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/blk-cgroup.h>
19 #include <linux/highmem.h>
20 #include <linux/sched/sysctl.h>
21 #include <linux/blk-crypto.h>
22
23 #include <trace/events/block.h>
24 #include "blk.h"
25 #include "blk-rq-qos.h"
26
27 /*
28 * Test patch to inline a certain number of bi_io_vec's inside the bio
29 * itself, to shrink a bio data allocation from two mempool calls to one
30 */
31 #define BIO_INLINE_VECS 4
32
33 /*
34 * if you change this list, also change bvec_alloc or things will
35 * break badly! cannot be bigger than what you can fit into an
36 * unsigned short
37 */
38 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
39 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
40 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
41 };
42 #undef BV
43
44 /*
45 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
46 * IO code that does not need private memory pools.
47 */
48 struct bio_set fs_bio_set;
49 EXPORT_SYMBOL(fs_bio_set);
50
51 /*
52 * Our slab pool management
53 */
54 struct bio_slab {
55 struct kmem_cache *slab;
56 unsigned int slab_ref;
57 unsigned int slab_size;
58 char name[8];
59 };
60 static DEFINE_MUTEX(bio_slab_lock);
61 static struct bio_slab *bio_slabs;
62 static unsigned int bio_slab_nr, bio_slab_max;
63
bio_find_or_create_slab(unsigned int extra_size)64 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
65 {
66 unsigned int sz = sizeof(struct bio) + extra_size;
67 struct kmem_cache *slab = NULL;
68 struct bio_slab *bslab, *new_bio_slabs;
69 unsigned int new_bio_slab_max;
70 unsigned int i, entry = -1;
71
72 mutex_lock(&bio_slab_lock);
73
74 i = 0;
75 while (i < bio_slab_nr) {
76 bslab = &bio_slabs[i];
77
78 if (!bslab->slab && entry == -1)
79 entry = i;
80 else if (bslab->slab_size == sz) {
81 slab = bslab->slab;
82 bslab->slab_ref++;
83 break;
84 }
85 i++;
86 }
87
88 if (slab)
89 goto out_unlock;
90
91 if (bio_slab_nr == bio_slab_max && entry == -1) {
92 new_bio_slab_max = bio_slab_max << 1;
93 new_bio_slabs = krealloc(bio_slabs,
94 new_bio_slab_max * sizeof(struct bio_slab),
95 GFP_KERNEL);
96 if (!new_bio_slabs)
97 goto out_unlock;
98 bio_slab_max = new_bio_slab_max;
99 bio_slabs = new_bio_slabs;
100 }
101 if (entry == -1)
102 entry = bio_slab_nr++;
103
104 bslab = &bio_slabs[entry];
105
106 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
107 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
108 SLAB_HWCACHE_ALIGN, NULL);
109 if (!slab)
110 goto out_unlock;
111
112 bslab->slab = slab;
113 bslab->slab_ref = 1;
114 bslab->slab_size = sz;
115 out_unlock:
116 mutex_unlock(&bio_slab_lock);
117 return slab;
118 }
119
bio_put_slab(struct bio_set * bs)120 static void bio_put_slab(struct bio_set *bs)
121 {
122 struct bio_slab *bslab = NULL;
123 unsigned int i;
124
125 mutex_lock(&bio_slab_lock);
126
127 for (i = 0; i < bio_slab_nr; i++) {
128 if (bs->bio_slab == bio_slabs[i].slab) {
129 bslab = &bio_slabs[i];
130 break;
131 }
132 }
133
134 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
135 goto out;
136
137 WARN_ON(!bslab->slab_ref);
138
139 if (--bslab->slab_ref)
140 goto out;
141
142 kmem_cache_destroy(bslab->slab);
143 bslab->slab = NULL;
144
145 out:
146 mutex_unlock(&bio_slab_lock);
147 }
148
bvec_nr_vecs(unsigned short idx)149 unsigned int bvec_nr_vecs(unsigned short idx)
150 {
151 return bvec_slabs[--idx].nr_vecs;
152 }
153
bvec_free(mempool_t * pool,struct bio_vec * bv,unsigned int idx)154 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
155 {
156 if (!idx)
157 return;
158 idx--;
159
160 BIO_BUG_ON(idx >= BVEC_POOL_NR);
161
162 if (idx == BVEC_POOL_MAX) {
163 mempool_free(bv, pool);
164 } else {
165 struct biovec_slab *bvs = bvec_slabs + idx;
166
167 kmem_cache_free(bvs->slab, bv);
168 }
169 }
170
bvec_alloc(gfp_t gfp_mask,int nr,unsigned long * idx,mempool_t * pool)171 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
172 mempool_t *pool)
173 {
174 struct bio_vec *bvl;
175
176 /*
177 * see comment near bvec_array define!
178 */
179 switch (nr) {
180 case 1:
181 *idx = 0;
182 break;
183 case 2 ... 4:
184 *idx = 1;
185 break;
186 case 5 ... 16:
187 *idx = 2;
188 break;
189 case 17 ... 64:
190 *idx = 3;
191 break;
192 case 65 ... 128:
193 *idx = 4;
194 break;
195 case 129 ... BIO_MAX_PAGES:
196 *idx = 5;
197 break;
198 default:
199 return NULL;
200 }
201
202 /*
203 * idx now points to the pool we want to allocate from. only the
204 * 1-vec entry pool is mempool backed.
205 */
206 if (*idx == BVEC_POOL_MAX) {
207 fallback:
208 bvl = mempool_alloc(pool, gfp_mask);
209 } else {
210 struct biovec_slab *bvs = bvec_slabs + *idx;
211 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
212
213 /*
214 * Make this allocation restricted and don't dump info on
215 * allocation failures, since we'll fallback to the mempool
216 * in case of failure.
217 */
218 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
219
220 /*
221 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
222 * is set, retry with the 1-entry mempool
223 */
224 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
225 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
226 *idx = BVEC_POOL_MAX;
227 goto fallback;
228 }
229 }
230
231 (*idx)++;
232 return bvl;
233 }
234
bio_uninit(struct bio * bio)235 void bio_uninit(struct bio *bio)
236 {
237 #ifdef CONFIG_BLK_CGROUP
238 if (bio->bi_blkg) {
239 blkg_put(bio->bi_blkg);
240 bio->bi_blkg = NULL;
241 }
242 #endif
243 if (bio_integrity(bio))
244 bio_integrity_free(bio);
245
246 bio_crypt_free_ctx(bio);
247 }
248 EXPORT_SYMBOL(bio_uninit);
249
bio_free(struct bio * bio)250 static void bio_free(struct bio *bio)
251 {
252 struct bio_set *bs = bio->bi_pool;
253 void *p;
254
255 bio_uninit(bio);
256
257 if (bs) {
258 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
259
260 /*
261 * If we have front padding, adjust the bio pointer before freeing
262 */
263 p = bio;
264 p -= bs->front_pad;
265
266 mempool_free(p, &bs->bio_pool);
267 } else {
268 /* Bio was allocated by bio_kmalloc() */
269 kfree(bio);
270 }
271 }
272
273 /*
274 * Users of this function have their own bio allocation. Subsequently,
275 * they must remember to pair any call to bio_init() with bio_uninit()
276 * when IO has completed, or when the bio is released.
