Loading...
Note: File does not exist in v3.1.
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-integrity.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/blk-crypto.h>
20#include <linux/xarray.h>
21
22#include <trace/events/block.h>
23#include "blk.h"
24#include "blk-rq-qos.h"
25#include "blk-cgroup.h"
26
27#define ALLOC_CACHE_THRESHOLD 16
28#define ALLOC_CACHE_MAX 256
29
30struct bio_alloc_cache {
31 struct bio *free_list;
32 struct bio *free_list_irq;
33 unsigned int nr;
34 unsigned int nr_irq;
35};
36
37static struct biovec_slab {
38 int nr_vecs;
39 char *name;
40 struct kmem_cache *slab;
41} bvec_slabs[] __read_mostly = {
42 { .nr_vecs = 16, .name = "biovec-16" },
43 { .nr_vecs = 64, .name = "biovec-64" },
44 { .nr_vecs = 128, .name = "biovec-128" },
45 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
46};
47
48static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
49{
50 switch (nr_vecs) {
51 /* smaller bios use inline vecs */
52 case 5 ... 16:
53 return &bvec_slabs[0];
54 case 17 ... 64:
55 return &bvec_slabs[1];
56 case 65 ... 128:
57 return &bvec_slabs[2];
58 case 129 ... BIO_MAX_VECS:
59 return &bvec_slabs[3];
60 default:
61 BUG();
62 return NULL;
63 }
64}
65
66/*
67 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
68 * IO code that does not need private memory pools.
69 */
70struct bio_set fs_bio_set;
71EXPORT_SYMBOL(fs_bio_set);
72
73/*
74 * Our slab pool management
75 */
76struct bio_slab {
77 struct kmem_cache *slab;
78 unsigned int slab_ref;
79 unsigned int slab_size;
80 char name[8];
81};
82static DEFINE_MUTEX(bio_slab_lock);
83static DEFINE_XARRAY(bio_slabs);
84
85static struct bio_slab *create_bio_slab(unsigned int size)
86{
87 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
88
89 if (!bslab)
90 return NULL;
91
92 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
93 bslab->slab = kmem_cache_create(bslab->name, size,
94 ARCH_KMALLOC_MINALIGN,
95 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
96 if (!bslab->slab)
97 goto fail_alloc_slab;
98
99 bslab->slab_ref = 1;
100 bslab->slab_size = size;
101
102 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
103 return bslab;
104
105 kmem_cache_destroy(bslab->slab);
106
107fail_alloc_slab:
108 kfree(bslab);
109 return NULL;
110}
111
112static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
113{
114 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
115}
116
117static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
118{
119 unsigned int size = bs_bio_slab_size(bs);
120 struct bio_slab *bslab;
121
122 mutex_lock(&bio_slab_lock);
123 bslab = xa_load(&bio_slabs, size);
124 if (bslab)
125 bslab->slab_ref++;
126 else
127 bslab = create_bio_slab(size);
128 mutex_unlock(&bio_slab_lock);
129
130 if (bslab)
131 return bslab->slab;
132 return NULL;
133}
134
135static void bio_put_slab(struct bio_set *bs)
136{
137 struct bio_slab *bslab = NULL;
138 unsigned int slab_size = bs_bio_slab_size(bs);
139
140 mutex_lock(&bio_slab_lock);
141
142 bslab = xa_load(&bio_slabs, slab_size);
143 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
144 goto out;
145
146 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
147
148 WARN_ON(!bslab->slab_ref);
149
150 if (--bslab->slab_ref)
151 goto out;
152
153 xa_erase(&bio_slabs, slab_size);
154
155 kmem_cache_destroy(bslab->slab);
156 kfree(bslab);
157
158out:
159 mutex_unlock(&bio_slab_lock);
160}
161
162void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
163{
164 BUG_ON(nr_vecs > BIO_MAX_VECS);
165
166 if (nr_vecs == BIO_MAX_VECS)
167 mempool_free(bv, pool);
168 else if (nr_vecs > BIO_INLINE_VECS)
169 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
170}
171
172/*
173 * Make the first allocation restricted and don't dump info on allocation
174 * failures, since we'll fall back to the mempool in case of failure.
175 */
176static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
177{
178 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
179 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
180}
181
182struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
183 gfp_t gfp_mask)
184{
185 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
186
187 if (WARN_ON_ONCE(!bvs))
188 return NULL;
189
190 /*
191 * Upgrade the nr_vecs request to take full advantage of the allocation.
192 * We also rely on this in the bvec_free path.
193 */
194 *nr_vecs = bvs->nr_vecs;
195
196 /*
197 * Try a slab allocation first for all smaller allocations. If that
198 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
199 * The mempool is sized to handle up to BIO_MAX_VECS entries.
200 */
201 if (*nr_vecs < BIO_MAX_VECS) {
202 struct bio_vec *bvl;
203
204 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
205 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
206 return bvl;
207 *nr_vecs = BIO_MAX_VECS;
208 }
209
210 return mempool_alloc(pool, gfp_mask);
211}
212
213void bio_uninit(struct bio *bio)
214{
215#ifdef CONFIG_BLK_CGROUP
216 if (bio->bi_blkg) {
217 blkg_put(bio->bi_blkg);
218 bio->bi_blkg = NULL;
219 }
220#endif
221 if (bio_integrity(bio))
222 bio_integrity_free(bio);
223
224 bio_crypt_free_ctx(bio);
225}
226EXPORT_SYMBOL(bio_uninit);
227
228static void bio_free(struct bio *bio)
229{
230 struct bio_set *bs = bio->bi_pool;
231 void *p = bio;
232
233 WARN_ON_ONCE(!bs);
234
235 bio_uninit(bio);
236 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
237 mempool_free(p - bs->front_pad, &bs->bio_pool);
238}
239
240/*
241 * Users of this function have their own bio allocation. Subsequently,
242 * they must remember to pair any call to bio_init() with bio_uninit()
243 * when IO has completed, or when the bio is released.
244 */
245void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
246 unsigned short max_vecs, blk_opf_t opf)
247{
248 bio->bi_next = NULL;
249 bio->bi_bdev = bdev;
250 bio->bi_opf = opf;
251 bio->bi_flags = 0;
252 bio->bi_ioprio = 0;
253 bio->bi_write_hint = 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
348/**
349 * bio_chain_and_submit - submit a bio after chaining it to another one
350 * @prev: bio to chain and submit
351 * @new: bio to chain to
352 *
353 * If @prev is non-NULL, chain it to @new and submit it.
354 *
355 * Return: @new.
