_next;
1022 if (!src)
1023 break;
1025 src_iter = src->bi_iter;
1026 }
1028 if (!dst_iter.bi_size) {
1029 dst = dst->bi_next;
1030 if (!dst)
1031 break;
1033 dst_iter = dst->bi_iter;
1034 }
1036 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1037 }
1038 }
1039 EXPORT_SYMBOL(bio_list_copy_data);
1041 struct bio_map_data {
1042 int is_our_pages;
1043 struct iov_iter iter;
1044 struct iovec iov[];
1045 };
1047 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
1048 gfp_t gfp_mask)
1049 {
1050 struct bio_map_data *bmd;
1051 if (data->nr_segs > UIO_MAXIOV)
1052 return NULL;
1054 bmd = kmalloc(sizeof(struct bio_map_data) +
1055 sizeof(struct iovec) * data->nr_segs, gfp_mask);
1056 if (!bmd)
1057 return NULL;
1058 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1059 bmd->iter = *data;
1060 bmd->iter.iov = bmd->iov;
1061 return bmd;
1062 }
1064 /**
1065 * bio_copy_from_iter - copy all pages from iov_iter to bio
1066 * @bio: The &struct bio which describes the I/O as destination
1067 * @iter: iov_iter as source
1068 *
1069 * Copy all pages from iov_iter to bio.
1070 * Returns 0 on success, or error on failure.
1071 */
1072 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
1073 {
1074 int i;
1075 struct bio_vec *bvec;
1077 bio_for_each_segment_all(bvec, bio, i) {
1078 ssize_t ret;
1080 ret = copy_page_from_iter(bvec->bv_page,
1081 bvec->bv_offset,
1082 bvec->bv_len,
1083 iter);
1085 if (!iov_iter_count(iter))
1086 break;
1088 if (ret < bvec->bv_len)
1089 return -EFAULT;
1090 }
1092 return 0;
1093 }
1095 /**
1096 * bio_copy_to_iter - copy all pages from bio to iov_iter
1097 * @bio: The &struct bio which describes the I/O as source
1098 * @iter: iov_iter as destination
1099 *
1100 * Copy all pages from bio to iov_iter.
1101 * Returns 0 on success, or error on failure.
1102 */
1103 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1104 {
1105 int i;
1106 struct bio_vec *bvec;
1108 bio_for_each_segment_all(bvec, bio, i) {
1109 ssize_t ret;
1111 ret = copy_page_to_iter(bvec->bv_page,
1112 bvec->bv_offset,
1113 bvec->bv_len,
1114 &iter);
1116 if (!iov_iter_count(&iter))
1117 break;
1119 if (ret < bvec->bv_len)
1120 return -EFAULT;
1121 }
1123 return 0;
1124 }
1126 void bio_free_pages(struct bio *bio)
1127 {
1128 struct bio_vec *bvec;
1129 int i;
1131 bio_for_each_segment_all(bvec, bio, i)
1132 __free_page(bvec->bv_page);
1133 }
1134 EXPORT_SYMBOL(bio_free_pages);
1136 /**
1137 * bio_uncopy_user - finish previously mapped bio
1138 * @bio: bio being terminated
1139 *
1140 * Free pages allocated from bio_copy_user_iov() and write back data
1141 * to user space in case of a read.
1142 */
1143 int bio_uncopy_user(struct bio *bio)
1144 {
1145 struct bio_map_data *bmd = bio->bi_private;
1146 int ret = 0;
1148 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1149 /*
1150 * if we're in a workqueue, the request is orphaned, so
1151 * don't copy into a random user address space, just free
1152 * and return -EINTR so user space doesn't expect any data.