277 */
bio_init(struct bio * bio,struct bio_vec * table,unsigned short max_vecs)278 void bio_init(struct bio *bio, struct bio_vec *table,
279 unsigned short max_vecs)
280 {
281 memset(bio, 0, sizeof(*bio));
282 atomic_set(&bio->__bi_remaining, 1);
283 atomic_set(&bio->__bi_cnt, 1);
284
285 bio->bi_io_vec = table;
286 bio->bi_max_vecs = max_vecs;
287 }
288 EXPORT_SYMBOL(bio_init);
289
290 /**
291 * bio_reset - reinitialize a bio
292 * @bio: bio to reset
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 */
bio_reset(struct bio * bio)300 void bio_reset(struct bio *bio)
301 {
302 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
303
304 bio_uninit(bio);
305
306 memset(bio, 0, BIO_RESET_BYTES);
307 bio->bi_flags = flags;
308 atomic_set(&bio->__bi_remaining, 1);
309 }
310 EXPORT_SYMBOL(bio_reset);
311
__bio_chain_endio(struct bio * bio)312 static 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
bio_chain_endio(struct bio * bio)322 static 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 */
bio_chain(struct bio * bio,struct bio * parent)338 void 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 }
346 EXPORT_SYMBOL(bio_chain);
347
bio_alloc_rescue(struct work_struct * work)348 static void bio_alloc_rescue(struct work_struct *work)
349 {
350 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
351 struct bio *bio;
352
353 while (1) {
354 spin_lock(&bs->rescue_lock);
355 bio = bio_list_pop(&bs->rescue_list);
356 spin_unlock(&bs->rescue_lock);
357
358 if (!bio)
359 break;
360
361 submit_bio_noacct(bio);
362 }
363 }
364
punt_bios_to_rescuer(struct bio_set * bs)365 static void punt_bios_to_rescuer(struct bio_set *bs)
366 {
367 struct bio_list punt, nopunt;
368 struct bio *bio;
369
370 if (WARN_ON_ONCE(!bs->rescue_workqueue))
371 return;
372 /*
373 * In order to guarantee forward progress we must punt only bios that
374 * were allocated from this bio_set; otherwise, if there was a bio on
375 * there for a stacking driver higher up in the stack, processing it
376 * could require allocating bios from this bio_set, and doing that from
377 * our own rescuer would be bad.
378 *
379 * Since bio lists are singly linked, pop them all instead of trying to
380 * remove from the middle of the list:
381 */
382
383 bio_list_init(&punt);
384 bio_list_init(&nopunt);
385
386 while ((bio = bio_list_pop(¤t->bio_list[0])))
387 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
388 current->bio_list[0] = nopunt;
389
390 bio_list_init(&nopunt);
391 while ((bio = bio_list_pop(¤t->bio_list[1])))
392 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
393 current->bio_list[1] = nopunt;
394
395 spin_lock(&bs->rescue_lock);
396 bio_list_merge(&bs->rescue_list, &punt);
397 spin_unlock(&bs->rescue_lock);
398
399 queue_work(bs->rescue_workqueue, &bs->rescue_work);
400 }
401
402 /**
403 * bio_alloc_bioset - allocate a bio for I/O
404 * @gfp_mask: the GFP_* mask given to the slab allocator
405 * @nr_iovecs: number of iovecs to pre-allocate
406 * @bs: the bio_set to allocate from.
407 *
408 * Description:
409 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
410 * backed by the @bs's mempool.
411 *
412 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
413 * always be able to allocate a bio. This is due to the mempool guarantees.
414 * To make this work, callers must never allocate more than 1 bio at a time
415 * from this pool. Callers that need to allocate more than 1 bio must always
416 * submit the previously allocated bio for IO before attempting to allocate
417 * a new one. Failure to do so can cause deadlocks under memory pressure.
418 *
419 * Note that when running under submit_bio_noacct() (i.e. any block
420 * driver), bios are not submitted until after you return - see the code in
421 * submit_bio_noacct() that converts recursion into iteration, to prevent
422 * stack overflows.
423 *
424 * This would normally mean allocating multiple bios under
425 * submit_bio_noacct() would be susceptible to deadlocks, but we have
426 * deadlock avoidance code that resubmits any blocked bios from a rescuer
427 * thread.
428 *
429 * However, we do not guarantee forward progress for allocations from other
430 * mempools. Doing multiple allocations from the same mempool under
431 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
432 * for per bio allocations.
433 *
434 * RETURNS:
435 * Pointer to new bio on success, NULL on failure.
436 */
bio_alloc_bioset(gfp_t gfp_mask,unsigned int nr_iovecs,struct bio_set * bs)437 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
438 struct bio_set *bs)
439 {
440 gfp_t saved_gfp = gfp_mask;
441 unsigned front_pad;
442 unsigned inline_vecs;
443 struct bio_vec *bvl = NULL;
444 struct bio *bio;
445 void *p;
446
447 if (!bs) {
448 if (nr_iovecs > UIO_MAXIOV)
449 return NULL;
450
451 p = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask);
452 front_pad = 0;
453 inline_vecs = nr_iovecs;
454 } else {
455 /* should not use nobvec bioset for nr_iovecs > 0 */
456 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
457 nr_iovecs > 0))
458 return NULL;
459 /*
460 * submit_bio_noacct() converts recursion to iteration; this
461 * means if we're running beneath it, any bios we allocate and
462 * submit will not be submitted (and thus freed) until after we
463 * return.
464 *
465 * This exposes us to a potential deadlock if we allocate
466 * multiple bios from the same bio_set() while running
467 * underneath submit_bio_noacct(). If we were to allocate
468 * multiple bios (say a stacking block driver that was splitting
469 * bios), we would deadlock if we exhausted the mempool's
470 * reserve.
471 *
472 * We solve this, and guarantee forward progress, with a rescuer
473 * workqueue per bio_set. If we go to allocate and there are
474 * bios on current->bio_list, we first try the allocation
475 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
476 * bios we would be blocking to the rescuer workqueue before
477 * we retry with the original gfp_flags.