356 */
357struct bio *bio_chain_and_submit(struct bio *prev, struct bio *new)
358{
359 if (prev) {
360 bio_chain(prev, new);
361 submit_bio(prev);
362 }
363 return new;
364}
365
366struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
367 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
368{
369 return bio_chain_and_submit(bio, bio_alloc(bdev, nr_pages, opf, gfp));
370}
371EXPORT_SYMBOL_GPL(blk_next_bio);
372
373static void bio_alloc_rescue(struct work_struct *work)
374{
375 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
376 struct bio *bio;
377
378 while (1) {
379 spin_lock(&bs->rescue_lock);
380 bio = bio_list_pop(&bs->rescue_list);
381 spin_unlock(&bs->rescue_lock);
382
383 if (!bio)
384 break;
385
386 submit_bio_noacct(bio);
387 }
388}
389
390static void punt_bios_to_rescuer(struct bio_set *bs)
391{
392 struct bio_list punt, nopunt;
393 struct bio *bio;
394
395 if (WARN_ON_ONCE(!bs->rescue_workqueue))
396 return;
397 /*
398 * In order to guarantee forward progress we must punt only bios that
399 * were allocated from this bio_set; otherwise, if there was a bio on
400 * there for a stacking driver higher up in the stack, processing it
401 * could require allocating bios from this bio_set, and doing that from
402 * our own rescuer would be bad.
403 *
404 * Since bio lists are singly linked, pop them all instead of trying to
405 * remove from the middle of the list:
406 */
407
408 bio_list_init(&punt);
409 bio_list_init(&nopunt);
410
411 while ((bio = bio_list_pop(¤t->bio_list[0])))
412 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
413 current->bio_list[0] = nopunt;
414
415 bio_list_init(&nopunt);
416 while ((bio = bio_list_pop(¤t->bio_list[1])))
417 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
418 current->bio_list[1] = nopunt;
419
420 spin_lock(&bs->rescue_lock);
421 bio_list_merge(&bs->rescue_list, &punt);
422 spin_unlock(&bs->rescue_lock);
423
424 queue_work(bs->rescue_workqueue, &bs->rescue_work);
425}
426
427static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
428{
429 unsigned long flags;
430
431 /* cache->free_list must be empty */
432 if (WARN_ON_ONCE(cache->free_list))
433 return;
434
435 local_irq_save(flags);
436 cache->free_list = cache->free_list_irq;
437 cache->free_list_irq = NULL;
438 cache->nr += cache->nr_irq;
439 cache->nr_irq = 0;
440 local_irq_restore(flags);
441}
442
443static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
444 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
445 struct bio_set *bs)
446{
447 struct bio_alloc_cache *cache;
448 struct bio *bio;
449
450 cache = per_cpu_ptr(bs->cache, get_cpu());
451 if (!cache->free_list) {
452 if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
453 bio_alloc_irq_cache_splice(cache);
454 if (!cache->free_list) {
455 put_cpu();
456 return NULL;
457 }
458 }
459 bio = cache->free_list;
460 cache->free_list = bio->bi_next;
461 cache->nr--;
462 put_cpu();
463
464 bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
465 bio->bi_pool = bs;
466 return bio;
467}
468
469/**
470 * bio_alloc_bioset - allocate a bio for I/O
471 * @bdev: block device to allocate the bio for (can be %NULL)
472 * @nr_vecs: number of bvecs to pre-allocate
473 * @opf: operation and flags for bio
474 * @gfp_mask: the GFP_* mask given to the slab allocator
475 * @bs: the bio_set to allocate from.
476 *
477 * Allocate a bio from the mempools in @bs.
478 *
479 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
480 * allocate a bio. This is due to the mempool guarantees. To make this work,
481 * callers must never allocate more than 1 bio at a time from the general pool.
482 * Callers that need to allocate more than 1 bio must always submit the
483 * previously allocated bio for IO before attempting to allocate a new one.
484 * Failure to do so can cause deadlocks under memory pressure.
485 *
486 * Note that when running under submit_bio_noacct() (i.e. any block driver),
487 * bios are not submitted until after you return - see the code in
488 * submit_bio_noacct() that converts recursion into iteration, to prevent
489 * stack overflows.
490 *
491 * This would normally mean allocating multiple bios under submit_bio_noacct()
492 * would be susceptible to deadlocks, but we have
493 * deadlock avoidance code that resubmits any blocked bios from a rescuer
494 * thread.
495 *
496 * However, we do not guarantee forward progress for allocations from other
497 * mempools. Doing multiple allocations from the same mempool under
498 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
499 * for per bio allocations.
500 *
501 * Returns: Pointer to new bio on success, NULL on failure.
502 */
503struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
504 blk_opf_t opf, gfp_t gfp_mask,
505 struct bio_set *bs)
506{
507 gfp_t saved_gfp = gfp_mask;
508 struct bio *bio;
509 void *p;
510
511 /* should not use nobvec bioset for nr_vecs > 0 */
512 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
513 return NULL;
514
515 if (opf & REQ_ALLOC_CACHE) {
516 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
517 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
518 gfp_mask, bs);
519 if (bio)
520 return bio;
521 /*
522 * No cached bio available, bio returned below marked with
523 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
524 */
525 } else {
526 opf &= ~REQ_ALLOC_CACHE;
527 }
528 }
529
530 /*
531 * submit_bio_noacct() converts recursion to iteration; this means if
532 * we're running beneath it, any bios we allocate and submit will not be
533 * submitted (and thus freed) until after we return.
534 *
535 * This exposes us to a potential deadlock if we allocate multiple bios
536 * from the same bio_set() while running underneath submit_bio_noacct().
537 * If we were to allocate multiple bios (say a stacking block driver
538 * that was splitting bios), we would deadlock if we exhausted the
539 * mempool's reserve.
540 *
541 * We solve this, and guarantee forward progress, with a rescuer
542 * workqueue per bio_set. If we go to allocate and there are bios on
543 * current->bio_list, we first try the allocation without
544 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
545 * blocking to the rescuer workqueue before we retry with the original
546 * gfp_flags.