1153 */
1154 if (!current->mm)
1155 ret = -EINTR;
1156 else if (bio_data_dir(bio) == READ)
1157 ret = bio_copy_to_iter(bio, bmd->iter);
1158 if (bmd->is_our_pages)
1159 bio_free_pages(bio);
1160 }
1161 kfree(bmd);
1162 bio_put(bio);
1163 return ret;
1164 }
1166 /**
1167 * bio_copy_user_iov - copy user data to bio
1168 * @q: destination block queue
1169 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1170 * @iter: iovec iterator
1171 * @gfp_mask: memory allocation flags
1172 *
1173 * Prepares and returns a bio for indirect user io, bouncing data
1174 * to/from kernel pages as necessary. Must be paired with
1175 * call bio_uncopy_user() on io completion.
1176 */
1177 struct bio *bio_copy_user_iov(struct request_queue *q,
1178 struct rq_map_data *map_data,
1179 struct iov_iter *iter,
1180 gfp_t gfp_mask)
1181 {
1182 struct bio_map_data *bmd;
1183 struct page *page;
1184 struct bio *bio;
1185 int i = 0, ret;
1186 int nr_pages;
1187 unsigned int len = iter->count;
1188 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1190 bmd = bio_alloc_map_data(iter, gfp_mask);
1191 if (!bmd)
1192 return ERR_PTR(-ENOMEM);
1194 /*
1195 * We need to do a deep copy of the iov_iter including the iovecs.
1196 * The caller provided iov might point to an on-stack or otherwise
1197 * shortlived one.
1198 */
1199 bmd->is_our_pages = map_data ? 0 : 1;
1201 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1202 if (nr_pages > BIO_MAX_PAGES)
1203 nr_pages = BIO_MAX_PAGES;
1205 ret = -ENOMEM;
1206 bio = bio_kmalloc(gfp_mask, nr_pages);
1207 if (!bio)
1208 goto out_bmd;
1210 ret = 0;
1212 if (map_data) {
1213 nr_pages = 1 << map_data->page_order;
1214 i = map_data->offset / PAGE_SIZE;
1215 }
1216 while (len) {
1217 unsigned int bytes = PAGE_SIZE;
1219 bytes -= offset;
1221 if (bytes > len)
1222 bytes = len;
1224 if (map_data) {
1225 if (i == map_data->nr_entries * nr_pages) {
1226 ret = -ENOMEM;
1227 break;
1228 }
1230 page = map_data->pages[i / nr_pages];
1231 page += (i % nr_pages);
1233 i++;
1234 } else {
1235 page = alloc_page(q->bounce_gfp | gfp_mask);
1236 if (!page) {
1237 ret = -ENOMEM;
1238 break;
1239 }
1240 }
1242 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1243 break;
1245 len -= bytes;
1246 offset = 0;
1247 }
1249 if (ret)
1250 goto cleanup;
1252 if (map_data)
1253 map_data->offset += bio->bi_iter.bi_size;
1255 /*
1256 * success
1257 */
1258 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1259 (map_data && map_data->from_user)) {
1260 ret = bio_copy_from_iter(bio, iter);
1261 if (ret)
1262 goto cleanup;
1263 } else {
1264 iov_iter_advance(iter, bio->bi_iter.bi_size);
1265 }
1267 bio->bi_private = bmd;
1268 if (map_data && map_data->null_mapped)
1269 bio_set_flag(bio, BIO_NULL_MAPPED);
1270 return bio;
1271 cleanup:
1272 if (!map_data)
1273 bio_free_pages(bio);
1274 bio_put(bio);
1275 out_bmd:
1276 kfree(bmd);
1277 return ERR_PTR(ret);
1278 }
1280 /**
1281 * bio_map_user_iov - map user iovec into bio
1282 * @q: the struct request_queue for the bio
1283 * @iter: iovec iterator
1284 * @gfp_mask: memory allocation flags
1285 *
1286 * Map the user space address into a bio suitable for io to a block
1287 * device. Returns an error pointer in case of error.
