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(size: sizeof(*bslab), GFP_KERNEL);
89
90 if (!bslab)
91 return NULL;
92
93 snprintf(buf: bslab->name, size: sizeof(bslab->name), fmt: "bio-%d", size);
94 bslab->slab = kmem_cache_create(name: 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(entry: xa_store(&bio_slabs, index: size, entry: bslab, GFP_KERNEL)))
104 return bslab;
105
106 kmem_cache_destroy(s: bslab->slab);
107
108fail_alloc_slab:
109 kfree(objp: 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, index: size);
125 if (bslab)
126 bslab->slab_ref++;
127 else
128 bslab = create_bio_slab(size);
129 mutex_unlock(lock: &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, index: 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, index: slab_size);
155
156 kmem_cache_destroy(s: bslab->slab);
157 kfree(objp: bslab);
158
159out:
160 mutex_unlock(lock: &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(element: bv, pool);
169 else if (nr_vecs > BIO_INLINE_VECS)
170 kmem_cache_free(s: biovec_slab(nr_vecs)->slab, objp: 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: *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(cachep: bvs->slab, flags: bvec_alloc_gfp(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(blkg: 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(pool: &bs->bvec_pool, bv: bio->bi_io_vec, nr_vecs: bio->bi_max_vecs);
238 mempool_free(element: p - bs->front_pad, pool: &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(v: &bio->__bi_remaining, i: 1);
279 atomic_set(v: &bio->__bi_cnt, i: 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(v: &bio->__bi_remaining, i: 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(bio: 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_vecs: nr_pages, opf, gfp_mask: 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(lock: &bs->rescue_lock);
369 bio = bio_list_pop(bl: &bs->rescue_list);
370 spin_unlock(lock: &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(bl: &punt);
398 bio_list_init(bl: &nopunt);
399
400 while ((bio = bio_list_pop(bl: &current->bio_list[0])))
401 bio_list_add(bl: bio->bi_pool == bs ? &punt : &nopunt, bio);
402 current->bio_list[0] = nopunt;
403
404 bio_list_init(bl: &nopunt);
405 while ((bio = bio_list_pop(bl: &current->bio_list[1])))
406 bio_list_add(bl: bio->bi_pool == bs ? &punt : &nopunt, bio);
407 current->bio_list[1] = nopunt;
408
409 spin_lock(lock: &bs->rescue_lock);
410 bio_list_merge(bl: &bs->rescue_list, bl2: &punt);
411 spin_unlock(lock: &bs->rescue_lock);
412
413 queue_work(wq: bs->rescue_workqueue, work: &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: 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(bl: &current->bio_list[0]) ||
539 !bio_list_empty(bl: &current->bio_list[1])) &&
540 bs->rescue_workqueue)
541 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
542
543 p = mempool_alloc(pool: &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(pool: &bs->bio_pool, gfp_mask);
548 }
549 if (unlikely(!p))
550 return NULL;
551 if (!mempool_is_saturated(pool: &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(pool: &bs->bvec_pool, nr_vecs: &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(pool: &bs->bvec_pool, nr_vecs: &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(element: p, pool: &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), flags: 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(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(page: bv.bv_page, start: bv.bv_offset + offset,
651 size: 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(bdev: 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, new_size: 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, nr: -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(state: CPUHP_BIO_DEAD, node: &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, nr: -1U);
755 }
756 free_percpu(pdata: 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(v: &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, bit: 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: bio_src, bit: BIO_REMAPPED))
821 bio_set_flag(bio, bit: BIO_REMAPPED);
822 bio_clone_blkg_association(dst: bio, src: bio_src);
823 }
824
825 if (bio_crypt_clone(dst: bio, src: bio_src, gfp_mask: gfp) < 0)
826 return -ENOMEM;
827 if (bio_integrity(bio: 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(vec1: bv, page))
916 return false;
917 if (!zone_device_pages_have_same_pgmap(a: bv->bv_page, b: 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 (bv->bv_len + len > queue_max_segment_size(q))
948 return false;
949 return bvec_try_merge_page(bv, page, len, off: 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 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
970 return 0;
971
972 if (((bio->bi_iter.bi_size + len) >> SECTOR_SHIFT) > max_sectors)
973 return 0;
974
975 if (bio->bi_vcnt > 0) {
976 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
977
978 if (bvec_try_merge_hw_page(q, bv, page, len, offset,
979 same_page)) {
980 bio->bi_iter.bi_size += len;
981 return len;
982 }
983
984 if (bio->bi_vcnt >=
985 min(bio->bi_max_vecs, queue_max_segments(q)))
986 return 0;
987
988 /*
989 * If the queue doesn't support SG gaps and adding this segment
990 * would create a gap, disallow it.