547 */
548 if (current->bio_list &&
549 (!bio_list_empty(¤t->bio_list[0]) ||
550 !bio_list_empty(¤t->bio_list[1])) &&
551 bs->rescue_workqueue)
552 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
553
554 p = mempool_alloc(&bs->bio_pool, gfp_mask);
555 if (!p && gfp_mask != saved_gfp) {
556 punt_bios_to_rescuer(bs);
557 gfp_mask = saved_gfp;
558 p = mempool_alloc(&bs->bio_pool, gfp_mask);
559 }
560 if (unlikely(!p))
561 return NULL;
562 if (!mempool_is_saturated(&bs->bio_pool))
563 opf &= ~REQ_ALLOC_CACHE;
564
565 bio = p + bs->front_pad;
566 if (nr_vecs > BIO_INLINE_VECS) {
567 struct bio_vec *bvl = NULL;
568
569 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
570 if (!bvl && gfp_mask != saved_gfp) {
571 punt_bios_to_rescuer(bs);
572 gfp_mask = saved_gfp;
573 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
574 }
575 if (unlikely(!bvl))
576 goto err_free;
577
578 bio_init(bio, bdev, bvl, nr_vecs, opf);
579 } else if (nr_vecs) {
580 bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
581 } else {
582 bio_init(bio, bdev, NULL, 0, opf);
583 }
584
585 bio->bi_pool = bs;
586 return bio;
587
588err_free:
589 mempool_free(p, &bs->bio_pool);
590 return NULL;
591}
592EXPORT_SYMBOL(bio_alloc_bioset);
593
594/**
595 * bio_kmalloc - kmalloc a bio
596 * @nr_vecs: number of bio_vecs to allocate
597 * @gfp_mask: the GFP_* mask given to the slab allocator
598 *
599 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
600 * using bio_init() before use. To free a bio returned from this function use
601 * kfree() after calling bio_uninit(). A bio returned from this function can
602 * be reused by calling bio_uninit() before calling bio_init() again.
603 *
604 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
605 * function are not backed by a mempool can fail. Do not use this function
606 * for allocations in the file system I/O path.
607 *
608 * Returns: Pointer to new bio on success, NULL on failure.
609 */
610struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
611{
612 struct bio *bio;
613
614 if (nr_vecs > UIO_MAXIOV)
615 return NULL;
616 return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
617}
618EXPORT_SYMBOL(bio_kmalloc);
619
620void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
621{
622 struct bio_vec bv;
623 struct bvec_iter iter;
624
625 __bio_for_each_segment(bv, bio, iter, start)
626 memzero_bvec(&bv);
627}
628EXPORT_SYMBOL(zero_fill_bio_iter);
629
630/**
631 * bio_truncate - truncate the bio to small size of @new_size
632 * @bio: the bio to be truncated
633 * @new_size: new size for truncating the bio
634 *
635 * Description:
636 * Truncate the bio to new size of @new_size. If bio_op(bio) is
637 * REQ_OP_READ, zero the truncated part. This function should only
638 * be used for handling corner cases, such as bio eod.
639 */
640static void bio_truncate(struct bio *bio, unsigned new_size)
641{
642 struct bio_vec bv;
643 struct bvec_iter iter;
644 unsigned int done = 0;
645 bool truncated = false;
646
647 if (new_size >= bio->bi_iter.bi_size)
648 return;
649
650 if (bio_op(bio) != REQ_OP_READ)
651 goto exit;
652
653 bio_for_each_segment(bv, bio, iter) {
654 if (done + bv.bv_len > new_size) {
655 unsigned offset;
656
657 if (!truncated)
658 offset = new_size - done;
659 else
660 offset = 0;
661 zero_user(bv.bv_page, bv.bv_offset + offset,
662 bv.bv_len - offset);
663 truncated = true;
664 }
665 done += bv.bv_len;
666 }
667
668 exit:
669 /*
670 * Don't touch bvec table here and make it really immutable, since
671 * fs bio user has to retrieve all pages via bio_for_each_segment_all
672 * in its .end_bio() callback.
673 *
674 * It is enough to truncate bio by updating .bi_size since we can make
675 * correct bvec with the updated .bi_size for drivers.
676 */
677 bio->bi_iter.bi_size = new_size;
678}
679
680/**
681 * guard_bio_eod - truncate a BIO to fit the block device
682 * @bio: bio to truncate
683 *
684 * This allows us to do IO even on the odd last sectors of a device, even if the
685 * block size is some multiple of the physical sector size.
686 *
687 * We'll just truncate the bio to the size of the device, and clear the end of
688 * the buffer head manually. Truly out-of-range accesses will turn into actual
689 * I/O errors, this only handles the "we need to be able to do I/O at the final
690 * sector" case.
691 */
692void guard_bio_eod(struct bio *bio)
693{
694 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
695
696 if (!maxsector)
697 return;
698
699 /*
700 * If the *whole* IO is past the end of the device,
701 * let it through, and the IO layer will turn it into
702 * an EIO.
703 */
704 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
705 return;
706
707 maxsector -= bio->bi_iter.bi_sector;
708 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
709 return;
710
711 bio_truncate(bio, maxsector << 9);
712}
713
714static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
715 unsigned int nr)
716{
717 unsigned int i = 0;
718 struct bio *bio;
719
720 while ((bio = cache->free_list) != NULL) {
721 cache->free_list = bio->bi_next;
722 cache->nr--;
723 bio_free(bio);
724 if (++i == nr)
725 break;
726 }
727 return i;
728}
729
730static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
731 unsigned int nr)
732{
733 nr -= __bio_alloc_cache_prune(cache, nr);
734 if (!READ_ONCE(cache->free_list)) {
735 bio_alloc_irq_cache_splice(cache);
736 __bio_alloc_cache_prune(cache, nr);
737 }
738}
739
740static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
741{
742 struct bio_set *bs;
743
744 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
745 if (bs->cache) {
746 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
747
748 bio_alloc_cache_prune(cache, -1U);
749 }
750 return 0;
751}
752
753static void bio_alloc_cache_destroy(struct bio_set *bs)
754{
755 int cpu;
756
757 if (!bs->cache)
758 return;
759
760 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
761 for_each_possible_cpu(cpu) {
762 struct bio_alloc_cache *cache;
763
764 cache = per_cpu_ptr(bs->cache, cpu);
765 bio_alloc_cache_prune(cache, -1U);
766 }
767 free_percpu(bs->cache);
768 bs->cache = NULL;
769}
770
771static inline void bio_put_percpu_cache(struct bio *bio)
772{
773 struct bio_alloc_cache *cache;
774
775 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
776 if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX)
777 goto out_free;
778
779 if (in_task()) {
780 bio_uninit(bio);
781 bio->bi_next = cache->free_list;
782 /* Not necessary but helps not to iopoll already freed bios */
783 bio->bi_bdev = NULL;
784 cache->free_list = bio;
785 cache->nr++;
786 } else if (in_hardirq()) {
787 lockdep_assert_irqs_disabled();
788
789 bio_uninit(bio);
790 bio->bi_next = cache->free_list_irq;
791 cache->free_list_irq = bio;
792 cache->nr_irq++;
793 } else {
794 goto out_free;
795 }
796 put_cpu();
797 return;
798out_free:
799 put_cpu();
800 bio_free(bio);
801}
802
803/**
804 * bio_put - release a reference to a bio
805 * @bio: bio to release reference to
806 *
807 * Description:
808 * Put a reference to a &struct bio, either one you have gotten with
809 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
810 **/
811void bio_put(struct bio *bio)
812{
813 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
814 BUG_ON(!atomic_read(&bio->__bi_cnt));
815 if (!atomic_dec_and_test(&bio->__bi_cnt))
816 return;
817 }
818 if (bio->bi_opf & REQ_ALLOC_CACHE)
819 bio_put_percpu_cache(bio);
820 else
821 bio_free(bio);
822}
823EXPORT_SYMBOL(bio_put);
824
825static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
826{
827 bio_set_flag(bio, BIO_CLONED);
828 bio->bi_ioprio = bio_src->bi_ioprio;
829 bio->bi_write_hint = bio_src->bi_write_hint;
830 bio->bi_iter = bio_src->bi_iter;
831
832 if (bio->bi_bdev) {
833 if (bio->bi_bdev == bio_src->bi_bdev &&
834 bio_flagged(bio_src, BIO_REMAPPED))
835 bio_set_flag(bio, BIO_REMAPPED);
836 bio_clone_blkg_association(bio, bio_src);
837 }
838
839 if (bio_crypt_clone(bio, bio_src, gfp) < 0)
840 return -ENOMEM;
841 if (bio_integrity(bio_src) &&
842 bio_integrity_clone(bio, bio_src, gfp) < 0)
843 return -ENOMEM;
844 return 0;
845}
846
847/**
848 * bio_alloc_clone - clone a bio that shares the original bio's biovec
849 * @bdev: block_device to clone onto
850 * @bio_src: bio to clone from
851 * @gfp: allocation priority
852 * @bs: bio_set to allocate from
853 *
854 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
855 * bio, but not the actual data it points to.
