Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 */
5#include <linux/mm.h>
6#include <linux/swap.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/uio.h>
10#include <linux/iocontext.h>
11#include <linux/slab.h>
12#include <linux/init.h>
13#include <linux/kernel.h>
14#include <linux/export.h>
15#include <linux/mempool.h>
16#include <linux/workqueue.h>
17#include <linux/cgroup.h>
18#include <linux/highmem.h>
19#include <linux/sched/sysctl.h>
20#include <linux/blk-crypto.h>
21#include <linux/xarray.h>
22
23#include <trace/events/block.h>
24#include "blk.h"
25#include "blk-rq-qos.h"
26#include "blk-cgroup.h"
27
28#define ALLOC_CACHE_THRESHOLD 16
29#define ALLOC_CACHE_MAX 256
30
31struct bio_alloc_cache {
32 struct bio *free_list;
33 struct bio *free_list_irq;
34 unsigned int nr;
35 unsigned int nr_irq;
36};
37
38static struct biovec_slab {
39 int nr_vecs;
40 char *name;
41 struct kmem_cache *slab;
42} bvec_slabs[] __read_mostly = {
43 { .nr_vecs = 16, .name = "biovec-16" },
44 { .nr_vecs = 64, .name = "biovec-64" },
45 { .nr_vecs = 128, .name = "biovec-128" },
46 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
47};
48
49static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
50{
51 switch (nr_vecs) {
52 /* smaller bios use inline vecs */
53 case 5 ... 16:
54 return &bvec_slabs[0];
55 case 17 ... 64:
56 return &bvec_slabs[1];
57 case 65 ... 128:
58 return &bvec_slabs[2];
59 case 129 ... BIO_MAX_VECS:
60 return &bvec_slabs[3];
61 default:
62 BUG();
63 return NULL;
64 }
65}
66
67/*
68 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
69 * IO code that does not need private memory pools.
70 */
71struct bio_set fs_bio_set;
72EXPORT_SYMBOL(fs_bio_set);
73
74/*
75 * Our slab pool management
76 */
77struct bio_slab {
78 struct kmem_cache *slab;
79 unsigned int slab_ref;
80 unsigned int slab_size;
81 char name[8];
82};
83static DEFINE_MUTEX(bio_slab_lock);
84static DEFINE_XARRAY(bio_slabs);
85
86static struct bio_slab *create_bio_slab(unsigned int size)
87{
88 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
89
90 if (!bslab)
91 return NULL;
92
93 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
94 bslab->slab = kmem_cache_create(bslab->name, size,
95 ARCH_KMALLOC_MINALIGN,
96 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
97 if (!bslab->slab)
98 goto fail_alloc_slab;
99
100 bslab->slab_ref = 1;
101 bslab->slab_size = size;
102
103 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
104 return bslab;
105
106 kmem_cache_destroy(bslab->slab);
107
108fail_alloc_slab:
109 kfree(bslab);
110 return NULL;
111}
112
113static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
114{
115 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
116}
117
118static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
119{
120 unsigned int size = bs_bio_slab_size(bs);
121 struct bio_slab *bslab;
122
123 mutex_lock(&bio_slab_lock);
124 bslab = xa_load(&bio_slabs, size);
125 if (bslab)
126 bslab->slab_ref++;
127 else
128 bslab = create_bio_slab(size);
129 mutex_unlock(&bio_slab_lock);
130
131 if (bslab)
132 return bslab->slab;
133 return NULL;
134}
135
136static void bio_put_slab(struct bio_set *bs)
137{
138 struct bio_slab *bslab = NULL;
139 unsigned int slab_size = bs_bio_slab_size(bs);
140
141 mutex_lock(&bio_slab_lock);
142
143 bslab = xa_load(&bio_slabs, slab_size);
144 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 goto out;
146
147 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
148
149 WARN_ON(!bslab->slab_ref);
150
151 if (--bslab->slab_ref)
152 goto out;
153
154 xa_erase(&bio_slabs, slab_size);
155
156 kmem_cache_destroy(bslab->slab);
157 kfree(bslab);
158
159out:
160 mutex_unlock(&bio_slab_lock);
161}
162
163void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
164{
165 BUG_ON(nr_vecs > BIO_MAX_VECS);
166
167 if (nr_vecs == BIO_MAX_VECS)
168 mempool_free(bv, pool);
169 else if (nr_vecs > BIO_INLINE_VECS)
170 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
171}
172
173/*
174 * Make the first allocation restricted and don't dump info on allocation
175 * failures, since we'll fall back to the mempool in case of failure.
176 */
177static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
178{
179 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
180 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
181}
182
183struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
184 gfp_t gfp_mask)
185{
186 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
187
188 if (WARN_ON_ONCE(!bvs))
189 return NULL;
190
191 /*
192 * Upgrade the nr_vecs request to take full advantage of the allocation.
193 * We also rely on this in the bvec_free path.
194 */
195 *nr_vecs = bvs->nr_vecs;
196
197 /*
198 * Try a slab allocation first for all smaller allocations. If that
199 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
200 * The mempool is sized to handle up to BIO_MAX_VECS entries.
201 */
202 if (*nr_vecs < BIO_MAX_VECS) {
203 struct bio_vec *bvl;
204
205 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
206 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
207 return bvl;
208 *nr_vecs = BIO_MAX_VECS;
209 }
210
211 return mempool_alloc(pool, gfp_mask);
212}
213
214void bio_uninit(struct bio *bio)
215{
216#ifdef CONFIG_BLK_CGROUP
217 if (bio->bi_blkg) {
218 blkg_put(bio->bi_blkg);
219 bio->bi_blkg = NULL;
220 }
221#endif
222 if (bio_integrity(bio))
223 bio_integrity_free(bio);
224
225 bio_crypt_free_ctx(bio);
226}
227EXPORT_SYMBOL(bio_uninit);
228
229static void bio_free(struct bio *bio)
230{
231 struct bio_set *bs = bio->bi_pool;
232 void *p = bio;
233
234 WARN_ON_ONCE(!bs);
235
236 bio_uninit(bio);
237 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
238 mempool_free(p - bs->front_pad, &bs->bio_pool);
239}
240
241/*
242 * Users of this function have their own bio allocation. Subsequently,
243 * they must remember to pair any call to bio_init() with bio_uninit()
244 * when IO has completed, or when the bio is released.
245 */
246void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
247 unsigned short max_vecs, blk_opf_t opf)
248{
249 bio->bi_next = NULL;
250 bio->bi_bdev = bdev;
251 bio->bi_opf = opf;
252 bio->bi_flags = 0;
253 bio->bi_ioprio = 0;
254 bio->bi_status = 0;
255 bio->bi_iter.bi_sector = 0;
256 bio->bi_iter.bi_size = 0;
257 bio->bi_iter.bi_idx = 0;
258 bio->bi_iter.bi_bvec_done = 0;
259 bio->bi_end_io = NULL;
260 bio->bi_private = NULL;
261#ifdef CONFIG_BLK_CGROUP
262 bio->bi_blkg = NULL;
263 bio->bi_issue.value = 0;
264 if (bdev)
265 bio_associate_blkg(bio);
266#ifdef CONFIG_BLK_CGROUP_IOCOST
267 bio->bi_iocost_cost = 0;
268#endif
269#endif
270#ifdef CONFIG_BLK_INLINE_ENCRYPTION
271 bio->bi_crypt_context = NULL;
272#endif
273#ifdef CONFIG_BLK_DEV_INTEGRITY
274 bio->bi_integrity = NULL;
275#endif
276 bio->bi_vcnt = 0;
277
278 atomic_set(&bio->__bi_remaining, 1);
279 atomic_set(&bio->__bi_cnt, 1);
280 bio->bi_cookie = BLK_QC_T_NONE;
281
282 bio->bi_max_vecs = max_vecs;
283 bio->bi_io_vec = table;
284 bio->bi_pool = NULL;
285}
286EXPORT_SYMBOL(bio_init);
287
288/**
289 * bio_reset - reinitialize a bio
290 * @bio: bio to reset
291 * @bdev: block device to use the bio for
292 * @opf: operation and flags for bio
293 *
294 * Description:
295 * After calling bio_reset(), @bio will be in the same state as a freshly
296 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
297 * preserved are the ones that are initialized by bio_alloc_bioset(). See
298 * comment in struct bio.
299 */
300void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
301{
302 bio_uninit(bio);
303 memset(bio, 0, BIO_RESET_BYTES);
304 atomic_set(&bio->__bi_remaining, 1);
305 bio->bi_bdev = bdev;
306 if (bio->bi_bdev)
307 bio_associate_blkg(bio);
308 bio->bi_opf = opf;
309}
310EXPORT_SYMBOL(bio_reset);
311
312static struct bio *__bio_chain_endio(struct bio *bio)
313{
314 struct bio *parent = bio->bi_private;
315
316 if (bio->bi_status && !parent->bi_status)
317 parent->bi_status = bio->bi_status;
318 bio_put(bio);
319 return parent;
320}
321
322static void bio_chain_endio(struct bio *bio)
323{
324 bio_endio(__bio_chain_endio(bio));
325}
326
327/**
328 * bio_chain - chain bio completions
329 * @bio: the target bio
330 * @parent: the parent bio of @bio
331 *
332 * The caller won't have a bi_end_io called when @bio completes - instead,
333 * @parent's bi_end_io won't be called until both @parent and @bio have
334 * completed; the chained bio will also be freed when it completes.
335 *
336 * The caller must not set bi_private or bi_end_io in @bio.
337 */
338void bio_chain(struct bio *bio, struct bio *parent)
339{
340 BUG_ON(bio->bi_private || bio->bi_end_io);
341
342 bio->bi_private = parent;
343 bio->bi_end_io = bio_chain_endio;
344 bio_inc_remaining(parent);
345}
346EXPORT_SYMBOL(bio_chain);
347
348struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
349 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
350{
351 struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
352
353 if (bio) {
354 bio_chain(bio, new);
355 submit_bio(bio);
356 }
357
358 return new;
359}
360EXPORT_SYMBOL_GPL(blk_next_bio);
361
362static void bio_alloc_rescue(struct work_struct *work)
363{
364 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
365 struct bio *bio;
366
367 while (1) {
368 spin_lock(&bs->rescue_lock);
369 bio = bio_list_pop(&bs->rescue_list);
370 spin_unlock(&bs->rescue_lock);
371
372 if (!bio)
373 break;
374
375 submit_bio_noacct(bio);
376 }
377}
378
379static void punt_bios_to_rescuer(struct bio_set *bs)
380{
381 struct bio_list punt, nopunt;
382 struct bio *bio;
383
384 if (WARN_ON_ONCE(!bs->rescue_workqueue))
385 return;
386 /*
387 * In order to guarantee forward progress we must punt only bios that
388 * were allocated from this bio_set; otherwise, if there was a bio on
389 * there for a stacking driver higher up in the stack, processing it
390 * could require allocating bios from this bio_set, and doing that from
391 * our own rescuer would be bad.
392 *
393 * Since bio lists are singly linked, pop them all instead of trying to
394 * remove from the middle of the list:
395 */
396
397 bio_list_init(&punt);
398 bio_list_init(&nopunt);
399
400 while ((bio = bio_list_pop(¤t->bio_list[0])))
401 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
402 current->bio_list[0] = nopunt;
403
404 bio_list_init(&nopunt);
405 while ((bio = bio_list_pop(¤t->bio_list[1])))
406 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
407 current->bio_list[1] = nopunt;
408
409 spin_lock(&bs->rescue_lock);
410 bio_list_merge(&bs->rescue_list, &punt);
411 spin_unlock(&bs->rescue_lock);
412
413 queue_work(bs->rescue_workqueue, &bs->rescue_work);
414}
415
416static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
417{
418 unsigned long flags;
419
420 /* cache->free_list must be empty */
421 if (WARN_ON_ONCE(cache->free_list))
422 return;
423
424 local_irq_save(flags);
425 cache->free_list = cache->free_list_irq;
426 cache->free_list_irq = NULL;
427 cache->nr += cache->nr_irq;
428 cache->nr_irq = 0;
429 local_irq_restore(flags);
430}
431
432static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
433 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
434 struct bio_set *bs)
435{
436 struct bio_alloc_cache *cache;
437 struct bio *bio;
438
439 cache = per_cpu_ptr(bs->cache, get_cpu());
440 if (!cache->free_list) {
441 if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
442 bio_alloc_irq_cache_splice(cache);
443 if (!cache->free_list) {
444 put_cpu();
445 return NULL;
446 }
447 }
448 bio = cache->free_list;
449 cache->free_list = bio->bi_next;
450 cache->nr--;
451 put_cpu();
452
453 bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
454 bio->bi_pool = bs;
455 return bio;
456}
457
458/**
459 * bio_alloc_bioset - allocate a bio for I/O
460 * @bdev: block device to allocate the bio for (can be %NULL)
461 * @nr_vecs: number of bvecs to pre-allocate
462 * @opf: operation and flags for bio
463 * @gfp_mask: the GFP_* mask given to the slab allocator
464 * @bs: the bio_set to allocate from.
465 *
466 * Allocate a bio from the mempools in @bs.
467 *
468 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
469 * allocate a bio. This is due to the mempool guarantees. To make this work,
470 * callers must never allocate more than 1 bio at a time from the general pool.
471 * Callers that need to allocate more than 1 bio must always submit the
472 * previously allocated bio for IO before attempting to allocate a new one.
473 * Failure to do so can cause deadlocks under memory pressure.
474 *
475 * Note that when running under submit_bio_noacct() (i.e. any block driver),
476 * bios are not submitted until after you return - see the code in
477 * submit_bio_noacct() that converts recursion into iteration, to prevent
478 * stack overflows.
479 *
480 * This would normally mean allocating multiple bios under submit_bio_noacct()
481 * would be susceptible to deadlocks, but we have
482 * deadlock avoidance code that resubmits any blocked bios from a rescuer
483 * thread.
