xref: /linux/block/bio.c (revision bb9707077b4ee5f77bc9939b057ff8a0d410296f)
1 /*
2  * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public Licens
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
16  *
17  */
18 #include <linux/mm.h>
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
31 
32 #include <trace/events/block.h>
33 
34 /*
35  * Test patch to inline a certain number of bi_io_vec's inside the bio
36  * itself, to shrink a bio data allocation from two mempool calls to one
37  */
38 #define BIO_INLINE_VECS		4
39 
40 /*
41  * if you change this list, also change bvec_alloc or things will
42  * break badly! cannot be bigger than what you can fit into an
43  * unsigned short
44  */
45 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
47 	BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48 };
49 #undef BV
50 
51 /*
52  * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53  * IO code that does not need private memory pools.
54  */
55 struct bio_set *fs_bio_set;
56 EXPORT_SYMBOL(fs_bio_set);
57 
58 /*
59  * Our slab pool management
60  */
61 struct bio_slab {
62 	struct kmem_cache *slab;
63 	unsigned int slab_ref;
64 	unsigned int slab_size;
65 	char name[8];
66 };
67 static DEFINE_MUTEX(bio_slab_lock);
68 static struct bio_slab *bio_slabs;
69 static unsigned int bio_slab_nr, bio_slab_max;
70 
71 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72 {
73 	unsigned int sz = sizeof(struct bio) + extra_size;
74 	struct kmem_cache *slab = NULL;
75 	struct bio_slab *bslab, *new_bio_slabs;
76 	unsigned int new_bio_slab_max;
77 	unsigned int i, entry = -1;
78 
79 	mutex_lock(&bio_slab_lock);
80 
81 	i = 0;
82 	while (i < bio_slab_nr) {
83 		bslab = &bio_slabs[i];
84 
85 		if (!bslab->slab && entry == -1)
86 			entry = i;
87 		else if (bslab->slab_size == sz) {
88 			slab = bslab->slab;
89 			bslab->slab_ref++;
90 			break;
91 		}
92 		i++;
93 	}
94 
95 	if (slab)
96 		goto out_unlock;
97 
98 	if (bio_slab_nr == bio_slab_max && entry == -1) {
99 		new_bio_slab_max = bio_slab_max << 1;
100 		new_bio_slabs = krealloc(bio_slabs,
101 					 new_bio_slab_max * sizeof(struct bio_slab),
102 					 GFP_KERNEL);
103 		if (!new_bio_slabs)
104 			goto out_unlock;
105 		bio_slab_max = new_bio_slab_max;
106 		bio_slabs = new_bio_slabs;
107 	}
108 	if (entry == -1)
109 		entry = bio_slab_nr++;
110 
111 	bslab = &bio_slabs[entry];
112 
113 	snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
114 	slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
115 				 SLAB_HWCACHE_ALIGN, NULL);
116 	if (!slab)
117 		goto out_unlock;
118 
119 	bslab->slab = slab;
120 	bslab->slab_ref = 1;
121 	bslab->slab_size = sz;
122 out_unlock:
123 	mutex_unlock(&bio_slab_lock);
124 	return slab;
125 }
126 
127 static void bio_put_slab(struct bio_set *bs)
128 {
129 	struct bio_slab *bslab = NULL;
130 	unsigned int i;
131 
132 	mutex_lock(&bio_slab_lock);
133 
134 	for (i = 0; i < bio_slab_nr; i++) {
135 		if (bs->bio_slab == bio_slabs[i].slab) {
136 			bslab = &bio_slabs[i];
137 			break;
138 		}
139 	}
140 
141 	if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 		goto out;
143 
144 	WARN_ON(!bslab->slab_ref);
145 
146 	if (--bslab->slab_ref)
147 		goto out;
148 
149 	kmem_cache_destroy(bslab->slab);
150 	bslab->slab = NULL;
151 
152 out:
153 	mutex_unlock(&bio_slab_lock);
154 }
155 
156 unsigned int bvec_nr_vecs(unsigned short idx)
157 {
158 	return bvec_slabs[idx].nr_vecs;
159 }
160 
161 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
162 {
163 	BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
164 
165 	if (idx == BIOVEC_MAX_IDX)
166 		mempool_free(bv, pool);
167 	else {
168 		struct biovec_slab *bvs = bvec_slabs + idx;
169 
170 		kmem_cache_free(bvs->slab, bv);
171 	}
172 }
173 
174 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
175 			   mempool_t *pool)
176 {
177 	struct bio_vec *bvl;
178 
179 	/*
180 	 * see comment near bvec_array define!
181 	 */
182 	switch (nr) {
183 	case 1:
184 		*idx = 0;
185 		break;
186 	case 2 ... 4:
187 		*idx = 1;
188 		break;
189 	case 5 ... 16:
190 		*idx = 2;
191 		break;
192 	case 17 ... 64:
193 		*idx = 3;
194 		break;
195 	case 65 ... 128:
196 		*idx = 4;
197 		break;
198 	case 129 ... BIO_MAX_PAGES:
199 		*idx = 5;
200 		break;
201 	default:
202 		return NULL;
203 	}
204 
205 	/*
206 	 * idx now points to the pool we want to allocate from. only the
207 	 * 1-vec entry pool is mempool backed.
208 	 */
209 	if (*idx == BIOVEC_MAX_IDX) {
210 fallback:
211 		bvl = mempool_alloc(pool, gfp_mask);
212 	} else {
213 		struct biovec_slab *bvs = bvec_slabs + *idx;
214 		gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
215 
216 		/*
217 		 * Make this allocation restricted and don't dump info on
218 		 * allocation failures, since we'll fallback to the mempool
219 		 * in case of failure.
220 		 */
221 		__gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
222 
223 		/*
224 		 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
225 		 * is set, retry with the 1-entry mempool
226 		 */
227 		bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
228 		if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
229 			*idx = BIOVEC_MAX_IDX;
230 			goto fallback;
231 		}
232 	}
233 
234 	return bvl;
235 }
236 
237 static void __bio_free(struct bio *bio)
238 {
239 	bio_disassociate_task(bio);
240 
241 	if (bio_integrity(bio))
242 		bio_integrity_free(bio);
243 }
244 
245 static void bio_free(struct bio *bio)
246 {
247 	struct bio_set *bs = bio->bi_pool;
248 	void *p;
249 
250 	__bio_free(bio);
251 
252 	if (bs) {
253 		if (bio_flagged(bio, BIO_OWNS_VEC))
254 			bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
255 
256 		/*
257 		 * If we have front padding, adjust the bio pointer before freeing
258 		 */
259 		p = bio;
260 		p -= bs->front_pad;
261 
262 		mempool_free(p, bs->bio_pool);
263 	} else {
264 		/* Bio was allocated by bio_kmalloc() */
265 		kfree(bio);
266 	}
267 }
268 
269 void bio_init(struct bio *bio)
270 {
271 	memset(bio, 0, sizeof(*bio));
272 	atomic_set(&bio->__bi_remaining, 1);
273 	atomic_set(&bio->__bi_cnt, 1);
274 }
275 EXPORT_SYMBOL(bio_init);
276 
277 /**
278  * bio_reset - reinitialize a bio
279  * @bio:	bio to reset
280  *
281  * Description:
282  *   After calling bio_reset(), @bio will be in the same state as a freshly
283  *   allocated bio returned bio bio_alloc_bioset() - the only fields that are
284  *   preserved are the ones that are initialized by bio_alloc_bioset(). See
285  *   comment in struct bio.
286  */
287 void bio_reset(struct bio *bio)
288 {
289 	unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
290 
291 	__bio_free(bio);
292 
293 	memset(bio, 0, BIO_RESET_BYTES);
294 	bio->bi_flags = flags;
295 	atomic_set(&bio->__bi_remaining, 1);
296 }
297 EXPORT_SYMBOL(bio_reset);
298 
299 static struct bio *__bio_chain_endio(struct bio *bio)
300 {
301 	struct bio *parent = bio->bi_private;
302 
303 	if (!parent->bi_error)
304 		parent->bi_error = bio->bi_error;
305 	bio_put(bio);
306 	return parent;
307 }
308 
309 static void bio_chain_endio(struct bio *bio)
310 {
311 	bio_endio(__bio_chain_endio(bio));
312 }
313 
314 /**
315  * bio_chain - chain bio completions
316  * @bio: the target bio
317  * @parent: the @bio's parent bio
318  *
319  * The caller won't have a bi_end_io called when @bio completes - instead,
320  * @parent's bi_end_io won't be called until both @parent and @bio have
321  * completed; the chained bio will also be freed when it completes.