856 *
857 * The caller must ensure that the return bio is not freed before @bio_src.
858 */
859struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
860 gfp_t gfp, struct bio_set *bs)
861{
862 struct bio *bio;
863
864 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
865 if (!bio)
866 return NULL;
867
868 if (__bio_clone(bio, bio_src, gfp) < 0) {
869 bio_put(bio);
870 return NULL;
871 }
872 bio->bi_io_vec = bio_src->bi_io_vec;
873
874 return bio;
875}
876EXPORT_SYMBOL(bio_alloc_clone);
877
878/**
879 * bio_init_clone - clone a bio that shares the original bio's biovec
880 * @bdev: block_device to clone onto
881 * @bio: bio to clone into
882 * @bio_src: bio to clone from
883 * @gfp: allocation priority
884 *
885 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
886 * The caller owns the returned bio, but not the actual data it points to.
887 *
888 * The caller must ensure that @bio_src is not freed before @bio.
889 */
890int bio_init_clone(struct block_device *bdev, struct bio *bio,
891 struct bio *bio_src, gfp_t gfp)
892{
893 int ret;
894
895 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
896 ret = __bio_clone(bio, bio_src, gfp);
897 if (ret)
898 bio_uninit(bio);
899 return ret;
900}
901EXPORT_SYMBOL(bio_init_clone);
902
903/**
904 * bio_full - check if the bio is full
905 * @bio: bio to check
906 * @len: length of one segment to be added
907 *
908 * Return true if @bio is full and one segment with @len bytes can't be
909 * added to the bio, otherwise return false
910 */
911static inline bool bio_full(struct bio *bio, unsigned len)
912{
913 if (bio->bi_vcnt >= bio->bi_max_vecs)
914 return true;
915 if (bio->bi_iter.bi_size > UINT_MAX - len)
916 return true;
917 return false;
918}
919
920static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page,
921 unsigned int len, unsigned int off, bool *same_page)
922{
923 size_t bv_end = bv->bv_offset + bv->bv_len;
924 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
925 phys_addr_t page_addr = page_to_phys(page);
926
927 if (vec_end_addr + 1 != page_addr + off)
928 return false;
929 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
930 return false;
931 if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
932 return false;
933
934 *same_page = ((vec_end_addr & PAGE_MASK) == ((page_addr + off) &
935 PAGE_MASK));
936 if (!*same_page) {
937 if (IS_ENABLED(CONFIG_KMSAN))
938 return false;
939 if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE)
940 return false;
941 }
942
943 bv->bv_len += len;
944 return true;
945}
946
947/*
948 * Try to merge a page into a segment, while obeying the hardware segment
949 * size limit. This is not for normal read/write bios, but for passthrough
950 * or Zone Append operations that we can't split.
951 */
952bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
953 struct page *page, unsigned len, unsigned offset,
954 bool *same_page)
955{
956 unsigned long mask = queue_segment_boundary(q);
957 phys_addr_t addr1 = bvec_phys(bv);
958 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
959
960 if ((addr1 | mask) != (addr2 | mask))
961 return false;
962 if (len > queue_max_segment_size(q) - bv->bv_len)
963 return false;
964 return bvec_try_merge_page(bv, page, len, offset, same_page);
965}
966
967/**
968 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
969 * @q: the target queue
970 * @bio: destination bio
971 * @page: page to add
972 * @len: vec entry length
973 * @offset: vec entry offset
974 * @max_sectors: maximum number of sectors that can be added
975 * @same_page: return if the segment has been merged inside the same page
976 *
977 * Add a page to a bio while respecting the hardware max_sectors, max_segment
978 * and gap limitations.
979 */
980int bio_add_hw_page(struct request_queue *q, struct bio *bio,
981 struct page *page, unsigned int len, unsigned int offset,
982 unsigned int max_sectors, bool *same_page)
983{
984 unsigned int max_size = max_sectors << SECTOR_SHIFT;
985
986 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
987 return 0;
988
989 len = min3(len, max_size, queue_max_segment_size(q));
990 if (len > max_size - bio->bi_iter.bi_size)
991 return 0;
992
993 if (bio->bi_vcnt > 0) {
994 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
995
996 if (bvec_try_merge_hw_page(q, bv, page, len, offset,
997 same_page)) {
998 bio->bi_iter.bi_size += len;
999 return len;
1000 }
1001
1002 if (bio->bi_vcnt >=
1003 min(bio->bi_max_vecs, queue_max_segments(q)))
1004 return 0;
1005
1006 /*
1007 * If the queue doesn't support SG gaps and adding this segment
1008 * would create a gap, disallow it.