484 *
485 * However, we do not guarantee forward progress for allocations from other
486 * mempools. Doing multiple allocations from the same mempool under
487 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
488 * for per bio allocations.
489 *
490 * Returns: Pointer to new bio on success, NULL on failure.
491 */
492struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
493 blk_opf_t opf, gfp_t gfp_mask,
494 struct bio_set *bs)
495{
496 gfp_t saved_gfp = gfp_mask;
497 struct bio *bio;
498 void *p;
499
500 /* should not use nobvec bioset for nr_vecs > 0 */
501 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
502 return NULL;
503
504 if (opf & REQ_ALLOC_CACHE) {
505 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
506 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
507 gfp_mask, bs);
508 if (bio)
509 return bio;
510 /*
511 * No cached bio available, bio returned below marked with
512 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
513 */
514 } else {
515 opf &= ~REQ_ALLOC_CACHE;
516 }
517 }
518
519 /*
520 * submit_bio_noacct() converts recursion to iteration; this means if
521 * we're running beneath it, any bios we allocate and submit will not be
522 * submitted (and thus freed) until after we return.
523 *
524 * This exposes us to a potential deadlock if we allocate multiple bios
525 * from the same bio_set() while running underneath submit_bio_noacct().
526 * If we were to allocate multiple bios (say a stacking block driver
527 * that was splitting bios), we would deadlock if we exhausted the
528 * mempool's reserve.
529 *
530 * We solve this, and guarantee forward progress, with a rescuer
531 * workqueue per bio_set. If we go to allocate and there are bios on
532 * current->bio_list, we first try the allocation without
533 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
534 * blocking to the rescuer workqueue before we retry with the original
535 * gfp_flags.
536 */
537 if (current->bio_list &&
538 (!bio_list_empty(¤t->bio_list[0]) ||
539 !bio_list_empty(¤t->bio_list[1])) &&
540 bs->rescue_workqueue)
541 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
542
543 p = mempool_alloc(&bs->bio_pool, gfp_mask);
544 if (!p && gfp_mask != saved_gfp) {
545 punt_bios_to_rescuer(bs);
546 gfp_mask = saved_gfp;
547 p = mempool_alloc(&bs->bio_pool, gfp_mask);
548 }
549 if (unlikely(!p))
550 return NULL;
551 if (!mempool_is_saturated(&bs->bio_pool))
552 opf &= ~REQ_ALLOC_CACHE;
553
554 bio = p + bs->front_pad;
555 if (nr_vecs > BIO_INLINE_VECS) {
556 struct bio_vec *bvl = NULL;
557
558 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
559 if (!bvl && gfp_mask != saved_gfp) {
560 punt_bios_to_rescuer(bs);
561 gfp_mask = saved_gfp;
562 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
563 }
564 if (unlikely(!bvl))
565 goto err_free;
566
567 bio_init(bio, bdev, bvl, nr_vecs, opf);
568 } else if (nr_vecs) {
569 bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
570 } else {
571 bio_init(bio, bdev, NULL, 0, opf);
572 }
573
574 bio->bi_pool = bs;
575 return bio;
576
577err_free:
578 mempool_free(p, &bs->bio_pool);
579 return NULL;
580}
581EXPORT_SYMBOL(bio_alloc_bioset);
582
583/**
584 * bio_kmalloc - kmalloc a bio
585 * @nr_vecs: number of bio_vecs to allocate
586 * @gfp_mask: the GFP_* mask given to the slab allocator
587 *
588 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
589 * using bio_init() before use. To free a bio returned from this function use
590 * kfree() after calling bio_uninit(). A bio returned from this function can
591 * be reused by calling bio_uninit() before calling bio_init() again.
592 *
593 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
594 * function are not backed by a mempool can fail. Do not use this function
595 * for allocations in the file system I/O path.
596 *
597 * Returns: Pointer to new bio on success, NULL on failure.
598 */
599struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
600{
601 struct bio *bio;
602
603 if (nr_vecs > UIO_MAXIOV)
604 return NULL;
605 return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
606}
607EXPORT_SYMBOL(bio_kmalloc);
608
609void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
610{
611 struct bio_vec bv;
612 struct bvec_iter iter;
613
614 __bio_for_each_segment(bv, bio, iter, start)
615 memzero_bvec(&bv);
616}
617EXPORT_SYMBOL(zero_fill_bio_iter);
618
619/**
620 * bio_truncate - truncate the bio to small size of @new_size
621 * @bio: the bio to be truncated
622 * @new_size: new size for truncating the bio
623 *
624 * Description:
625 * Truncate the bio to new size of @new_size. If bio_op(bio) is
626 * REQ_OP_READ, zero the truncated part. This function should only
627 * be used for handling corner cases, such as bio eod.
628 */
629static void bio_truncate(struct bio *bio, unsigned new_size)
630{
631 struct bio_vec bv;
632 struct bvec_iter iter;
633 unsigned int done = 0;
634 bool truncated = false;
635
636 if (new_size >= bio->bi_iter.bi_size)
637 return;
638
639 if (bio_op(bio) != REQ_OP_READ)
640 goto exit;
641
642 bio_for_each_segment(bv, bio, iter) {
643 if (done + bv.bv_len > new_size) {
644 unsigned offset;
645
646 if (!truncated)
647 offset = new_size - done;
648 else
649 offset = 0;
650 zero_user(bv.bv_page, bv.bv_offset + offset,
651 bv.bv_len - offset);
652 truncated = true;
653 }
654 done += bv.bv_len;
655 }
656
657 exit:
658 /*
659 * Don't touch bvec table here and make it really immutable, since
660 * fs bio user has to retrieve all pages via bio_for_each_segment_all
661 * in its .end_bio() callback.
662 *
663 * It is enough to truncate bio by updating .bi_size since we can make
664 * correct bvec with the updated .bi_size for drivers.
665 */
666 bio->bi_iter.bi_size = new_size;
667}
668
669/**
670 * guard_bio_eod - truncate a BIO to fit the block device
671 * @bio: bio to truncate
672 *
673 * This allows us to do IO even on the odd last sectors of a device, even if the
674 * block size is some multiple of the physical sector size.
675 *
676 * We'll just truncate the bio to the size of the device, and clear the end of
677 * the buffer head manually. Truly out-of-range accesses will turn into actual
678 * I/O errors, this only handles the "we need to be able to do I/O at the final
679 * sector" case.
680 */
681void guard_bio_eod(struct bio *bio)
682{
683 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
684
685 if (!maxsector)
686 return;
687
688 /*
689 * If the *whole* IO is past the end of the device,
690 * let it through, and the IO layer will turn it into
691 * an EIO.
692 */
693 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
694 return;
695
696 maxsector -= bio->bi_iter.bi_sector;
697 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
698 return;
699
700 bio_truncate(bio, maxsector << 9);
701}
702
703static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
704 unsigned int nr)
705{
706 unsigned int i = 0;
707 struct bio *bio;
708
709 while ((bio = cache->free_list) != NULL) {
710 cache->free_list = bio->bi_next;
711 cache->nr--;
712 bio_free(bio);
713 if (++i == nr)
714 break;
715 }
716 return i;
717}
718
719static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
720 unsigned int nr)
721{
722 nr -= __bio_alloc_cache_prune(cache, nr);
723 if (!READ_ONCE(cache->free_list)) {
724 bio_alloc_irq_cache_splice(cache);
725 __bio_alloc_cache_prune(cache, nr);
726 }
727}
728
729static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
730{
731 struct bio_set *bs;
732
733 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
734 if (bs->cache) {
735 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
736
737 bio_alloc_cache_prune(cache, -1U);
738 }
739 return 0;
740}
741
742static void bio_alloc_cache_destroy(struct bio_set *bs)
743{
744 int cpu;
745
746 if (!bs->cache)
747 return;
748
749 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
750 for_each_possible_cpu(cpu) {
751 struct bio_alloc_cache *cache;
752
753 cache = per_cpu_ptr(bs->cache, cpu);
754 bio_alloc_cache_prune(cache, -1U);
755 }
756 free_percpu(bs->cache);
757 bs->cache = NULL;
758}
759
760static inline void bio_put_percpu_cache(struct bio *bio)
761{
762 struct bio_alloc_cache *cache;
763
764 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
765 if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) {
766 put_cpu();
767 bio_free(bio);
768 return;
769 }
770
771 bio_uninit(bio);
772
773 if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
774 bio->bi_next = cache->free_list;
775 bio->bi_bdev = NULL;
776 cache->free_list = bio;
777 cache->nr++;
778 } else {
779 unsigned long flags;
780
781 local_irq_save(flags);
782 bio->bi_next = cache->free_list_irq;
783 cache->free_list_irq = bio;
784 cache->nr_irq++;
785 local_irq_restore(flags);
786 }
787 put_cpu();
788}
789
790/**
791 * bio_put - release a reference to a bio
792 * @bio: bio to release reference to
793 *
794 * Description:
795 * Put a reference to a &struct bio, either one you have gotten with
796 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
797 **/
798void bio_put(struct bio *bio)
799{
800 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
801 BUG_ON(!atomic_read(&bio->__bi_cnt));
802 if (!atomic_dec_and_test(&bio->__bi_cnt))
803 return;
804 }
805 if (bio->bi_opf & REQ_ALLOC_CACHE)
806 bio_put_percpu_cache(bio);
807 else
808 bio_free(bio);
809}
810EXPORT_SYMBOL(bio_put);
811
812static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
813{
814 bio_set_flag(bio, BIO_CLONED);
815 bio->bi_ioprio = bio_src->bi_ioprio;
816 bio->bi_iter = bio_src->bi_iter;
817
818 if (bio->bi_bdev) {
819 if (bio->bi_bdev == bio_src->bi_bdev &&
820 bio_flagged(bio_src, BIO_REMAPPED))
821 bio_set_flag(bio, BIO_REMAPPED);
822 bio_clone_blkg_association(bio, bio_src);
823 }
824
825 if (bio_crypt_clone(bio, bio_src, gfp) < 0)
826 return -ENOMEM;
827 if (bio_integrity(bio_src) &&
828 bio_integrity_clone(bio, bio_src, gfp) < 0)
829 return -ENOMEM;
830 return 0;
831}
832
833/**
834 * bio_alloc_clone - clone a bio that shares the original bio's biovec
835 * @bdev: block_device to clone onto
836 * @bio_src: bio to clone from
837 * @gfp: allocation priority
838 * @bs: bio_set to allocate from
839 *
840 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
841 * bio, but not the actual data it points to.
842 *
843 * The caller must ensure that the return bio is not freed before @bio_src.
844 */
845struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
846 gfp_t gfp, struct bio_set *bs)
847{
848 struct bio *bio;
849
850 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
851 if (!bio)
852 return NULL;
853
854 if (__bio_clone(bio, bio_src, gfp) < 0) {
855 bio_put(bio);
856 return NULL;
857 }
858 bio->bi_io_vec = bio_src->bi_io_vec;
859
860 return bio;
861}
862EXPORT_SYMBOL(bio_alloc_clone);
863
864/**
865 * bio_init_clone - clone a bio that shares the original bio's biovec
866 * @bdev: block_device to clone onto
867 * @bio: bio to clone into
868 * @bio_src: bio to clone from
869 * @gfp: allocation priority
870 *
871 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
872 * The caller owns the returned bio, but not the actual data it points to.
873 *
874 * The caller must ensure that @bio_src is not freed before @bio.
875 */
876int bio_init_clone(struct block_device *bdev, struct bio *bio,
877 struct bio *bio_src, gfp_t gfp)
878{
879 int ret;
880
881 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
882 ret = __bio_clone(bio, bio_src, gfp);
883 if (ret)
884 bio_uninit(bio);
885 return ret;
886}
887EXPORT_SYMBOL(bio_init_clone);
888
889/**
890 * bio_full - check if the bio is full
891 * @bio: bio to check
892 * @len: length of one segment to be added
893 *
894 * Return true if @bio is full and one segment with @len bytes can't be
895 * added to the bio, otherwise return false
896 */
897static inline bool bio_full(struct bio *bio, unsigned len)
898{
899 if (bio->bi_vcnt >= bio->bi_max_vecs)
900 return true;
901 if (bio->bi_iter.bi_size > UINT_MAX - len)
902 return true;
903 return false;
904}
905
906static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page,
907 unsigned int len, unsigned int off, bool *same_page)
908{
909 size_t bv_end = bv->bv_offset + bv->bv_len;
910 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
911 phys_addr_t page_addr = page_to_phys(page);
912
913 if (vec_end_addr + 1 != page_addr + off)
914 return false;
915 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
916 return false;
917 if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
918 return false;
919
920 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
921 if (!*same_page) {
922 if (IS_ENABLED(CONFIG_KMSAN))
923 return false;
924 if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE)
925 return false;
926 }
927
928 bv->bv_len += len;
929 return true;
930}
931
932/*
933 * Try to merge a page into a segment, while obeying the hardware segment
934 * size limit. This is not for normal read/write bios, but for passthrough
935 * or Zone Append operations that we can't split.
936 */
937bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
938 struct page *page, unsigned len, unsigned offset,
939 bool *same_page)
940{
941 unsigned long mask = queue_segment_boundary(q);
942 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
943 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
944
945 if ((addr1 | mask) != (addr2 | mask))
946 return false;
947 if (len > queue_max_segment_size(q) - bv->bv_len)
948 return false;
949 return bvec_try_merge_page(bv, page, len, offset, same_page);
950}
951
952/**
953 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
954 * @q: the target queue
955 * @bio: destination bio
956 * @page: page to add
957 * @len: vec entry length
958 * @offset: vec entry offset
959 * @max_sectors: maximum number of sectors that can be added
960 * @same_page: return if the segment has been merged inside the same page
961 *
962 * Add a page to a bio while respecting the hardware max_sectors, max_segment
963 * and gap limitations.