322  *
323  * The caller must not set bi_private or bi_end_io in @bio.
324  */
325 void bio_chain(struct bio *bio, struct bio *parent)
326 {
327 	BUG_ON(bio->bi_private || bio->bi_end_io);
328 
329 	bio->bi_private = parent;
330 	bio->bi_end_io	= bio_chain_endio;
331 	bio_inc_remaining(parent);
332 }
333 EXPORT_SYMBOL(bio_chain);
334 
335 static void bio_alloc_rescue(struct work_struct *work)
336 {
337 	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
338 	struct bio *bio;
339 
340 	while (1) {
341 		spin_lock(&bs->rescue_lock);
342 		bio = bio_list_pop(&bs->rescue_list);
343 		spin_unlock(&bs->rescue_lock);
344 
345 		if (!bio)
346 			break;
347 
348 		generic_make_request(bio);
349 	}
350 }
351 
352 static void punt_bios_to_rescuer(struct bio_set *bs)
353 {
354 	struct bio_list punt, nopunt;
355 	struct bio *bio;
356 
357 	/*
358 	 * In order to guarantee forward progress we must punt only bios that
359 	 * were allocated from this bio_set; otherwise, if there was a bio on
360 	 * there for a stacking driver higher up in the stack, processing it
361 	 * could require allocating bios from this bio_set, and doing that from
362 	 * our own rescuer would be bad.
363 	 *
364 	 * Since bio lists are singly linked, pop them all instead of trying to
365 	 * remove from the middle of the list:
366 	 */
367 
368 	bio_list_init(&punt);
369 	bio_list_init(&nopunt);
370 
371 	while ((bio = bio_list_pop(current->bio_list)))
372 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
373 
374 	*current->bio_list = nopunt;
375 
376 	spin_lock(&bs->rescue_lock);
377 	bio_list_merge(&bs->rescue_list, &punt);
378 	spin_unlock(&bs->rescue_lock);
379 
380 	queue_work(bs->rescue_workqueue, &bs->rescue_work);
381 }
382 
383 /**
384  * bio_alloc_bioset - allocate a bio for I/O
385  * @gfp_mask:   the GFP_ mask given to the slab allocator
386  * @nr_iovecs:	number of iovecs to pre-allocate
387  * @bs:		the bio_set to allocate from.
388  *
389  * Description:
390  *   If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
391  *   backed by the @bs's mempool.
392  *
393  *   When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
394  *   always be able to allocate a bio. This is due to the mempool guarantees.
395  *   To make this work, callers must never allocate more than 1 bio at a time
396  *   from this pool. Callers that need to allocate more than 1 bio must always
397  *   submit the previously allocated bio for IO before attempting to allocate
398  *   a new one. Failure to do so can cause deadlocks under memory pressure.
399  *
400  *   Note that when running under generic_make_request() (i.e. any block
401  *   driver), bios are not submitted until after you return - see the code in
402  *   generic_make_request() that converts recursion into iteration, to prevent
403  *   stack overflows.
404  *
405  *   This would normally mean allocating multiple bios under
406  *   generic_make_request() would be susceptible to deadlocks, but we have
407  *   deadlock avoidance code that resubmits any blocked bios from a rescuer
408  *   thread.
409  *
410  *   However, we do not guarantee forward progress for allocations from other
411  *   mempools. Doing multiple allocations from the same mempool under
412  *   generic_make_request() should be avoided - instead, use bio_set's front_pad
413  *   for per bio allocations.
414  *
415  *   RETURNS:
416  *   Pointer to new bio on success, NULL on failure.
417  */
418 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
419 {
420 	gfp_t saved_gfp = gfp_mask;
421 	unsigned front_pad;
422 	unsigned inline_vecs;
423 	unsigned long idx = BIO_POOL_NONE;
424 	struct bio_vec *bvl = NULL;
425 	struct bio *bio;
426 	void *p;
427 
428 	if (!bs) {
429 		if (nr_iovecs > UIO_MAXIOV)
430 			return NULL;
431 
432 		p = kmalloc(sizeof(struct bio) +
433 			    nr_iovecs * sizeof(struct bio_vec),
434 			    gfp_mask);
435 		front_pad = 0;
436 		inline_vecs = nr_iovecs;
437 	} else {
438 		/* should not use nobvec bioset for nr_iovecs > 0 */
439 		if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
440 			return NULL;
441 		/*
442 		 * generic_make_request() converts recursion to iteration; this
443 		 * means if we're running beneath it, any bios we allocate and
444 		 * submit will not be submitted (and thus freed) until after we
445 		 * return.
446 		 *
447 		 * This exposes us to a potential deadlock if we allocate
448 		 * multiple bios from the same bio_set() while running
449 		 * underneath generic_make_request(). If we were to allocate
450 		 * multiple bios (say a stacking block driver that was splitting
451 		 * bios), we would deadlock if we exhausted the mempool's
452 		 * reserve.
453 		 *
454 		 * We solve this, and guarantee forward progress, with a rescuer
455 		 * workqueue per bio_set. If we go to allocate and there are
456 		 * bios on current->bio_list, we first try the allocation
457 		 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
458 		 * bios we would be blocking to the rescuer workqueue before
459 		 * we retry with the original gfp_flags.
460 		 */
461 
462 		if (current->bio_list && !bio_list_empty(current->bio_list))
463 			gfp_mask &= ~__GFP_DIRECT_RECLAIM;
464 
465 		p = mempool_alloc(bs->bio_pool, gfp_mask);
466 		if (!p && gfp_mask != saved_gfp) {
467 			punt_bios_to_rescuer(bs);
468 			gfp_mask = saved_gfp;
469 			p = mempool_alloc(bs->bio_pool, gfp_mask);
470 		}
471 
472 		front_pad = bs->front_pad;
473 		inline_vecs = BIO_INLINE_VECS;
474 	}
475 
476 	if (unlikely(!p))
477 		return NULL;
478 
479 	bio = p + front_pad;
480 	bio_init(bio);
481 
482 	if (nr_iovecs > inline_vecs) {
483 		bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
484 		if (!bvl && gfp_mask != saved_gfp) {
485 			punt_bios_to_rescuer(bs);
486 			gfp_mask = saved_gfp;
487 			bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
488 		}
489 
490 		if (unlikely(!bvl))
491 			goto err_free;
492 
493 		bio_set_flag(bio, BIO_OWNS_VEC);
494 	} else if (nr_iovecs) {
495 		bvl = bio->bi_inline_vecs;
496 	}
497 
498 	bio->bi_pool = bs;
499 	bio->bi_flags |= idx << BIO_POOL_OFFSET;
500 	bio->bi_max_vecs = nr_iovecs;
501 	bio->bi_io_vec = bvl;
502 	return bio;
503 
504 err_free:
505 	mempool_free(p, bs->bio_pool);
506 	return NULL;
507 }
508 EXPORT_SYMBOL(bio_alloc_bioset);
509 
510 void zero_fill_bio(struct bio *bio)
511 {
512 	unsigned long flags;
513 	struct bio_vec bv;
514 	struct bvec_iter iter;
515 
516 	bio_for_each_segment(bv, bio, iter) {
517 		char *data = bvec_kmap_irq(&bv, &flags);
518 		memset(data, 0, bv.bv_len);
519 		flush_dcache_page(bv.bv_page);
520 		bvec_kunmap_irq(data, &flags);
521 	}
522 }
523 EXPORT_SYMBOL(zero_fill_bio);
524 
525 /**
526  * bio_put - release a reference to a bio
527  * @bio:   bio to release reference to
528  *
529  * Description:
530  *   Put a reference to a &struct bio, either one you have gotten with
531  *   bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
532  **/
533 void bio_put(struct bio *bio)
534 {
535 	if (!bio_flagged(bio, BIO_REFFED))
536 		bio_free(bio);
537 	else {
538 		BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
539 
540 		/*
541 		 * last put frees it
542 		 */
543 		if (atomic_dec_and_test(&bio->__bi_cnt))
544 			bio_free(bio);
545 	}
546 }
547 EXPORT_SYMBOL(bio_put);
548 
549 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
550 {
551 	if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
552 		blk_recount_segments(q, bio);
553 
554 	return bio->bi_phys_segments;
555 }
556 EXPORT_SYMBOL(bio_phys_segments);
557 
558 /**
559  * 	__bio_clone_fast - clone a bio that shares the original bio's biovec
560  * 	@bio: destination bio
561  * 	@bio_src: bio to clone
562  *
563  *	Clone a &bio. Caller will own the returned bio, but not
564  *	the actual data it points to. Reference count of returned
565  * 	bio will be one.