1009 */
1010 if (bvec_gap_to_prev(&q->limits, bv, offset))
1011 return 0;
1012 }
1013
1014 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
1015 bio->bi_vcnt++;
1016 bio->bi_iter.bi_size += len;
1017 return len;
1018}
1019
1020/**
1021 * bio_add_hw_folio - attempt to add a folio to a bio with hw constraints
1022 * @q: the target queue
1023 * @bio: destination bio
1024 * @folio: folio to add
1025 * @len: vec entry length
1026 * @offset: vec entry offset in the folio
1027 * @max_sectors: maximum number of sectors that can be added
1028 * @same_page: return if the segment has been merged inside the same folio
1029 *
1030 * Add a folio to a bio while respecting the hardware max_sectors, max_segment
1031 * and gap limitations.
1032 */
1033int bio_add_hw_folio(struct request_queue *q, struct bio *bio,
1034 struct folio *folio, size_t len, size_t offset,
1035 unsigned int max_sectors, bool *same_page)
1036{
1037 if (len > UINT_MAX || offset > UINT_MAX)
1038 return 0;
1039 return bio_add_hw_page(q, bio, folio_page(folio, 0), len, offset,
1040 max_sectors, same_page);
1041}
1042
1043/**
1044 * bio_add_pc_page - attempt to add page to passthrough bio
1045 * @q: the target queue
1046 * @bio: destination bio
1047 * @page: page to add
1048 * @len: vec entry length
1049 * @offset: vec entry offset
1050 *
1051 * Attempt to add a page to the bio_vec maplist. This can fail for a
1052 * number of reasons, such as the bio being full or target block device
1053 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1054 * so it is always possible to add a single page to an empty bio.
1055 *
1056 * This should only be used by passthrough bios.
1057 */
1058int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1059 struct page *page, unsigned int len, unsigned int offset)
1060{
1061 bool same_page = false;
1062 return bio_add_hw_page(q, bio, page, len, offset,
1063 queue_max_hw_sectors(q), &same_page);
1064}
1065EXPORT_SYMBOL(bio_add_pc_page);
1066
1067/**
1068 * __bio_add_page - add page(s) to a bio in a new segment
1069 * @bio: destination bio
1070 * @page: start page to add
1071 * @len: length of the data to add, may cross pages
1072 * @off: offset of the data relative to @page, may cross pages
1073 *
1074 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1075 * that @bio has space for another bvec.
1076 */
1077void __bio_add_page(struct bio *bio, struct page *page,
1078 unsigned int len, unsigned int off)
1079{
1080 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1081 WARN_ON_ONCE(bio_full(bio, len));
1082
1083 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
1084 bio->bi_iter.bi_size += len;
1085 bio->bi_vcnt++;
1086}
1087EXPORT_SYMBOL_GPL(__bio_add_page);
1088
1089/**
1090 * bio_add_page - attempt to add page(s) to bio
1091 * @bio: destination bio
1092 * @page: start page to add
1093 * @len: vec entry length, may cross pages
1094 * @offset: vec entry offset relative to @page, may cross pages
1095 *
1096 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1097 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1098 */
1099int bio_add_page(struct bio *bio, struct page *page,
1100 unsigned int len, unsigned int offset)
1101{
1102 bool same_page = false;
1103
1104 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1105 return 0;
1106 if (bio->bi_iter.bi_size > UINT_MAX - len)
1107 return 0;
1108
1109 if (bio->bi_vcnt > 0 &&
1110 bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1111 page, len, offset, &same_page)) {
1112 bio->bi_iter.bi_size += len;
1113 return len;
1114 }
1115
1116 if (bio->bi_vcnt >= bio->bi_max_vecs)
1117 return 0;
1118 __bio_add_page(bio, page, len, offset);
1119 return len;
1120}
1121EXPORT_SYMBOL(bio_add_page);
1122
1123void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1124 size_t off)
1125{
1126 WARN_ON_ONCE(len > UINT_MAX);
1127 WARN_ON_ONCE(off > UINT_MAX);
1128 __bio_add_page(bio, &folio->page, len, off);
1129}
1130EXPORT_SYMBOL_GPL(bio_add_folio_nofail);
1131
1132/**
1133 * bio_add_folio - Attempt to add part of a folio to a bio.
1134 * @bio: BIO to add to.
1135 * @folio: Folio to add.
1136 * @len: How many bytes from the folio to add.
1137 * @off: First byte in this folio to add.
1138 *
1139 * Filesystems that use folios can call this function instead of calling
1140 * bio_add_page() for each page in the folio. If @off is bigger than
1141 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1142 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1143 *
1144 * Return: Whether the addition was successful.
1145 */
1146bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1147 size_t off)
1148{
1149 if (len > UINT_MAX || off > UINT_MAX)
1150 return false;
1151 return bio_add_page(bio, &folio->page, len, off) > 0;
1152}
1153EXPORT_SYMBOL(bio_add_folio);
1154
1155void __bio_release_pages(struct bio *bio, bool mark_dirty)
1156{
1157 struct folio_iter fi;
1158
1159 bio_for_each_folio_all(fi, bio) {
1160 size_t nr_pages;
1161
1162 if (mark_dirty) {
1163 folio_lock(fi.folio);
1164 folio_mark_dirty(fi.folio);
1165 folio_unlock(fi.folio);
1166 }
1167 nr_pages = (fi.offset + fi.length - 1) / PAGE_SIZE -
1168 fi.offset / PAGE_SIZE + 1;
1169 unpin_user_folio(fi.folio, nr_pages);
1170 }
1171}
1172EXPORT_SYMBOL_GPL(__bio_release_pages);
1173
1174void bio_iov_bvec_set(struct bio *bio, const struct iov_iter *iter)
1175{
1176 WARN_ON_ONCE(bio->bi_max_vecs);
1177
1178 bio->bi_vcnt = iter->nr_segs;
1179 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1180 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1181 bio->bi_iter.bi_size = iov_iter_count(iter);
1182 bio_set_flag(bio, BIO_CLONED);
1183}
1184
1185static int bio_iov_add_folio(struct bio *bio, struct folio *folio, size_t len,
1186 size_t offset)
1187{
1188 bool same_page = false;
1189
1190 if (WARN_ON_ONCE(bio->bi_iter.bi_size > UINT_MAX - len))
1191 return -EIO;
1192
1193 if (bio->bi_vcnt > 0 &&
1194 bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1195 folio_page(folio, 0), len, offset,
1196 &same_page)) {
1197 bio->bi_iter.bi_size += len;
1198 if (same_page && bio_flagged(bio, BIO_PAGE_PINNED))
1199 unpin_user_folio(folio, 1);
1200 return 0;
1201 }
1202 bio_add_folio_nofail(bio, folio, len, offset);
1203 return 0;
1204}
1205
1206static unsigned int get_contig_folio_len(unsigned int *num_pages,
1207 struct page **pages, unsigned int i,
1208 struct folio *folio, size_t left,
1209 size_t offset)
1210{
1211 size_t bytes = left;
1212 size_t contig_sz = min_t(size_t, PAGE_SIZE - offset, bytes);
1213 unsigned int j;
1214
1215 /*
1216 * We might COW a single page in the middle of
1217 * a large folio, so we have to check that all
1218 * pages belong to the same folio.