1288 */
1289 struct bio *bio_map_user_iov(struct request_queue *q,
1290 struct iov_iter *iter,
1291 gfp_t gfp_mask)
1292 {
1293 int j;
1294 struct bio *bio;
1295 int ret;
1296 struct bio_vec *bvec;
1298 if (!iov_iter_count(iter))
1299 return ERR_PTR(-EINVAL);
1301 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
1302 if (!bio)
1303 return ERR_PTR(-ENOMEM);
1305 while (iov_iter_count(iter)) {
1306 struct page **pages;
1307 ssize_t bytes;
1308 size_t offs, added = 0;
1309 int npages;
1311 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
1312 if (unlikely(bytes <= 0)) {
1313 ret = bytes ? bytes : -EFAULT;
1314 goto out_unmap;
1315 }
1317 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
1319 if (unlikely(offs & queue_dma_alignment(q))) {
1320 ret = -EINVAL;
1321 j = 0;
1322 } else {
1323 for (j = 0; j < npages; j++) {
1324 struct page *page = pages[j];
1325 unsigned int n = PAGE_SIZE - offs;
1326 unsigned short prev_bi_vcnt = bio->bi_vcnt;
1328 if (n > bytes)
1329 n = bytes;
1331 if (!bio_add_pc_page(q, bio, page, n, offs))
1332 break;
1334 /*
1335 * check if vector was merged with previous
1336 * drop page reference if needed
1337 */
1338 if (bio->bi_vcnt == prev_bi_vcnt)
1339 put_page(page);
1341 added += n;
1342 bytes -= n;
1343 offs = 0;
1344 }
1345 iov_iter_advance(iter, added);
1346 }
1347 /*
1348 * release the pages we didn't map into the bio, if any
1349 */
1350 while (j < npages)
1351 put_page(pages[j++]);
1352 kvfree(pages);
1353 /* couldn't stuff something into bio? */
1354 if (bytes)
1355 break;
1356 }
1358 bio_set_flag(bio, BIO_USER_MAPPED);
1360 /*
1361 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1362 * it would normally disappear when its bi_end_io is run.
1363 * however, we need it for the unmap, so grab an extra
1364 * reference to it
1365 */
1366 bio_get(bio);
1367 return bio;
1369 out_unmap:
1370 bio_for_each_segment_all(bvec, bio, j) {
1371 put_page(bvec->bv_page);
1372 }
1373 bio_put(bio);
1374 return ERR_PTR(ret);
1375 }
1377 static void __bio_unmap_user(struct bio *bio)
1378 {
1379 struct bio_vec *bvec;
1380 int i;
1382 /*
1383 * make sure we dirty pages we wrote to
1384 */
1385 bio_for_each_segment_all(bvec, bio, i) {
1386 if (bio_data_dir(bio) == READ)
1387 set_page_dirty_lock(bvec->bv_page);
1389 put_page(bvec->bv_page);
1390 }
1392 bio_put(bio);
1393 }
1395 /**
1396 * bio_unmap_user - unmap a bio
1397 * @bio: the bio being unmapped
1398 *
1399 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1400 * process context.
1401 *
1402 * bio_unmap_user() may sleep.
1403 */
1404 void bio_unmap_user(struct bio *bio)
1405 {
1406 __bio_unmap_user(bio);
1407 bio_put(bio);
1408 }
1410 static void bio_map_kern_endio(struct bio *bio)
1411 {
1412 bio_put(bio);
1413 }
1415 /**
1416 * bio_map_kern - map kernel address into bio
1417 * @q: the struct request_queue for the bio
1418 * @data: pointer to buffer to map
1419 * @len: length in bytes
1420 * @gfp_mask: allocation flags for bio allocation
1421 *
1422 * Map the kernel address into a bio suitable for io to a block
1423 * device. Returns an error pointer in case of error.