566  *
567  * 	Caller must ensure that @bio_src is not freed before @bio.
568  */
569 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
570 {
571 	BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
572 
573 	/*
574 	 * most users will be overriding ->bi_bdev with a new target,
575 	 * so we don't set nor calculate new physical/hw segment counts here
576 	 */
577 	bio->bi_bdev = bio_src->bi_bdev;
578 	bio_set_flag(bio, BIO_CLONED);
579 	bio->bi_rw = bio_src->bi_rw;
580 	bio->bi_iter = bio_src->bi_iter;
581 	bio->bi_io_vec = bio_src->bi_io_vec;
582 }
583 EXPORT_SYMBOL(__bio_clone_fast);
584 
585 /**
586  *	bio_clone_fast - clone a bio that shares the original bio's biovec
587  *	@bio: bio to clone
588  *	@gfp_mask: allocation priority
589  *	@bs: bio_set to allocate from
590  *
591  * 	Like __bio_clone_fast, only also allocates the returned bio
592  */
593 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
594 {
595 	struct bio *b;
596 
597 	b = bio_alloc_bioset(gfp_mask, 0, bs);
598 	if (!b)
599 		return NULL;
600 
601 	__bio_clone_fast(b, bio);
602 
603 	if (bio_integrity(bio)) {
604 		int ret;
605 
606 		ret = bio_integrity_clone(b, bio, gfp_mask);
607 
608 		if (ret < 0) {
609 			bio_put(b);
610 			return NULL;
611 		}
612 	}
613 
614 	return b;
615 }
616 EXPORT_SYMBOL(bio_clone_fast);
617 
618 /**
619  * 	bio_clone_bioset - clone a bio
620  * 	@bio_src: bio to clone
621  *	@gfp_mask: allocation priority
622  *	@bs: bio_set to allocate from
623  *
624  *	Clone bio. Caller will own the returned bio, but not the actual data it
625  *	points to. Reference count of returned bio will be one.
626  */
627 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
628 			     struct bio_set *bs)
629 {
630 	struct bvec_iter iter;
631 	struct bio_vec bv;
632 	struct bio *bio;
633 
634 	/*
635 	 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
636 	 * bio_src->bi_io_vec to bio->bi_io_vec.
637 	 *
638 	 * We can't do that anymore, because:
639 	 *
640 	 *  - The point of cloning the biovec is to produce a bio with a biovec
641 	 *    the caller can modify: bi_idx and bi_bvec_done should be 0.
642 	 *
643 	 *  - The original bio could've had more than BIO_MAX_PAGES biovecs; if
644 	 *    we tried to clone the whole thing bio_alloc_bioset() would fail.
645 	 *    But the clone should succeed as long as the number of biovecs we
646 	 *    actually need to allocate is fewer than BIO_MAX_PAGES.
647 	 *
648 	 *  - Lastly, bi_vcnt should not be looked at or relied upon by code
649 	 *    that does not own the bio - reason being drivers don't use it for
650 	 *    iterating over the biovec anymore, so expecting it to be kept up
651 	 *    to date (i.e. for clones that share the parent biovec) is just
652 	 *    asking for trouble and would force extra work on
653 	 *    __bio_clone_fast() anyways.
654 	 */
655 
656 	bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
657 	if (!bio)
658 		return NULL;
659 
660 	bio->bi_bdev		= bio_src->bi_bdev;
661 	bio->bi_rw		= bio_src->bi_rw;
662 	bio->bi_iter.bi_sector	= bio_src->bi_iter.bi_sector;
663 	bio->bi_iter.bi_size	= bio_src->bi_iter.bi_size;
664 
665 	if (bio->bi_rw & REQ_DISCARD)
666 		goto integrity_clone;
667 
668 	if (bio->bi_rw & REQ_WRITE_SAME) {
669 		bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
670 		goto integrity_clone;
671 	}
672 
673 	bio_for_each_segment(bv, bio_src, iter)
674 		bio->bi_io_vec[bio->bi_vcnt++] = bv;
675 
676 integrity_clone:
677 	if (bio_integrity(bio_src)) {
678 		int ret;
679 
680 		ret = bio_integrity_clone(bio, bio_src, gfp_mask);
681 		if (ret < 0) {
682 			bio_put(bio);
683 			return NULL;
684 		}
685 	}
686 
687 	return bio;
688 }
689 EXPORT_SYMBOL(bio_clone_bioset);
690 
691 /**
692  *	bio_add_pc_page	-	attempt to add page to bio
693  *	@q: the target queue
694  *	@bio: destination bio
695  *	@page: page to add
696  *	@len: vec entry length
697  *	@offset: vec entry offset
698  *
699  *	Attempt to add a page to the bio_vec maplist. This can fail for a
700  *	number of reasons, such as the bio being full or target block device
701  *	limitations. The target block device must allow bio's up to PAGE_SIZE,
702  *	so it is always possible to add a single page to an empty bio.
703  *
704  *	This should only be used by REQ_PC bios.
705  */
706 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
707 		    *page, unsigned int len, unsigned int offset)
708 {
709 	int retried_segments = 0;
710 	struct bio_vec *bvec;
711 
712 	/*
713 	 * cloned bio must not modify vec list
714 	 */
715 	if (unlikely(bio_flagged(bio, BIO_CLONED)))
716 		return 0;
717 
718 	if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
719 		return 0;
720 
721 	/*
722 	 * For filesystems with a blocksize smaller than the pagesize
723 	 * we will often be called with the same page as last time and
724 	 * a consecutive offset.  Optimize this special case.
725 	 */
726 	if (bio->bi_vcnt > 0) {
727 		struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
728 
729 		if (page == prev->bv_page &&
730 		    offset == prev->bv_offset + prev->bv_len) {
731 			prev->bv_len += len;
732 			bio->bi_iter.bi_size += len;
733 			goto done;
734 		}
735 
736 		/*
737 		 * If the queue doesn't support SG gaps and adding this
738 		 * offset would create a gap, disallow it.
739 		 */
740 		if (bvec_gap_to_prev(q, prev, offset))
741 			return 0;
742 	}
743 
744 	if (bio->bi_vcnt >= bio->bi_max_vecs)
745 		return 0;
746 
747 	/*
748 	 * setup the new entry, we might clear it again later if we
749 	 * cannot add the page
750 	 */
751 	bvec = &bio->bi_io_vec[bio->bi_vcnt];
752 	bvec->bv_page = page;
753 	bvec->bv_len = len;
754 	bvec->bv_offset = offset;
755 	bio->bi_vcnt++;
756 	bio->bi_phys_segments++;
757 	bio->bi_iter.bi_size += len;
758 
759 	/*
760 	 * Perform a recount if the number of segments is greater
761 	 * than queue_max_segments(q).