991 */
992 if (bvec_gap_to_prev(lim: &q->limits, bprv: bv, offset))
993 return 0;
994 }
995
996 bvec_set_page(bv: &bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
997 bio->bi_vcnt++;
998 bio->bi_iter.bi_size += len;
999 return len;
1000}
1001
1002/**
1003 * bio_add_pc_page - attempt to add page to passthrough bio
1004 * @q: the target queue
1005 * @bio: destination bio
1006 * @page: page to add
1007 * @len: vec entry length
1008 * @offset: vec entry offset
1009 *
1010 * Attempt to add a page to the bio_vec maplist. This can fail for a
1011 * number of reasons, such as the bio being full or target block device
1012 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1013 * so it is always possible to add a single page to an empty bio.
1014 *
1015 * This should only be used by passthrough bios.
1016 */
1017int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1018 struct page *page, unsigned int len, unsigned int offset)
1019{
1020 bool same_page = false;
1021 return bio_add_hw_page(q, bio, page, len, offset,
1022 max_sectors: queue_max_hw_sectors(q), same_page: &same_page);
1023}
1024EXPORT_SYMBOL(bio_add_pc_page);
1025
1026/**
1027 * bio_add_zone_append_page - attempt to add page to zone-append bio
1028 * @bio: destination bio
1029 * @page: page to add
1030 * @len: vec entry length
1031 * @offset: vec entry offset
1032 *
1033 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1034 * for a zone-append request. This can fail for a number of reasons, such as the
1035 * bio being full or the target block device is not a zoned block device or
1036 * other limitations of the target block device. The target block device must
1037 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1038 * to an empty bio.
1039 *
1040 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1041 */
1042int bio_add_zone_append_page(struct bio *bio, struct page *page,
1043 unsigned int len, unsigned int offset)
1044{
1045 struct request_queue *q = bdev_get_queue(bdev: bio->bi_bdev);
1046 bool same_page = false;
1047
1048 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1049 return 0;
1050
1051 if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1052 return 0;
1053
1054 return bio_add_hw_page(q, bio, page, len, offset,
1055 max_sectors: queue_max_zone_append_sectors(q), same_page: &same_page);
1056}
1057EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1058
1059/**
1060 * __bio_add_page - add page(s) to a bio in a new segment
1061 * @bio: destination bio
1062 * @page: start page to add
1063 * @len: length of the data to add, may cross pages
1064 * @off: offset of the data relative to @page, may cross pages
1065 *
1066 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1067 * that @bio has space for another bvec.