478 */
479
480 if (current->bio_list &&
481 (!bio_list_empty(¤t->bio_list[0]) ||
482 !bio_list_empty(¤t->bio_list[1])) &&
483 bs->rescue_workqueue)
484 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
485
486 p = mempool_alloc(&bs->bio_pool, gfp_mask);
487 if (!p && gfp_mask != saved_gfp) {
488 punt_bios_to_rescuer(bs);
489 gfp_mask = saved_gfp;
490 p = mempool_alloc(&bs->bio_pool, gfp_mask);
491 }
492
493 front_pad = bs->front_pad;
494 inline_vecs = BIO_INLINE_VECS;
495 }
496
497 if (unlikely(!p))
498 return NULL;
499
500 bio = p + front_pad;
501 bio_init(bio, NULL, 0);
502
503 if (nr_iovecs > inline_vecs) {
504 unsigned long idx = 0;
505
506 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
507 if (!bvl && gfp_mask != saved_gfp) {
508 punt_bios_to_rescuer(bs);
509 gfp_mask = saved_gfp;
510 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
511 }
512
513 if (unlikely(!bvl))
514 goto err_free;
515
516 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
517 } else if (nr_iovecs) {
518 bvl = bio->bi_inline_vecs;
519 }
520
521 bio->bi_pool = bs;
522 bio->bi_max_vecs = nr_iovecs;
523 bio->bi_io_vec = bvl;
524 return bio;
525
526 err_free:
527 mempool_free(p, &bs->bio_pool);
528 return NULL;
529 }
530 EXPORT_SYMBOL(bio_alloc_bioset);
531
zero_fill_bio_iter(struct bio * bio,struct bvec_iter start)532 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
533 {
534 unsigned long flags;
535 struct bio_vec bv;
536 struct bvec_iter iter;
537
538 __bio_for_each_segment(bv, bio, iter, start) {
539 char *data = bvec_kmap_irq(&bv, &flags);
540 memset(data, 0, bv.bv_len);
541 flush_dcache_page(bv.bv_page);
542 bvec_kunmap_irq(data, &flags);
543 }
544 }
545 EXPORT_SYMBOL(zero_fill_bio_iter);
546
547 /**
548 * bio_truncate - truncate the bio to small size of @new_size
549 * @bio: the bio to be truncated
550 * @new_size: new size for truncating the bio
551 *
552 * Description:
553 * Truncate the bio to new size of @new_size. If bio_op(bio) is
554 * REQ_OP_READ, zero the truncated part. This function should only
555 * be used for handling corner cases, such as bio eod.
556 */
bio_truncate(struct bio * bio,unsigned new_size)557 void bio_truncate(struct bio *bio, unsigned new_size)
558 {
559 struct bio_vec bv;
560 struct bvec_iter iter;
561 unsigned int done = 0;
562 bool truncated = false;
563
564 if (new_size >= bio->bi_iter.bi_size)
565 return;
566
567 if (bio_op(bio) != REQ_OP_READ)
568 goto exit;
569
570 bio_for_each_segment(bv, bio, iter) {
571 if (done + bv.bv_len > new_size) {
572 unsigned offset;
573
574 if (!truncated)
575 offset = new_size - done;
576 else
577 offset = 0;
578 zero_user(bv.bv_page, bv.bv_offset + offset,
579 bv.bv_len - offset);
580 truncated = true;
581 }
582 done += bv.bv_len;
583 }
584
585 exit:
586 /*
587 * Don't touch bvec table here and make it really immutable, since
588 * fs bio user has to retrieve all pages via bio_for_each_segment_all
589 * in its .end_bio() callback.
590 *
591 * It is enough to truncate bio by updating .bi_size since we can make
592 * correct bvec with the updated .bi_size for drivers.
593 */
594 bio->bi_iter.bi_size = new_size;
595 }
596
597 /**
598 * guard_bio_eod - truncate a BIO to fit the block device
599 * @bio: bio to truncate
600 *
601 * This allows us to do IO even on the odd last sectors of a device, even if the
602 * block size is some multiple of the physical sector size.
603 *
604 * We'll just truncate the bio to the size of the device, and clear the end of
605 * the buffer head manually. Truly out-of-range accesses will turn into actual
606 * I/O errors, this only handles the "we need to be able to do I/O at the final
607 * sector" case.
608 */
guard_bio_eod(struct bio * bio)609 void guard_bio_eod(struct bio *bio)
610 {
611 sector_t maxsector;
612 struct hd_struct *part;
613
614 rcu_read_lock();
615 part = __disk_get_part(bio->bi_disk, bio->bi_partno);
616 if (part)
617 maxsector = part_nr_sects_read(part);
618 else
619 maxsector = get_capacity(bio->bi_disk);
620 rcu_read_unlock();
621
622 if (!maxsector)
623 return;
624
625 /*
626 * If the *whole* IO is past the end of the device,
627 * let it through, and the IO layer will turn it into
628 * an EIO.
629 */
630 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
631 return;
632
633 maxsector -= bio->bi_iter.bi_sector;
634 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
635 return;
636
637 bio_truncate(bio, maxsector << 9);
638 }
639
640 /**
641 * bio_put - release a reference to a bio
642 * @bio: bio to release reference to
643 *
644 * Description:
645 * Put a reference to a &struct bio, either one you have gotten with
646 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
647 **/
bio_put(struct bio * bio)648 void bio_put(struct bio *bio)
649 {
650 if (!bio_flagged(bio, BIO_REFFED))
651 bio_free(bio);
652 else {
653 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
654
655 /*
656 * last put frees it
657 */
658 if (atomic_dec_and_test(&bio->__bi_cnt))
659 bio_free(bio);
660 }
661 }
662 EXPORT_SYMBOL(bio_put);
663
664 /**
665 * __bio_clone_fast - clone a bio that shares the original bio's biovec
666 * @bio: destination bio
667 * @bio_src: bio to clone
668 *
669 * Clone a &bio. Caller will own the returned bio, but not
670 * the actual data it points to. Reference count of returned
671 * bio will be one.
672 *
673 * Caller must ensure that @bio_src is not freed before @bio.