1219 */
1220 bytes -= contig_sz;
1221 for (j = i + 1; j < i + *num_pages; j++) {
1222 size_t next = min_t(size_t, PAGE_SIZE, bytes);
1223
1224 if (page_folio(pages[j]) != folio ||
1225 pages[j] != pages[j - 1] + 1) {
1226 break;
1227 }
1228 contig_sz += next;
1229 bytes -= next;
1230 }
1231 *num_pages = j - i;
1232
1233 return contig_sz;
1234}
1235
1236#define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1237
1238/**
1239 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1240 * @bio: bio to add pages to
1241 * @iter: iov iterator describing the region to be mapped
1242 *
1243 * Extracts pages from *iter and appends them to @bio's bvec array. The pages
1244 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.
1245 * For a multi-segment *iter, this function only adds pages from the next
1246 * non-empty segment of the iov iterator.
1247 */
1248static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1249{
1250 iov_iter_extraction_t extraction_flags = 0;
1251 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1252 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1253 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1254 struct page **pages = (struct page **)bv;
1255 ssize_t size;
1256 unsigned int num_pages, i = 0;
1257 size_t offset, folio_offset, left, len;
1258 int ret = 0;
1259
1260 /*
1261 * Move page array up in the allocated memory for the bio vecs as far as
1262 * possible so that we can start filling biovecs from the beginning
1263 * without overwriting the temporary page array.
1264 */
1265 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1266 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1267
1268 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1269 extraction_flags |= ITER_ALLOW_P2PDMA;
1270
1271 /*
1272 * Each segment in the iov is required to be a block size multiple.
1273 * However, we may not be able to get the entire segment if it spans
1274 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1275 * result to ensure the bio's total size is correct. The remainder of
1276 * the iov data will be picked up in the next bio iteration.
1277 */
1278 size = iov_iter_extract_pages(iter, &pages,
1279 UINT_MAX - bio->bi_iter.bi_size,
1280 nr_pages, extraction_flags, &offset);
1281 if (unlikely(size <= 0))
1282 return size ? size : -EFAULT;
1283
1284 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1285
1286 if (bio->bi_bdev) {
1287 size_t trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1288 iov_iter_revert(iter, trim);
1289 size -= trim;
1290 }
1291
1292 if (unlikely(!size)) {
1293 ret = -EFAULT;
1294 goto out;
1295 }
1296
1297 for (left = size, i = 0; left > 0; left -= len, i += num_pages) {
1298 struct page *page = pages[i];
1299 struct folio *folio = page_folio(page);
1300
1301 folio_offset = ((size_t)folio_page_idx(folio, page) <<
1302 PAGE_SHIFT) + offset;
1303
1304 len = min(folio_size(folio) - folio_offset, left);
1305
1306 num_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1307
1308 if (num_pages > 1)
1309 len = get_contig_folio_len(&num_pages, pages, i,
1310 folio, left, offset);
1311
1312 bio_iov_add_folio(bio, folio, len, folio_offset);
1313 offset = 0;
1314 }
1315
1316 iov_iter_revert(iter, left);
1317out:
1318 while (i < nr_pages)
1319 bio_release_page(bio, pages[i++]);
1320
1321 return ret;
1322}
1323
1324/**
1325 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1326 * @bio: bio to add pages to
1327 * @iter: iov iterator describing the region to be added
1328 *
1329 * This takes either an iterator pointing to user memory, or one pointing to
1330 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1331 * map them into the kernel. On IO completion, the caller should put those
1332 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1333 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1334 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1335 * completed by a call to ->ki_complete() or returns with an error other than
1336 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1337 * on IO completion. If it isn't, then pages should be released.
1338 *
1339 * The function tries, but does not guarantee, to pin as many pages as
1340 * fit into the bio, or are requested in @iter, whatever is smaller. If
1341 * MM encounters an error pinning the requested pages, it stops. Error
1342 * is returned only if 0 pages could be pinned.
1343 */
1344int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1345{
1346 int ret = 0;
1347
1348 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1349 return -EIO;
1350
1351 if (iov_iter_is_bvec(iter)) {
1352 bio_iov_bvec_set(bio, iter);
1353 iov_iter_advance(iter, bio->bi_iter.bi_size);
1354 return 0;
1355 }
1356
1357 if (iov_iter_extract_will_pin(iter))
1358 bio_set_flag(bio, BIO_PAGE_PINNED);
1359 do {
1360 ret = __bio_iov_iter_get_pages(bio, iter);
1361 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1362
1363 return bio->bi_vcnt ? 0 : ret;
1364}
1365EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1366
1367static void submit_bio_wait_endio(struct bio *bio)
1368{
1369 complete(bio->bi_private);
1370}
1371
1372/**
1373 * submit_bio_wait - submit a bio, and wait until it completes
1374 * @bio: The &struct bio which describes the I/O
1375 *
1376 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1377 * bio_endio() on failure.
1378 *
1379 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1380 * result in bio reference to be consumed. The caller must drop the reference
1381 * on his own.