1424 */
1425 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1426 gfp_t gfp_mask)
1427 {
1428 unsigned long kaddr = (unsigned long)data;
1429 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1430 unsigned long start = kaddr >> PAGE_SHIFT;
1431 const int nr_pages = end - start;
1432 int offset, i;
1433 struct bio *bio;
1435 bio = bio_kmalloc(gfp_mask, nr_pages);
1436 if (!bio)
1437 return ERR_PTR(-ENOMEM);
1439 offset = offset_in_page(kaddr);
1440 for (i = 0; i < nr_pages; i++) {
1441 unsigned int bytes = PAGE_SIZE - offset;
1443 if (len <= 0)
1444 break;
1446 if (bytes > len)
1447 bytes = len;
1449 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1450 offset) < bytes) {
1451 /* we don't support partial mappings */
1452 bio_put(bio);
1453 return ERR_PTR(-EINVAL);
1454 }
1456 data += bytes;
1457 len -= bytes;
1458 offset = 0;
1459 }
1461 bio->bi_end_io = bio_map_kern_endio;
1462 return bio;
1463 }
1464 EXPORT_SYMBOL(bio_map_kern);
1466 static void bio_copy_kern_endio(struct bio *bio)
1467 {
1468 bio_free_pages(bio);
1469 bio_put(bio);
1470 }
1472 static void bio_copy_kern_endio_read(struct bio *bio)
1473 {
1474 char *p = bio->bi_private;
1475 struct bio_vec *bvec;
1476 int i;
1478 bio_for_each_segment_all(bvec, bio, i) {
1479 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1480 p += bvec->bv_len;
1481 }
1483 bio_copy_kern_endio(bio);
1484 }
1486 /**
1487 * bio_copy_kern - copy kernel address into bio
1488 * @q: the struct request_queue for the bio
1489 * @data: pointer to buffer to copy
1490 * @len: length in bytes
1491 * @gfp_mask: allocation flags for bio and page allocation
1492 * @reading: data direction is READ
1493 *
1494 * copy the kernel address into a bio suitable for io to a block
1495 * device. Returns an error pointer in case of error.
1496 */
1497 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1498 gfp_t gfp_mask, int reading)
1499 {
1500 unsigned long kaddr = (unsigned long)data;
1501 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1502 unsigned long start = kaddr >> PAGE_SHIFT;
1503 struct bio *bio;
1504 void *p = data;
1505 int nr_pages = 0;
1507 /*
1508 * Overflow, abort
1509 */
1510 if (end < start)
1511 return ERR_PTR(-EINVAL);
1513 nr_pages = end - start;
1514 bio = bio_kmalloc(gfp_mask, nr_pages);
1515 if (!bio)
1516 return ERR_PTR(-ENOMEM);
1518 while (len) {
1519 struct page *page;
1520 unsigned int bytes = PAGE_SIZE;
1522 if (bytes > len)
1523 bytes = len;
1525 page = alloc_page(q->bounce_gfp | gfp_mask);
1526 if (!page)
1527 goto cleanup;
1529 if (!reading)
1530 memcpy(page_address(page), p, bytes);
1532 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1533 break;
1535 len -= bytes;
1536 p += bytes;
1537 }
1539 if (reading) {
1540 bio->bi_end_io = bio_copy_kern_endio_read;
1541 bio->bi_private = data;
1542 } else {
1543 bio->bi_end_io = bio_copy_kern_endio;
1544 }
1546 return bio;
1548 cleanup:
1549 bio_free_pages(bio);
1550 bio_put(bio);
1551 return ERR_PTR(-ENOMEM);
1552 }
1554 /*
1555 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1556 * for performing direct-IO in BIOs.
1557 *
1558 * The problem is that we cannot run set_page_dirty() from interrupt context
1559 * because the required locks are not interrupt-safe. So what we can do is to
1560 * mark the pages dirty _before_ performing IO. And in interrupt context,
1561 * check that the pages are still dirty. If so, fine. If not, redirty them
1562 * in process context.
1563 *
1564 * We special-case compound pages here: normally this means reads into hugetlb
1565 * pages. The logic in here doesn't really work right for compound pages
1566 * because the VM does not uniformly chase down the head page in all cases.
1567 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1568 * handle them at all. So we skip compound pages here at an early stage.
1569 *
1570 * Note that this code is very hard to test under normal circumstances because
1571 * direct-io pins the pages with get_user_pages(). This makes
1572 * is_page_cache_freeable return false, and the VM will not clean the pages.