964 */
965int bio_add_hw_page(struct request_queue *q, struct bio *bio,
966 struct page *page, unsigned int len, unsigned int offset,
967 unsigned int max_sectors, bool *same_page)
968{
969 unsigned int max_size = max_sectors << SECTOR_SHIFT;
970
971 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
972 return 0;
973
974 len = min3(len, max_size, queue_max_segment_size(q));
975 if (len > max_size - bio->bi_iter.bi_size)
976 return 0;
977
978 if (bio->bi_vcnt > 0) {
979 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
980
981 if (bvec_try_merge_hw_page(q, bv, page, len, offset,
982 same_page)) {
983 bio->bi_iter.bi_size += len;
984 return len;
985 }
986
987 if (bio->bi_vcnt >=
988 min(bio->bi_max_vecs, queue_max_segments(q)))
989 return 0;
990
991 /*
992 * If the queue doesn't support SG gaps and adding this segment
993 * would create a gap, disallow it.
994 */
995 if (bvec_gap_to_prev(&q->limits, bv, offset))
996 return 0;
997 }
998
999 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
1000 bio->bi_vcnt++;
1001 bio->bi_iter.bi_size += len;
1002 return len;
1003}
1004
1005/**
1006 * bio_add_pc_page - attempt to add page to passthrough bio
1007 * @q: the target queue
1008 * @bio: destination bio
1009 * @page: page to add
1010 * @len: vec entry length
1011 * @offset: vec entry offset
1012 *
1013 * Attempt to add a page to the bio_vec maplist. This can fail for a
1014 * number of reasons, such as the bio being full or target block device
1015 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1016 * so it is always possible to add a single page to an empty bio.
1017 *
1018 * This should only be used by passthrough bios.
1019 */
1020int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1021 struct page *page, unsigned int len, unsigned int offset)
1022{
1023 bool same_page = false;
1024 return bio_add_hw_page(q, bio, page, len, offset,
1025 queue_max_hw_sectors(q), &same_page);
1026}
1027EXPORT_SYMBOL(bio_add_pc_page);
1028
1029/**
1030 * bio_add_zone_append_page - attempt to add page to zone-append bio
1031 * @bio: destination bio
1032 * @page: page to add
1033 * @len: vec entry length
1034 * @offset: vec entry offset
1035 *
1036 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1037 * for a zone-append request. This can fail for a number of reasons, such as the
1038 * bio being full or the target block device is not a zoned block device or
1039 * other limitations of the target block device. The target block device must
1040 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1041 * to an empty bio.
1042 *
1043 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1044 */
1045int bio_add_zone_append_page(struct bio *bio, struct page *page,
1046 unsigned int len, unsigned int offset)
1047{
1048 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1049 bool same_page = false;
1050
1051 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1052 return 0;
1053
1054 if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1055 return 0;
1056
1057 return bio_add_hw_page(q, bio, page, len, offset,
1058 queue_max_zone_append_sectors(q), &same_page);
1059}
1060EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1061
1062/**
1063 * __bio_add_page - add page(s) to a bio in a new segment
1064 * @bio: destination bio
1065 * @page: start page to add
1066 * @len: length of the data to add, may cross pages
1067 * @off: offset of the data relative to @page, may cross pages
1068 *
1069 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1070 * that @bio has space for another bvec.
1071 */
1072void __bio_add_page(struct bio *bio, struct page *page,
1073 unsigned int len, unsigned int off)
1074{
1075 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1076 WARN_ON_ONCE(bio_full(bio, len));
1077
1078 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
1079 bio->bi_iter.bi_size += len;
1080 bio->bi_vcnt++;
1081}
1082EXPORT_SYMBOL_GPL(__bio_add_page);
1083
1084/**
1085 * bio_add_page - attempt to add page(s) to bio
1086 * @bio: destination bio
1087 * @page: start page to add
1088 * @len: vec entry length, may cross pages
1089 * @offset: vec entry offset relative to @page, may cross pages
1090 *
1091 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1092 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1093 */
1094int bio_add_page(struct bio *bio, struct page *page,
1095 unsigned int len, unsigned int offset)
1096{
1097 bool same_page = false;
1098
1099 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1100 return 0;
1101 if (bio->bi_iter.bi_size > UINT_MAX - len)
1102 return 0;
1103
1104 if (bio->bi_vcnt > 0 &&
1105 bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1106 page, len, offset, &same_page)) {
1107 bio->bi_iter.bi_size += len;
1108 return len;
1109 }
1110
1111 if (bio->bi_vcnt >= bio->bi_max_vecs)
1112 return 0;
1113 __bio_add_page(bio, page, len, offset);
1114 return len;
1115}
1116EXPORT_SYMBOL(bio_add_page);
1117
1118void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1119 size_t off)
1120{
1121 WARN_ON_ONCE(len > UINT_MAX);
1122 WARN_ON_ONCE(off > UINT_MAX);
1123 __bio_add_page(bio, &folio->page, len, off);
1124}
1125
1126/**
1127 * bio_add_folio - Attempt to add part of a folio to a bio.
1128 * @bio: BIO to add to.
1129 * @folio: Folio to add.
1130 * @len: How many bytes from the folio to add.
1131 * @off: First byte in this folio to add.
1132 *
1133 * Filesystems that use folios can call this function instead of calling
1134 * bio_add_page() for each page in the folio. If @off is bigger than
1135 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1136 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1137 *
1138 * Return: Whether the addition was successful.
1139 */
1140bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1141 size_t off)
1142{
1143 if (len > UINT_MAX || off > UINT_MAX)
1144 return false;
1145 return bio_add_page(bio, &folio->page, len, off) > 0;
1146}
1147EXPORT_SYMBOL(bio_add_folio);
1148
1149void __bio_release_pages(struct bio *bio, bool mark_dirty)
1150{
1151 struct folio_iter fi;
1152
1153 bio_for_each_folio_all(fi, bio) {
1154 struct page *page;
1155 size_t done = 0;
1156
1157 if (mark_dirty) {
1158 folio_lock(fi.folio);
1159 folio_mark_dirty(fi.folio);
1160 folio_unlock(fi.folio);
1161 }
1162 page = folio_page(fi.folio, fi.offset / PAGE_SIZE);
1163 do {
1164 bio_release_page(bio, page++);
1165 done += PAGE_SIZE;
1166 } while (done < fi.length);
1167 }
1168}
1169EXPORT_SYMBOL_GPL(__bio_release_pages);
1170
1171void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1172{
1173 size_t size = iov_iter_count(iter);
1174
1175 WARN_ON_ONCE(bio->bi_max_vecs);
1176
1177 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1178 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1179 size_t max_sectors = queue_max_zone_append_sectors(q);
1180
1181 size = min(size, max_sectors << SECTOR_SHIFT);
1182 }
1183
1184 bio->bi_vcnt = iter->nr_segs;
1185 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1186 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1187 bio->bi_iter.bi_size = size;
1188 bio_set_flag(bio, BIO_CLONED);
1189}
1190
1191static int bio_iov_add_page(struct bio *bio, struct page *page,
1192 unsigned int len, unsigned int offset)
1193{
1194 bool same_page = false;
1195
1196 if (WARN_ON_ONCE(bio->bi_iter.bi_size > UINT_MAX - len))
1197 return -EIO;
1198
1199 if (bio->bi_vcnt > 0 &&
1200 bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1201 page, len, offset, &same_page)) {
1202 bio->bi_iter.bi_size += len;
1203 if (same_page)
1204 bio_release_page(bio, page);
1205 return 0;
1206 }
1207 __bio_add_page(bio, page, len, offset);
1208 return 0;
1209}
1210
1211static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1212 unsigned int len, unsigned int offset)
1213{
1214 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1215 bool same_page = false;
1216
1217 if (bio_add_hw_page(q, bio, page, len, offset,
1218 queue_max_zone_append_sectors(q), &same_page) != len)
1219 return -EINVAL;
1220 if (same_page)
1221 bio_release_page(bio, page);
1222 return 0;
1223}
1224
1225#define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1226
1227/**
1228 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1229 * @bio: bio to add pages to
1230 * @iter: iov iterator describing the region to be mapped
1231 *
1232 * Extracts pages from *iter and appends them to @bio's bvec array. The pages
1233 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.
1234 * For a multi-segment *iter, this function only adds pages from the next
1235 * non-empty segment of the iov iterator.
1236 */
1237static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1238{
1239 iov_iter_extraction_t extraction_flags = 0;
1240 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1241 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1242 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1243 struct page **pages = (struct page **)bv;
1244 ssize_t size, left;
1245 unsigned len, i = 0;
1246 size_t offset;
1247 int ret = 0;
1248
1249 /*
1250 * Move page array up in the allocated memory for the bio vecs as far as
1251 * possible so that we can start filling biovecs from the beginning
1252 * without overwriting the temporary page array.
1253 */
1254 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1255 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1256
1257 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1258 extraction_flags |= ITER_ALLOW_P2PDMA;
1259
1260 /*
1261 * Each segment in the iov is required to be a block size multiple.
1262 * However, we may not be able to get the entire segment if it spans
1263 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1264 * result to ensure the bio's total size is correct. The remainder of
1265 * the iov data will be picked up in the next bio iteration.
1266 */
1267 size = iov_iter_extract_pages(iter, &pages,
1268 UINT_MAX - bio->bi_iter.bi_size,
1269 nr_pages, extraction_flags, &offset);
1270 if (unlikely(size <= 0))
1271 return size ? size : -EFAULT;
1272
1273 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1274
1275 if (bio->bi_bdev) {
1276 size_t trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1277 iov_iter_revert(iter, trim);
1278 size -= trim;
1279 }
1280
1281 if (unlikely(!size)) {
1282 ret = -EFAULT;
1283 goto out;
1284 }
1285
1286 for (left = size, i = 0; left > 0; left -= len, i++) {
1287 struct page *page = pages[i];
1288
1289 len = min_t(size_t, PAGE_SIZE - offset, left);
1290 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1291 ret = bio_iov_add_zone_append_page(bio, page, len,
1292 offset);
1293 if (ret)
1294 break;
1295 } else
1296 bio_iov_add_page(bio, page, len, offset);
1297
1298 offset = 0;
1299 }
1300
1301 iov_iter_revert(iter, left);
1302out:
1303 while (i < nr_pages)
1304 bio_release_page(bio, pages[i++]);
1305
1306 return ret;
1307}
1308
1309/**
1310 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1311 * @bio: bio to add pages to
1312 * @iter: iov iterator describing the region to be added
1313 *
1314 * This takes either an iterator pointing to user memory, or one pointing to
1315 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1316 * map them into the kernel. On IO completion, the caller should put those
1317 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1318 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1319 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1320 * completed by a call to ->ki_complete() or returns with an error other than
1321 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1322 * on IO completion. If it isn't, then pages should be released.
1323 *
1324 * The function tries, but does not guarantee, to pin as many pages as
1325 * fit into the bio, or are requested in @iter, whatever is smaller. If
1326 * MM encounters an error pinning the requested pages, it stops. Error
1327 * is returned only if 0 pages could be pinned.
1328 */
1329int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1330{
1331 int ret = 0;
1332
1333 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1334 return -EIO;
1335
1336 if (iov_iter_is_bvec(iter)) {
1337 bio_iov_bvec_set(bio, iter);
1338 iov_iter_advance(iter, bio->bi_iter.bi_size);
1339 return 0;
1340 }
1341
1342 if (iov_iter_extract_will_pin(iter))
1343 bio_set_flag(bio, BIO_PAGE_PINNED);
1344 do {
1345 ret = __bio_iov_iter_get_pages(bio, iter);
1346 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1347
1348 return bio->bi_vcnt ? 0 : ret;
1349}
1350EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1351
1352static void submit_bio_wait_endio(struct bio *bio)
1353{
1354 complete(bio->bi_private);
1355}
1356
1357/**
1358 * submit_bio_wait - submit a bio, and wait until it completes
1359 * @bio: The &struct bio which describes the I/O
1360 *
1361 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1362 * bio_endio() on failure.
1363 *
1364 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1365 * result in bio reference to be consumed. The caller must drop the reference
1366 * on his own.
1367 */
1368int submit_bio_wait(struct bio *bio)
1369{
1370 DECLARE_COMPLETION_ONSTACK_MAP(done,
1371 bio->bi_bdev->bd_disk->lockdep_map);
1372 unsigned long hang_check;
1373
1374 bio->bi_private = &done;
1375 bio->bi_end_io = submit_bio_wait_endio;
1376 bio->bi_opf |= REQ_SYNC;
1377 submit_bio(bio);
1378
1379 /* Prevent hang_check timer from firing at us during very long I/O */
1380 hang_check = sysctl_hung_task_timeout_secs;
1381 if (hang_check)
1382 while (!wait_for_completion_io_timeout(&done,
1383 hang_check * (HZ/2)))
1384 ;
1385 else
1386 wait_for_completion_io(&done);
1387
1388 return blk_status_to_errno(bio->bi_status);
1389}
1390EXPORT_SYMBOL(submit_bio_wait);
1391
1392void __bio_advance(struct bio *bio, unsigned bytes)
1393{
1394 if (bio_integrity(bio))
1395 bio_integrity_advance(bio, bytes);
1396
1397 bio_crypt_advance(bio, bytes);
1398 bio_advance_iter(bio, &bio->bi_iter, bytes);
1399}
1400EXPORT_SYMBOL(__bio_advance);
1401
1402void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1403 struct bio *src, struct bvec_iter *src_iter)
1404{
1405 while (src_iter->bi_size && dst_iter->bi_size) {
1406 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1407 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1408 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1409 void *src_buf = bvec_kmap_local(&src_bv);
1410 void *dst_buf = bvec_kmap_local(&dst_bv);
1411
1412 memcpy(dst_buf, src_buf, bytes);
1413
1414 kunmap_local(dst_buf);
1415 kunmap_local(src_buf);
1416
1417 bio_advance_iter_single(src, src_iter, bytes);
1418 bio_advance_iter_single(dst, dst_iter, bytes);
1419 }
1420}
1421EXPORT_SYMBOL(bio_copy_data_iter);
1422
1423/**
1424 * bio_copy_data - copy contents of data buffers from one bio to another
1425 * @src: source bio
1426 * @dst: destination bio
1427 *
1428 * Stops when it reaches the end of either @src or @dst - that is, copies
1429 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1430 */
1431void bio_copy_data(struct bio *dst, struct bio *src)
1432{
1433 struct bvec_iter src_iter = src->bi_iter;
1434 struct bvec_iter dst_iter = dst->bi_iter;
1435
1436 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1437}
1438EXPORT_SYMBOL(bio_copy_data);
1439
1440void bio_free_pages(struct bio *bio)
1441{
1442 struct bio_vec *bvec;
1443 struct bvec_iter_all iter_all;
1444
1445 bio_for_each_segment_all(bvec, bio, iter_all)
1446 __free_page(bvec->bv_page);
1447}
1448EXPORT_SYMBOL(bio_free_pages);
1449
1450/*
1451 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1452 * for performing direct-IO in BIOs.