674 */
__bio_clone_fast(struct bio * bio,struct bio * bio_src)675 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
676 {
677 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
678
679 /*
680 * most users will be overriding ->bi_disk with a new target,
681 * so we don't set nor calculate new physical/hw segment counts here
682 */
683 bio->bi_disk = bio_src->bi_disk;
684 bio->bi_partno = bio_src->bi_partno;
685 bio_set_flag(bio, BIO_CLONED);
686 if (bio_flagged(bio_src, BIO_THROTTLED))
687 bio_set_flag(bio, BIO_THROTTLED);
688 bio->bi_opf = bio_src->bi_opf;
689 bio->bi_ioprio = bio_src->bi_ioprio;
690 bio->bi_write_hint = bio_src->bi_write_hint;
691 bio->bi_iter = bio_src->bi_iter;
692 bio->bi_io_vec = bio_src->bi_io_vec;
693
694 bio_clone_blkg_association(bio, bio_src);
695 blkcg_bio_issue_init(bio);
696 }
697 EXPORT_SYMBOL(__bio_clone_fast);
698
699 /**
700 * bio_clone_fast - clone a bio that shares the original bio's biovec
701 * @bio: bio to clone
702 * @gfp_mask: allocation priority
703 * @bs: bio_set to allocate from
704 *
705 * Like __bio_clone_fast, only also allocates the returned bio
706 */
bio_clone_fast(struct bio * bio,gfp_t gfp_mask,struct bio_set * bs)707 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
708 {
709 struct bio *b;
710
711 b = bio_alloc_bioset(gfp_mask, 0, bs);
712 if (!b)
713 return NULL;
714
715 __bio_clone_fast(b, bio);
716
717 if (bio_crypt_clone(b, bio, gfp_mask) < 0)
718 goto err_put;
719
720 if (bio_integrity(bio) &&
721 bio_integrity_clone(b, bio, gfp_mask) < 0)
722 goto err_put;
723
724 return b;
725
726 err_put:
727 bio_put(b);
728 return NULL;
729 }
730 EXPORT_SYMBOL(bio_clone_fast);
731
bio_devname(struct bio * bio,char * buf)732 const char *bio_devname(struct bio *bio, char *buf)
733 {
734 return disk_name(bio->bi_disk, bio->bi_partno, buf);
735 }
736 EXPORT_SYMBOL(bio_devname);
737
page_is_mergeable(const struct bio_vec * bv,struct page * page,unsigned int len,unsigned int off,bool * same_page)738 static inline bool page_is_mergeable(const struct bio_vec *bv,
739 struct page *page, unsigned int len, unsigned int off,
740 bool *same_page)
741 {
742 size_t bv_end = bv->bv_offset + bv->bv_len;
743 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
744 phys_addr_t page_addr = page_to_phys(page);
745
746 if (vec_end_addr + 1 != page_addr + off)
747 return false;
748 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
749 return false;
750
751 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
752 if (*same_page)
753 return true;
754 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
755 }
756
757 /*
758 * Try to merge a page into a segment, while obeying the hardware segment
759 * size limit. This is not for normal read/write bios, but for passthrough
760 * or Zone Append operations that we can't split.
761 */
bio_try_merge_hw_seg(struct request_queue * q,struct bio * bio,struct page * page,unsigned len,unsigned offset,bool * same_page)762 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
763 struct page *page, unsigned len,
764 unsigned offset, bool *same_page)
765 {
766 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
767 unsigned long mask = queue_segment_boundary(q);
768 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
769 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
770
771 if ((addr1 | mask) != (addr2 | mask))
772 return false;
773 if (bv->bv_len + len > queue_max_segment_size(q))
774 return false;
775 return __bio_try_merge_page(bio, page, len, offset, same_page);
776 }
777
778 /**
779 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
780 * @q: the target queue
781 * @bio: destination bio
782 * @page: page to add
783 * @len: vec entry length
784 * @offset: vec entry offset
785 * @max_sectors: maximum number of sectors that can be added
786 * @same_page: return if the segment has been merged inside the same page
787 *
788 * Add a page to a bio while respecting the hardware max_sectors, max_segment
789 * and gap limitations.
790 */
bio_add_hw_page(struct request_queue * q,struct bio * bio,struct page * page,unsigned int len,unsigned int offset,unsigned int max_sectors,bool * same_page)791 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
792 struct page *page, unsigned int len, unsigned int offset,
793 unsigned int max_sectors, bool *same_page)
794 {
795 struct bio_vec *bvec;
796
797 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
798 return 0;
799
800 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
801 return 0;
802
803 if (bio->bi_vcnt > 0) {
804 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
805 return len;
806
807 /*
808 * If the queue doesn't support SG gaps and adding this segment
809 * would create a gap, disallow it.
810 */
811 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
812 if (bvec_gap_to_prev(q, bvec, offset))
813 return 0;
814 }
815
816 if (bio_full(bio, len))
817 return 0;
818
819 if (bio->bi_vcnt >= queue_max_segments(q))
820 return 0;
821
822 bvec = &bio->bi_io_vec[bio->bi_vcnt];
823 bvec->bv_page = page;
824 bvec->bv_len = len;
825 bvec->bv_offset = offset;
826 bio->bi_vcnt++;
827 bio->bi_iter.bi_size += len;
828 return len;
829 }
830
831 /**
832 * bio_add_pc_page - attempt to add page to passthrough bio
833 * @q: the target queue
834 * @bio: destination bio
835 * @page: page to add
836 * @len: vec entry length
837 * @offset: vec entry offset
838 *
839 * Attempt to add a page to the bio_vec maplist. This can fail for a
840 * number of reasons, such as the bio being full or target block device
841 * limitations. The target block device must allow bio's up to PAGE_SIZE,
842 * so it is always possible to add a single page to an empty bio.
843 *
844 * This should only be used by passthrough bios.
845 */
bio_add_pc_page(struct request_queue * q,struct bio * bio,struct page * page,unsigned int len,unsigned int offset)846 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
847 struct page *page, unsigned int len, unsigned int offset)
848 {
849 bool same_page = false;
850 return bio_add_hw_page(q, bio, page, len, offset,
851 queue_max_hw_sectors(q), &same_page);
852 }
853 EXPORT_SYMBOL(bio_add_pc_page);
854
855 /**
856 * __bio_try_merge_page - try appending data to an existing bvec.
857 * @bio: destination bio
858 * @page: start page to add
859 * @len: length of the data to add
860 * @off: offset of the data relative to @page
861 * @same_page: return if the segment has been merged inside the same page
862 *
863 * Try to add the data at @page + @off to the last bvec of @bio. This is a
864 * useful optimisation for file systems with a block size smaller than the
865 * page size.
866 *
867 * Warn if (@len, @off) crosses pages in case that @same_page is true.
868 *
869 * Return %true on success or %false on failure.