1573 * But other code (eg, flusher threads) could clean the pages if they are mapped
1574 * pagecache.
1575 *
1576 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1577 * deferred bio dirtying paths.
1578 */
1580 /*
1581 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1582 */
1583 void bio_set_pages_dirty(struct bio *bio)
1584 {
1585 struct bio_vec *bvec;
1586 int i;
1588 bio_for_each_segment_all(bvec, bio, i) {
1589 if (!PageCompound(bvec->bv_page))
1590 set_page_dirty_lock(bvec->bv_page);
1591 }
1592 }
1593 EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1595 static void bio_release_pages(struct bio *bio)
1596 {
1597 struct bio_vec *bvec;
1598 int i;
1600 bio_for_each_segment_all(bvec, bio, i)
1601 put_page(bvec->bv_page);
1602 }
1604 /*
1605 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1606 * If they are, then fine. If, however, some pages are clean then they must
1607 * have been written out during the direct-IO read. So we take another ref on
1608 * the BIO and re-dirty the pages in process context.
1609 *
1610 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1611 * here on. It will run one put_page() against each page and will run one
1612 * bio_put() against the BIO.
1613 */
1615 static void bio_dirty_fn(struct work_struct *work);
1617 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1618 static DEFINE_SPINLOCK(bio_dirty_lock);
1619 static struct bio *bio_dirty_list;
1621 /*
1622 * This runs in process context
1623 */
1624 static void bio_dirty_fn(struct work_struct *work)
1625 {
1626 struct bio *bio, *next;
1628 spin_lock_irq(&bio_dirty_lock);
1629 next = bio_dirty_list;
1630 bio_dirty_list = NULL;
1631 spin_unlock_irq(&bio_dirty_lock);
1633 while ((bio = next) != NULL) {
1634 next = bio->bi_private;
1636 bio_set_pages_dirty(bio);
1637 bio_release_pages(bio);
1638 bio_put(bio);
1639 }
1640 }
1642 void bio_check_pages_dirty(struct bio *bio)
1643 {
1644 struct bio_vec *bvec;
1645 unsigned long flags;
1646 int i;
1648 bio_for_each_segment_all(bvec, bio, i) {
1649 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1650 goto defer;
1651 }
1653 bio_release_pages(bio);
1654 bio_put(bio);
1655 return;
1656 defer:
1657 spin_lock_irqsave(&bio_dirty_lock, flags);
1658 bio->bi_private = bio_dirty_list;
1659 bio_dirty_list = bio;
1660 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1661 schedule_work(&bio_dirty_work);
1662 }
1663 EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1665 void generic_start_io_acct(struct request_queue *q, int op,
1666 unsigned long sectors, struct hd_struct *part)
1667 {
1668 const int sgrp = op_stat_group(op);
1669 int cpu = part_stat_lock();
1671 part_round_stats(q, cpu, part);
1672 part_stat_inc(cpu, part, ios[sgrp]);
1673 part_stat_add(cpu, part, sectors[sgrp], sectors);
1674 part_inc_in_flight(q, part, op_is_write(op));
1676 part_stat_unlock();
1677 }
1678 EXPORT_SYMBOL(generic_start_io_acct);
1680 void generic_end_io_acct(struct request_queue *q, int req_op,
1681 struct hd_struct *part, unsigned long start_time)
1682 {
1683 unsigned long duration = jiffies - start_time;
1684 const int sgrp = op_stat_group(req_op);
1685 int cpu = part_stat_lock();
1687 part_stat_add(cpu, part, nsecs[sgrp], jiffies_to_nsecs(duration));
1688 part_round_stats(q, cpu, part);
1689 part_dec_in_flight(q, part, op_is_write(req_op));
1691 part_stat_unlock();
1692 }
1693 EXPORT_SYMBOL(generic_end_io_acct);
1695 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1696 void bio_flush_dcache_pages(struct bio *bi)
1697 {
1698 struct bio_vec bvec;
1699 struct bvec_iter iter;
1701 bio_for_each_segment(bvec, bi, iter)
1702 flush_dcache_page(bvec.bv_page);
1703 }
1704 EXPORT_SYMBOL(bio_flush_dcache_pages);
1705 #endif
1707 static inline bool bio_remaining_done(struct bio *bio)
1708 {
1709 /*
1710 * If we're not chaining, then ->__bi_remaining is always 1 and
1711 * we always end io on the first invocation.