1453 *
1454 * The problem is that we cannot run folio_mark_dirty() from interrupt context
1455 * because the required locks are not interrupt-safe. So what we can do is to
1456 * mark the pages dirty _before_ performing IO. And in interrupt context,
1457 * check that the pages are still dirty. If so, fine. If not, redirty them
1458 * in process context.
1459 *
1460 * Note that this code is very hard to test under normal circumstances because
1461 * direct-io pins the pages with get_user_pages(). This makes
1462 * is_page_cache_freeable return false, and the VM will not clean the pages.
1463 * But other code (eg, flusher threads) could clean the pages if they are mapped
1464 * pagecache.
1465 *
1466 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1467 * deferred bio dirtying paths.
1468 */
1469
1470/*
1471 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1472 */
1473void bio_set_pages_dirty(struct bio *bio)
1474{
1475 struct folio_iter fi;
1476
1477 bio_for_each_folio_all(fi, bio) {
1478 folio_lock(fi.folio);
1479 folio_mark_dirty(fi.folio);
1480 folio_unlock(fi.folio);
1481 }
1482}
1483EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1484
1485/*
1486 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1487 * If they are, then fine. If, however, some pages are clean then they must
1488 * have been written out during the direct-IO read. So we take another ref on
1489 * the BIO and re-dirty the pages in process context.
1490 *
1491 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1492 * here on. It will unpin each page and will run one bio_put() against the
1493 * BIO.
1494 */
1495
1496static void bio_dirty_fn(struct work_struct *work);
1497
1498static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1499static DEFINE_SPINLOCK(bio_dirty_lock);
1500static struct bio *bio_dirty_list;
1501
1502/*
1503 * This runs in process context
1504 */
1505static void bio_dirty_fn(struct work_struct *work)
1506{
1507 struct bio *bio, *next;
1508
1509 spin_lock_irq(&bio_dirty_lock);
1510 next = bio_dirty_list;
1511 bio_dirty_list = NULL;
1512 spin_unlock_irq(&bio_dirty_lock);
1513
1514 while ((bio = next) != NULL) {
1515 next = bio->bi_private;
1516
1517 bio_release_pages(bio, true);
1518 bio_put(bio);
1519 }
1520}
1521
1522void bio_check_pages_dirty(struct bio *bio)
1523{
1524 struct folio_iter fi;
1525 unsigned long flags;
1526
1527 bio_for_each_folio_all(fi, bio) {
1528 if (!folio_test_dirty(fi.folio))
1529 goto defer;
1530 }
1531
1532 bio_release_pages(bio, false);
1533 bio_put(bio);
1534 return;
1535defer:
1536 spin_lock_irqsave(&bio_dirty_lock, flags);
1537 bio->bi_private = bio_dirty_list;
1538 bio_dirty_list = bio;
1539 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1540 schedule_work(&bio_dirty_work);
1541}
1542EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1543
1544static inline bool bio_remaining_done(struct bio *bio)
1545{
1546 /*
1547 * If we're not chaining, then ->__bi_remaining is always 1 and
1548 * we always end io on the first invocation.
1549 */
1550 if (!bio_flagged(bio, BIO_CHAIN))
1551 return true;
1552
1553 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1554
1555 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1556 bio_clear_flag(bio, BIO_CHAIN);
1557 return true;
1558 }
1559
1560 return false;
1561}
1562
1563/**
1564 * bio_endio - end I/O on a bio
1565 * @bio: bio
1566 *
1567 * Description:
1568 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1569 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1570 * bio unless they own it and thus know that it has an end_io function.
1571 *
1572 * bio_endio() can be called several times on a bio that has been chained
1573 * using bio_chain(). The ->bi_end_io() function will only be called the
1574 * last time.
1575 **/
1576void bio_endio(struct bio *bio)
1577{
1578again:
1579 if (!bio_remaining_done(bio))
1580 return;
1581 if (!bio_integrity_endio(bio))
1582 return;
1583
1584 rq_qos_done_bio(bio);
1585
1586 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1587 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1588 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1589 }
1590
1591 /*
1592 * Need to have a real endio function for chained bios, otherwise
1593 * various corner cases will break (like stacking block devices that
1594 * save/restore bi_end_io) - however, we want to avoid unbounded
1595 * recursion and blowing the stack. Tail call optimization would
1596 * handle this, but compiling with frame pointers also disables
1597 * gcc's sibling call optimization.
1598 */
1599 if (bio->bi_end_io == bio_chain_endio) {
1600 bio = __bio_chain_endio(bio);
1601 goto again;
1602 }
1603
1604 blk_throtl_bio_endio(bio);
1605 /* release cgroup info */
1606 bio_uninit(bio);
1607 if (bio->bi_end_io)
1608 bio->bi_end_io(bio);
1609}
1610EXPORT_SYMBOL(bio_endio);
1611
1612/**
1613 * bio_split - split a bio
1614 * @bio: bio to split
1615 * @sectors: number of sectors to split from the front of @bio
1616 * @gfp: gfp mask
1617 * @bs: bio set to allocate from
1618 *
1619 * Allocates and returns a new bio which represents @sectors from the start of
1620 * @bio, and updates @bio to represent the remaining sectors.
1621 *
1622 * Unless this is a discard request the newly allocated bio will point
1623 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1624 * neither @bio nor @bs are freed before the split bio.
1625 */
1626struct bio *bio_split(struct bio *bio, int sectors,
1627 gfp_t gfp, struct bio_set *bs)
1628{
1629 struct bio *split;
1630
1631 BUG_ON(sectors <= 0);
1632 BUG_ON(sectors >= bio_sectors(bio));
1633
1634 /* Zone append commands cannot be split */
1635 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1636 return NULL;
1637
1638 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1639 if (!split)
1640 return NULL;
1641
1642 split->bi_iter.bi_size = sectors << 9;
1643
1644 if (bio_integrity(split))
1645 bio_integrity_trim(split);
1646
1647 bio_advance(bio, split->bi_iter.bi_size);
1648
1649 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1650 bio_set_flag(split, BIO_TRACE_COMPLETION);
1651
1652 return split;
1653}
1654EXPORT_SYMBOL(bio_split);
1655
1656/**
1657 * bio_trim - trim a bio
1658 * @bio: bio to trim
1659 * @offset: number of sectors to trim from the front of @bio
1660 * @size: size we want to trim @bio to, in sectors
1661 *
1662 * This function is typically used for bios that are cloned and submitted
1663 * to the underlying device in parts.
1664 */
1665void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1666{
1667 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1668 offset + size > bio_sectors(bio)))
1669 return;
1670
1671 size <<= 9;
1672 if (offset == 0 && size == bio->bi_iter.bi_size)
1673 return;
1674
1675 bio_advance(bio, offset << 9);
1676 bio->bi_iter.bi_size = size;
1677
1678 if (bio_integrity(bio))
1679 bio_integrity_trim(bio);
1680}
1681EXPORT_SYMBOL_GPL(bio_trim);
1682
1683/*
1684 * create memory pools for biovec's in a bio_set.
1685 * use the global biovec slabs created for general use.
1686 */
1687int biovec_init_pool(mempool_t *pool, int pool_entries)
1688{
1689 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1690
1691 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1692}
1693
1694/*
1695 * bioset_exit - exit a bioset initialized with bioset_init()
1696 *
1697 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1698 * kzalloc()).
1699 */
1700void bioset_exit(struct bio_set *bs)
1701{
1702 bio_alloc_cache_destroy(bs);
1703 if (bs->rescue_workqueue)
1704 destroy_workqueue(bs->rescue_workqueue);
1705 bs->rescue_workqueue = NULL;
1706
1707 mempool_exit(&bs->bio_pool);
1708 mempool_exit(&bs->bvec_pool);
1709
1710 bioset_integrity_free(bs);
1711 if (bs->bio_slab)
1712 bio_put_slab(bs);
1713 bs->bio_slab = NULL;
1714}
1715EXPORT_SYMBOL(bioset_exit);
1716
1717/**
1718 * bioset_init - Initialize a bio_set
1719 * @bs: pool to initialize
1720 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1721 * @front_pad: Number of bytes to allocate in front of the returned bio
1722 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1723 * and %BIOSET_NEED_RESCUER
1724 *
1725 * Description:
1726 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1727 * to ask for a number of bytes to be allocated in front of the bio.
1728 * Front pad allocation is useful for embedding the bio inside
1729 * another structure, to avoid allocating extra data to go with the bio.
1730 * Note that the bio must be embedded at the END of that structure always,
1731 * or things will break badly.
1732 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1733 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1734 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1735 * to dispatch queued requests when the mempool runs out of space.
1736 *
1737 */
1738int bioset_init(struct bio_set *bs,
1739 unsigned int pool_size,
1740 unsigned int front_pad,
1741 int flags)
1742{
1743 bs->front_pad = front_pad;
1744 if (flags & BIOSET_NEED_BVECS)
1745 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1746 else
1747 bs->back_pad = 0;
1748
1749 spin_lock_init(&bs->rescue_lock);
1750 bio_list_init(&bs->rescue_list);
1751 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1752
1753 bs->bio_slab = bio_find_or_create_slab(bs);
1754 if (!bs->bio_slab)
1755 return -ENOMEM;
1756
1757 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1758 goto bad;
1759
1760 if ((flags & BIOSET_NEED_BVECS) &&
1761 biovec_init_pool(&bs->bvec_pool, pool_size))
1762 goto bad;
1763
1764 if (flags & BIOSET_NEED_RESCUER) {
1765 bs->rescue_workqueue = alloc_workqueue("bioset",
1766 WQ_MEM_RECLAIM, 0);
1767 if (!bs->rescue_workqueue)
1768 goto bad;
1769 }
1770 if (flags & BIOSET_PERCPU_CACHE) {
1771 bs->cache = alloc_percpu(struct bio_alloc_cache);
1772 if (!bs->cache)
1773 goto bad;
1774 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1775 }
1776
1777 return 0;
1778bad:
1779 bioset_exit(bs);
1780 return -ENOMEM;
1781}
1782EXPORT_SYMBOL(bioset_init);
1783
1784static int __init init_bio(void)
1785{
1786 int i;
1787
1788 BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1789
1790 bio_integrity_init();
1791
1792 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1793 struct biovec_slab *bvs = bvec_slabs + i;
1794
1795 bvs->slab = kmem_cache_create(bvs->name,
1796 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1797 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1798 }
1799
1800 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1801 bio_cpu_dead);
1802
1803 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1804 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1805 panic("bio: can't allocate bios\n");
1806
1807 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1808 panic("bio: can't create integrity pool\n");
1809
1810 return 0;
1811}
1812subsys_initcall(init_bio);
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
32#include <trace/events/block.h>
33
34/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
40/*
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
43 * unsigned short
44 */
45#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
47 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48};
49#undef BV
50
51/*
52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
54 */
55struct bio_set *fs_bio_set;
56EXPORT_SYMBOL(fs_bio_set);
57
58/*
59 * Our slab pool management
60 */
61struct bio_slab {
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
65 char name[8];
66};
67static DEFINE_MUTEX(bio_slab_lock);
68static struct bio_slab *bio_slabs;
69static unsigned int bio_slab_nr, bio_slab_max;
70
71static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72{
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
75 struct bio_slab *bslab, *new_bio_slabs;
76 unsigned int new_bio_slab_max;
77 unsigned int i, entry = -1;
78
79 mutex_lock(&bio_slab_lock);
80
81 i = 0;
82 while (i < bio_slab_nr) {
83 bslab = &bio_slabs[i];
84
85 if (!bslab->slab && entry == -1)
86 entry = i;
87 else if (bslab->slab_size == sz) {
88 slab = bslab->slab;
89 bslab->slab_ref++;
90 break;
91 }
92 i++;
93 }
94
95 if (slab)
96 goto out_unlock;
97
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
99 new_bio_slab_max = bio_slab_max << 1;
100 new_bio_slabs = krealloc(bio_slabs,
101 new_bio_slab_max * sizeof(struct bio_slab),
102 GFP_KERNEL);
103 if (!new_bio_slabs)
104 goto out_unlock;
105 bio_slab_max = new_bio_slab_max;
106 bio_slabs = new_bio_slabs;
107 }
108 if (entry == -1)
109 entry = bio_slab_nr++;
110
111 bslab = &bio_slabs[entry];
112
113 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
114 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
115 SLAB_HWCACHE_ALIGN, NULL);
116 if (!slab)
117 goto out_unlock;
118
119 bslab->slab = slab;
120 bslab->slab_ref = 1;
121 bslab->slab_size = sz;
122out_unlock:
123 mutex_unlock(&bio_slab_lock);
124 return slab;
125}
126
127static void bio_put_slab(struct bio_set *bs)
128{
129 struct bio_slab *bslab = NULL;
130 unsigned int i;
131
132 mutex_lock(&bio_slab_lock);
133
134 for (i = 0; i < bio_slab_nr; i++) {
135 if (bs->bio_slab == bio_slabs[i].slab) {
136 bslab = &bio_slabs[i];
137 break;
138 }
139 }
140
141 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 goto out;
143
144 WARN_ON(!bslab->slab_ref);
145
146 if (--bslab->slab_ref)
147 goto out;
148
149 kmem_cache_destroy(bslab->slab);
150 bslab->slab = NULL;
151
152out:
153 mutex_unlock(&bio_slab_lock);
154}
155
156unsigned int bvec_nr_vecs(unsigned short idx)
157{
158 return bvec_slabs[idx].nr_vecs;
159}
160
161void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
162{
163 if (!idx)
164 return;
165 idx--;
166
167 BIO_BUG_ON(idx >= BVEC_POOL_NR);
168
169 if (idx == BVEC_POOL_MAX) {
170 mempool_free(bv, pool);
171 } else {
172 struct biovec_slab *bvs = bvec_slabs + idx;
173
174 kmem_cache_free(bvs->slab, bv);
175 }
176}
177
178struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
179 mempool_t *pool)
180{
181 struct bio_vec *bvl;
182
183 /*
184 * see comment near bvec_array define!