870 */
__bio_try_merge_page(struct bio * bio,struct page * page,unsigned int len,unsigned int off,bool * same_page)871 bool __bio_try_merge_page(struct bio *bio, struct page *page,
872 unsigned int len, unsigned int off, bool *same_page)
873 {
874 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
875 return false;
876
877 if (bio->bi_vcnt > 0) {
878 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
879
880 if (page_is_mergeable(bv, page, len, off, same_page)) {
881 if (bio->bi_iter.bi_size > UINT_MAX - len) {
882 *same_page = false;
883 return false;
884 }
885 bv->bv_len += len;
886 bio->bi_iter.bi_size += len;
887 return true;
888 }
889 }
890 return false;
891 }
892 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
893
894 /**
895 * __bio_add_page - add page(s) to a bio in a new segment
896 * @bio: destination bio
897 * @page: start page to add
898 * @len: length of the data to add, may cross pages
899 * @off: offset of the data relative to @page, may cross pages
900 *
901 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
902 * that @bio has space for another bvec.
903 */
__bio_add_page(struct bio * bio,struct page * page,unsigned int len,unsigned int off)904 void __bio_add_page(struct bio *bio, struct page *page,
905 unsigned int len, unsigned int off)
906 {
907 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
908
909 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
910 WARN_ON_ONCE(bio_full(bio, len));
911
912 bv->bv_page = page;
913 bv->bv_offset = off;
914 bv->bv_len = len;
915
916 bio->bi_iter.bi_size += len;
917 bio->bi_vcnt++;
918
919 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
920 bio_set_flag(bio, BIO_WORKINGSET);
921 }
922 EXPORT_SYMBOL_GPL(__bio_add_page);
923
924 /**
925 * bio_add_page - attempt to add page(s) to bio
926 * @bio: destination bio
927 * @page: start page to add
928 * @len: vec entry length, may cross pages
929 * @offset: vec entry offset relative to @page, may cross pages
930 *
931 * Attempt to add page(s) to the bio_vec maplist. This will only fail
932 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
933 */
bio_add_page(struct bio * bio,struct page * page,unsigned int len,unsigned int offset)934 int bio_add_page(struct bio *bio, struct page *page,
935 unsigned int len, unsigned int offset)
936 {
937 bool same_page = false;
938
939 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
940 if (bio_full(bio, len))
941 return 0;
942 __bio_add_page(bio, page, len, offset);
943 }
944 return len;
945 }
946 EXPORT_SYMBOL(bio_add_page);
947
bio_release_pages(struct bio * bio,bool mark_dirty)948 void bio_release_pages(struct bio *bio, bool mark_dirty)
949 {
950 struct bvec_iter_all iter_all;
951 struct bio_vec *bvec;
952
953 if (bio_flagged(bio, BIO_NO_PAGE_REF))
954 return;
955
956 bio_for_each_segment_all(bvec, bio, iter_all) {
957 if (mark_dirty && !PageCompound(bvec->bv_page))
958 set_page_dirty_lock(bvec->bv_page);
959 put_page(bvec->bv_page);
960 }
961 }
962 EXPORT_SYMBOL_GPL(bio_release_pages);
963
__bio_iov_bvec_add_pages(struct bio * bio,struct iov_iter * iter)964 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter)
965 {
966 const struct bio_vec *bv = iter->bvec;
967 unsigned int len;
968 size_t size;
969
970 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len))
971 return -EINVAL;
972
973 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count);
974 size = bio_add_page(bio, bv->bv_page, len,
975 bv->bv_offset + iter->iov_offset);
976 if (unlikely(size != len))
977 return -EINVAL;
978 iov_iter_advance(iter, size);
979 return 0;
980 }
981
bio_put_pages(struct page ** pages,size_t size,size_t off)982 static void bio_put_pages(struct page **pages, size_t size, size_t off)
983 {
984 size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE);
985
986 for (i = 0; i < nr; i++)
987 put_page(pages[i]);
988 }
989
990 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
991
992 /**
993 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
994 * @bio: bio to add pages to
995 * @iter: iov iterator describing the region to be mapped
996 *
997 * Pins pages from *iter and appends them to @bio's bvec array. The
998 * pages will have to be released using put_page() when done.
999 * For multi-segment *iter, this function only adds pages from the
1000 * next non-empty segment of the iov iterator.
1001 */
__bio_iov_iter_get_pages(struct bio * bio,struct iov_iter * iter)1002 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1003 {
1004 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1005 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1006 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1007 struct page **pages = (struct page **)bv;
1008 bool same_page = false;
1009 ssize_t size, left;
1010 unsigned len, i;
1011 size_t offset;
1012
1013 /*
1014 * Move page array up in the allocated memory for the bio vecs as far as
1015 * possible so that we can start filling biovecs from the beginning
1016 * without overwriting the temporary page array.
1017 */
1018 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1019 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1020
1021 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1022 if (unlikely(size <= 0))
1023 return size ? size : -EFAULT;
1024
1025 for (left = size, i = 0; left > 0; left -= len, i++) {
1026 struct page *page = pages[i];
1027
1028 len = min_t(size_t, PAGE_SIZE - offset, left);
1029
1030 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1031 if (same_page)
1032 put_page(page);
1033 } else {
1034 if (WARN_ON_ONCE(bio_full(bio, len))) {
1035 bio_put_pages(pages + i, left, offset);
1036 return -EINVAL;
1037 }
1038 __bio_add_page(bio, page, len, offset);
1039 }
1040 offset = 0;
1041 }
1042
1043 iov_iter_advance(iter, size);
1044 return 0;
1045 }
1046
__bio_iov_append_get_pages(struct bio * bio,struct iov_iter * iter)1047 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
1048 {
1049 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1050 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1051 struct request_queue *q = bio->bi_disk->queue;
1052 unsigned int max_append_sectors = queue_max_zone_append_sectors(q);
1053 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1054 struct page **pages = (struct page **)bv;
1055 ssize_t size, left;
1056 unsigned len, i;
1057 size_t offset;
1058 int ret = 0;
1059
1060 /*
1061 * Move page array up in the allocated memory for the bio vecs as far as
1062 * possible so that we can start filling biovecs from the beginning
1063 * without overwriting the temporary page array.
1064 */
1065 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1066 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1067
1068 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1069 if (unlikely(size <= 0))
1070 return size ? size : -EFAULT;
1071
1072 for (left = size, i = 0; left > 0; left -= len, i++) {
1073 struct page *page = pages[i];
1074 bool same_page = false;
1075
1076 len = min_t(size_t, PAGE_SIZE - offset, left);
1077 if (bio_add_hw_page(q, bio, page, len, offset,
1078 max_append_sectors, &same_page) != len) {
1079 bio_put_pages(pages + i, left, offset);
1080 ret = -EINVAL;
1081 break;
1082 }
1083 if (same_page)
1084 put_page(page);
1085 offset = 0;
1086 }
1087
1088 iov_iter_advance(iter, size - left);
1089 return ret;
1090 }
1091
1092 /**
1093 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1094 * @bio: bio to add pages to
1095 * @iter: iov iterator describing the region to be added
1096 *
1097 * This takes either an iterator pointing to user memory, or one pointing to
1098 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1099 * map them into the kernel. On IO completion, the caller should put those
1100 * pages. If we're adding kernel pages, and the caller told us it's safe to
1101 * do so, we just have to add the pages to the bio directly. We don't grab an
1102 * extra reference to those pages (the user should already have that), and we
1103 * don't put the page on IO completion. The caller needs to check if the bio is
1104 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
1105 * released.