1712 */
1713 if (!bio_flagged(bio, BIO_CHAIN))
1714 return true;
1716 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1718 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1719 bio_clear_flag(bio, BIO_CHAIN);
1720 return true;
1721 }
1723 return false;
1724 }
1726 /**
1727 * bio_endio - end I/O on a bio
1728 * @bio: bio
1729 *
1730 * Description:
1731 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1732 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1733 * bio unless they own it and thus know that it has an end_io function.
1734 *
1735 * bio_endio() can be called several times on a bio that has been chained
1736 * using bio_chain(). The ->bi_end_io() function will only be called the
1737 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1738 * generated if BIO_TRACE_COMPLETION is set.
1739 **/
1740 void bio_endio(struct bio *bio)
1741 {
1742 again:
1743 if (!bio_remaining_done(bio))
1744 return;
1745 if (!bio_integrity_endio(bio))
1746 return;
1748 if (bio->bi_disk)
1749 rq_qos_done_bio(bio->bi_disk->queue, bio);
1751 /*
1752 * Need to have a real endio function for chained bios, otherwise
1753 * various corner cases will break (like stacking block devices that
1754 * save/restore bi_end_io) - however, we want to avoid unbounded
1755 * recursion and blowing the stack. Tail call optimization would
1756 * handle this, but compiling with frame pointers also disables
1757 * gcc's sibling call optimization.
1758 */
1759 if (bio->bi_end_io == bio_chain_endio) {
1760 bio = __bio_chain_endio(bio);
1761 goto again;
1762 }
1764 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1765 trace_block_bio_complete(bio->bi_disk->queue, bio,
1766 blk_status_to_errno(bio->bi_status));
1767 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1768 }
1770 blk_throtl_bio_endio(bio);
1771 /* release cgroup info */
1772 bio_uninit(bio);
1773 if (bio->bi_end_io)
1774 bio->bi_end_io(bio);
1775 }
1776 EXPORT_SYMBOL(bio_endio);
1778 /**
1779 * bio_split - split a bio
1780 * @bio: bio to split
1781 * @sectors: number of sectors to split from the front of @bio
1782 * @gfp: gfp mask
1783 * @bs: bio set to allocate from
1784 *
1785 * Allocates and returns a new bio which represents @sectors from the start of
1786 * @bio, and updates @bio to represent the remaining sectors.
1787 *
1788 * Unless this is a discard request the newly allocated bio will point
1789 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1790 * @bio is not freed before the split.
1791 */
1792 struct bio *bio_split(struct bio *bio, int sectors,
1793 gfp_t gfp, struct bio_set *bs)
1794 {
1795 struct bio *split;
1797 BUG_ON(sectors <= 0);
1798 BUG_ON(sectors >= bio_sectors(bio));
1800 split = bio_clone_fast(bio, gfp, bs);
1801 if (!split)
1802 return NULL;
1804 split->bi_iter.bi_size = sectors << 9;
1806 if (bio_integrity(split))
1807 bio_integrity_trim(split);
1809 bio_advance(bio, split->bi_iter.bi_size);
1810 bio->bi_iter.bi_done = 0;
1812 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1813 bio_set_flag(split, BIO_TRACE_COMPLETION);
1815 return split;
1816 }
1817 EXPORT_SYMBOL(bio_split);
1819 /**
1820 * bio_trim - trim a bio
1821 * @bio: bio to trim
1822 * @offset: number of sectors to trim from the front of @bio
1823 * @size: size we want to trim @bio to, in sectors
1824 */
1825 void bio_trim(struct bio *bio, int offset, int size)
1826 {
1827 /* 'bio' is a cloned bio which we need to trim to match
1828 * the given offset and size.
