xref: /openbmc/linux/block/bio.c (revision ddcf35d397976421a4ec1d0d00fbcc027a8cb034)
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 #include <linux/blk-cgroup.h>
32 
33 #include <trace/events/block.h>
34 #include "blk.h"
35 #include "blk-rq-qos.h"
36 
37 /*
38  * Test patch to inline a certain number of bi_io_vec's inside the bio
39  * itself, to shrink a bio data allocation from two mempool calls to one
40  */
41 #define BIO_INLINE_VECS		4
42 
43 /*
44  * if you change this list, also change bvec_alloc or things will
45  * break badly! cannot be bigger than what you can fit into an
46  * unsigned short
47  */
48 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
49 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
50 	BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
51 };
52 #undef BV
53 
54 /*
55  * fs_bio_set is the bio_set containing bio and iovec memory pools used by
56  * IO code that does not need private memory pools.
57  */
58 struct bio_set fs_bio_set;
59 EXPORT_SYMBOL(fs_bio_set);
60 
61 /*
62  * Our slab pool management
63  */
64 struct bio_slab {
65 	struct kmem_cache *slab;
66 	unsigned int slab_ref;
67 	unsigned int slab_size;
68 	char name[8];
69 };
70 static DEFINE_MUTEX(bio_slab_lock);
71 static struct bio_slab *bio_slabs;
72 static unsigned int bio_slab_nr, bio_slab_max;
73 
74 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
75 {
76 	unsigned int sz = sizeof(struct bio) + extra_size;
77 	struct kmem_cache *slab = NULL;
78 	struct bio_slab *bslab, *new_bio_slabs;
79 	unsigned int new_bio_slab_max;
80 	unsigned int i, entry = -1;
81 
82 	mutex_lock(&bio_slab_lock);
83 
84 	i = 0;
85 	while (i < bio_slab_nr) {
86 		bslab = &bio_slabs[i];
87 
88 		if (!bslab->slab && entry == -1)
89 			entry = i;
90 		else if (bslab->slab_size == sz) {
91 			slab = bslab->slab;
92 			bslab->slab_ref++;
93 			break;
94 		}
95 		i++;
96 	}
97 
98 	if (slab)
99 		goto out_unlock;
100 
101 	if (bio_slab_nr == bio_slab_max && entry == -1) {
102 		new_bio_slab_max = bio_slab_max << 1;
103 		new_bio_slabs = krealloc(bio_slabs,
104 					 new_bio_slab_max * sizeof(struct bio_slab),
105 					 GFP_KERNEL);
106 		if (!new_bio_slabs)
107 			goto out_unlock;
108 		bio_slab_max = new_bio_slab_max;
109 		bio_slabs = new_bio_slabs;
110 	}
111 	if (entry == -1)
112 		entry = bio_slab_nr++;
113 
114 	bslab = &bio_slabs[entry];
115 
116 	snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
117 	slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
118 				 SLAB_HWCACHE_ALIGN, NULL);
119 	if (!slab)
120 		goto out_unlock;
121 
122 	bslab->slab = slab;
123 	bslab->slab_ref = 1;
124 	bslab->slab_size = sz;
125 out_unlock:
126 	mutex_unlock(&bio_slab_lock);
127 	return slab;
128 }
129 
130 static void bio_put_slab(struct bio_set *bs)
131 {
132 	struct bio_slab *bslab = NULL;
133 	unsigned int i;
134 
135 	mutex_lock(&bio_slab_lock);
136 
137 	for (i = 0; i < bio_slab_nr; i++) {
138 		if (bs->bio_slab == bio_slabs[i].slab) {
139 			bslab = &bio_slabs[i];
140 			break;
141 		}
142 	}
143 
144 	if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 		goto out;
146 
147 	WARN_ON(!bslab->slab_ref);
148 
149 	if (--bslab->slab_ref)
150 		goto out;
151 
152 	kmem_cache_destroy(bslab->slab);
153 	bslab->slab = NULL;
154 
155 out:
156 	mutex_unlock(&bio_slab_lock);
157 }
158 
159 unsigned int bvec_nr_vecs(unsigned short idx)
160 {
161 	return bvec_slabs[idx].nr_vecs;
162 }
163 
164 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
165 {
166 	if (!idx)
167 		return;
168 	idx--;
169 
170 	BIO_BUG_ON(idx >= BVEC_POOL_NR);
171 
172 	if (idx == BVEC_POOL_MAX) {
173 		mempool_free(bv, pool);
174 	} else {
175 		struct biovec_slab *bvs = bvec_slabs + idx;
176 
177 		kmem_cache_free(bvs->slab, bv);
178 	}
179 }
180 
181 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
182 			   mempool_t *pool)
183 {
184 	struct bio_vec *bvl;
185 
186 	/*
187 	 * see comment near bvec_array define!