762 	 */
763 
764 	while (bio->bi_phys_segments > queue_max_segments(q)) {
765 
766 		if (retried_segments)
767 			goto failed;
768 
769 		retried_segments = 1;
770 		blk_recount_segments(q, bio);
771 	}
772 
773 	/* If we may be able to merge these biovecs, force a recount */
774 	if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
775 		bio_clear_flag(bio, BIO_SEG_VALID);
776 
777  done:
778 	return len;
779 
780  failed:
781 	bvec->bv_page = NULL;
782 	bvec->bv_len = 0;
783 	bvec->bv_offset = 0;
784 	bio->bi_vcnt--;
785 	bio->bi_iter.bi_size -= len;
786 	blk_recount_segments(q, bio);
787 	return 0;
788 }
789 EXPORT_SYMBOL(bio_add_pc_page);
790 
791 /**
792  *	bio_add_page	-	attempt to add page to bio
793  *	@bio: destination bio
794  *	@page: page to add
795  *	@len: vec entry length
796  *	@offset: vec entry offset
797  *
798  *	Attempt to add a page to the bio_vec maplist. This will only fail
799  *	if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
800  */
801 int bio_add_page(struct bio *bio, struct page *page,
802 		 unsigned int len, unsigned int offset)
803 {
804 	struct bio_vec *bv;
805 
806 	/*
807 	 * cloned bio must not modify vec list
808 	 */
809 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
810 		return 0;
811 
812 	/*
813 	 * For filesystems with a blocksize smaller than the pagesize
814 	 * we will often be called with the same page as last time and
815 	 * a consecutive offset.  Optimize this special case.
816 	 */
817 	if (bio->bi_vcnt > 0) {
818 		bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
819 
820 		if (page == bv->bv_page &&
821 		    offset == bv->bv_offset + bv->bv_len) {
822 			bv->bv_len += len;
823 			goto done;
824 		}
825 	}
826 
827 	if (bio->bi_vcnt >= bio->bi_max_vecs)
828 		return 0;
829 
830 	bv		= &bio->bi_io_vec[bio->bi_vcnt];
831 	bv->bv_page	= page;
832 	bv->bv_len	= len;
833 	bv->bv_offset	= offset;
834 
835 	bio->bi_vcnt++;
836 done:
837 	bio->bi_iter.bi_size += len;
838 	return len;
839 }
840 EXPORT_SYMBOL(bio_add_page);
841 
842 struct submit_bio_ret {
843 	struct completion event;
844 	int error;
845 };
846 
847 static void submit_bio_wait_endio(struct bio *bio)
848 {
849 	struct submit_bio_ret *ret = bio->bi_private;
850 
851 	ret->error = bio->bi_error;
852 	complete(&ret->event);
853 }
854 
855 /**
856  * submit_bio_wait - submit a bio, and wait until it completes
857  * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
858  * @bio: The &struct bio which describes the I/O
859  *
860  * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
861  * bio_endio() on failure.
862  */
863 int submit_bio_wait(int rw, struct bio *bio)
864 {
865 	struct submit_bio_ret ret;
866 
867 	rw |= REQ_SYNC;
868 	init_completion(&ret.event);
869 	bio->bi_private = &ret;
870 	bio->bi_end_io = submit_bio_wait_endio;
871 	submit_bio(rw, bio);
872 	wait_for_completion_io(&ret.event);
873 
874 	return ret.error;
875 }
876 EXPORT_SYMBOL(submit_bio_wait);
877 
878 /**
879  * bio_advance - increment/complete a bio by some number of bytes
880  * @bio:	bio to advance
881  * @bytes:	number of bytes to complete
882  *
883  * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
884  * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
885  * be updated on the last bvec as well.
886  *
887  * @bio will then represent the remaining, uncompleted portion of the io.
888  */
889 void bio_advance(struct bio *bio, unsigned bytes)
890 {
891 	if (bio_integrity(bio))
892 		bio_integrity_advance(bio, bytes);
893 
894 	bio_advance_iter(bio, &bio->bi_iter, bytes);
895 }
896 EXPORT_SYMBOL(bio_advance);
897 
898 /**
899  * bio_alloc_pages - allocates a single page for each bvec in a bio
900  * @bio: bio to allocate pages for
901  * @gfp_mask: flags for allocation
902  *
903  * Allocates pages up to @bio->bi_vcnt.
904  *
905  * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
906  * freed.
907  */
908 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
909 {
910 	int i;
911 	struct bio_vec *bv;
912 
913 	bio_for_each_segment_all(bv, bio, i) {
914 		bv->bv_page = alloc_page(gfp_mask);
915 		if (!bv->bv_page) {
916 			while (--bv >= bio->bi_io_vec)
917 				__free_page(bv->bv_page);
918 			return -ENOMEM;
919 		}
920 	}
921 
922 	return 0;
923 }
924 EXPORT_SYMBOL(bio_alloc_pages);
925 
926 /**
927  * bio_copy_data - copy contents of data buffers from one chain of bios to
928  * another
929  * @src: source bio list
930  * @dst: destination bio list
931  *
932  * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
933  * @src and @dst as linked lists of bios.
934  *
935  * Stops when it reaches the end of either @src or @dst - that is, copies
936  * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
937  */
938 void bio_copy_data(struct bio *dst, struct bio *src)
939 {
940 	struct bvec_iter src_iter, dst_iter;
941 	struct bio_vec src_bv, dst_bv;
942 	void *src_p, *dst_p;
943 	unsigned bytes;
944 
945 	src_iter = src->bi_iter;
946 	dst_iter = dst->bi_iter;
947 
948 	while (1) {
949 		if (!src_iter.bi_size) {
950 			src = src->bi_next;
951 			if (!src)
952 				break;
953 
954 			src_iter = src->bi_iter;
955 		}
956 
957 		if (!dst_iter.bi_size) {
958 			dst = dst->bi_next;
959 			if (!dst)
960 				break;
961 
962 			dst_iter = dst->bi_iter;
963 		}
964 
965 		src_bv = bio_iter_iovec(src, src_iter);
966 		dst_bv = bio_iter_iovec(dst, dst_iter);
967 
968 		bytes = min(src_bv.bv_len, dst_bv.bv_len);
969 
970 		src_p = kmap_atomic(src_bv.bv_page);
971 		dst_p = kmap_atomic(dst_bv.bv_page);
972 
973 		memcpy(dst_p + dst_bv.bv_offset,
974 		       src_p + src_bv.bv_offset,
975 		       bytes);
976 
977 		kunmap_atomic(dst_p);
978 		kunmap_atomic(src_p);
979 
980 		bio_advance_iter(src, &src_iter, bytes);
981 		bio_advance_iter(dst, &dst_iter, bytes);
982 	}
983 }
984 EXPORT_SYMBOL(bio_copy_data);
985 
986 struct bio_map_data {
987 	int is_our_pages;
988 	struct iov_iter iter;
989 	struct iovec iov[];
990 };
991 
992 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
993 					       gfp_t gfp_mask)
994 {
995 	if (iov_count > UIO_MAXIOV)
996 		return NULL;
997 
998 	return kmalloc(sizeof(struct bio_map_data) +
999 		       sizeof(struct iovec) * iov_count, gfp_mask);
1000 }
1001 
1002 /**
1003  * bio_copy_from_iter - copy all pages from iov_iter to bio
1004  * @bio: The &struct bio which describes the I/O as destination
1005  * @iter: iov_iter as source
1006  *
1007  * Copy all pages from iov_iter to bio.
1008  * Returns 0 on success, or error on failure.
1009  */
1010 static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
1011 {
1012 	int i;
1013 	struct bio_vec *bvec;
1014 
1015 	bio_for_each_segment_all(bvec, bio, i) {
1016 		ssize_t ret;
1017 
1018 		ret = copy_page_from_iter(bvec->bv_page,
1019 					  bvec->bv_offset,
1020 					  bvec->bv_len,
1021 					  &iter);
1022 
1023 		if (!iov_iter_count(&iter))
1024 			break;
1025 
1026 		if (ret < bvec->bv_len)
1027 			return -EFAULT;
1028 	}
1029 
1030 	return 0;
1031 }
1032 
1033 /**
1034  * bio_copy_to_iter - copy all pages from bio to iov_iter
1035  * @bio: The &struct bio which describes the I/O as source
1036  * @iter: iov_iter as destination
1037  *
1038  * Copy all pages from bio to iov_iter.
1039  * Returns 0 on success, or error on failure.