1829 */
1831 size <<= 9;
1832 if (offset == 0 && size == bio->bi_iter.bi_size)
1833 return;
1835 bio_clear_flag(bio, BIO_SEG_VALID);
1837 bio_advance(bio, offset << 9);
1839 bio->bi_iter.bi_size = size;
1841 if (bio_integrity(bio))
1842 bio_integrity_trim(bio);
1844 }
1845 EXPORT_SYMBOL_GPL(bio_trim);
1847 /*
1848 * create memory pools for biovec's in a bio_set.
1849 * use the global biovec slabs created for general use.
1850 */
1851 int biovec_init_pool(mempool_t *pool, int pool_entries)
1852 {
1853 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1855 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1856 }
1858 /*
1859 * bioset_exit - exit a bioset initialized with bioset_init()
1860 *
1861 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1862 * kzalloc()).
1863 */
1864 void bioset_exit(struct bio_set *bs)
1865 {
1866 if (bs->rescue_workqueue)
1867 destroy_workqueue(bs->rescue_workqueue);
1868 bs->rescue_workqueue = NULL;
1870 mempool_exit(&bs->bio_pool);
1871 mempool_exit(&bs->bvec_pool);
1873 bioset_integrity_free(bs);
1874 if (bs->bio_slab)
1875 bio_put_slab(bs);
1876 bs->bio_slab = NULL;
1877 }
1878 EXPORT_SYMBOL(bioset_exit);
1880 /**
1881 * bioset_init - Initialize a bio_set
1882 * @bs: pool to initialize
1883 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1884 * @front_pad: Number of bytes to allocate in front of the returned bio
1885 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1886 * and %BIOSET_NEED_RESCUER
1887 *
1888 * Description:
1889 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1890 * to ask for a number of bytes to be allocated in front of the bio.
1891 * Front pad allocation is useful for embedding the bio inside
1892 * another structure, to avoid allocating extra data to go with the bio.
1893 * Note that the bio must be embedded at the END of that structure always,
1894 * or things will break badly.
1895 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1896 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1897 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1898 * dispatch queued requests when the mempool runs out of space.
1899 *
1900 */
1901 int bioset_init(struct bio_set *bs,
1902 unsigned int pool_size,
1903 unsigned int front_pad,
1904 int flags)
1905 {
1906 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1908 bs->front_pad = front_pad;
1910 spin_lock_init(&bs->rescue_lock);
1911 bio_list_init(&bs->rescue_list);
1912 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1914 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1915 if (!bs->bio_slab)
1916 return -ENOMEM;
1918 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1919 goto bad;
1921 if ((flags & BIOSET_NEED_BVECS) &&
1922 biovec_init_pool(&bs->bvec_pool, pool_size))
1923 goto bad;
1925 if (!(flags & BIOSET_NEED_RESCUER))
1926 return 0;
1928 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1929 if (!bs->rescue_workqueue)
1930 goto bad;
1932 return 0;
1933 bad:
1934 bioset_exit(bs);
1935 return -ENOMEM;
1936 }
1937 EXPORT_SYMBOL(bioset_init);
1939 /*
1940 * Initialize and setup a new bio_set, based on the settings from
1941 * another bio_set.
1942 */
1943 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1944 {
1945 int flags;
1947 flags = 0;
1948 if (src->bvec_pool.min_nr)
1949 flags |= BIOSET_NEED_BVECS;
1950 if (src->rescue_workqueue)
1951 flags |= BIOSET_NEED_RESCUER;
1953 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1954 }
1955 EXPORT_SYMBOL(bioset_init_from_src);
1957 #ifdef CONFIG_BLK_CGROUP
1959 #ifdef CONFIG_MEMCG
1960 /**
1961 * bio_associate_blkcg_from_page - associate a bio with the page's blkcg
1962 * @bio: target bio
1963 * @page: the page to lookup the blkcg from
1964 *
1965 * Associate @bio with the blkcg from @page's owning memcg. This works like
1966 * every other associate function wrt references.