188 	 */
189 	switch (nr) {
190 	case 1:
191 		*idx = 0;
192 		break;
193 	case 2 ... 4:
194 		*idx = 1;
195 		break;
196 	case 5 ... 16:
197 		*idx = 2;
198 		break;
199 	case 17 ... 64:
200 		*idx = 3;
201 		break;
202 	case 65 ... 128:
203 		*idx = 4;
204 		break;
205 	case 129 ... BIO_MAX_PAGES:
206 		*idx = 5;
207 		break;
208 	default:
209 		return NULL;
210 	}
211 
212 	/*
213 	 * idx now points to the pool we want to allocate from. only the
214 	 * 1-vec entry pool is mempool backed.
215 	 */
216 	if (*idx == BVEC_POOL_MAX) {
217 fallback:
218 		bvl = mempool_alloc(pool, gfp_mask);
219 	} else {
220 		struct biovec_slab *bvs = bvec_slabs + *idx;
221 		gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
222 
223 		/*
224 		 * Make this allocation restricted and don't dump info on
225 		 * allocation failures, since we'll fallback to the mempool
226 		 * in case of failure.
227 		 */
228 		__gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
229 
230 		/*
231 		 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
232 		 * is set, retry with the 1-entry mempool
233 		 */
234 		bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
235 		if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
236 			*idx = BVEC_POOL_MAX;
237 			goto fallback;
238 		}
239 	}
240 
241 	(*idx)++;
242 	return bvl;
243 }
244 
245 void bio_uninit(struct bio *bio)
246 {
247 	bio_disassociate_task(bio);
248 }
249 EXPORT_SYMBOL(bio_uninit);
250 
251 static void bio_free(struct bio *bio)
252 {
253 	struct bio_set *bs = bio->bi_pool;
254 	void *p;
255 
256 	bio_uninit(bio);
257 
258 	if (bs) {
259 		bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
260 
261 		/*
262 		 * If we have front padding, adjust the bio pointer before freeing
263 		 */
264 		p = bio;
265 		p -= bs->front_pad;
266 
267 		mempool_free(p, &bs->bio_pool);
268 	} else {
269 		/* Bio was allocated by bio_kmalloc() */
270 		kfree(bio);
271 	}
272 }
273 
274 /*
275  * Users of this function have their own bio allocation. Subsequently,
276  * they must remember to pair any call to bio_init() with bio_uninit()
277  * when IO has completed, or when the bio is released.
278  */
279 void bio_init(struct bio *bio, struct bio_vec *table,
280 	      unsigned short max_vecs)
281 {
282 	memset(bio, 0, sizeof(*bio));
283 	atomic_set(&bio->__bi_remaining, 1);
284 	atomic_set(&bio->__bi_cnt, 1);
285 
286 	bio->bi_io_vec = table;
287 	bio->bi_max_vecs = max_vecs;
288 }
289 EXPORT_SYMBOL(bio_init);
290 
291 /**
292  * bio_reset - reinitialize a bio
293  * @bio:	bio to reset
294  *
295  * Description:
296  *   After calling bio_reset(), @bio will be in the same state as a freshly
297  *   allocated bio returned bio bio_alloc_bioset() - the only fields that are
298  *   preserved are the ones that are initialized by bio_alloc_bioset(). See
299  *   comment in struct bio.