1040  */
1041 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1042 {
1043 	int i;
1044 	struct bio_vec *bvec;
1045 
1046 	bio_for_each_segment_all(bvec, bio, i) {
1047 		ssize_t ret;
1048 
1049 		ret = copy_page_to_iter(bvec->bv_page,
1050 					bvec->bv_offset,
1051 					bvec->bv_len,
1052 					&iter);
1053 
1054 		if (!iov_iter_count(&iter))
1055 			break;
1056 
1057 		if (ret < bvec->bv_len)
1058 			return -EFAULT;
1059 	}
1060 
1061 	return 0;
1062 }
1063 
1064 static void bio_free_pages(struct bio *bio)
1065 {
1066 	struct bio_vec *bvec;
1067 	int i;
1068 
1069 	bio_for_each_segment_all(bvec, bio, i)
1070 		__free_page(bvec->bv_page);
1071 }
1072 
1073 /**
1074  *	bio_uncopy_user	-	finish previously mapped bio
1075  *	@bio: bio being terminated
1076  *
1077  *	Free pages allocated from bio_copy_user_iov() and write back data
1078  *	to user space in case of a read.
1079  */
1080 int bio_uncopy_user(struct bio *bio)
1081 {
1082 	struct bio_map_data *bmd = bio->bi_private;
1083 	int ret = 0;
1084 
1085 	if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1086 		/*
1087 		 * if we're in a workqueue, the request is orphaned, so
1088 		 * don't copy into a random user address space, just free
1089 		 * and return -EINTR so user space doesn't expect any data.
1090 		 */
1091 		if (!current->mm)
1092 			ret = -EINTR;
1093 		else if (bio_data_dir(bio) == READ)
1094 			ret = bio_copy_to_iter(bio, bmd->iter);
1095 		if (bmd->is_our_pages)
1096 			bio_free_pages(bio);
1097 	}
1098 	kfree(bmd);
1099 	bio_put(bio);
1100 	return ret;
1101 }
1102 EXPORT_SYMBOL(bio_uncopy_user);
1103 
1104 /**
1105  *	bio_copy_user_iov	-	copy user data to bio
1106  *	@q:		destination block queue
1107  *	@map_data:	pointer to the rq_map_data holding pages (if necessary)
1108  *	@iter:		iovec iterator
1109  *	@gfp_mask:	memory allocation flags
1110  *
1111  *	Prepares and returns a bio for indirect user io, bouncing data
1112  *	to/from kernel pages as necessary. Must be paired with
1113  *	call bio_uncopy_user() on io completion.
1114  */
1115 struct bio *bio_copy_user_iov(struct request_queue *q,
1116 			      struct rq_map_data *map_data,
1117 			      const struct iov_iter *iter,
1118 			      gfp_t gfp_mask)
1119 {
1120 	struct bio_map_data *bmd;
1121 	struct page *page;
1122 	struct bio *bio;
1123 	int i, ret;
1124 	int nr_pages = 0;
1125 	unsigned int len = iter->count;
1126 	unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1127 
1128 	for (i = 0; i < iter->nr_segs; i++) {
1129 		unsigned long uaddr;
1130 		unsigned long end;
1131 		unsigned long start;
1132 
1133 		uaddr = (unsigned long) iter->iov[i].iov_base;
1134 		end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1135 			>> PAGE_SHIFT;
1136 		start = uaddr >> PAGE_SHIFT;
1137 
1138 		/*
1139 		 * Overflow, abort
1140 		 */
1141 		if (end < start)
1142 			return ERR_PTR(-EINVAL);
1143 
1144 		nr_pages += end - start;
1145 	}
1146 
1147 	if (offset)
1148 		nr_pages++;
1149 
1150 	bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
1151 	if (!bmd)
1152 		return ERR_PTR(-ENOMEM);
1153 
1154 	/*
1155 	 * We need to do a deep copy of the iov_iter including the iovecs.
1156 	 * The caller provided iov might point to an on-stack or otherwise
1157 	 * shortlived one.
1158 	 */
1159 	bmd->is_our_pages = map_data ? 0 : 1;
1160 	memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1161 	iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1162 			iter->nr_segs, iter->count);
1163 
1164 	ret = -ENOMEM;
1165 	bio = bio_kmalloc(gfp_mask, nr_pages);
1166 	if (!bio)
1167 		goto out_bmd;
1168 
1169 	if (iter->type & WRITE)
1170 		bio->bi_rw |= REQ_WRITE;
1171 
1172 	ret = 0;
1173 
1174 	if (map_data) {
1175 		nr_pages = 1 << map_data->page_order;
1176 		i = map_data->offset / PAGE_SIZE;
1177 	}
1178 	while (len) {
1179 		unsigned int bytes = PAGE_SIZE;
1180 
1181 		bytes -= offset;
1182 
1183 		if (bytes > len)
1184 			bytes = len;
1185 
1186 		if (map_data) {
1187 			if (i == map_data->nr_entries * nr_pages) {
1188 				ret = -ENOMEM;
1189 				break;
1190 			}
1191 
1192 			page = map_data->pages[i / nr_pages];
1193 			page += (i % nr_pages);
1194 
1195 			i++;
1196 		} else {
1197 			page = alloc_page(q->bounce_gfp | gfp_mask);
1198 			if (!page) {
1199 				ret = -ENOMEM;
1200 				break;
1201 			}
1202 		}
1203 
1204 		if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1205 			break;
1206 
1207 		len -= bytes;
1208 		offset = 0;
1209 	}
1210 
1211 	if (ret)
1212 		goto cleanup;
1213 
1214 	/*
1215 	 * success
1216 	 */
1217 	if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1218 	    (map_data && map_data->from_user)) {
1219 		ret = bio_copy_from_iter(bio, *iter);
1220 		if (ret)
1221 			goto cleanup;
1222 	}
1223 
1224 	bio->bi_private = bmd;
1225 	return bio;
1226 cleanup:
1227 	if (!map_data)
1228 		bio_free_pages(bio);
1229 	bio_put(bio);
1230 out_bmd:
1231 	kfree(bmd);
1232 	return ERR_PTR(ret);
1233 }
1234 
1235 /**
1236  *	bio_map_user_iov - map user iovec into bio
1237  *	@q:		the struct request_queue for the bio
1238  *	@iter:		iovec iterator
1239  *	@gfp_mask:	memory allocation flags
1240  *
1241  *	Map the user space address into a bio suitable for io to a block
1242  *	device. Returns an error pointer in case of error.
1243  */
1244 struct bio *bio_map_user_iov(struct request_queue *q,
1245 			     const struct iov_iter *iter,
1246 			     gfp_t gfp_mask)
1247 {
1248 	int j;
1249 	int nr_pages = 0;
1250 	struct page **pages;
1251 	struct bio *bio;
1252 	int cur_page = 0;
1253 	int ret, offset;
1254 	struct iov_iter i;
1255 	struct iovec iov;
1256 
1257 	iov_for_each(iov, i, *iter) {
1258 		unsigned long uaddr = (unsigned long) iov.iov_base;
1259 		unsigned long len = iov.iov_len;
1260 		unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1261 		unsigned long start = uaddr >> PAGE_SHIFT;
1262 
1263 		/*
1264 		 * Overflow, abort
1265 		 */
1266 		if (end < start)
1267 			return ERR_PTR(-EINVAL);
1268 
1269 		nr_pages += end - start;
1270 		/*
1271 		 * buffer must be aligned to at least hardsector size for now
1272 		 */
1273 		if (uaddr & queue_dma_alignment(q))
1274 			return ERR_PTR(-EINVAL);
1275 	}
1276 
1277 	if (!nr_pages)
1278 		return ERR_PTR(-EINVAL);
1279 
1280 	bio = bio_kmalloc(gfp_mask, nr_pages);
1281 	if (!bio)
1282 		return ERR_PTR(-ENOMEM);
1283 
1284 	ret = -ENOMEM;
1285 	pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1286 	if (!pages)
1287 		goto out;
1288 
1289 	iov_for_each(iov, i, *iter) {
1290 		unsigned long uaddr = (unsigned long) iov.iov_base;
1291 		unsigned long len = iov.iov_len;
1292 		unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1293 		unsigned long start = uaddr >> PAGE_SHIFT;
1294 		const int local_nr_pages = end - start;
1295 		const int page_limit = cur_page + local_nr_pages;
1296 
1297 		ret = get_user_pages_fast(uaddr, local_nr_pages,
1298 				(iter->type & WRITE) != WRITE,
1299 				&pages[cur_page]);
1300 		if (ret < local_nr_pages) {
1301 			ret = -EFAULT;
1302 			goto out_unmap;
1303 		}
1304 
1305 		offset = offset_in_page(uaddr);
1306 		for (j = cur_page; j < page_limit; j++) {
1307 			unsigned int bytes = PAGE_SIZE - offset;
1308 
1309 			if (len <= 0)
1310 				break;
1311 
1312 			if (bytes > len)
1313 				bytes = len;
1314 
1315 			/*
1316 			 * sorry...