1068 */
1069void __bio_add_page(struct bio *bio, struct page *page,
1070 unsigned int len, unsigned int off)
1071{
1072 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1073 WARN_ON_ONCE(bio_full(bio, len));
1074
1075 bvec_set_page(bv: &bio->bi_io_vec[bio->bi_vcnt], page, len, offset: off);
1076 bio->bi_iter.bi_size += len;
1077 bio->bi_vcnt++;
1078}
1079EXPORT_SYMBOL_GPL(__bio_add_page);
1080
1081/**
1082 * bio_add_page - attempt to add page(s) to bio
1083 * @bio: destination bio
1084 * @page: start page to add
1085 * @len: vec entry length, may cross pages
1086 * @offset: vec entry offset relative to @page, may cross pages
1087 *
1088 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1089 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1090 */
1091int bio_add_page(struct bio *bio, struct page *page,
1092 unsigned int len, unsigned int offset)
1093{
1094 bool same_page = false;
1095
1096 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1097 return 0;
1098 if (bio->bi_iter.bi_size > UINT_MAX - len)
1099 return 0;
1100
1101 if (bio->bi_vcnt > 0 &&
1102 bvec_try_merge_page(bv: &bio->bi_io_vec[bio->bi_vcnt - 1],
1103 page, len, off: offset, same_page: &same_page)) {
1104 bio->bi_iter.bi_size += len;
1105 return len;
1106 }
1107
1108 if (bio->bi_vcnt >= bio->bi_max_vecs)
1109 return 0;
1110 __bio_add_page(bio, page, len, offset);
1111 return len;
1112}
1113EXPORT_SYMBOL(bio_add_page);
1114
1115void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1116 size_t off)
1117{
1118 WARN_ON_ONCE(len > UINT_MAX);
1119 WARN_ON_ONCE(off > UINT_MAX);
1120 __bio_add_page(bio, &folio->page, len, off);
1121}
1122
1123/**
1124 * bio_add_folio - Attempt to add part of a folio to a bio.
1125 * @bio: BIO to add to.
1126 * @folio: Folio to add.
1127 * @len: How many bytes from the folio to add.
1128 * @off: First byte in this folio to add.
1129 *
1130 * Filesystems that use folios can call this function instead of calling
1131 * bio_add_page() for each page in the folio. If @off is bigger than
1132 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1133 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1134 *
1135 * Return: Whether the addition was successful.
1136 */
1137bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1138 size_t off)
1139{
1140 if (len > UINT_MAX || off > UINT_MAX)
1141 return false;
1142 return bio_add_page(bio, &folio->page, len, off) > 0;
1143}
1144EXPORT_SYMBOL(bio_add_folio);
1145
1146void __bio_release_pages(struct bio *bio, bool mark_dirty)
1147{
1148 struct bvec_iter_all iter_all;
1149 struct bio_vec *bvec;
1150
1151 bio_for_each_segment_all(bvec, bio, iter_all) {
1152 if (mark_dirty && !PageCompound(page: bvec->bv_page))
1153 set_page_dirty_lock(bvec->bv_page);
1154 bio_release_page(bio, page: bvec->bv_page);
1155 }
1156}
1157EXPORT_SYMBOL_GPL(__bio_release_pages);
1158
1159void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1160{
1161 size_t size = iov_iter_count(i: iter);
1162
1163 WARN_ON_ONCE(bio->bi_max_vecs);
1164
1165 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1166 struct request_queue *q = bdev_get_queue(bdev: bio->bi_bdev);
1167 size_t max_sectors = queue_max_zone_append_sectors(q);
1168
1169 size = min(size, max_sectors << SECTOR_SHIFT);
1170 }
1171
1172 bio->bi_vcnt = iter->nr_segs;
1173 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1174 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1175 bio->bi_iter.bi_size = size;
1176 bio_set_flag(bio, bit: BIO_CLONED);
1177}
1178
1179static int bio_iov_add_page(struct bio *bio, struct page *page,
1180 unsigned int len, unsigned int offset)
1181{
1182 bool same_page = false;
1183
1184 if (WARN_ON_ONCE(bio->bi_iter.bi_size > UINT_MAX - len))
1185 return -EIO;
1186
1187 if (bio->bi_vcnt > 0 &&
1188 bvec_try_merge_page(bv: &bio->bi_io_vec[bio->bi_vcnt - 1],
1189 page, len, off: offset, same_page: &same_page)) {
1190 bio->bi_iter.bi_size += len;
1191 if (same_page)
1192 bio_release_page(bio, page);
1193 return 0;
1194 }
1195 __bio_add_page(bio, page, len, offset);
1196 return 0;
1197}
1198
1199static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1200 unsigned int len, unsigned int offset)
1201{
1202 struct request_queue *q = bdev_get_queue(bdev: bio->bi_bdev);
1203 bool same_page = false;
1204
1205 if (bio_add_hw_page(q, bio, page, len, offset,
1206 max_sectors: queue_max_zone_append_sectors(q), same_page: &same_page) != len)
1207 return -EINVAL;
1208 if (same_page)
1209 bio_release_page(bio, page);
1210 return 0;
1211}
1212
1213#define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1214
1215/**
1216 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1217 * @bio: bio to add pages to
1218 * @iter: iov iterator describing the region to be mapped
1219 *
1220 * Extracts pages from *iter and appends them to @bio's bvec array. The pages
1221 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.