300  */
301 void bio_reset(struct bio *bio)
302 {
303 	unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
304 
305 	bio_uninit(bio);
306 
307 	memset(bio, 0, BIO_RESET_BYTES);
308 	bio->bi_flags = flags;
309 	atomic_set(&bio->__bi_remaining, 1);
310 }
311 EXPORT_SYMBOL(bio_reset);
312 
313 static struct bio *__bio_chain_endio(struct bio *bio)
314 {
315 	struct bio *parent = bio->bi_private;
316 
317 	if (!parent->bi_status)
318 		parent->bi_status = bio->bi_status;
319 	bio_put(bio);
320 	return parent;
321 }
322 
323 static void bio_chain_endio(struct bio *bio)
324 {
325 	bio_endio(__bio_chain_endio(bio));
326 }
327 
328 /**
329  * bio_chain - chain bio completions
330  * @bio: the target bio
331  * @parent: the @bio's parent bio
332  *
333  * The caller won't have a bi_end_io called when @bio completes - instead,
334  * @parent's bi_end_io won't be called until both @parent and @bio have
335  * completed; the chained bio will also be freed when it completes.
336  *
337  * The caller must not set bi_private or bi_end_io in @bio.
338  */
339 void bio_chain(struct bio *bio, struct bio *parent)
340 {
341 	BUG_ON(bio->bi_private || bio->bi_end_io);
342 
343 	bio->bi_private = parent;
344 	bio->bi_end_io	= bio_chain_endio;
345 	bio_inc_remaining(parent);
346 }
347 EXPORT_SYMBOL(bio_chain);
348 
349 static void bio_alloc_rescue(struct work_struct *work)
350 {
351 	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
352 	struct bio *bio;
353 
354 	while (1) {
355 		spin_lock(&bs->rescue_lock);
356 		bio = bio_list_pop(&bs->rescue_list);
357 		spin_unlock(&bs->rescue_lock);
358 
359 		if (!bio)
360 			break;
361 
362 		generic_make_request(bio);
363 	}
364 }
365 
366 static void punt_bios_to_rescuer(struct bio_set *bs)
367 {
368 	struct bio_list punt, nopunt;
369 	struct bio *bio;
370 
371 	if (WARN_ON_ONCE(!bs->rescue_workqueue))
372 		return;
373 	/*
374 	 * In order to guarantee forward progress we must punt only bios that
375 	 * were allocated from this bio_set; otherwise, if there was a bio on
376 	 * there for a stacking driver higher up in the stack, processing it
377 	 * could require allocating bios from this bio_set, and doing that from
378 	 * our own rescuer would be bad.
379 	 *
380 	 * Since bio lists are singly linked, pop them all instead of trying to
381 	 * remove from the middle of the list:
382 	 */
383 
384 	bio_list_init(&punt);
385 	bio_list_init(&nopunt);
386 
387 	while ((bio = bio_list_pop(&current->bio_list[0])))
388 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
389 	current->bio_list[0] = nopunt;
390 
391 	bio_list_init(&nopunt);
392 	while ((bio = bio_list_pop(&current->bio_list[1])))
393 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
394 	current->bio_list[1] = nopunt;
395 
396 	spin_lock(&bs->rescue_lock);
397 	bio_list_merge(&bs->rescue_list, &punt);
398 	spin_unlock(&bs->rescue_lock);
399 
400 	queue_work(bs->rescue_workqueue, &bs->rescue_work);
401 }
402 
403 /**
404  * bio_alloc_bioset - allocate a bio for I/O
405  * @gfp_mask:   the GFP_* mask given to the slab allocator
406  * @nr_iovecs:	number of iovecs to pre-allocate
407  * @bs:		the bio_set to allocate from.
408  *
409  * Description:
410  *   If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
411  *   backed by the @bs's mempool.
412  *
413  *   When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
414  *   always be able to allocate a bio. This is due to the mempool guarantees.
415  *   To make this work, callers must never allocate more than 1 bio at a time
416  *   from this pool. Callers that need to allocate more than 1 bio must always
417  *   submit the previously allocated bio for IO before attempting to allocate
418  *   a new one. Failure to do so can cause deadlocks under memory pressure.
419  *
420  *   Note that when running under generic_make_request() (i.e. any block
421  *   driver), bios are not submitted until after you return - see the code in
422  *   generic_make_request() that converts recursion into iteration, to prevent
423  *   stack overflows.
424  *
425  *   This would normally mean allocating multiple bios under
426  *   generic_make_request() would be susceptible to deadlocks, but we have
427  *   deadlock avoidance code that resubmits any blocked bios from a rescuer
428  *   thread.