1967 */
1968 int bio_associate_blkcg_from_page(struct bio *bio, struct page *page)
1969 {
1970 struct cgroup_subsys_state *blkcg_css;
1972 if (unlikely(bio->bi_css))
1973 return -EBUSY;
1974 if (!page->mem_cgroup)
1975 return 0;
1976 blkcg_css = cgroup_get_e_css(page->mem_cgroup->css.cgroup,
1977 &io_cgrp_subsys);
1978 bio->bi_css = blkcg_css;
1979 return 0;
1980 }
1981 #endif /* CONFIG_MEMCG */
1983 /**
1984 * bio_associate_blkcg - associate a bio with the specified blkcg
1985 * @bio: target bio
1986 * @blkcg_css: css of the blkcg to associate
1987 *
1988 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1989 * treat @bio as if it were issued by a task which belongs to the blkcg.
1990 *
1991 * This function takes an extra reference of @blkcg_css which will be put
1992 * when @bio is released. The caller must own @bio and is responsible for
1993 * synchronizing calls to this function.
1994 */
1995 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1996 {
1997 if (unlikely(bio->bi_css))
1998 return -EBUSY;
1999 css_get(blkcg_css);
2000 bio->bi_css = blkcg_css;
2001 return 0;
2002 }
2003 EXPORT_SYMBOL_GPL(bio_associate_blkcg);
2005 /**
2006 * bio_associate_blkg - associate a bio with the specified blkg
2007 * @bio: target bio
2008 * @blkg: the blkg to associate
2009 *
2010 * Associate @bio with the blkg specified by @blkg. This is the queue specific
2011 * blkcg information associated with the @bio, a reference will be taken on the
2012 * @blkg and will be freed when the bio is freed.
2013 */
2014 int bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg)
2015 {
2016 if (unlikely(bio->bi_blkg))
2017 return -EBUSY;
2018 if (!blkg_try_get(blkg))
2019 return -ENODEV;
2020 bio->bi_blkg = blkg;
2021 return 0;
2022 }
2024 /**
2025 * bio_disassociate_task - undo bio_associate_current()
2026 * @bio: target bio
2027 */
2028 void bio_disassociate_task(struct bio *bio)
2029 {
2030 if (bio->bi_ioc) {
2031 put_io_context(bio->bi_ioc);
2032 bio->bi_ioc = NULL;
2033 }
2034 if (bio->bi_css) {
2035 css_put(bio->bi_css);
2036 bio->bi_css = NULL;
2037 }
2038 if (bio->bi_blkg) {
2039 blkg_put(bio->bi_blkg);
2040 bio->bi_blkg = NULL;
2041 }
2042 }
2044 /**
2045 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2046 * @dst: destination bio
2047 * @src: source bio
2048 */
2049 void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2050 {
2051 if (src->bi_css)
2052 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2053 }
2054 EXPORT_SYMBOL_GPL(bio_clone_blkcg_association);
2055 #endif /* CONFIG_BLK_CGROUP */
2057 static void __init biovec_init_slabs(void)
2058 {
2059 int i;
2061 for (i = 0; i < BVEC_POOL_NR; i++) {
2062 int size;
2063 struct biovec_slab *bvs = bvec_slabs + i;
2065 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2066 bvs->slab = NULL;
2067 continue;
2068 }
2070 size = bvs->nr_vecs * sizeof(struct bio_vec);
2071 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2072 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2073 }
2074 }
2076 static int __init init_bio(void)
2077 {
2078 bio_slab_max = 2;
2079 bio_slab_nr = 0;
2080 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab),
2081 GFP_KERNEL);
2082 if (!bio_slabs)
2083 panic("bio: can't allocate bios\n");
2085 bio_integrity_init();
2086 biovec_init_slabs();
2088 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
2089 panic("bio: can't allocate bios\n");
2091 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
2092 panic("bio: can't create integrity pool\n");
2094 return 0;
2095 }
2096 subsys_initcall(init_bio);