1222 * For a multi-segment *iter, this function only adds pages from the next
1223 * non-empty segment of the iov iterator.
1224 */
1225static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1226{
1227 iov_iter_extraction_t extraction_flags = 0;
1228 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1229 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1230 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1231 struct page **pages = (struct page **)bv;
1232 ssize_t size, left;
1233 unsigned len, i = 0;
1234 size_t offset;
1235 int ret = 0;
1236
1237 /*
1238 * Move page array up in the allocated memory for the bio vecs as far as
1239 * possible so that we can start filling biovecs from the beginning
1240 * without overwriting the temporary page array.
1241 */
1242 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1243 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1244
1245 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1246 extraction_flags |= ITER_ALLOW_P2PDMA;
1247
1248 /*
1249 * Each segment in the iov is required to be a block size multiple.
1250 * However, we may not be able to get the entire segment if it spans
1251 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1252 * result to ensure the bio's total size is correct. The remainder of
1253 * the iov data will be picked up in the next bio iteration.
1254 */
1255 size = iov_iter_extract_pages(i: iter, pages: &pages,
1256 UINT_MAX - bio->bi_iter.bi_size,
1257 maxpages: nr_pages, extraction_flags, offset0: &offset);
1258 if (unlikely(size <= 0))
1259 return size ? size : -EFAULT;
1260
1261 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1262
1263 if (bio->bi_bdev) {
1264 size_t trim = size & (bdev_logical_block_size(bdev: bio->bi_bdev) - 1);
1265 iov_iter_revert(i: iter, bytes: trim);
1266 size -= trim;
1267 }
1268
1269 if (unlikely(!size)) {
1270 ret = -EFAULT;
1271 goto out;
1272 }
1273
1274 for (left = size, i = 0; left > 0; left -= len, i++) {
1275 struct page *page = pages[i];
1276
1277 len = min_t(size_t, PAGE_SIZE - offset, left);
1278 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1279 ret = bio_iov_add_zone_append_page(bio, page, len,
1280 offset);
1281 if (ret)
1282 break;
1283 } else
1284 bio_iov_add_page(bio, page, len, offset);
1285
1286 offset = 0;
1287 }
1288
1289 iov_iter_revert(i: iter, bytes: left);
1290out:
1291 while (i < nr_pages)
1292 bio_release_page(bio, page: pages[i++]);
1293
1294 return ret;
1295}
1296
1297/**
1298 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1299 * @bio: bio to add pages to
1300 * @iter: iov iterator describing the region to be added
1301 *
1302 * This takes either an iterator pointing to user memory, or one pointing to
1303 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1304 * map them into the kernel. On IO completion, the caller should put those
1305 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1306 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1307 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1308 * completed by a call to ->ki_complete() or returns with an error other than
1309 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1310 * on IO completion. If it isn't, then pages should be released.
1311 *
1312 * The function tries, but does not guarantee, to pin as many pages as
1313 * fit into the bio, or are requested in @iter, whatever is smaller. If
1314 * MM encounters an error pinning the requested pages, it stops. Error
1315 * is returned only if 0 pages could be pinned.
1316 */
1317int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1318{
1319 int ret = 0;
1320
1321 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1322 return -EIO;
1323
1324 if (iov_iter_is_bvec(i: iter)) {
1325 bio_iov_bvec_set(bio, iter);
1326 iov_iter_advance(i: iter, bytes: bio->bi_iter.bi_size);
1327 return 0;
1328 }
1329
1330 if (iov_iter_extract_will_pin(iter))
1331 bio_set_flag(bio, bit: BIO_PAGE_PINNED);
1332 do {
1333 ret = __bio_iov_iter_get_pages(bio, iter);
1334 } while (!ret && iov_iter_count(i: iter) && !bio_full(bio, len: 0));
1335
1336 return bio->bi_vcnt ? 0 : ret;
1337}
1338EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1339
1340static void submit_bio_wait_endio(struct bio *bio)
1341{
1342 complete(bio->bi_private);
1343}
1344
1345/**
1346 * submit_bio_wait - submit a bio, and wait until it completes
1347 * @bio: The &struct bio which describes the I/O
1348 *
1349 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1350 * bio_endio() on failure.