1317 			 */
1318 			if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1319 					    bytes)
1320 				break;
1321 
1322 			len -= bytes;
1323 			offset = 0;
1324 		}
1325 
1326 		cur_page = j;
1327 		/*
1328 		 * release the pages we didn't map into the bio, if any
1329 		 */
1330 		while (j < page_limit)
1331 			put_page(pages[j++]);
1332 	}
1333 
1334 	kfree(pages);
1335 
1336 	/*
1337 	 * set data direction, and check if mapped pages need bouncing
1338 	 */
1339 	if (iter->type & WRITE)
1340 		bio->bi_rw |= REQ_WRITE;
1341 
1342 	bio_set_flag(bio, BIO_USER_MAPPED);
1343 
1344 	/*
1345 	 * subtle -- if __bio_map_user() ended up bouncing a bio,
1346 	 * it would normally disappear when its bi_end_io is run.
1347 	 * however, we need it for the unmap, so grab an extra
1348 	 * reference to it
1349 	 */
1350 	bio_get(bio);
1351 	return bio;
1352 
1353  out_unmap:
1354 	for (j = 0; j < nr_pages; j++) {
1355 		if (!pages[j])
1356 			break;
1357 		put_page(pages[j]);
1358 	}
1359  out:
1360 	kfree(pages);
1361 	bio_put(bio);
1362 	return ERR_PTR(ret);
1363 }
1364 
1365 static void __bio_unmap_user(struct bio *bio)
1366 {
1367 	struct bio_vec *bvec;
1368 	int i;
1369 
1370 	/*
1371 	 * make sure we dirty pages we wrote to
1372 	 */
1373 	bio_for_each_segment_all(bvec, bio, i) {
1374 		if (bio_data_dir(bio) == READ)
1375 			set_page_dirty_lock(bvec->bv_page);
1376 
1377 		put_page(bvec->bv_page);
1378 	}
1379 
1380 	bio_put(bio);
1381 }
1382 
1383 /**
1384  *	bio_unmap_user	-	unmap a bio
1385  *	@bio:		the bio being unmapped
1386  *
1387  *	Unmap a bio previously mapped by bio_map_user(). Must be called with
1388  *	a process context.
1389  *
1390  *	bio_unmap_user() may sleep.
1391  */
1392 void bio_unmap_user(struct bio *bio)
1393 {
1394 	__bio_unmap_user(bio);
1395 	bio_put(bio);
1396 }
1397 EXPORT_SYMBOL(bio_unmap_user);
1398 
1399 static void bio_map_kern_endio(struct bio *bio)
1400 {
1401 	bio_put(bio);
1402 }
1403 
1404 /**
1405  *	bio_map_kern	-	map kernel address into bio
1406  *	@q: the struct request_queue for the bio
1407  *	@data: pointer to buffer to map
1408  *	@len: length in bytes
1409  *	@gfp_mask: allocation flags for bio allocation
1410  *
1411  *	Map the kernel address into a bio suitable for io to a block
1412  *	device. Returns an error pointer in case of error.
1413  */
1414 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1415 			 gfp_t gfp_mask)
1416 {
1417 	unsigned long kaddr = (unsigned long)data;
1418 	unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1419 	unsigned long start = kaddr >> PAGE_SHIFT;
1420 	const int nr_pages = end - start;
1421 	int offset, i;
1422 	struct bio *bio;
1423 
1424 	bio = bio_kmalloc(gfp_mask, nr_pages);
1425 	if (!bio)
1426 		return ERR_PTR(-ENOMEM);
1427 
1428 	offset = offset_in_page(kaddr);
1429 	for (i = 0; i < nr_pages; i++) {
1430 		unsigned int bytes = PAGE_SIZE - offset;
1431 
1432 		if (len <= 0)
1433 			break;
1434 
1435 		if (bytes > len)
1436 			bytes = len;
1437 
1438 		if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1439 				    offset) < bytes) {
1440 			/* we don't support partial mappings */
1441 			bio_put(bio);
1442 			return ERR_PTR(-EINVAL);
1443 		}
1444 
1445 		data += bytes;
1446 		len -= bytes;
1447 		offset = 0;
1448 	}
1449 
1450 	bio->bi_end_io = bio_map_kern_endio;
1451 	return bio;
1452 }
1453 EXPORT_SYMBOL(bio_map_kern);
1454 
1455 static void bio_copy_kern_endio(struct bio *bio)
1456 {
1457 	bio_free_pages(bio);
1458 	bio_put(bio);
1459 }
1460 
1461 static void bio_copy_kern_endio_read(struct bio *bio)
1462 {
1463 	char *p = bio->bi_private;
1464 	struct bio_vec *bvec;
1465 	int i;
1466 
1467 	bio_for_each_segment_all(bvec, bio, i) {
1468 		memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1469 		p += bvec->bv_len;
1470 	}
1471 
1472 	bio_copy_kern_endio(bio);
1473 }
1474 
1475 /**
1476  *	bio_copy_kern	-	copy kernel address into bio
1477  *	@q: the struct request_queue for the bio
1478  *	@data: pointer to buffer to copy
1479  *	@len: length in bytes
1480  *	@gfp_mask: allocation flags for bio and page allocation
1481  *	@reading: data direction is READ
1482  *
1483  *	copy the kernel address into a bio suitable for io to a block
1484  *	device. Returns an error pointer in case of error.
1485  */
1486 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1487 			  gfp_t gfp_mask, int reading)
1488 {
1489 	unsigned long kaddr = (unsigned long)data;
1490 	unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1491 	unsigned long start = kaddr >> PAGE_SHIFT;
1492 	struct bio *bio;
1493 	void *p = data;
1494 	int nr_pages = 0;
1495 
1496 	/*
1497 	 * Overflow, abort
1498 	 */
1499 	if (end < start)
1500 		return ERR_PTR(-EINVAL);
1501 
1502 	nr_pages = end - start;
1503 	bio = bio_kmalloc(gfp_mask, nr_pages);
1504 	if (!bio)
1505 		return ERR_PTR(-ENOMEM);
1506 
1507 	while (len) {
1508 		struct page *page;
1509 		unsigned int bytes = PAGE_SIZE;
1510 
1511 		if (bytes > len)
1512 			bytes = len;
1513 
1514 		page = alloc_page(q->bounce_gfp | gfp_mask);
1515 		if (!page)
1516 			goto cleanup;
1517 
1518 		if (!reading)
1519 			memcpy(page_address(page), p, bytes);
1520 
1521 		if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1522 			break;
1523 
1524 		len -= bytes;
1525 		p += bytes;
1526 	}
1527 
1528 	if (reading) {
1529 		bio->bi_end_io = bio_copy_kern_endio_read;
1530 		bio->bi_private = data;
1531 	} else {
1532 		bio->bi_end_io = bio_copy_kern_endio;
1533 		bio->bi_rw |= REQ_WRITE;
1534 	}
1535 
1536 	return bio;
1537 
1538 cleanup:
1539 	bio_free_pages(bio);
1540 	bio_put(bio);
1541 	return ERR_PTR(-ENOMEM);
1542 }
1543 EXPORT_SYMBOL(bio_copy_kern);
1544 
1545 /*
1546  * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1547  * for performing direct-IO in BIOs.
1548  *
1549  * The problem is that we cannot run set_page_dirty() from interrupt context
1550  * because the required locks are not interrupt-safe.  So what we can do is to
1551  * mark the pages dirty _before_ performing IO.  And in interrupt context,
1552  * check that the pages are still dirty.   If so, fine.  If not, redirty them
1553  * in process context.
1554  *
1555  * We special-case compound pages here: normally this means reads into hugetlb
1556  * pages.  The logic in here doesn't really work right for compound pages
1557  * because the VM does not uniformly chase down the head page in all cases.
1558  * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1559  * handle them at all.  So we skip compound pages here at an early stage.