1106 *
1107 * The function tries, but does not guarantee, to pin as many pages as
1108 * fit into the bio, or are requested in @iter, whatever is smaller. If
1109 * MM encounters an error pinning the requested pages, it stops. Error
1110 * is returned only if 0 pages could be pinned.
1111 */
bio_iov_iter_get_pages(struct bio * bio,struct iov_iter * iter)1112 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1113 {
1114 const bool is_bvec = iov_iter_is_bvec(iter);
1115 int ret;
1116
1117 if (WARN_ON_ONCE(bio->bi_vcnt))
1118 return -EINVAL;
1119
1120 do {
1121 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1122 if (WARN_ON_ONCE(is_bvec))
1123 return -EINVAL;
1124 ret = __bio_iov_append_get_pages(bio, iter);
1125 } else {
1126 if (is_bvec)
1127 ret = __bio_iov_bvec_add_pages(bio, iter);
1128 else
1129 ret = __bio_iov_iter_get_pages(bio, iter);
1130 }
1131 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1132
1133 if (is_bvec)
1134 bio_set_flag(bio, BIO_NO_PAGE_REF);
1135 return bio->bi_vcnt ? 0 : ret;
1136 }
1137 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1138
submit_bio_wait_endio(struct bio * bio)1139 static void submit_bio_wait_endio(struct bio *bio)
1140 {
1141 complete(bio->bi_private);
1142 }
1143
1144 /**
1145 * submit_bio_wait - submit a bio, and wait until it completes
1146 * @bio: The &struct bio which describes the I/O
1147 *
1148 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1149 * bio_endio() on failure.
1150 *
1151 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1152 * result in bio reference to be consumed. The caller must drop the reference
1153 * on his own.
1154 */
submit_bio_wait(struct bio * bio)1155 int submit_bio_wait(struct bio *bio)
1156 {
1157 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
1158 unsigned long hang_check;
1159
1160 bio->bi_private = &done;
1161 bio->bi_end_io = submit_bio_wait_endio;
1162 bio->bi_opf |= REQ_SYNC;
1163 submit_bio(bio);
1164
1165 /* Prevent hang_check timer from firing at us during very long I/O */
1166 hang_check = sysctl_hung_task_timeout_secs;
1167 if (hang_check)
1168 while (!wait_for_completion_io_timeout(&done,
1169 hang_check * (HZ/2)))
1170 ;
1171 else
1172 wait_for_completion_io(&done);
1173
1174 return blk_status_to_errno(bio->bi_status);
1175 }
1176 EXPORT_SYMBOL(submit_bio_wait);
1177
1178 /**
1179 * bio_advance - increment/complete a bio by some number of bytes
1180 * @bio: bio to advance
1181 * @bytes: number of bytes to complete
1182 *
1183 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1184 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1185 * be updated on the last bvec as well.
1186 *
1187 * @bio will then represent the remaining, uncompleted portion of the io.
1188 */
bio_advance(struct bio * bio,unsigned bytes)1189 void bio_advance(struct bio *bio, unsigned bytes)
1190 {
1191 if (bio_integrity(bio))
1192 bio_integrity_advance(bio, bytes);
1193
1194 bio_crypt_advance(bio, bytes);
1195 bio_advance_iter(bio, &bio->bi_iter, bytes);
1196 }
1197 EXPORT_SYMBOL(bio_advance);
1198
bio_copy_data_iter(struct bio * dst,struct bvec_iter * dst_iter,struct bio * src,struct bvec_iter * src_iter)1199 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1200 struct bio *src, struct bvec_iter *src_iter)
1201 {
1202 struct bio_vec src_bv, dst_bv;
1203 void *src_p, *dst_p;
1204 unsigned bytes;
1205
1206 while (src_iter->bi_size && dst_iter->bi_size) {
1207 src_bv = bio_iter_iovec(src, *src_iter);
1208 dst_bv = bio_iter_iovec(dst, *dst_iter);
1209
1210 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1211
1212 src_p = kmap_atomic(src_bv.bv_page);
1213 dst_p = kmap_atomic(dst_bv.bv_page);
1214
1215 memcpy(dst_p + dst_bv.bv_offset,
1216 src_p + src_bv.bv_offset,
1217 bytes);
1218
1219 kunmap_atomic(dst_p);
1220 kunmap_atomic(src_p);
1221
1222 flush_dcache_page(dst_bv.bv_page);
1223
1224 bio_advance_iter(src, src_iter, bytes);
1225 bio_advance_iter(dst, dst_iter, bytes);
1226 }
1227 }
1228 EXPORT_SYMBOL(bio_copy_data_iter);
1229
1230 /**
1231 * bio_copy_data - copy contents of data buffers from one bio to another
1232 * @src: source bio
1233 * @dst: destination bio
1234 *
1235 * Stops when it reaches the end of either @src or @dst - that is, copies
1236 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1237 */
bio_copy_data(struct bio * dst,struct bio * src)1238 void bio_copy_data(struct bio *dst, struct bio *src)
1239 {
1240 struct bvec_iter src_iter = src->bi_iter;
1241 struct bvec_iter dst_iter = dst->bi_iter;
1242
1243 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1244 }
1245 EXPORT_SYMBOL(bio_copy_data);
1246
1247 /**
1248 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1249 * another
1250 * @src: source bio list
1251 * @dst: destination bio list
1252 *
1253 * Stops when it reaches the end of either the @src list or @dst list - that is,
1254 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1255 * bios).