1351 *
1352 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1353 * result in bio reference to be consumed. The caller must drop the reference
1354 * on his own.
1355 */
1356int submit_bio_wait(struct bio *bio)
1357{
1358 DECLARE_COMPLETION_ONSTACK_MAP(done,
1359 bio->bi_bdev->bd_disk->lockdep_map);
1360 unsigned long hang_check;
1361
1362 bio->bi_private = &done;
1363 bio->bi_end_io = submit_bio_wait_endio;
1364 bio->bi_opf |= REQ_SYNC;
1365 submit_bio(bio);
1366
1367 /* Prevent hang_check timer from firing at us during very long I/O */
1368 hang_check = sysctl_hung_task_timeout_secs;
1369 if (hang_check)
1370 while (!wait_for_completion_io_timeout(x: &done,
1371 timeout: hang_check * (HZ/2)))
1372 ;
1373 else
1374 wait_for_completion_io(&done);
1375
1376 return blk_status_to_errno(status: bio->bi_status);
1377}
1378EXPORT_SYMBOL(submit_bio_wait);
1379
1380void __bio_advance(struct bio *bio, unsigned bytes)
1381{
1382 if (bio_integrity(bio))
1383 bio_integrity_advance(bio, bytes);
1384
1385 bio_crypt_advance(bio, bytes);
1386 bio_advance_iter(bio, iter: &bio->bi_iter, bytes);
1387}
1388EXPORT_SYMBOL(__bio_advance);
1389
1390void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1391 struct bio *src, struct bvec_iter *src_iter)
1392{
1393 while (src_iter->bi_size && dst_iter->bi_size) {
1394 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1395 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1396 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1397 void *src_buf = bvec_kmap_local(bvec: &src_bv);
1398 void *dst_buf = bvec_kmap_local(bvec: &dst_bv);
1399
1400 memcpy(dst_buf, src_buf, bytes);
1401
1402 kunmap_local(dst_buf);
1403 kunmap_local(src_buf);
1404
1405 bio_advance_iter_single(bio: src, iter: src_iter, bytes);
1406 bio_advance_iter_single(bio: dst, iter: dst_iter, bytes);
1407 }
1408}
1409EXPORT_SYMBOL(bio_copy_data_iter);
1410
1411/**
1412 * bio_copy_data - copy contents of data buffers from one bio to another
1413 * @src: source bio
1414 * @dst: destination bio
1415 *
1416 * Stops when it reaches the end of either @src or @dst - that is, copies
1417 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1418 */
1419void bio_copy_data(struct bio *dst, struct bio *src)
1420{
1421 struct bvec_iter src_iter = src->bi_iter;
1422 struct bvec_iter dst_iter = dst->bi_iter;
1423
1424 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1425}
1426EXPORT_SYMBOL(bio_copy_data);
1427
1428void bio_free_pages(struct bio *bio)
1429{
1430 struct bio_vec *bvec;
1431 struct bvec_iter_all iter_all;
1432
1433 bio_for_each_segment_all(bvec, bio, iter_all)
1434 __free_page(bvec->bv_page);
1435}
1436EXPORT_SYMBOL(bio_free_pages);
1437
1438/*
1439 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1440 * for performing direct-IO in BIOs.
1441 *
1442 * The problem is that we cannot run set_page_dirty() from interrupt context
1443 * because the required locks are not interrupt-safe. So what we can do is to
1444 * mark the pages dirty _before_ performing IO. And in interrupt context,
1445 * check that the pages are still dirty. If so, fine. If not, redirty them
1446 * in process context.