1560  *
1561  * Note that this code is very hard to test under normal circumstances because
1562  * direct-io pins the pages with get_user_pages().  This makes
1563  * is_page_cache_freeable return false, and the VM will not clean the pages.
1564  * But other code (eg, flusher threads) could clean the pages if they are mapped
1565  * pagecache.
1566  *
1567  * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1568  * deferred bio dirtying paths.
1569  */
1570 
1571 /*
1572  * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1573  */
1574 void bio_set_pages_dirty(struct bio *bio)
1575 {
1576 	struct bio_vec *bvec;
1577 	int i;
1578 
1579 	bio_for_each_segment_all(bvec, bio, i) {
1580 		struct page *page = bvec->bv_page;
1581 
1582 		if (page && !PageCompound(page))
1583 			set_page_dirty_lock(page);
1584 	}
1585 }
1586 
1587 static void bio_release_pages(struct bio *bio)
1588 {
1589 	struct bio_vec *bvec;
1590 	int i;
1591 
1592 	bio_for_each_segment_all(bvec, bio, i) {
1593 		struct page *page = bvec->bv_page;
1594 
1595 		if (page)
1596 			put_page(page);
1597 	}
1598 }
1599 
1600 /*
1601  * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1602  * If they are, then fine.  If, however, some pages are clean then they must
1603  * have been written out during the direct-IO read.  So we take another ref on
1604  * the BIO and the offending pages and re-dirty the pages in process context.
1605  *
1606  * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1607  * here on.  It will run one put_page() against each page and will run one
1608  * bio_put() against the BIO.
1609  */
1610 
1611 static void bio_dirty_fn(struct work_struct *work);
1612 
1613 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1614 static DEFINE_SPINLOCK(bio_dirty_lock);
1615 static struct bio *bio_dirty_list;
1616 
1617 /*
1618  * This runs in process context
1619  */
1620 static void bio_dirty_fn(struct work_struct *work)
1621 {
1622 	unsigned long flags;
1623 	struct bio *bio;
1624 
1625 	spin_lock_irqsave(&bio_dirty_lock, flags);
1626 	bio = bio_dirty_list;
1627 	bio_dirty_list = NULL;
1628 	spin_unlock_irqrestore(&bio_dirty_lock, flags);
1629 
1630 	while (bio) {
1631 		struct bio *next = bio->bi_private;
1632 
1633 		bio_set_pages_dirty(bio);
1634 		bio_release_pages(bio);
1635 		bio_put(bio);
1636 		bio = next;
1637 	}
1638 }
1639 
1640 void bio_check_pages_dirty(struct bio *bio)
1641 {
1642 	struct bio_vec *bvec;
1643 	int nr_clean_pages = 0;
1644 	int i;
1645 
1646 	bio_for_each_segment_all(bvec, bio, i) {
1647 		struct page *page = bvec->bv_page;
1648 
1649 		if (PageDirty(page) || PageCompound(page)) {
1650 			put_page(page);
1651 			bvec->bv_page = NULL;
1652 		} else {
1653 			nr_clean_pages++;
1654 		}
1655 	}
1656 
1657 	if (nr_clean_pages) {
1658 		unsigned long flags;
1659 
1660 		spin_lock_irqsave(&bio_dirty_lock, flags);
1661 		bio->bi_private = bio_dirty_list;
1662 		bio_dirty_list = bio;
1663 		spin_unlock_irqrestore(&bio_dirty_lock, flags);
1664 		schedule_work(&bio_dirty_work);
1665 	} else {
1666 		bio_put(bio);
1667 	}
1668 }
1669 
1670 void generic_start_io_acct(int rw, unsigned long sectors,
1671 			   struct hd_struct *part)
1672 {
1673 	int cpu = part_stat_lock();
1674 
1675 	part_round_stats(cpu, part);
1676 	part_stat_inc(cpu, part, ios[rw]);
1677 	part_stat_add(cpu, part, sectors[rw], sectors);
1678 	part_inc_in_flight(part, rw);
1679 
1680 	part_stat_unlock();
1681 }
1682 EXPORT_SYMBOL(generic_start_io_acct);
1683 
1684 void generic_end_io_acct(int rw, struct hd_struct *part,
1685 			 unsigned long start_time)
1686 {
1687 	unsigned long duration = jiffies - start_time;
1688 	int cpu = part_stat_lock();
1689 
1690 	part_stat_add(cpu, part, ticks[rw], duration);
1691 	part_round_stats(cpu, part);
1692 	part_dec_in_flight(part, rw);
1693 
1694 	part_stat_unlock();
1695 }
1696 EXPORT_SYMBOL(generic_end_io_acct);
1697 
1698 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1699 void bio_flush_dcache_pages(struct bio *bi)
1700 {
1701 	struct bio_vec bvec;
1702 	struct bvec_iter iter;
1703 
1704 	bio_for_each_segment(bvec, bi, iter)
1705 		flush_dcache_page(bvec.bv_page);
1706 }
1707 EXPORT_SYMBOL(bio_flush_dcache_pages);
1708 #endif
1709 
1710 static inline bool bio_remaining_done(struct bio *bio)
1711 {
1712 	/*
1713 	 * If we're not chaining, then ->__bi_remaining is always 1 and
1714 	 * we always end io on the first invocation.
1715 	 */
1716 	if (!bio_flagged(bio, BIO_CHAIN))
1717 		return true;
1718 
1719 	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1720 
1721 	if (atomic_dec_and_test(&bio->__bi_remaining)) {
1722 		bio_clear_flag(bio, BIO_CHAIN);
1723 		return true;
1724 	}
1725 
1726 	return false;
1727 }
1728 
1729 /**
1730  * bio_endio - end I/O on a bio
1731  * @bio:	bio
1732  *
1733  * Description:
1734  *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1735  *   way to end I/O on a bio. No one should call bi_end_io() directly on a
1736  *   bio unless they own it and thus know that it has an end_io function.
1737  **/
1738 void bio_endio(struct bio *bio)
1739 {
1740 again:
1741 	if (!bio_remaining_done(bio))
1742 		return;
1743 
1744 	/*
1745 	 * Need to have a real endio function for chained bios, otherwise
1746 	 * various corner cases will break (like stacking block devices that
1747 	 * save/restore bi_end_io) - however, we want to avoid unbounded
1748 	 * recursion and blowing the stack. Tail call optimization would
1749 	 * handle this, but compiling with frame pointers also disables
1750 	 * gcc's sibling call optimization.
1751 	 */
1752 	if (bio->bi_end_io == bio_chain_endio) {
1753 		bio = __bio_chain_endio(bio);
1754 		goto again;
1755 	}
1756 
1757 	if (bio->bi_end_io)
1758 		bio->bi_end_io(bio);
1759 }
1760 EXPORT_SYMBOL(bio_endio);
1761 
1762 /**
1763  * bio_split - split a bio
1764  * @bio:	bio to split
1765  * @sectors:	number of sectors to split from the front of @bio
1766  * @gfp:	gfp mask
1767  * @bs:		bio set to allocate from
1768  *
1769  * Allocates and returns a new bio which represents @sectors from the start of
1770  * @bio, and updates @bio to represent the remaining sectors.
1771  *
1772  * Unless this is a discard request the newly allocated bio will point
1773  * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1774  * @bio is not freed before the split.
1775  */
1776 struct bio *bio_split(struct bio *bio, int sectors,
1777 		      gfp_t gfp, struct bio_set *bs)
1778 {
1779 	struct bio *split = NULL;
1780 
1781 	BUG_ON(sectors <= 0);
1782 	BUG_ON(sectors >= bio_sectors(bio));
1783 
1784 	/*
1785 	 * Discards need a mutable bio_vec to accommodate the payload
1786 	 * required by the DSM TRIM and UNMAP commands.