185 */
186 switch (nr) {
187 case 1:
188 *idx = 0;
189 break;
190 case 2 ... 4:
191 *idx = 1;
192 break;
193 case 5 ... 16:
194 *idx = 2;
195 break;
196 case 17 ... 64:
197 *idx = 3;
198 break;
199 case 65 ... 128:
200 *idx = 4;
201 break;
202 case 129 ... BIO_MAX_PAGES:
203 *idx = 5;
204 break;
205 default:
206 return NULL;
207 }
208
209 /*
210 * idx now points to the pool we want to allocate from. only the
211 * 1-vec entry pool is mempool backed.
212 */
213 if (*idx == BVEC_POOL_MAX) {
214fallback:
215 bvl = mempool_alloc(pool, gfp_mask);
216 } else {
217 struct biovec_slab *bvs = bvec_slabs + *idx;
218 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
219
220 /*
221 * Make this allocation restricted and don't dump info on
222 * allocation failures, since we'll fallback to the mempool
223 * in case of failure.
224 */
225 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
226
227 /*
228 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
229 * is set, retry with the 1-entry mempool
230 */
231 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
232 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
233 *idx = BVEC_POOL_MAX;
234 goto fallback;
235 }
236 }
237
238 (*idx)++;
239 return bvl;
240}
241
242static void __bio_free(struct bio *bio)
243{
244 bio_disassociate_task(bio);
245
246 if (bio_integrity(bio))
247 bio_integrity_free(bio);
248}
249
250static void bio_free(struct bio *bio)
251{
252 struct bio_set *bs = bio->bi_pool;
253 void *p;
254
255 __bio_free(bio);
256
257 if (bs) {
258 bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
259
260 /*
261 * If we have front padding, adjust the bio pointer before freeing
262 */
263 p = bio;
264 p -= bs->front_pad;
265
266 mempool_free(p, bs->bio_pool);
267 } else {
268 /* Bio was allocated by bio_kmalloc() */
269 kfree(bio);
270 }
271}
272
273void bio_init(struct bio *bio, struct bio_vec *table,
274 unsigned short max_vecs)
275{
276 memset(bio, 0, sizeof(*bio));
277 atomic_set(&bio->__bi_remaining, 1);
278 atomic_set(&bio->__bi_cnt, 1);
279
280 bio->bi_io_vec = table;
281 bio->bi_max_vecs = max_vecs;
282}
283EXPORT_SYMBOL(bio_init);
284
285/**
286 * bio_reset - reinitialize a bio
287 * @bio: bio to reset
288 *
289 * Description:
290 * After calling bio_reset(), @bio will be in the same state as a freshly
291 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
292 * preserved are the ones that are initialized by bio_alloc_bioset(). See
293 * comment in struct bio.
294 */
295void bio_reset(struct bio *bio)
296{
297 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
298
299 __bio_free(bio);
300
301 memset(bio, 0, BIO_RESET_BYTES);
302 bio->bi_flags = flags;
303 atomic_set(&bio->__bi_remaining, 1);
304}
305EXPORT_SYMBOL(bio_reset);
306
307static struct bio *__bio_chain_endio(struct bio *bio)
308{
309 struct bio *parent = bio->bi_private;
310
311 if (!parent->bi_error)
312 parent->bi_error = bio->bi_error;
313 bio_put(bio);
314 return parent;
315}
316
317static void bio_chain_endio(struct bio *bio)
318{
319 bio_endio(__bio_chain_endio(bio));
320}
321
322/**
323 * bio_chain - chain bio completions
324 * @bio: the target bio
325 * @parent: the @bio's parent bio
326 *
327 * The caller won't have a bi_end_io called when @bio completes - instead,
328 * @parent's bi_end_io won't be called until both @parent and @bio have
329 * completed; the chained bio will also be freed when it completes.
330 *
331 * The caller must not set bi_private or bi_end_io in @bio.
332 */
333void bio_chain(struct bio *bio, struct bio *parent)
334{
335 BUG_ON(bio->bi_private || bio->bi_end_io);
336
337 bio->bi_private = parent;
338 bio->bi_end_io = bio_chain_endio;
339 bio_inc_remaining(parent);
340}
341EXPORT_SYMBOL(bio_chain);
342
343static void bio_alloc_rescue(struct work_struct *work)
344{
345 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
346 struct bio *bio;
347
348 while (1) {
349 spin_lock(&bs->rescue_lock);
350 bio = bio_list_pop(&bs->rescue_list);
351 spin_unlock(&bs->rescue_lock);
352
353 if (!bio)
354 break;
355
356 generic_make_request(bio);
357 }
358}
359
360static void punt_bios_to_rescuer(struct bio_set *bs)
361{
362 struct bio_list punt, nopunt;
363 struct bio *bio;
364
365 /*
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.
371 *
372 * Since bio lists are singly linked, pop them all instead of trying to
373 * remove from the middle of the list:
374 */
375
376 bio_list_init(&punt);
377 bio_list_init(&nopunt);
378
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;
382
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;
387
388 spin_lock(&bs->rescue_lock);
389 bio_list_merge(&bs->rescue_list, &punt);
390 spin_unlock(&bs->rescue_lock);
391
392 queue_work(bs->rescue_workqueue, &bs->rescue_work);
393}
394
395/**
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.
400 *
401 * Description:
402 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
403 * backed by the @bs's mempool.
404 *
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.
411 *
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
415 * stack overflows.
416 *
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
420 * thread.
421 *
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.
426 *
427 * RETURNS:
428 * Pointer to new bio on success, NULL on failure.
429 */
430struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
431{
432 gfp_t saved_gfp = gfp_mask;
433 unsigned front_pad;
434 unsigned inline_vecs;
435 struct bio_vec *bvl = NULL;
436 struct bio *bio;
437 void *p;
438
439 if (!bs) {
440 if (nr_iovecs > UIO_MAXIOV)
441 return NULL;
442
443 p = kmalloc(sizeof(struct bio) +
444 nr_iovecs * sizeof(struct bio_vec),
445 gfp_mask);
446 front_pad = 0;
447 inline_vecs = nr_iovecs;
448 } else {
449 /* should not use nobvec bioset for nr_iovecs > 0 */
450 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
451 return NULL;
452 /*
453 * generic_make_request() converts recursion to iteration; this
454 * means if we're running beneath it, any bios we allocate and
455 * submit will not be submitted (and thus freed) until after we
456 * return.
457 *
458 * This exposes us to a potential deadlock if we allocate
459 * multiple bios from the same bio_set() while running
460 * underneath generic_make_request(). If we were to allocate
461 * multiple bios (say a stacking block driver that was splitting
462 * bios), we would deadlock if we exhausted the mempool's
463 * reserve.
464 *
465 * We solve this, and guarantee forward progress, with a rescuer
466 * workqueue per bio_set. If we go to allocate and there are
467 * bios on current->bio_list, we first try the allocation
468 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
469 * bios we would be blocking to the rescuer workqueue before
470 * we retry with the original gfp_flags.
471 */
472
473 if (current->bio_list &&
474 (!bio_list_empty(¤t->bio_list[0]) ||
475 !bio_list_empty(¤t->bio_list[1])))
476 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
477
478 p = mempool_alloc(bs->bio_pool, gfp_mask);
479 if (!p && gfp_mask != saved_gfp) {
480 punt_bios_to_rescuer(bs);
481 gfp_mask = saved_gfp;
482 p = mempool_alloc(bs->bio_pool, gfp_mask);
483 }
484
485 front_pad = bs->front_pad;
486 inline_vecs = BIO_INLINE_VECS;
487 }
488
489 if (unlikely(!p))
490 return NULL;
491
492 bio = p + front_pad;
493 bio_init(bio, NULL, 0);
494
495 if (nr_iovecs > inline_vecs) {
496 unsigned long idx = 0;
497
498 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
499 if (!bvl && gfp_mask != saved_gfp) {
500 punt_bios_to_rescuer(bs);
501 gfp_mask = saved_gfp;
502 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
503 }
504
505 if (unlikely(!bvl))
506 goto err_free;
507
508 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
509 } else if (nr_iovecs) {
510 bvl = bio->bi_inline_vecs;
511 }
512
513 bio->bi_pool = bs;
514 bio->bi_max_vecs = nr_iovecs;
515 bio->bi_io_vec = bvl;
516 return bio;
517
518err_free:
519 mempool_free(p, bs->bio_pool);
520 return NULL;
521}
522EXPORT_SYMBOL(bio_alloc_bioset);
523
524void zero_fill_bio(struct bio *bio)
525{
526 unsigned long flags;
527 struct bio_vec bv;
528 struct bvec_iter iter;
529
530 bio_for_each_segment(bv, bio, iter) {
531 char *data = bvec_kmap_irq(&bv, &flags);
532 memset(data, 0, bv.bv_len);
533 flush_dcache_page(bv.bv_page);
534 bvec_kunmap_irq(data, &flags);
535 }
536}
537EXPORT_SYMBOL(zero_fill_bio);
538
539/**
540 * bio_put - release a reference to a bio
541 * @bio: bio to release reference to
542 *
543 * Description:
544 * Put a reference to a &struct bio, either one you have gotten with
545 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
546 **/
547void bio_put(struct bio *bio)
548{
549 if (!bio_flagged(bio, BIO_REFFED))
550 bio_free(bio);
551 else {
552 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
553
554 /*
555 * last put frees it
556 */
557 if (atomic_dec_and_test(&bio->__bi_cnt))
558 bio_free(bio);
559 }
560}
561EXPORT_SYMBOL(bio_put);
562
563inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
564{
565 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
566 blk_recount_segments(q, bio);
567
568 return bio->bi_phys_segments;
569}
570EXPORT_SYMBOL(bio_phys_segments);
571
572/**
573 * __bio_clone_fast - clone a bio that shares the original bio's biovec
574 * @bio: destination bio
575 * @bio_src: bio to clone
576 *
577 * Clone a &bio. Caller will own the returned bio, but not
578 * the actual data it points to. Reference count of returned
579 * bio will be one.
580 *
581 * Caller must ensure that @bio_src is not freed before @bio.
582 */
583void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
584{
585 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
586
587 /*
588 * most users will be overriding ->bi_bdev with a new target,
589 * so we don't set nor calculate new physical/hw segment counts here
590 */
591 bio->bi_bdev = bio_src->bi_bdev;
592 bio_set_flag(bio, BIO_CLONED);
593 bio->bi_opf = bio_src->bi_opf;
594 bio->bi_iter = bio_src->bi_iter;
595 bio->bi_io_vec = bio_src->bi_io_vec;
596
597 bio_clone_blkcg_association(bio, bio_src);
598}
599EXPORT_SYMBOL(__bio_clone_fast);
600
601/**
602 * bio_clone_fast - clone a bio that shares the original bio's biovec
603 * @bio: bio to clone
604 * @gfp_mask: allocation priority
605 * @bs: bio_set to allocate from
606 *
607 * Like __bio_clone_fast, only also allocates the returned bio
608 */
609struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
610{
611 struct bio *b;
612
613 b = bio_alloc_bioset(gfp_mask, 0, bs);
614 if (!b)
615 return NULL;
616
617 __bio_clone_fast(b, bio);
618
619 if (bio_integrity(bio)) {
620 int ret;
621
622 ret = bio_integrity_clone(b, bio, gfp_mask);
623
624 if (ret < 0) {
625 bio_put(b);
626 return NULL;
627 }
628 }
629
630 return b;
631}
632EXPORT_SYMBOL(bio_clone_fast);
633
634/**
635 * bio_clone_bioset - clone a bio
636 * @bio_src: bio to clone
637 * @gfp_mask: allocation priority
638 * @bs: bio_set to allocate from
639 *
640 * Clone bio. Caller will own the returned bio, but not the actual data it
641 * points to. Reference count of returned bio will be one.
642 */
643struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
644 struct bio_set *bs)
645{
646 struct bvec_iter iter;
647 struct bio_vec bv;
648 struct bio *bio;
649
650 /*
651 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
652 * bio_src->bi_io_vec to bio->bi_io_vec.
653 *
654 * We can't do that anymore, because:
655 *
656 * - The point of cloning the biovec is to produce a bio with a biovec
657 * the caller can modify: bi_idx and bi_bvec_done should be 0.
658 *
659 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
660 * we tried to clone the whole thing bio_alloc_bioset() would fail.
661 * But the clone should succeed as long as the number of biovecs we
662 * actually need to allocate is fewer than BIO_MAX_PAGES.
663 *
664 * - Lastly, bi_vcnt should not be looked at or relied upon by code
665 * that does not own the bio - reason being drivers don't use it for
666 * iterating over the biovec anymore, so expecting it to be kept up
667 * to date (i.e. for clones that share the parent biovec) is just
668 * asking for trouble and would force extra work on
669 * __bio_clone_fast() anyways.