1256 */
bio_list_copy_data(struct bio * dst,struct bio * src)1257 void bio_list_copy_data(struct bio *dst, struct bio *src)
1258 {
1259 struct bvec_iter src_iter = src->bi_iter;
1260 struct bvec_iter dst_iter = dst->bi_iter;
1261
1262 while (1) {
1263 if (!src_iter.bi_size) {
1264 src = src->bi_next;
1265 if (!src)
1266 break;
1267
1268 src_iter = src->bi_iter;
1269 }
1270
1271 if (!dst_iter.bi_size) {
1272 dst = dst->bi_next;
1273 if (!dst)
1274 break;
1275
1276 dst_iter = dst->bi_iter;
1277 }
1278
1279 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1280 }
1281 }
1282 EXPORT_SYMBOL(bio_list_copy_data);
1283
bio_free_pages(struct bio * bio)1284 void bio_free_pages(struct bio *bio)
1285 {
1286 struct bio_vec *bvec;
1287 struct bvec_iter_all iter_all;
1288
1289 bio_for_each_segment_all(bvec, bio, iter_all)
1290 __free_page(bvec->bv_page);
1291 }
1292 EXPORT_SYMBOL(bio_free_pages);
1293
1294 /*
1295 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1296 * for performing direct-IO in BIOs.
1297 *
1298 * The problem is that we cannot run set_page_dirty() from interrupt context
1299 * because the required locks are not interrupt-safe. So what we can do is to
1300 * mark the pages dirty _before_ performing IO. And in interrupt context,
1301 * check that the pages are still dirty. If so, fine. If not, redirty them
1302 * in process context.
1303 *
1304 * We special-case compound pages here: normally this means reads into hugetlb
1305 * pages. The logic in here doesn't really work right for compound pages
1306 * because the VM does not uniformly chase down the head page in all cases.
1307 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1308 * handle them at all. So we skip compound pages here at an early stage.
1309 *
1310 * Note that this code is very hard to test under normal circumstances because
1311 * direct-io pins the pages with get_user_pages(). This makes
1312 * is_page_cache_freeable return false, and the VM will not clean the pages.
1313 * But other code (eg, flusher threads) could clean the pages if they are mapped
1314 * pagecache.
1315 *
1316 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1317 * deferred bio dirtying paths.
1318 */
1319
1320 /*
1321 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1322 */
bio_set_pages_dirty(struct bio * bio)1323 void bio_set_pages_dirty(struct bio *bio)
1324 {
1325 struct bio_vec *bvec;
1326 struct bvec_iter_all iter_all;
1327
1328 bio_for_each_segment_all(bvec, bio, iter_all) {
1329 if (!PageCompound(bvec->bv_page))
1330 set_page_dirty_lock(bvec->bv_page);
1331 }
1332 }
1333
1334 /*
1335 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1336 * If they are, then fine. If, however, some pages are clean then they must
1337 * have been written out during the direct-IO read. So we take another ref on
1338 * the BIO and re-dirty the pages in process context.
1339 *
1340 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1341 * here on. It will run one put_page() against each page and will run one
1342 * bio_put() against the BIO.
1343 */
1344
1345 static void bio_dirty_fn(struct work_struct *work);
1346
1347 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1348 static DEFINE_SPINLOCK(bio_dirty_lock);
1349 static struct bio *bio_dirty_list;
1350
1351 /*
1352 * This runs in process context
1353 */
bio_dirty_fn(struct work_struct * work)1354 static void bio_dirty_fn(struct work_struct *work)
1355 {
1356 struct bio *bio, *next;
1357
1358 spin_lock_irq(&bio_dirty_lock);
1359 next = bio_dirty_list;
1360 bio_dirty_list = NULL;
1361 spin_unlock_irq(&bio_dirty_lock);
1362
1363 while ((bio = next) != NULL) {
1364 next = bio->bi_private;
1365
1366 bio_release_pages(bio, true);
1367 bio_put(bio);
1368 }
1369 }
1370
bio_check_pages_dirty(struct bio * bio)1371 void bio_check_pages_dirty(struct bio *bio)
1372 {
1373 struct bio_vec *bvec;
1374 unsigned long flags;
1375 struct bvec_iter_all iter_all;
1376
1377 bio_for_each_segment_all(bvec, bio, iter_all) {
1378 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1379 goto defer;
1380 }
1381
1382 bio_release_pages(bio, false);
1383 bio_put(bio);
1384 return;
1385 defer:
1386 spin_lock_irqsave(&bio_dirty_lock, flags);
1387 bio->bi_private = bio_dirty_list;
1388 bio_dirty_list = bio;
1389 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1390 schedule_work(&bio_dirty_work);
1391 }
1392
bio_remaining_done(struct bio * bio)1393 static inline bool bio_remaining_done(struct bio *bio)
1394 {
1395 /*
1396 * If we're not chaining, then ->__bi_remaining is always 1 and
1397 * we always end io on the first invocation.
1398 */
1399 if (!bio_flagged(bio, BIO_CHAIN))
1400 return true;
1401
1402 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1403
1404 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1405 bio_clear_flag(bio, BIO_CHAIN);
1406 return true;
1407 }
1408
1409 return false;
1410 }
1411
1412 /**
1413 * bio_endio - end I/O on a bio
1414 * @bio: bio
1415 *
1416 * Description:
1417 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1418 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1419 * bio unless they own it and thus know that it has an end_io function.
1420 *
1421 * bio_endio() can be called several times on a bio that has been chained
1422 * using bio_chain(). The ->bi_end_io() function will only be called the
1423 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1424 * generated if BIO_TRACE_COMPLETION is set.
1425 **/
bio_endio(struct bio * bio)1426 void bio_endio(struct bio *bio)
1427 {
1428 again:
1429 if (!bio_remaining_done(bio))
1430 return;
1431 if (!bio_integrity_endio(bio))
1432 return;
1433
1434 if (bio->bi_disk && bio_flagged(bio, BIO_TRACKED))
1435 rq_qos_done_bio(bio->bi_disk->queue, bio);
1436
1437 /*
1438 * Need to have a real endio function for chained bios, otherwise
1439 * various corner cases will break (like stacking block devices that
1440 * save/restore bi_end_io) - however, we want to avoid unbounded
1441 * recursion and blowing the stack. Tail call optimization would
1442 * handle this, but compiling with frame pointers also disables
1443 * gcc's sibling call optimization.
1444 */
1445 if (bio->bi_end_io == bio_chain_endio) {
1446 bio = __bio_chain_endio(bio);
1447 goto again;
1448 }
1449
1450 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1451 trace_block_bio_complete(bio->bi_disk->queue, bio);
1452 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1453 }
1454
1455 blk_throtl_bio_endio(bio);
1456 /* release cgroup info */
1457 bio_uninit(bio);
1458 if (bio->bi_end_io)
1459 bio->bi_end_io(bio);
1460 }
1461 EXPORT_SYMBOL(bio_endio);
1462
1463 /**
1464 * bio_split - split a bio
1465 * @bio: bio to split
1466 * @sectors: number of sectors to split from the front of @bio
1467 * @gfp: gfp mask
1468 * @bs: bio set to allocate from
1469 *
1470 * Allocates and returns a new bio which represents @sectors from the start of
1471 * @bio, and updates @bio to represent the remaining sectors.