1787 	 */
1788 	if (bio->bi_rw & REQ_DISCARD)
1789 		split = bio_clone_bioset(bio, gfp, bs);
1790 	else
1791 		split = bio_clone_fast(bio, gfp, bs);
1792 
1793 	if (!split)
1794 		return NULL;
1795 
1796 	split->bi_iter.bi_size = sectors << 9;
1797 
1798 	if (bio_integrity(split))
1799 		bio_integrity_trim(split, 0, sectors);
1800 
1801 	bio_advance(bio, split->bi_iter.bi_size);
1802 
1803 	return split;
1804 }
1805 EXPORT_SYMBOL(bio_split);
1806 
1807 /**
1808  * bio_trim - trim a bio
1809  * @bio:	bio to trim
1810  * @offset:	number of sectors to trim from the front of @bio
1811  * @size:	size we want to trim @bio to, in sectors
1812  */
1813 void bio_trim(struct bio *bio, int offset, int size)
1814 {
1815 	/* 'bio' is a cloned bio which we need to trim to match
1816 	 * the given offset and size.
1817 	 */
1818 
1819 	size <<= 9;
1820 	if (offset == 0 && size == bio->bi_iter.bi_size)
1821 		return;
1822 
1823 	bio_clear_flag(bio, BIO_SEG_VALID);
1824 
1825 	bio_advance(bio, offset << 9);
1826 
1827 	bio->bi_iter.bi_size = size;
1828 }
1829 EXPORT_SYMBOL_GPL(bio_trim);
1830 
1831 /*
1832  * create memory pools for biovec's in a bio_set.
1833  * use the global biovec slabs created for general use.
1834  */
1835 mempool_t *biovec_create_pool(int pool_entries)
1836 {
1837 	struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1838 
1839 	return mempool_create_slab_pool(pool_entries, bp->slab);
1840 }
1841 
1842 void bioset_free(struct bio_set *bs)
1843 {
1844 	if (bs->rescue_workqueue)
1845 		destroy_workqueue(bs->rescue_workqueue);
1846 
1847 	if (bs->bio_pool)
1848 		mempool_destroy(bs->bio_pool);
1849 
1850 	if (bs->bvec_pool)
1851 		mempool_destroy(bs->bvec_pool);
1852 
1853 	bioset_integrity_free(bs);
1854 	bio_put_slab(bs);
1855 
1856 	kfree(bs);
1857 }
1858 EXPORT_SYMBOL(bioset_free);
1859 
1860 static struct bio_set *__bioset_create(unsigned int pool_size,
1861 				       unsigned int front_pad,
1862 				       bool create_bvec_pool)
1863 {
1864 	unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1865 	struct bio_set *bs;
1866 
1867 	bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1868 	if (!bs)
1869 		return NULL;
1870 
1871 	bs->front_pad = front_pad;
1872 
1873 	spin_lock_init(&bs->rescue_lock);
1874 	bio_list_init(&bs->rescue_list);
1875 	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1876 
1877 	bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1878 	if (!bs->bio_slab) {
1879 		kfree(bs);
1880 		return NULL;
1881 	}
1882 
1883 	bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1884 	if (!bs->bio_pool)
1885 		goto bad;
1886 
1887 	if (create_bvec_pool) {
1888 		bs->bvec_pool = biovec_create_pool(pool_size);
1889 		if (!bs->bvec_pool)
1890 			goto bad;
1891 	}
1892 
1893 	bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1894 	if (!bs->rescue_workqueue)
1895 		goto bad;
1896 
1897 	return bs;
1898 bad:
1899 	bioset_free(bs);
1900 	return NULL;
1901 }
1902 
1903 /**
1904  * bioset_create  - Create a bio_set
1905  * @pool_size:	Number of bio and bio_vecs to cache in the mempool
1906  * @front_pad:	Number of bytes to allocate in front of the returned bio
1907  *
1908  * Description:
1909  *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1910  *    to ask for a number of bytes to be allocated in front of the bio.
1911  *    Front pad allocation is useful for embedding the bio inside
1912  *    another structure, to avoid allocating extra data to go with the bio.
1913  *    Note that the bio must be embedded at the END of that structure always,
1914  *    or things will break badly.
1915  */
1916 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1917 {
1918 	return __bioset_create(pool_size, front_pad, true);
1919 }
1920 EXPORT_SYMBOL(bioset_create);
1921 
1922 /**
1923  * bioset_create_nobvec  - Create a bio_set without bio_vec mempool
1924  * @pool_size:	Number of bio to cache in the mempool
1925  * @front_pad:	Number of bytes to allocate in front of the returned bio
1926  *
1927  * Description:
1928  *    Same functionality as bioset_create() except that mempool is not
1929  *    created for bio_vecs. Saving some memory for bio_clone_fast() users.
1930  */
1931 struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1932 {
1933 	return __bioset_create(pool_size, front_pad, false);
1934 }
1935 EXPORT_SYMBOL(bioset_create_nobvec);
1936 
1937 #ifdef CONFIG_BLK_CGROUP
1938 
1939 /**
1940  * bio_associate_blkcg - associate a bio with the specified blkcg
1941  * @bio: target bio
1942  * @blkcg_css: css of the blkcg to associate
1943  *
1944  * Associate @bio with the blkcg specified by @blkcg_css.  Block layer will
1945  * treat @bio as if it were issued by a task which belongs to the blkcg.
1946  *
1947  * This function takes an extra reference of @blkcg_css which will be put
1948  * when @bio is released.  The caller must own @bio and is responsible for
1949  * synchronizing calls to this function.
1950  */
1951 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1952 {
1953 	if (unlikely(bio->bi_css))
1954 		return -EBUSY;
1955 	css_get(blkcg_css);
1956 	bio->bi_css = blkcg_css;
1957 	return 0;
1958 }
1959 EXPORT_SYMBOL_GPL(bio_associate_blkcg);
1960 
1961 /**
1962  * bio_associate_current - associate a bio with %current
1963  * @bio: target bio
1964  *
1965  * Associate @bio with %current if it hasn't been associated yet.  Block
1966  * layer will treat @bio as if it were issued by %current no matter which
1967  * task actually issues it.
1968  *
1969  * This function takes an extra reference of @task's io_context and blkcg
1970  * which will be put when @bio is released.  The caller must own @bio,
1971  * ensure %current->io_context exists, and is responsible for synchronizing
1972  * calls to this function.
1973  */
1974 int bio_associate_current(struct bio *bio)
1975 {
1976 	struct io_context *ioc;
1977 
1978 	if (bio->bi_css)
1979 		return -EBUSY;
1980 
1981 	ioc = current->io_context;
1982 	if (!ioc)
1983 		return -ENOENT;
1984 
1985 	get_io_context_active(ioc);
1986 	bio->bi_ioc = ioc;
1987 	bio->bi_css = task_get_css(current, io_cgrp_id);
1988 	return 0;
1989 }
1990 EXPORT_SYMBOL_GPL(bio_associate_current);
1991 
1992 /**
1993  * bio_disassociate_task - undo bio_associate_current()
1994  * @bio: target bio
1995  */
1996 void bio_disassociate_task(struct bio *bio)
1997 {
1998 	if (bio->bi_ioc) {
1999 		put_io_context(bio->bi_ioc);
2000 		bio->bi_ioc = NULL;
2001 	}
2002 	if (bio->bi_css) {
2003 		css_put(bio->bi_css);
2004 		bio->bi_css = NULL;
2005 	}
2006 }
2007 
2008 #endif /* CONFIG_BLK_CGROUP */
2009 
2010 static void __init biovec_init_slabs(void)
2011 {
2012 	int i;
2013 
2014 	for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2015 		int size;
2016 		struct biovec_slab *bvs = bvec_slabs + i;
2017 
2018 		if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2019 			bvs->slab = NULL;
2020 			continue;
2021 		}
2022 
2023 		size = bvs->nr_vecs * sizeof(struct bio_vec);
2024 		bvs->slab = kmem_cache_create(bvs->name, size, 0,
2025                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2026 	}
2027 }
2028 
2029 static int __init init_bio(void)
2030 {
2031 	bio_slab_max = 2;
2032 	bio_slab_nr = 0;
2033 	bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2034 	if (!bio_slabs)
2035 		panic("bio: can't allocate bios\n");
2036 
2037 	bio_integrity_init();
2038 	biovec_init_slabs();
2039 
2040 	fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2041 	if (!fs_bio_set)
2042 		panic("bio: can't allocate bios\n");
2043 
2044 	if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2045 		panic("bio: can't create integrity pool\n");
2046 
2047 	return 0;
2048 }
2049 subsys_initcall(init_bio);
2050