670 */
671
672 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
673 if (!bio)
674 return NULL;
675 bio->bi_bdev = bio_src->bi_bdev;
676 bio->bi_opf = bio_src->bi_opf;
677 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
678 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
679
680 switch (bio_op(bio)) {
681 case REQ_OP_DISCARD:
682 case REQ_OP_SECURE_ERASE:
683 case REQ_OP_WRITE_ZEROES:
684 break;
685 case REQ_OP_WRITE_SAME:
686 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
687 break;
688 default:
689 bio_for_each_segment(bv, bio_src, iter)
690 bio->bi_io_vec[bio->bi_vcnt++] = bv;
691 break;
692 }
693
694 if (bio_integrity(bio_src)) {
695 int ret;
696
697 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
698 if (ret < 0) {
699 bio_put(bio);
700 return NULL;
701 }
702 }
703
704 bio_clone_blkcg_association(bio, bio_src);
705
706 return bio;
707}
708EXPORT_SYMBOL(bio_clone_bioset);
709
710/**
711 * bio_add_pc_page - attempt to add page to bio
712 * @q: the target queue
713 * @bio: destination bio
714 * @page: page to add
715 * @len: vec entry length
716 * @offset: vec entry offset
717 *
718 * Attempt to add a page to the bio_vec maplist. This can fail for a
719 * number of reasons, such as the bio being full or target block device
720 * limitations. The target block device must allow bio's up to PAGE_SIZE,
721 * so it is always possible to add a single page to an empty bio.
722 *
723 * This should only be used by REQ_PC bios.
724 */
725int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
726 *page, unsigned int len, unsigned int offset)
727{
728 int retried_segments = 0;
729 struct bio_vec *bvec;
730
731 /*
732 * cloned bio must not modify vec list
733 */
734 if (unlikely(bio_flagged(bio, BIO_CLONED)))
735 return 0;
736
737 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
738 return 0;
739
740 /*
741 * For filesystems with a blocksize smaller than the pagesize
742 * we will often be called with the same page as last time and
743 * a consecutive offset. Optimize this special case.
744 */
745 if (bio->bi_vcnt > 0) {
746 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
747
748 if (page == prev->bv_page &&
749 offset == prev->bv_offset + prev->bv_len) {
750 prev->bv_len += len;
751 bio->bi_iter.bi_size += len;
752 goto done;
753 }
754
755 /*
756 * If the queue doesn't support SG gaps and adding this
757 * offset would create a gap, disallow it.
758 */
759 if (bvec_gap_to_prev(q, prev, offset))
760 return 0;
761 }
762
763 if (bio->bi_vcnt >= bio->bi_max_vecs)
764 return 0;
765
766 /*
767 * setup the new entry, we might clear it again later if we
768 * cannot add the page
769 */
770 bvec = &bio->bi_io_vec[bio->bi_vcnt];
771 bvec->bv_page = page;
772 bvec->bv_len = len;
773 bvec->bv_offset = offset;
774 bio->bi_vcnt++;
775 bio->bi_phys_segments++;
776 bio->bi_iter.bi_size += len;
777
778 /*
779 * Perform a recount if the number of segments is greater
780 * than queue_max_segments(q).
781 */
782
783 while (bio->bi_phys_segments > queue_max_segments(q)) {
784
785 if (retried_segments)
786 goto failed;
787
788 retried_segments = 1;
789 blk_recount_segments(q, bio);
790 }
791
792 /* If we may be able to merge these biovecs, force a recount */
793 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
794 bio_clear_flag(bio, BIO_SEG_VALID);
795
796 done:
797 return len;
798
799 failed:
800 bvec->bv_page = NULL;
801 bvec->bv_len = 0;
802 bvec->bv_offset = 0;
803 bio->bi_vcnt--;
804 bio->bi_iter.bi_size -= len;
805 blk_recount_segments(q, bio);
806 return 0;
807}
808EXPORT_SYMBOL(bio_add_pc_page);
809
810/**
811 * bio_add_page - attempt to add page to bio
812 * @bio: destination bio
813 * @page: page to add
814 * @len: vec entry length
815 * @offset: vec entry offset
816 *
817 * Attempt to add a page to the bio_vec maplist. This will only fail
818 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
819 */
820int bio_add_page(struct bio *bio, struct page *page,
821 unsigned int len, unsigned int offset)
822{
823 struct bio_vec *bv;
824
825 /*
826 * cloned bio must not modify vec list
827 */
828 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
829 return 0;
830
831 /*
832 * For filesystems with a blocksize smaller than the pagesize
833 * we will often be called with the same page as last time and
834 * a consecutive offset. Optimize this special case.
835 */
836 if (bio->bi_vcnt > 0) {
837 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
838
839 if (page == bv->bv_page &&
840 offset == bv->bv_offset + bv->bv_len) {
841 bv->bv_len += len;
842 goto done;
843 }
844 }
845
846 if (bio->bi_vcnt >= bio->bi_max_vecs)
847 return 0;
848
849 bv = &bio->bi_io_vec[bio->bi_vcnt];
850 bv->bv_page = page;
851 bv->bv_len = len;
852 bv->bv_offset = offset;
853
854 bio->bi_vcnt++;
855done:
856 bio->bi_iter.bi_size += len;
857 return len;
858}
859EXPORT_SYMBOL(bio_add_page);
860
861/**
862 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
863 * @bio: bio to add pages to
864 * @iter: iov iterator describing the region to be mapped
865 *
866 * Pins as many pages from *iter and appends them to @bio's bvec array. The
867 * pages will have to be released using put_page() when done.
868 */
869int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
870{
871 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
872 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
873 struct page **pages = (struct page **)bv;
874 size_t offset, diff;
875 ssize_t size;
876
877 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
878 if (unlikely(size <= 0))
879 return size ? size : -EFAULT;
880 nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE;
881
882 /*
883 * Deep magic below: We need to walk the pinned pages backwards
884 * because we are abusing the space allocated for the bio_vecs
885 * for the page array. Because the bio_vecs are larger than the
886 * page pointers by definition this will always work. But it also
887 * means we can't use bio_add_page, so any changes to it's semantics
888 * need to be reflected here as well.
889 */
890 bio->bi_iter.bi_size += size;
891 bio->bi_vcnt += nr_pages;
892
893 diff = (nr_pages * PAGE_SIZE - offset) - size;
894 while (nr_pages--) {
895 bv[nr_pages].bv_page = pages[nr_pages];
896 bv[nr_pages].bv_len = PAGE_SIZE;
897 bv[nr_pages].bv_offset = 0;
898 }
899
900 bv[0].bv_offset += offset;
901 bv[0].bv_len -= offset;
902 if (diff)
903 bv[bio->bi_vcnt - 1].bv_len -= diff;
904
905 iov_iter_advance(iter, size);
906 return 0;
907}
908EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
909
910struct submit_bio_ret {
911 struct completion event;
912 int error;
913};
914
915static void submit_bio_wait_endio(struct bio *bio)
916{
917 struct submit_bio_ret *ret = bio->bi_private;
918
919 ret->error = bio->bi_error;
920 complete(&ret->event);
921}
922
923/**
924 * submit_bio_wait - submit a bio, and wait until it completes
925 * @bio: The &struct bio which describes the I/O
926 *
927 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
928 * bio_endio() on failure.
929 */
930int submit_bio_wait(struct bio *bio)
931{
932 struct submit_bio_ret ret;
933
934 init_completion(&ret.event);
935 bio->bi_private = &ret;
936 bio->bi_end_io = submit_bio_wait_endio;
937 bio->bi_opf |= REQ_SYNC;
938 submit_bio(bio);
939 wait_for_completion_io(&ret.event);
940
941 return ret.error;
942}
943EXPORT_SYMBOL(submit_bio_wait);
944
945/**
946 * bio_advance - increment/complete a bio by some number of bytes
947 * @bio: bio to advance
948 * @bytes: number of bytes to complete
949 *
950 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
951 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
952 * be updated on the last bvec as well.
953 *
954 * @bio will then represent the remaining, uncompleted portion of the io.
955 */
956void bio_advance(struct bio *bio, unsigned bytes)
957{
958 if (bio_integrity(bio))
959 bio_integrity_advance(bio, bytes);
960
961 bio_advance_iter(bio, &bio->bi_iter, bytes);
962}
963EXPORT_SYMBOL(bio_advance);
964
965/**
966 * bio_alloc_pages - allocates a single page for each bvec in a bio
967 * @bio: bio to allocate pages for
968 * @gfp_mask: flags for allocation
969 *
970 * Allocates pages up to @bio->bi_vcnt.
971 *
972 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
973 * freed.
974 */
975int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
976{
977 int i;
978 struct bio_vec *bv;
979
980 bio_for_each_segment_all(bv, bio, i) {
981 bv->bv_page = alloc_page(gfp_mask);
982 if (!bv->bv_page) {
983 while (--bv >= bio->bi_io_vec)
984 __free_page(bv->bv_page);
985 return -ENOMEM;
986 }
987 }
988
989 return 0;
990}
991EXPORT_SYMBOL(bio_alloc_pages);
992
993/**
994 * bio_copy_data - copy contents of data buffers from one chain of bios to
995 * another
996 * @src: source bio list
997 * @dst: destination bio list
998 *
999 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
1000 * @src and @dst as linked lists of bios.
1001 *
1002 * Stops when it reaches the end of either @src or @dst - that is, copies
1003 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1004 */
1005void bio_copy_data(struct bio *dst, struct bio *src)
1006{
1007 struct bvec_iter src_iter, dst_iter;
1008 struct bio_vec src_bv, dst_bv;
1009 void *src_p, *dst_p;
1010 unsigned bytes;
1011
1012 src_iter = src->bi_iter;
1013 dst_iter = dst->bi_iter;
1014
1015 while (1) {
1016 if (!src_iter.bi_size) {
1017 src = src->bi_next;
1018 if (!src)
1019 break;
1020
1021 src_iter = src->bi_iter;
1022 }
1023
1024 if (!dst_iter.bi_size) {
1025 dst = dst->bi_next;
1026 if (!dst)
1027 break;
1028
1029 dst_iter = dst->bi_iter;
1030 }
1031
1032 src_bv = bio_iter_iovec(src, src_iter);
1033 dst_bv = bio_iter_iovec(dst, dst_iter);
1034
1035 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1036
1037 src_p = kmap_atomic(src_bv.bv_page);
1038 dst_p = kmap_atomic(dst_bv.bv_page);
1039
1040 memcpy(dst_p + dst_bv.bv_offset,
1041 src_p + src_bv.bv_offset,
1042 bytes);
1043
1044 kunmap_atomic(dst_p);
1045 kunmap_atomic(src_p);
1046
1047 bio_advance_iter(src, &src_iter, bytes);
1048 bio_advance_iter(dst, &dst_iter, bytes);
1049 }
1050}
1051EXPORT_SYMBOL(bio_copy_data);
1052
1053struct bio_map_data {
1054 int is_our_pages;
1055 struct iov_iter iter;
1056 struct iovec iov[];
1057};
1058
1059static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
1060 gfp_t gfp_mask)
1061{
1062 if (iov_count > UIO_MAXIOV)
1063 return NULL;
1064
1065 return kmalloc(sizeof(struct bio_map_data) +
1066 sizeof(struct iovec) * iov_count, gfp_mask);
1067}
1068
1069/**
1070 * bio_copy_from_iter - copy all pages from iov_iter to bio
1071 * @bio: The &struct bio which describes the I/O as destination
1072 * @iter: iov_iter as source
1073 *
1074 * Copy all pages from iov_iter to bio.
1075 * Returns 0 on success, or error on failure.
1076 */
1077static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
1078{
1079 int i;
1080 struct bio_vec *bvec;
1081
1082 bio_for_each_segment_all(bvec, bio, i) {
1083 ssize_t ret;
1084
1085 ret = copy_page_from_iter(bvec->bv_page,
1086 bvec->bv_offset,
1087 bvec->bv_len,
1088 &iter);
1089
1090 if (!iov_iter_count(&iter))
1091 break;
1092
1093 if (ret < bvec->bv_len)
1094 return -EFAULT;
1095 }
1096
1097 return 0;
1098}
1099
1100/**
1101 * bio_copy_to_iter - copy all pages from bio to iov_iter
1102 * @bio: The &struct bio which describes the I/O as source
1103 * @iter: iov_iter as destination
1104 *
1105 * Copy all pages from bio to iov_iter.
1106 * Returns 0 on success, or error on failure.
1107 */
1108static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1109{
1110 int i;
1111 struct bio_vec *bvec;
1112
1113 bio_for_each_segment_all(bvec, bio, i) {
1114 ssize_t ret;
1115
1116 ret = copy_page_to_iter(bvec->bv_page,
1117 bvec->bv_offset,
1118 bvec->bv_len,
1119 &iter);
1120
1121 if (!iov_iter_count(&iter))
1122 break;
1123
1124 if (ret < bvec->bv_len)
1125 return -EFAULT;
1126 }
1127
1128 return 0;
1129}
1130
1131void bio_free_pages(struct bio *bio)
1132{
1133 struct bio_vec *bvec;
1134 int i;
1135
1136 bio_for_each_segment_all(bvec, bio, i)
1137 __free_page(bvec->bv_page);
1138}
1139EXPORT_SYMBOL(bio_free_pages);
1140
1141/**
1142 * bio_uncopy_user - finish previously mapped bio
1143 * @bio: bio being terminated
1144 *
1145 * Free pages allocated from bio_copy_user_iov() and write back data
1146 * to user space in case of a read.
1147 */
1148int bio_uncopy_user(struct bio *bio)
1149{
1150 struct bio_map_data *bmd = bio->bi_private;
1151 int ret = 0;
1152
1153 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1154 /*
1155 * if we're in a workqueue, the request is orphaned, so
1156 * don't copy into a random user address space, just free
1157 * and return -EINTR so user space doesn't expect any data.