1447 *
1448 * We special-case compound pages here: normally this means reads into hugetlb
1449 * pages. The logic in here doesn't really work right for compound pages
1450 * because the VM does not uniformly chase down the head page in all cases.
1451 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1452 * handle them at all. So we skip compound pages here at an early stage.
1453 *
1454 * Note that this code is very hard to test under normal circumstances because
1455 * direct-io pins the pages with get_user_pages(). This makes
1456 * is_page_cache_freeable return false, and the VM will not clean the pages.
1457 * But other code (eg, flusher threads) could clean the pages if they are mapped
1458 * pagecache.
1459 *
1460 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1461 * deferred bio dirtying paths.
1462 */
1463
1464/*
1465 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1466 */
1467void bio_set_pages_dirty(struct bio *bio)
1468{
1469 struct bio_vec *bvec;
1470 struct bvec_iter_all iter_all;
1471
1472 bio_for_each_segment_all(bvec, bio, iter_all) {
1473 if (!PageCompound(page: bvec->bv_page))
1474 set_page_dirty_lock(bvec->bv_page);
1475 }
1476}
1477EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1478
1479/*
1480 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1481 * If they are, then fine. If, however, some pages are clean then they must
1482 * have been written out during the direct-IO read. So we take another ref on
1483 * the BIO and re-dirty the pages in process context.
1484 *
1485 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1486 * here on. It will unpin each page and will run one bio_put() against the
1487 * BIO.
1488 */
1489
1490static void bio_dirty_fn(struct work_struct *work);
1491
1492static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1493static DEFINE_SPINLOCK(bio_dirty_lock);
1494static struct bio *bio_dirty_list;
1495
1496/*
1497 * This runs in process context
1498 */
1499static void bio_dirty_fn(struct work_struct *work)
1500{
1501 struct bio *bio, *next;
1502
1503 spin_lock_irq(lock: &bio_dirty_lock);
1504 next = bio_dirty_list;
1505 bio_dirty_list = NULL;
1506 spin_unlock_irq(lock: &bio_dirty_lock);
1507
1508 while ((bio = next) != NULL) {
1509 next = bio->bi_private;
1510
1511 bio_release_pages(bio, mark_dirty: true);
1512 bio_put(bio);
1513 }
1514}
1515
1516void bio_check_pages_dirty(struct bio *bio)
1517{
1518 struct bio_vec *bvec;
1519 unsigned long flags;
1520 struct bvec_iter_all iter_all;
1521
1522 bio_for_each_segment_all(bvec, bio, iter_all) {
1523 if (!PageDirty(page: bvec->bv_page) && !PageCompound(page: bvec->bv_page))
1524 goto defer;
1525 }
1526
1527 bio_release_pages(bio, mark_dirty: false);
1528 bio_put(bio);
1529 return;
1530defer:
1531 spin_lock_irqsave(&bio_dirty_lock, flags);
1532 bio->bi_private = bio_dirty_list;
1533 bio_dirty_list = bio;
1534 spin_unlock_irqrestore(lock: &bio_dirty_lock, flags);
1535 schedule_work(work: &bio_dirty_work);
1536}
1537EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1538
1539static inline bool bio_remaining_done(struct bio *bio)
1540{
1541 /*
1542 * If we're not chaining, then ->__bi_remaining is always 1 and
1543 * we always end io on the first invocation.
1544 */
1545 if (!bio_flagged(bio, bit: BIO_CHAIN))
1546 return true;
1547
1548 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1549
1550 if (atomic_dec_and_test(v: &bio->__bi_remaining)) {
1551 bio_clear_flag(bio, bit: BIO_CHAIN);
1552 return true;
1553 }
1554
1555 return false;
1556}
1557
1558/**
1559 * bio_endio - end I/O on a bio
1560 * @bio: bio
1561 *
1562 * Description:
1563 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1564 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1565 * bio unless they own it and thus know that it has an end_io function.
1566 *
1567 * bio_endio() can be called several times on a bio that has been chained
1568 * using bio_chain(). The ->bi_end_io() function will only be called the
1569 * last time.