1382 */
1383int submit_bio_wait(struct bio *bio)
1384{
1385 DECLARE_COMPLETION_ONSTACK_MAP(done,
1386 bio->bi_bdev->bd_disk->lockdep_map);
1387
1388 bio->bi_private = &done;
1389 bio->bi_end_io = submit_bio_wait_endio;
1390 bio->bi_opf |= REQ_SYNC;
1391 submit_bio(bio);
1392 blk_wait_io(&done);
1393
1394 return blk_status_to_errno(bio->bi_status);
1395}
1396EXPORT_SYMBOL(submit_bio_wait);
1397
1398static void bio_wait_end_io(struct bio *bio)
1399{
1400 complete(bio->bi_private);
1401 bio_put(bio);
1402}
1403
1404/*
1405 * bio_await_chain - ends @bio and waits for every chained bio to complete
1406 */
1407void bio_await_chain(struct bio *bio)
1408{
1409 DECLARE_COMPLETION_ONSTACK_MAP(done,
1410 bio->bi_bdev->bd_disk->lockdep_map);
1411
1412 bio->bi_private = &done;
1413 bio->bi_end_io = bio_wait_end_io;
1414 bio_endio(bio);
1415 blk_wait_io(&done);
1416}
1417
1418void __bio_advance(struct bio *bio, unsigned bytes)
1419{
1420 if (bio_integrity(bio))
1421 bio_integrity_advance(bio, bytes);
1422
1423 bio_crypt_advance(bio, bytes);
1424 bio_advance_iter(bio, &bio->bi_iter, bytes);
1425}
1426EXPORT_SYMBOL(__bio_advance);
1427
1428void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1429 struct bio *src, struct bvec_iter *src_iter)
1430{
1431 while (src_iter->bi_size && dst_iter->bi_size) {
1432 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1433 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1434 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1435 void *src_buf = bvec_kmap_local(&src_bv);
1436 void *dst_buf = bvec_kmap_local(&dst_bv);
1437
1438 memcpy(dst_buf, src_buf, bytes);
1439
1440 kunmap_local(dst_buf);
1441 kunmap_local(src_buf);
1442
1443 bio_advance_iter_single(src, src_iter, bytes);
1444 bio_advance_iter_single(dst, dst_iter, bytes);
1445 }
1446}
1447EXPORT_SYMBOL(bio_copy_data_iter);
1448
1449/**
1450 * bio_copy_data - copy contents of data buffers from one bio to another
1451 * @src: source bio
1452 * @dst: destination bio
1453 *
1454 * Stops when it reaches the end of either @src or @dst - that is, copies
1455 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1456 */
1457void bio_copy_data(struct bio *dst, struct bio *src)
1458{
1459 struct bvec_iter src_iter = src->bi_iter;
1460 struct bvec_iter dst_iter = dst->bi_iter;
1461
1462 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1463}
1464EXPORT_SYMBOL(bio_copy_data);
1465
1466void bio_free_pages(struct bio *bio)
1467{
1468 struct bio_vec *bvec;
1469 struct bvec_iter_all iter_all;
1470
1471 bio_for_each_segment_all(bvec, bio, iter_all)
1472 __free_page(bvec->bv_page);
1473}
1474EXPORT_SYMBOL(bio_free_pages);
1475
1476/*
1477 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1478 * for performing direct-IO in BIOs.
1479 *
1480 * The problem is that we cannot run folio_mark_dirty() from interrupt context
1481 * because the required locks are not interrupt-safe. So what we can do is to
1482 * mark the pages dirty _before_ performing IO. And in interrupt context,
1483 * check that the pages are still dirty. If so, fine. If not, redirty them
1484 * in process context.
1485 *
1486 * Note that this code is very hard to test under normal circumstances because
1487 * direct-io pins the pages with get_user_pages(). This makes
1488 * is_page_cache_freeable return false, and the VM will not clean the pages.
1489 * But other code (eg, flusher threads) could clean the pages if they are mapped
1490 * pagecache.
1491 *
1492 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1493 * deferred bio dirtying paths.
1494 */
1495
1496/*
1497 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1498 */
1499void bio_set_pages_dirty(struct bio *bio)
1500{
1501 struct folio_iter fi;
1502
1503 bio_for_each_folio_all(fi, bio) {
1504 folio_lock(fi.folio);
1505 folio_mark_dirty(fi.folio);
1506 folio_unlock(fi.folio);
1507 }
1508}
1509EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1510
1511/*
1512 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1513 * If they are, then fine. If, however, some pages are clean then they must
1514 * have been written out during the direct-IO read. So we take another ref on
1515 * the BIO and re-dirty the pages in process context.
1516 *
1517 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1518 * here on. It will unpin each page and will run one bio_put() against the
1519 * BIO.
1520 */
1521
1522static void bio_dirty_fn(struct work_struct *work);
1523
1524static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1525static DEFINE_SPINLOCK(bio_dirty_lock);
1526static struct bio *bio_dirty_list;
1527
1528/*
1529 * This runs in process context
1530 */
1531static void bio_dirty_fn(struct work_struct *work)
1532{
1533 struct bio *bio, *next;
1534
1535 spin_lock_irq(&bio_dirty_lock);
1536 next = bio_dirty_list;
1537 bio_dirty_list = NULL;
1538 spin_unlock_irq(&bio_dirty_lock);
1539
1540 while ((bio = next) != NULL) {
1541 next = bio->bi_private;
1542
1543 bio_release_pages(bio, true);
1544 bio_put(bio);
1545 }
1546}
1547
1548void bio_check_pages_dirty(struct bio *bio)
1549{
1550 struct folio_iter fi;
1551 unsigned long flags;
1552
1553 bio_for_each_folio_all(fi, bio) {
1554 if (!folio_test_dirty(fi.folio))
1555 goto defer;
1556 }
1557
1558 bio_release_pages(bio, false);
1559 bio_put(bio);
1560 return;
1561defer:
1562 spin_lock_irqsave(&bio_dirty_lock, flags);
1563 bio->bi_private = bio_dirty_list;
1564 bio_dirty_list = bio;
1565 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1566 schedule_work(&bio_dirty_work);
1567}
1568EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1569
1570static inline bool bio_remaining_done(struct bio *bio)
1571{
1572 /*
1573 * If we're not chaining, then ->__bi_remaining is always 1 and
1574 * we always end io on the first invocation.
1575 */
1576 if (!bio_flagged(bio, BIO_CHAIN))
1577 return true;
1578
1579 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1580
1581 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1582 bio_clear_flag(bio, BIO_CHAIN);
1583 return true;
1584 }
1585
1586 return false;
1587}
1588
1589/**
1590 * bio_endio - end I/O on a bio
1591 * @bio: bio
1592 *
1593 * Description:
1594 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1595 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1596 * bio unless they own it and thus know that it has an end_io function.
1597 *
1598 * bio_endio() can be called several times on a bio that has been chained
1599 * using bio_chain(). The ->bi_end_io() function will only be called the
1600 * last time.
1601 **/
1602void bio_endio(struct bio *bio)
1603{
1604again:
1605 if (!bio_remaining_done(bio))
1606 return;
1607 if (!bio_integrity_endio(bio))
1608 return;
1609
1610 blk_zone_bio_endio(bio);
1611
1612 rq_qos_done_bio(bio);
1613
1614 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1615 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1616 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1617 }
1618
1619 /*
1620 * Need to have a real endio function for chained bios, otherwise
1621 * various corner cases will break (like stacking block devices that
1622 * save/restore bi_end_io) - however, we want to avoid unbounded
1623 * recursion and blowing the stack. Tail call optimization would
1624 * handle this, but compiling with frame pointers also disables
1625 * gcc's sibling call optimization.
1626 */
1627 if (bio->bi_end_io == bio_chain_endio) {
1628 bio = __bio_chain_endio(bio);
1629 goto again;
1630 }
1631
1632#ifdef CONFIG_BLK_CGROUP
1633 /*
1634 * Release cgroup info. We shouldn't have to do this here, but quite
1635 * a few callers of bio_init fail to call bio_uninit, so we cover up
1636 * for that here at least for now.