1158 */
1159 if (!current->mm)
1160 ret = -EINTR;
1161 else if (bio_data_dir(bio) == READ)
1162 ret = bio_copy_to_iter(bio, bmd->iter);
1163 if (bmd->is_our_pages)
1164 bio_free_pages(bio);
1165 }
1166 kfree(bmd);
1167 bio_put(bio);
1168 return ret;
1169}
1170
1171/**
1172 * bio_copy_user_iov - copy user data to bio
1173 * @q: destination block queue
1174 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1175 * @iter: iovec iterator
1176 * @gfp_mask: memory allocation flags
1177 *
1178 * Prepares and returns a bio for indirect user io, bouncing data
1179 * to/from kernel pages as necessary. Must be paired with
1180 * call bio_uncopy_user() on io completion.
1181 */
1182struct bio *bio_copy_user_iov(struct request_queue *q,
1183 struct rq_map_data *map_data,
1184 const struct iov_iter *iter,
1185 gfp_t gfp_mask)
1186{
1187 struct bio_map_data *bmd;
1188 struct page *page;
1189 struct bio *bio;
1190 int i, ret;
1191 int nr_pages = 0;
1192 unsigned int len = iter->count;
1193 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1194
1195 for (i = 0; i < iter->nr_segs; i++) {
1196 unsigned long uaddr;
1197 unsigned long end;
1198 unsigned long start;
1199
1200 uaddr = (unsigned long) iter->iov[i].iov_base;
1201 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1202 >> PAGE_SHIFT;
1203 start = uaddr >> PAGE_SHIFT;
1204
1205 /*
1206 * Overflow, abort
1207 */
1208 if (end < start)
1209 return ERR_PTR(-EINVAL);
1210
1211 nr_pages += end - start;
1212 }
1213
1214 if (offset)
1215 nr_pages++;
1216
1217 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
1218 if (!bmd)
1219 return ERR_PTR(-ENOMEM);
1220
1221 /*
1222 * We need to do a deep copy of the iov_iter including the iovecs.
1223 * The caller provided iov might point to an on-stack or otherwise
1224 * shortlived one.
1225 */
1226 bmd->is_our_pages = map_data ? 0 : 1;
1227 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1228 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1229 iter->nr_segs, iter->count);
1230
1231 ret = -ENOMEM;
1232 bio = bio_kmalloc(gfp_mask, nr_pages);
1233 if (!bio)
1234 goto out_bmd;
1235
1236 if (iter->type & WRITE)
1237 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1238
1239 ret = 0;
1240
1241 if (map_data) {
1242 nr_pages = 1 << map_data->page_order;
1243 i = map_data->offset / PAGE_SIZE;
1244 }
1245 while (len) {
1246 unsigned int bytes = PAGE_SIZE;
1247
1248 bytes -= offset;
1249
1250 if (bytes > len)
1251 bytes = len;
1252
1253 if (map_data) {
1254 if (i == map_data->nr_entries * nr_pages) {
1255 ret = -ENOMEM;
1256 break;
1257 }
1258
1259 page = map_data->pages[i / nr_pages];
1260 page += (i % nr_pages);
1261
1262 i++;
1263 } else {
1264 page = alloc_page(q->bounce_gfp | gfp_mask);
1265 if (!page) {
1266 ret = -ENOMEM;
1267 break;
1268 }
1269 }
1270
1271 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1272 break;
1273
1274 len -= bytes;
1275 offset = 0;
1276 }
1277
1278 if (ret)
1279 goto cleanup;
1280
1281 /*
1282 * success
1283 */
1284 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1285 (map_data && map_data->from_user)) {
1286 ret = bio_copy_from_iter(bio, *iter);
1287 if (ret)
1288 goto cleanup;
1289 }
1290
1291 bio->bi_private = bmd;
1292 return bio;
1293cleanup:
1294 if (!map_data)
1295 bio_free_pages(bio);
1296 bio_put(bio);
1297out_bmd:
1298 kfree(bmd);
1299 return ERR_PTR(ret);
1300}
1301
1302/**
1303 * bio_map_user_iov - map user iovec into bio
1304 * @q: the struct request_queue for the bio
1305 * @iter: iovec iterator
1306 * @gfp_mask: memory allocation flags
1307 *
1308 * Map the user space address into a bio suitable for io to a block
1309 * device. Returns an error pointer in case of error.
1310 */
1311struct bio *bio_map_user_iov(struct request_queue *q,
1312 const struct iov_iter *iter,
1313 gfp_t gfp_mask)
1314{
1315 int j;
1316 int nr_pages = 0;
1317 struct page **pages;
1318 struct bio *bio;
1319 int cur_page = 0;
1320 int ret, offset;
1321 struct iov_iter i;
1322 struct iovec iov;
1323
1324 iov_for_each(iov, i, *iter) {
1325 unsigned long uaddr = (unsigned long) iov.iov_base;
1326 unsigned long len = iov.iov_len;
1327 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1328 unsigned long start = uaddr >> PAGE_SHIFT;
1329
1330 /*
1331 * Overflow, abort
1332 */
1333 if (end < start)
1334 return ERR_PTR(-EINVAL);
1335
1336 nr_pages += end - start;
1337 /*
1338 * buffer must be aligned to at least logical block size for now
1339 */
1340 if (uaddr & queue_dma_alignment(q))
1341 return ERR_PTR(-EINVAL);
1342 }
1343
1344 if (!nr_pages)
1345 return ERR_PTR(-EINVAL);
1346
1347 bio = bio_kmalloc(gfp_mask, nr_pages);
1348 if (!bio)
1349 return ERR_PTR(-ENOMEM);
1350
1351 ret = -ENOMEM;
1352 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1353 if (!pages)
1354 goto out;
1355
1356 iov_for_each(iov, i, *iter) {
1357 unsigned long uaddr = (unsigned long) iov.iov_base;
1358 unsigned long len = iov.iov_len;
1359 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1360 unsigned long start = uaddr >> PAGE_SHIFT;
1361 const int local_nr_pages = end - start;
1362 const int page_limit = cur_page + local_nr_pages;
1363
1364 ret = get_user_pages_fast(uaddr, local_nr_pages,
1365 (iter->type & WRITE) != WRITE,
1366 &pages[cur_page]);
1367 if (ret < local_nr_pages) {
1368 ret = -EFAULT;
1369 goto out_unmap;
1370 }
1371
1372 offset = offset_in_page(uaddr);
1373 for (j = cur_page; j < page_limit; j++) {
1374 unsigned int bytes = PAGE_SIZE - offset;
1375
1376 if (len <= 0)
1377 break;
1378
1379 if (bytes > len)
1380 bytes = len;
1381
1382 /*
1383 * sorry...
1384 */
1385 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1386 bytes)
1387 break;
1388
1389 len -= bytes;
1390 offset = 0;
1391 }
1392
1393 cur_page = j;
1394 /*
1395 * release the pages we didn't map into the bio, if any
1396 */
1397 while (j < page_limit)
1398 put_page(pages[j++]);
1399 }
1400
1401 kfree(pages);
1402
1403 /*
1404 * set data direction, and check if mapped pages need bouncing
1405 */
1406 if (iter->type & WRITE)
1407 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1408
1409 bio_set_flag(bio, BIO_USER_MAPPED);
1410
1411 /*
1412 * subtle -- if __bio_map_user() ended up bouncing a bio,
1413 * it would normally disappear when its bi_end_io is run.
1414 * however, we need it for the unmap, so grab an extra
1415 * reference to it
1416 */
1417 bio_get(bio);
1418 return bio;
1419
1420 out_unmap:
1421 for (j = 0; j < nr_pages; j++) {
1422 if (!pages[j])
1423 break;
1424 put_page(pages[j]);
1425 }
1426 out:
1427 kfree(pages);
1428 bio_put(bio);
1429 return ERR_PTR(ret);
1430}
1431
1432static void __bio_unmap_user(struct bio *bio)
1433{
1434 struct bio_vec *bvec;
1435 int i;
1436
1437 /*
1438 * make sure we dirty pages we wrote to
1439 */
1440 bio_for_each_segment_all(bvec, bio, i) {
1441 if (bio_data_dir(bio) == READ)
1442 set_page_dirty_lock(bvec->bv_page);
1443
1444 put_page(bvec->bv_page);
1445 }
1446
1447 bio_put(bio);
1448}
1449
1450/**
1451 * bio_unmap_user - unmap a bio
1452 * @bio: the bio being unmapped
1453 *
1454 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1455 * a process context.
1456 *
1457 * bio_unmap_user() may sleep.
1458 */
1459void bio_unmap_user(struct bio *bio)
1460{
1461 __bio_unmap_user(bio);
1462 bio_put(bio);
1463}
1464
1465static void bio_map_kern_endio(struct bio *bio)
1466{
1467 bio_put(bio);
1468}
1469
1470/**
1471 * bio_map_kern - map kernel address into bio
1472 * @q: the struct request_queue for the bio
1473 * @data: pointer to buffer to map
1474 * @len: length in bytes
1475 * @gfp_mask: allocation flags for bio allocation
1476 *
1477 * Map the kernel address into a bio suitable for io to a block
1478 * device. Returns an error pointer in case of error.
1479 */
1480struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1481 gfp_t gfp_mask)
1482{
1483 unsigned long kaddr = (unsigned long)data;
1484 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1485 unsigned long start = kaddr >> PAGE_SHIFT;
1486 const int nr_pages = end - start;
1487 int offset, i;
1488 struct bio *bio;
1489
1490 bio = bio_kmalloc(gfp_mask, nr_pages);
1491 if (!bio)
1492 return ERR_PTR(-ENOMEM);
1493
1494 offset = offset_in_page(kaddr);
1495 for (i = 0; i < nr_pages; i++) {
1496 unsigned int bytes = PAGE_SIZE - offset;
1497
1498 if (len <= 0)
1499 break;
1500
1501 if (bytes > len)
1502 bytes = len;
1503
1504 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1505 offset) < bytes) {
1506 /* we don't support partial mappings */
1507 bio_put(bio);
1508 return ERR_PTR(-EINVAL);
1509 }
1510
1511 data += bytes;
1512 len -= bytes;
1513 offset = 0;
1514 }
1515
1516 bio->bi_end_io = bio_map_kern_endio;
1517 return bio;
1518}
1519EXPORT_SYMBOL(bio_map_kern);
1520
1521static void bio_copy_kern_endio(struct bio *bio)
1522{
1523 bio_free_pages(bio);
1524 bio_put(bio);
1525}
1526
1527static void bio_copy_kern_endio_read(struct bio *bio)
1528{
1529 char *p = bio->bi_private;
1530 struct bio_vec *bvec;
1531 int i;
1532
1533 bio_for_each_segment_all(bvec, bio, i) {
1534 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1535 p += bvec->bv_len;
1536 }
1537
1538 bio_copy_kern_endio(bio);
1539}
1540
1541/**
1542 * bio_copy_kern - copy kernel address into bio
1543 * @q: the struct request_queue for the bio
1544 * @data: pointer to buffer to copy
1545 * @len: length in bytes
1546 * @gfp_mask: allocation flags for bio and page allocation
1547 * @reading: data direction is READ
1548 *
1549 * copy the kernel address into a bio suitable for io to a block
1550 * device. Returns an error pointer in case of error.
1551 */
1552struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1553 gfp_t gfp_mask, int reading)
1554{
1555 unsigned long kaddr = (unsigned long)data;
1556 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1557 unsigned long start = kaddr >> PAGE_SHIFT;
1558 struct bio *bio;
1559 void *p = data;
1560 int nr_pages = 0;
1561
1562 /*
1563 * Overflow, abort
1564 */
1565 if (end < start)
1566 return ERR_PTR(-EINVAL);
1567
1568 nr_pages = end - start;
1569 bio = bio_kmalloc(gfp_mask, nr_pages);
1570 if (!bio)
1571 return ERR_PTR(-ENOMEM);
1572
1573 while (len) {
1574 struct page *page;
1575 unsigned int bytes = PAGE_SIZE;
1576
1577 if (bytes > len)
1578 bytes = len;
1579
1580 page = alloc_page(q->bounce_gfp | gfp_mask);
1581 if (!page)
1582 goto cleanup;
1583
1584 if (!reading)
1585 memcpy(page_address(page), p, bytes);
1586
1587 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1588 break;
1589
1590 len -= bytes;
1591 p += bytes;
1592 }
1593
1594 if (reading) {
1595 bio->bi_end_io = bio_copy_kern_endio_read;
1596 bio->bi_private = data;
1597 } else {
1598 bio->bi_end_io = bio_copy_kern_endio;
1599 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1600 }
1601
1602 return bio;
1603
1604cleanup:
1605 bio_free_pages(bio);
1606 bio_put(bio);
1607 return ERR_PTR(-ENOMEM);
1608}
1609
1610/*
1611 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1612 * for performing direct-IO in BIOs.
1613 *
1614 * The problem is that we cannot run set_page_dirty() from interrupt context
1615 * because the required locks are not interrupt-safe. So what we can do is to
1616 * mark the pages dirty _before_ performing IO. And in interrupt context,
1617 * check that the pages are still dirty. If so, fine. If not, redirty them
1618 * in process context.
1619 *
1620 * We special-case compound pages here: normally this means reads into hugetlb
1621 * pages. The logic in here doesn't really work right for compound pages
1622 * because the VM does not uniformly chase down the head page in all cases.
1623 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1624 * handle them at all. So we skip compound pages here at an early stage.
1625 *
1626 * Note that this code is very hard to test under normal circumstances because
1627 * direct-io pins the pages with get_user_pages(). This makes
1628 * is_page_cache_freeable return false, and the VM will not clean the pages.
1629 * But other code (eg, flusher threads) could clean the pages if they are mapped
1630 * pagecache.
1631 *
1632 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1633 * deferred bio dirtying paths.