1570 **/
1571void bio_endio(struct bio *bio)
1572{
1573again:
1574 if (!bio_remaining_done(bio))
1575 return;
1576 if (!bio_integrity_endio(bio))
1577 return;
1578
1579 rq_qos_done_bio(bio);
1580
1581 if (bio->bi_bdev && bio_flagged(bio, bit: BIO_TRACE_COMPLETION)) {
1582 trace_block_bio_complete(q: bdev_get_queue(bdev: bio->bi_bdev), bio);
1583 bio_clear_flag(bio, bit: BIO_TRACE_COMPLETION);
1584 }
1585
1586 /*
1587 * Need to have a real endio function for chained bios, otherwise
1588 * various corner cases will break (like stacking block devices that
1589 * save/restore bi_end_io) - however, we want to avoid unbounded
1590 * recursion and blowing the stack. Tail call optimization would
1591 * handle this, but compiling with frame pointers also disables
1592 * gcc's sibling call optimization.
1593 */
1594 if (bio->bi_end_io == bio_chain_endio) {
1595 bio = __bio_chain_endio(bio);
1596 goto again;
1597 }
1598
1599 blk_throtl_bio_endio(bio);
1600 /* release cgroup info */
1601 bio_uninit(bio);
1602 if (bio->bi_end_io)
1603 bio->bi_end_io(bio);
1604}
1605EXPORT_SYMBOL(bio_endio);
1606
1607/**
1608 * bio_split - split a bio
1609 * @bio: bio to split
1610 * @sectors: number of sectors to split from the front of @bio
1611 * @gfp: gfp mask
1612 * @bs: bio set to allocate from
1613 *
1614 * Allocates and returns a new bio which represents @sectors from the start of
1615 * @bio, and updates @bio to represent the remaining sectors.
1616 *
1617 * Unless this is a discard request the newly allocated bio will point
1618 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1619 * neither @bio nor @bs are freed before the split bio.
1620 */
1621struct bio *bio_split(struct bio *bio, int sectors,
1622 gfp_t gfp, struct bio_set *bs)
1623{
1624 struct bio *split;
1625
1626 BUG_ON(sectors <= 0);
1627 BUG_ON(sectors >= bio_sectors(bio));
1628
1629 /* Zone append commands cannot be split */
1630 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1631 return NULL;
1632
1633 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1634 if (!split)
1635 return NULL;
1636
1637 split->bi_iter.bi_size = sectors << 9;
1638
1639 if (bio_integrity(bio: split))
1640 bio_integrity_trim(split);
1641
1642 bio_advance(bio, nbytes: split->bi_iter.bi_size);
1643
1644 if (bio_flagged(bio, bit: BIO_TRACE_COMPLETION))
1645 bio_set_flag(bio: split, bit: BIO_TRACE_COMPLETION);
1646
1647 return split;
1648}
1649EXPORT_SYMBOL(bio_split);
1650
1651/**
1652 * bio_trim - trim a bio
1653 * @bio: bio to trim
1654 * @offset: number of sectors to trim from the front of @bio
1655 * @size: size we want to trim @bio to, in sectors
1656 *
1657 * This function is typically used for bios that are cloned and submitted
1658 * to the underlying device in parts.
1659 */
1660void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1661{
1662 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1663 offset + size > bio_sectors(bio)))
1664 return;
1665
1666 size <<= 9;
1667 if (offset == 0 && size == bio->bi_iter.bi_size)
1668 return;
1669
1670 bio_advance(bio, nbytes: offset << 9);
1671 bio->bi_iter.bi_size = size;
1672
1673 if (bio_integrity(bio))
1674 bio_integrity_trim(bio);
1675}
1676EXPORT_SYMBOL_GPL(bio_trim);
1677
1678/*
1679 * create memory pools for biovec's in a bio_set.
1680 * use the global biovec slabs created for general use.
1681 */
1682int biovec_init_pool(mempool_t *pool, int pool_entries)
1683{
1684 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1685
1686 return mempool_init_slab_pool(pool, min_nr: pool_entries, kc: bp->slab);
1687}
1688
1689/*
1690 * bioset_exit - exit a bioset initialized with bioset_init()
1691 *
1692 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1693 * kzalloc()).