1637 */
1638 if (bio->bi_blkg) {
1639 blkg_put(bio->bi_blkg);
1640 bio->bi_blkg = NULL;
1641 }
1642#endif
1643
1644 if (bio->bi_end_io)
1645 bio->bi_end_io(bio);
1646}
1647EXPORT_SYMBOL(bio_endio);
1648
1649/**
1650 * bio_split - split a bio
1651 * @bio: bio to split
1652 * @sectors: number of sectors to split from the front of @bio
1653 * @gfp: gfp mask
1654 * @bs: bio set to allocate from
1655 *
1656 * Allocates and returns a new bio which represents @sectors from the start of
1657 * @bio, and updates @bio to represent the remaining sectors.
1658 *
1659 * Unless this is a discard request the newly allocated bio will point
1660 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1661 * neither @bio nor @bs are freed before the split bio.
1662 */
1663struct bio *bio_split(struct bio *bio, int sectors,
1664 gfp_t gfp, struct bio_set *bs)
1665{
1666 struct bio *split;
1667
1668 if (WARN_ON_ONCE(sectors <= 0))
1669 return ERR_PTR(-EINVAL);
1670 if (WARN_ON_ONCE(sectors >= bio_sectors(bio)))
1671 return ERR_PTR(-EINVAL);
1672
1673 /* Zone append commands cannot be split */
1674 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1675 return ERR_PTR(-EINVAL);
1676
1677 /* atomic writes cannot be split */
1678 if (bio->bi_opf & REQ_ATOMIC)
1679 return ERR_PTR(-EINVAL);
1680
1681 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1682 if (!split)
1683 return ERR_PTR(-ENOMEM);
1684
1685 split->bi_iter.bi_size = sectors << 9;
1686
1687 if (bio_integrity(split))
1688 bio_integrity_trim(split);
1689
1690 bio_advance(bio, split->bi_iter.bi_size);
1691
1692 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1693 bio_set_flag(split, BIO_TRACE_COMPLETION);
1694
1695 return split;
1696}
1697EXPORT_SYMBOL(bio_split);
1698
1699/**
1700 * bio_trim - trim a bio
1701 * @bio: bio to trim
1702 * @offset: number of sectors to trim from the front of @bio
1703 * @size: size we want to trim @bio to, in sectors
1704 *
1705 * This function is typically used for bios that are cloned and submitted
1706 * to the underlying device in parts.
1707 */
1708void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1709{
1710 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1711 offset + size > bio_sectors(bio)))
1712 return;
1713
1714 size <<= 9;
1715 if (offset == 0 && size == bio->bi_iter.bi_size)
1716 return;
1717
1718 bio_advance(bio, offset << 9);
1719 bio->bi_iter.bi_size = size;
1720
1721 if (bio_integrity(bio))
1722 bio_integrity_trim(bio);
1723}
1724EXPORT_SYMBOL_GPL(bio_trim);
1725
1726/*
1727 * create memory pools for biovec's in a bio_set.
1728 * use the global biovec slabs created for general use.
1729 */
1730int biovec_init_pool(mempool_t *pool, int pool_entries)
1731{
1732 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1733
1734 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1735}
1736
1737/*
1738 * bioset_exit - exit a bioset initialized with bioset_init()
1739 *
1740 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1741 * kzalloc()).
1742 */
1743void bioset_exit(struct bio_set *bs)
1744{
1745 bio_alloc_cache_destroy(bs);
1746 if (bs->rescue_workqueue)
1747 destroy_workqueue(bs->rescue_workqueue);
1748 bs->rescue_workqueue = NULL;
1749
1750 mempool_exit(&bs->bio_pool);
1751 mempool_exit(&bs->bvec_pool);
1752
1753 bioset_integrity_free(bs);
1754 if (bs->bio_slab)
1755 bio_put_slab(bs);
1756 bs->bio_slab = NULL;
1757}
1758EXPORT_SYMBOL(bioset_exit);
1759
1760/**
1761 * bioset_init - Initialize a bio_set
1762 * @bs: pool to initialize
1763 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1764 * @front_pad: Number of bytes to allocate in front of the returned bio
1765 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1766 * and %BIOSET_NEED_RESCUER
1767 *
1768 * Description:
1769 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1770 * to ask for a number of bytes to be allocated in front of the bio.
1771 * Front pad allocation is useful for embedding the bio inside
1772 * another structure, to avoid allocating extra data to go with the bio.
1773 * Note that the bio must be embedded at the END of that structure always,
1774 * or things will break badly.
1775 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1776 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1777 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1778 * to dispatch queued requests when the mempool runs out of space.
1779 *
1780 */
1781int bioset_init(struct bio_set *bs,
1782 unsigned int pool_size,
1783 unsigned int front_pad,
1784 int flags)
1785{
1786 bs->front_pad = front_pad;
1787 if (flags & BIOSET_NEED_BVECS)
1788 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1789 else
1790 bs->back_pad = 0;
1791
1792 spin_lock_init(&bs->rescue_lock);
1793 bio_list_init(&bs->rescue_list);
1794 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1795
1796 bs->bio_slab = bio_find_or_create_slab(bs);
1797 if (!bs->bio_slab)
1798 return -ENOMEM;
1799
1800 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1801 goto bad;
1802
1803 if ((flags & BIOSET_NEED_BVECS) &&
1804 biovec_init_pool(&bs->bvec_pool, pool_size))
1805 goto bad;
1806
1807 if (flags & BIOSET_NEED_RESCUER) {
1808 bs->rescue_workqueue = alloc_workqueue("bioset",
1809 WQ_MEM_RECLAIM, 0);
1810 if (!bs->rescue_workqueue)
1811 goto bad;
1812 }
1813 if (flags & BIOSET_PERCPU_CACHE) {
1814 bs->cache = alloc_percpu(struct bio_alloc_cache);
1815 if (!bs->cache)
1816 goto bad;
1817 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1818 }
1819
1820 return 0;
1821bad:
1822 bioset_exit(bs);
1823 return -ENOMEM;
1824}
1825EXPORT_SYMBOL(bioset_init);
1826
1827static int __init init_bio(void)
1828{
1829 int i;
1830
1831 BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1832
1833 bio_integrity_init();
1834
1835 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1836 struct biovec_slab *bvs = bvec_slabs + i;
1837
1838 bvs->slab = kmem_cache_create(bvs->name,
1839 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1840 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1841 }
1842
1843 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1844 bio_cpu_dead);
1845
1846 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1847 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1848 panic("bio: can't allocate bios\n");
1849
1850 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1851 panic("bio: can't create integrity pool\n");
1852
1853 return 0;
1854}
1855subsys_initcall(init_bio);