1634 */
1635
1636/*
1637 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1638 */
1639void bio_set_pages_dirty(struct bio *bio)
1640{
1641 struct bio_vec *bvec;
1642 int i;
1643
1644 bio_for_each_segment_all(bvec, bio, i) {
1645 struct page *page = bvec->bv_page;
1646
1647 if (page && !PageCompound(page))
1648 set_page_dirty_lock(page);
1649 }
1650}
1651
1652static void bio_release_pages(struct bio *bio)
1653{
1654 struct bio_vec *bvec;
1655 int i;
1656
1657 bio_for_each_segment_all(bvec, bio, i) {
1658 struct page *page = bvec->bv_page;
1659
1660 if (page)
1661 put_page(page);
1662 }
1663}
1664
1665/*
1666 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1667 * If they are, then fine. If, however, some pages are clean then they must
1668 * have been written out during the direct-IO read. So we take another ref on
1669 * the BIO and the offending pages and re-dirty the pages in process context.
1670 *
1671 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1672 * here on. It will run one put_page() against each page and will run one
1673 * bio_put() against the BIO.
1674 */
1675
1676static void bio_dirty_fn(struct work_struct *work);
1677
1678static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1679static DEFINE_SPINLOCK(bio_dirty_lock);
1680static struct bio *bio_dirty_list;
1681
1682/*
1683 * This runs in process context
1684 */
1685static void bio_dirty_fn(struct work_struct *work)
1686{
1687 unsigned long flags;
1688 struct bio *bio;
1689
1690 spin_lock_irqsave(&bio_dirty_lock, flags);
1691 bio = bio_dirty_list;
1692 bio_dirty_list = NULL;
1693 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1694
1695 while (bio) {
1696 struct bio *next = bio->bi_private;
1697
1698 bio_set_pages_dirty(bio);
1699 bio_release_pages(bio);
1700 bio_put(bio);
1701 bio = next;
1702 }
1703}
1704
1705void bio_check_pages_dirty(struct bio *bio)
1706{
1707 struct bio_vec *bvec;
1708 int nr_clean_pages = 0;
1709 int i;
1710
1711 bio_for_each_segment_all(bvec, bio, i) {
1712 struct page *page = bvec->bv_page;
1713
1714 if (PageDirty(page) || PageCompound(page)) {
1715 put_page(page);
1716 bvec->bv_page = NULL;
1717 } else {
1718 nr_clean_pages++;
1719 }
1720 }
1721
1722 if (nr_clean_pages) {
1723 unsigned long flags;
1724
1725 spin_lock_irqsave(&bio_dirty_lock, flags);
1726 bio->bi_private = bio_dirty_list;
1727 bio_dirty_list = bio;
1728 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1729 schedule_work(&bio_dirty_work);
1730 } else {
1731 bio_put(bio);
1732 }
1733}
1734
1735void generic_start_io_acct(int rw, unsigned long sectors,
1736 struct hd_struct *part)
1737{
1738 int cpu = part_stat_lock();
1739
1740 part_round_stats(cpu, part);
1741 part_stat_inc(cpu, part, ios[rw]);
1742 part_stat_add(cpu, part, sectors[rw], sectors);
1743 part_inc_in_flight(part, rw);
1744
1745 part_stat_unlock();
1746}
1747EXPORT_SYMBOL(generic_start_io_acct);
1748
1749void generic_end_io_acct(int rw, struct hd_struct *part,
1750 unsigned long start_time)
1751{
1752 unsigned long duration = jiffies - start_time;
1753 int cpu = part_stat_lock();
1754
1755 part_stat_add(cpu, part, ticks[rw], duration);
1756 part_round_stats(cpu, part);
1757 part_dec_in_flight(part, rw);
1758
1759 part_stat_unlock();
1760}
1761EXPORT_SYMBOL(generic_end_io_acct);
1762
1763#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1764void bio_flush_dcache_pages(struct bio *bi)
1765{
1766 struct bio_vec bvec;
1767 struct bvec_iter iter;
1768
1769 bio_for_each_segment(bvec, bi, iter)
1770 flush_dcache_page(bvec.bv_page);
1771}
1772EXPORT_SYMBOL(bio_flush_dcache_pages);
1773#endif
1774
1775static inline bool bio_remaining_done(struct bio *bio)
1776{
1777 /*
1778 * If we're not chaining, then ->__bi_remaining is always 1 and
1779 * we always end io on the first invocation.
1780 */
1781 if (!bio_flagged(bio, BIO_CHAIN))
1782 return true;
1783
1784 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1785
1786 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1787 bio_clear_flag(bio, BIO_CHAIN);
1788 return true;
1789 }
1790
1791 return false;
1792}
1793
1794/**
1795 * bio_endio - end I/O on a bio
1796 * @bio: bio
1797 *
1798 * Description:
1799 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1800 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1801 * bio unless they own it and thus know that it has an end_io function.
1802 **/
1803void bio_endio(struct bio *bio)
1804{
1805again:
1806 if (!bio_remaining_done(bio))
1807 return;
1808
1809 /*
1810 * Need to have a real endio function for chained bios, otherwise
1811 * various corner cases will break (like stacking block devices that
1812 * save/restore bi_end_io) - however, we want to avoid unbounded
1813 * recursion and blowing the stack. Tail call optimization would
1814 * handle this, but compiling with frame pointers also disables
1815 * gcc's sibling call optimization.
1816 */
1817 if (bio->bi_end_io == bio_chain_endio) {
1818 bio = __bio_chain_endio(bio);
1819 goto again;
1820 }
1821
1822 if (bio->bi_end_io)
1823 bio->bi_end_io(bio);
1824}
1825EXPORT_SYMBOL(bio_endio);
1826
1827/**
1828 * bio_split - split a bio
1829 * @bio: bio to split
1830 * @sectors: number of sectors to split from the front of @bio
1831 * @gfp: gfp mask
1832 * @bs: bio set to allocate from
1833 *
1834 * Allocates and returns a new bio which represents @sectors from the start of
1835 * @bio, and updates @bio to represent the remaining sectors.
1836 *
1837 * Unless this is a discard request the newly allocated bio will point
1838 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1839 * @bio is not freed before the split.
1840 */
1841struct bio *bio_split(struct bio *bio, int sectors,
1842 gfp_t gfp, struct bio_set *bs)
1843{
1844 struct bio *split = NULL;
1845
1846 BUG_ON(sectors <= 0);
1847 BUG_ON(sectors >= bio_sectors(bio));
1848
1849 split = bio_clone_fast(bio, gfp, bs);
1850 if (!split)
1851 return NULL;
1852
1853 split->bi_iter.bi_size = sectors << 9;
1854
1855 if (bio_integrity(split))
1856 bio_integrity_trim(split, 0, sectors);
1857
1858 bio_advance(bio, split->bi_iter.bi_size);
1859
1860 return split;
1861}
1862EXPORT_SYMBOL(bio_split);
1863
1864/**
1865 * bio_trim - trim a bio
1866 * @bio: bio to trim
1867 * @offset: number of sectors to trim from the front of @bio
1868 * @size: size we want to trim @bio to, in sectors
1869 */
1870void bio_trim(struct bio *bio, int offset, int size)
1871{
1872 /* 'bio' is a cloned bio which we need to trim to match
1873 * the given offset and size.
1874 */
1875
1876 size <<= 9;
1877 if (offset == 0 && size == bio->bi_iter.bi_size)
1878 return;
1879
1880 bio_clear_flag(bio, BIO_SEG_VALID);
1881
1882 bio_advance(bio, offset << 9);
1883
1884 bio->bi_iter.bi_size = size;
1885}
1886EXPORT_SYMBOL_GPL(bio_trim);
1887
1888/*
1889 * create memory pools for biovec's in a bio_set.
1890 * use the global biovec slabs created for general use.
1891 */
1892mempool_t *biovec_create_pool(int pool_entries)
1893{
1894 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1895
1896 return mempool_create_slab_pool(pool_entries, bp->slab);
1897}
1898
1899void bioset_free(struct bio_set *bs)
1900{
1901 if (bs->rescue_workqueue)
1902 destroy_workqueue(bs->rescue_workqueue);
1903
1904 if (bs->bio_pool)
1905 mempool_destroy(bs->bio_pool);
1906
1907 if (bs->bvec_pool)
1908 mempool_destroy(bs->bvec_pool);
1909
1910 bioset_integrity_free(bs);
1911 bio_put_slab(bs);
1912
1913 kfree(bs);
1914}
1915EXPORT_SYMBOL(bioset_free);
1916
1917static struct bio_set *__bioset_create(unsigned int pool_size,
1918 unsigned int front_pad,
1919 bool create_bvec_pool)
1920{
1921 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1922 struct bio_set *bs;
1923
1924 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1925 if (!bs)
1926 return NULL;
1927
1928 bs->front_pad = front_pad;
1929
1930 spin_lock_init(&bs->rescue_lock);
1931 bio_list_init(&bs->rescue_list);
1932 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1933
1934 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1935 if (!bs->bio_slab) {
1936 kfree(bs);
1937 return NULL;
1938 }
1939
1940 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1941 if (!bs->bio_pool)
1942 goto bad;
1943
1944 if (create_bvec_pool) {
1945 bs->bvec_pool = biovec_create_pool(pool_size);
1946 if (!bs->bvec_pool)
1947 goto bad;
1948 }
1949
1950 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1951 if (!bs->rescue_workqueue)
1952 goto bad;
1953
1954 return bs;
1955bad:
1956 bioset_free(bs);
1957 return NULL;
1958}
1959
1960/**
1961 * bioset_create - Create a bio_set
1962 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1963 * @front_pad: Number of bytes to allocate in front of the returned bio
1964 *
1965 * Description:
1966 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1967 * to ask for a number of bytes to be allocated in front of the bio.
1968 * Front pad allocation is useful for embedding the bio inside
1969 * another structure, to avoid allocating extra data to go with the bio.
1970 * Note that the bio must be embedded at the END of that structure always,
1971 * or things will break badly.
1972 */
1973struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1974{
1975 return __bioset_create(pool_size, front_pad, true);
1976}
1977EXPORT_SYMBOL(bioset_create);
1978
1979/**
1980 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1981 * @pool_size: Number of bio to cache in the mempool
1982 * @front_pad: Number of bytes to allocate in front of the returned bio
1983 *
1984 * Description:
1985 * Same functionality as bioset_create() except that mempool is not
1986 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1987 */
1988struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1989{
1990 return __bioset_create(pool_size, front_pad, false);
1991}
1992EXPORT_SYMBOL(bioset_create_nobvec);
1993
1994#ifdef CONFIG_BLK_CGROUP
1995
1996/**
1997 * bio_associate_blkcg - associate a bio with the specified blkcg
1998 * @bio: target bio
1999 * @blkcg_css: css of the blkcg to associate
2000 *
2001 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
2002 * treat @bio as if it were issued by a task which belongs to the blkcg.
2003 *
2004 * This function takes an extra reference of @blkcg_css which will be put
2005 * when @bio is released. The caller must own @bio and is responsible for
2006 * synchronizing calls to this function.
2007 */
2008int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
2009{
2010 if (unlikely(bio->bi_css))
2011 return -EBUSY;
2012 css_get(blkcg_css);
2013 bio->bi_css = blkcg_css;
2014 return 0;
2015}
2016EXPORT_SYMBOL_GPL(bio_associate_blkcg);
2017
2018/**
2019 * bio_associate_current - associate a bio with %current
2020 * @bio: target bio
2021 *
2022 * Associate @bio with %current if it hasn't been associated yet. Block
2023 * layer will treat @bio as if it were issued by %current no matter which
2024 * task actually issues it.
2025 *
2026 * This function takes an extra reference of @task's io_context and blkcg
2027 * which will be put when @bio is released. The caller must own @bio,
2028 * ensure %current->io_context exists, and is responsible for synchronizing
2029 * calls to this function.
2030 */
2031int bio_associate_current(struct bio *bio)
2032{
2033 struct io_context *ioc;
2034
2035 if (bio->bi_css)
2036 return -EBUSY;
2037
2038 ioc = current->io_context;
2039 if (!ioc)
2040 return -ENOENT;
2041
2042 get_io_context_active(ioc);
2043 bio->bi_ioc = ioc;
2044 bio->bi_css = task_get_css(current, io_cgrp_id);
2045 return 0;
2046}
2047EXPORT_SYMBOL_GPL(bio_associate_current);
2048
2049/**
2050 * bio_disassociate_task - undo bio_associate_current()
2051 * @bio: target bio
2052 */
2053void bio_disassociate_task(struct bio *bio)
2054{
2055 if (bio->bi_ioc) {
2056 put_io_context(bio->bi_ioc);
2057 bio->bi_ioc = NULL;
2058 }
2059 if (bio->bi_css) {
2060 css_put(bio->bi_css);
2061 bio->bi_css = NULL;
2062 }
2063}
2064
2065/**
2066 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2067 * @dst: destination bio
2068 * @src: source bio
2069 */
2070void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2071{
2072 if (src->bi_css)
2073 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2074}
2075
2076#endif /* CONFIG_BLK_CGROUP */
2077
2078static void __init biovec_init_slabs(void)
2079{
2080 int i;
2081
2082 for (i = 0; i < BVEC_POOL_NR; i++) {
2083 int size;
2084 struct biovec_slab *bvs = bvec_slabs + i;
2085
2086 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2087 bvs->slab = NULL;
2088 continue;
2089 }
2090
2091 size = bvs->nr_vecs * sizeof(struct bio_vec);
2092 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2093 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2094 }
2095}
2096
2097static int __init init_bio(void)
2098{
2099 bio_slab_max = 2;
2100 bio_slab_nr = 0;
2101 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2102 if (!bio_slabs)
2103 panic("bio: can't allocate bios\n");
2104
2105 bio_integrity_init();
2106 biovec_init_slabs();
2107
2108 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2109 if (!fs_bio_set)
2110 panic("bio: can't allocate bios\n");
2111
2112 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2113 panic("bio: can't create integrity pool\n");
2114
2115 return 0;
2116}
2117subsys_initcall(init_bio);