1694 */
1695void bioset_exit(struct bio_set *bs)
1696{
1697 bio_alloc_cache_destroy(bs);
1698 if (bs->rescue_workqueue)
1699 destroy_workqueue(wq: bs->rescue_workqueue);
1700 bs->rescue_workqueue = NULL;
1701
1702 mempool_exit(pool: &bs->bio_pool);
1703 mempool_exit(pool: &bs->bvec_pool);
1704
1705 bioset_integrity_free(bs);
1706 if (bs->bio_slab)
1707 bio_put_slab(bs);
1708 bs->bio_slab = NULL;
1709}
1710EXPORT_SYMBOL(bioset_exit);
1711
1712/**
1713 * bioset_init - Initialize a bio_set
1714 * @bs: pool to initialize
1715 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1716 * @front_pad: Number of bytes to allocate in front of the returned bio
1717 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1718 * and %BIOSET_NEED_RESCUER
1719 *
1720 * Description:
1721 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1722 * to ask for a number of bytes to be allocated in front of the bio.
1723 * Front pad allocation is useful for embedding the bio inside
1724 * another structure, to avoid allocating extra data to go with the bio.
1725 * Note that the bio must be embedded at the END of that structure always,
1726 * or things will break badly.
1727 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1728 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1729 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1730 * to dispatch queued requests when the mempool runs out of space.
1731 *
1732 */
1733int bioset_init(struct bio_set *bs,
1734 unsigned int pool_size,
1735 unsigned int front_pad,
1736 int flags)
1737{
1738 bs->front_pad = front_pad;
1739 if (flags & BIOSET_NEED_BVECS)
1740 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1741 else
1742 bs->back_pad = 0;
1743
1744 spin_lock_init(&bs->rescue_lock);
1745 bio_list_init(bl: &bs->rescue_list);
1746 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1747
1748 bs->bio_slab = bio_find_or_create_slab(bs);
1749 if (!bs->bio_slab)
1750 return -ENOMEM;
1751
1752 if (mempool_init_slab_pool(pool: &bs->bio_pool, min_nr: pool_size, kc: bs->bio_slab))
1753 goto bad;
1754
1755 if ((flags & BIOSET_NEED_BVECS) &&
1756 biovec_init_pool(pool: &bs->bvec_pool, pool_entries: pool_size))
1757 goto bad;
1758
1759 if (flags & BIOSET_NEED_RESCUER) {
1760 bs->rescue_workqueue = alloc_workqueue(fmt: "bioset",
1761 flags: WQ_MEM_RECLAIM, max_active: 0);
1762 if (!bs->rescue_workqueue)
1763 goto bad;
1764 }
1765 if (flags & BIOSET_PERCPU_CACHE) {
1766 bs->cache = alloc_percpu(struct bio_alloc_cache);
1767 if (!bs->cache)
1768 goto bad;
1769 cpuhp_state_add_instance_nocalls(state: CPUHP_BIO_DEAD, node: &bs->cpuhp_dead);
1770 }
1771
1772 return 0;
1773bad:
1774 bioset_exit(bs);
1775 return -ENOMEM;
1776}
1777EXPORT_SYMBOL(bioset_init);
1778
1779static int __init init_bio(void)
1780{
1781 int i;
1782
1783 BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1784
1785 bio_integrity_init();
1786
1787 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1788 struct biovec_slab *bvs = bvec_slabs + i;
1789
1790 bvs->slab = kmem_cache_create(name: bvs->name,
1791 size: bvs->nr_vecs * sizeof(struct bio_vec), align: 0,
1792 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1793 }
1794
1795 cpuhp_setup_state_multi(state: CPUHP_BIO_DEAD, name: "block/bio:dead", NULL,
1796 teardown: bio_cpu_dead);
1797
1798 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1799 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1800 panic(fmt: "bio: can't allocate bios\n");
1801
1802 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1803 panic(fmt: "bio: can't create integrity pool\n");
1804
1805 return 0;
1806}
1807subsys_initcall(init_bio);
1808

source code of linux/block/bio.c