1472 *
1473 * Unless this is a discard request the newly allocated bio will point
1474 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1475 * neither @bio nor @bs are freed before the split bio.
1476 */
bio_split(struct bio * bio,int sectors,gfp_t gfp,struct bio_set * bs)1477 struct bio *bio_split(struct bio *bio, int sectors,
1478 gfp_t gfp, struct bio_set *bs)
1479 {
1480 struct bio *split;
1481
1482 BUG_ON(sectors <= 0);
1483 BUG_ON(sectors >= bio_sectors(bio));
1484
1485 /* Zone append commands cannot be split */
1486 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1487 return NULL;
1488
1489 split = bio_clone_fast(bio, gfp, bs);
1490 if (!split)
1491 return NULL;
1492
1493 split->bi_iter.bi_size = sectors << 9;
1494
1495 if (bio_integrity(split))
1496 bio_integrity_trim(split);
1497
1498 bio_advance(bio, split->bi_iter.bi_size);
1499
1500 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1501 bio_set_flag(split, BIO_TRACE_COMPLETION);
1502
1503 return split;
1504 }
1505 EXPORT_SYMBOL(bio_split);
1506
1507 /**
1508 * bio_trim - trim a bio
1509 * @bio: bio to trim
1510 * @offset: number of sectors to trim from the front of @bio
1511 * @size: size we want to trim @bio to, in sectors
1512 */
bio_trim(struct bio * bio,int offset,int size)1513 void bio_trim(struct bio *bio, int offset, int size)
1514 {
1515 /* 'bio' is a cloned bio which we need to trim to match
1516 * the given offset and size.
1517 */
1518
1519 size <<= 9;
1520 if (offset == 0 && size == bio->bi_iter.bi_size)
1521 return;
1522
1523 bio_advance(bio, offset << 9);
1524 bio->bi_iter.bi_size = size;
1525
1526 if (bio_integrity(bio))
1527 bio_integrity_trim(bio);
1528
1529 }
1530 EXPORT_SYMBOL_GPL(bio_trim);
1531
1532 /*
1533 * create memory pools for biovec's in a bio_set.
1534 * use the global biovec slabs created for general use.
1535 */
biovec_init_pool(mempool_t * pool,int pool_entries)1536 int biovec_init_pool(mempool_t *pool, int pool_entries)
1537 {
1538 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1539
1540 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1541 }
1542
1543 /*
1544 * bioset_exit - exit a bioset initialized with bioset_init()
1545 *
1546 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1547 * kzalloc()).
1548 */
bioset_exit(struct bio_set * bs)1549 void bioset_exit(struct bio_set *bs)
1550 {
1551 if (bs->rescue_workqueue)
1552 destroy_workqueue(bs->rescue_workqueue);
1553 bs->rescue_workqueue = NULL;
1554
1555 mempool_exit(&bs->bio_pool);
1556 mempool_exit(&bs->bvec_pool);
1557
1558 bioset_integrity_free(bs);
1559 if (bs->bio_slab)
1560 bio_put_slab(bs);
1561 bs->bio_slab = NULL;
1562 }
1563 EXPORT_SYMBOL(bioset_exit);
1564
1565 /**
1566 * bioset_init - Initialize a bio_set
1567 * @bs: pool to initialize
1568 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1569 * @front_pad: Number of bytes to allocate in front of the returned bio
1570 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1571 * and %BIOSET_NEED_RESCUER
1572 *
1573 * Description:
1574 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1575 * to ask for a number of bytes to be allocated in front of the bio.
1576 * Front pad allocation is useful for embedding the bio inside
1577 * another structure, to avoid allocating extra data to go with the bio.
1578 * Note that the bio must be embedded at the END of that structure always,
1579 * or things will break badly.
1580 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1581 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1582 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1583 * dispatch queued requests when the mempool runs out of space.
1584 *
1585 */
bioset_init(struct bio_set * bs,unsigned int pool_size,unsigned int front_pad,int flags)1586 int bioset_init(struct bio_set *bs,
1587 unsigned int pool_size,
1588 unsigned int front_pad,
1589 int flags)
1590 {
1591 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1592
1593 bs->front_pad = front_pad;
1594
1595 spin_lock_init(&bs->rescue_lock);
1596 bio_list_init(&bs->rescue_list);
1597 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1598
1599 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1600 if (!bs->bio_slab)
1601 return -ENOMEM;
1602
1603 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1604 goto bad;
1605
1606 if ((flags & BIOSET_NEED_BVECS) &&
1607 biovec_init_pool(&bs->bvec_pool, pool_size))
1608 goto bad;
1609
1610 if (!(flags & BIOSET_NEED_RESCUER))
1611 return 0;
1612
1613 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1614 if (!bs->rescue_workqueue)
1615 goto bad;
1616
1617 return 0;
1618 bad:
1619 bioset_exit(bs);
1620 return -ENOMEM;
1621 }
1622 EXPORT_SYMBOL(bioset_init);
1623
1624 /*
1625 * Initialize and setup a new bio_set, based on the settings from
1626 * another bio_set.
1627 */
bioset_init_from_src(struct bio_set * bs,struct bio_set * src)1628 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1629 {
1630 int flags;
1631
1632 flags = 0;
1633 if (src->bvec_pool.min_nr)
1634 flags |= BIOSET_NEED_BVECS;
1635 if (src->rescue_workqueue)
1636 flags |= BIOSET_NEED_RESCUER;
1637
1638 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1639 }
1640 EXPORT_SYMBOL(bioset_init_from_src);
1641
biovec_init_slabs(void)1642 static void __init biovec_init_slabs(void)
1643 {
1644 int i;
1645
1646 for (i = 0; i < BVEC_POOL_NR; i++) {
1647 int size;
1648 struct biovec_slab *bvs = bvec_slabs + i;
1649
1650 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1651 bvs->slab = NULL;
1652 continue;
1653 }
1654
1655 size = bvs->nr_vecs * sizeof(struct bio_vec);
1656 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1657 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1658 }
1659 }
1660
init_bio(void)1661 static int __init init_bio(void)
1662 {
1663 bio_slab_max = 2;
1664 bio_slab_nr = 0;
1665 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab),
1666 GFP_KERNEL);
1667
1668 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET);
1669
1670 if (!bio_slabs)
1671 panic("bio: can't allocate bios\n");
1672
1673 bio_integrity_init();
1674 biovec_init_slabs();
1675
1676 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
1677 panic("bio: can't allocate bios\n");
1678
1679 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1680 panic("bio: can't create integrity pool\n");
1681
1682 return 0;
1683 }
1684 subsys_initcall(init_bio);
1685