429  *
430  *   However, we do not guarantee forward progress for allocations from other
431  *   mempools. Doing multiple allocations from the same mempool under
432  *   generic_make_request() should be avoided - instead, use bio_set's front_pad
433  *   for per bio allocations.
434  *
435  *   RETURNS:
436  *   Pointer to new bio on success, NULL on failure.
437  */
438 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
439 			     struct bio_set *bs)
440 {
441 	gfp_t saved_gfp = gfp_mask;
442 	unsigned front_pad;
443 	unsigned inline_vecs;
444 	struct bio_vec *bvl = NULL;
445 	struct bio *bio;
446 	void *p;
447 
448 	if (!bs) {
449 		if (nr_iovecs > UIO_MAXIOV)
450 			return NULL;
451 
452 		p = kmalloc(sizeof(struct bio) +
453 			    nr_iovecs * sizeof(struct bio_vec),
454 			    gfp_mask);
455 		front_pad = 0;
456 		inline_vecs = nr_iovecs;
457 	} else {
458 		/* should not use nobvec bioset for nr_iovecs > 0 */
459 		if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) &&
460 				 nr_iovecs > 0))
461 			return NULL;
462 		/*
463 		 * generic_make_request() converts recursion to iteration; this
464 		 * means if we're running beneath it, any bios we allocate and
465 		 * submit will not be submitted (and thus freed) until after we
466 		 * return.
467 		 *
468 		 * This exposes us to a potential deadlock if we allocate
469 		 * multiple bios from the same bio_set() while running
470 		 * underneath generic_make_request(). If we were to allocate
471 		 * multiple bios (say a stacking block driver that was splitting
472 		 * bios), we would deadlock if we exhausted the mempool's
473 		 * reserve.
474 		 *
475 		 * We solve this, and guarantee forward progress, with a rescuer
476 		 * workqueue per bio_set. If we go to allocate and there are
477 		 * bios on current->bio_list, we first try the allocation
478 		 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
479 		 * bios we would be blocking to the rescuer workqueue before
480 		 * we retry with the original gfp_flags.
481 		 */
482 
483 		if (current->bio_list &&
484 		    (!bio_list_empty(&current->bio_list[0]) ||
485 		     !bio_list_empty(&current->bio_list[1])) &&
486 		    bs->rescue_workqueue)
487 			gfp_mask &= ~__GFP_DIRECT_RECLAIM;
488 
489 		p = mempool_alloc(&bs->bio_pool, gfp_mask);
490 		if (!p && gfp_mask != saved_gfp) {
491 			punt_bios_to_rescuer(bs);
492 			gfp_mask = saved_gfp;
493 			p = mempool_alloc(&bs->bio_pool, gfp_mask);
494 		}
495 
496 		front_pad = bs->front_pad;
497 		inline_vecs = BIO_INLINE_VECS;
498 	}
499 
500 	if (unlikely(!p))
501 		return NULL;
502 
503 	bio = p + front_pad;
504 	bio_init(bio, NULL, 0);
505 
506 	if (nr_iovecs > inline_vecs) {
507 		unsigned long idx = 0;
508 
509 		bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
510 		if (!bvl && gfp_mask != saved_gfp) {
511 			punt_bios_to_rescuer(bs);
512 			gfp_mask = saved_gfp;
513 			bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool);
514 		}
515 
516 		if (unlikely(!bvl))
517 			goto err_free;
518 
519 		bio->bi_flags |= idx << BVEC_POOL_OFFSET;
520 	} else if (nr_iovecs) {
521 		bvl = bio->bi_inline_vecs;
522 	}
523 
524 	bio->bi_pool = bs;
525 	bio->bi_max_vecs = nr_iovecs;
526 	bio->bi_io_vec = bvl;
527 	return bio;
528 
529 err_free:
530 	mempool_free(p, &bs->bio_pool);
531 	return NULL;
532 }
533 EXPORT_SYMBOL(bio_alloc_bioset);
534 
535 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
536 {
537 	unsigned long flags;
538 	struct bio_vec bv;
539 	struct bvec_iter iter;
540 
541 	__bio_for_each_segment(bv, bio, iter, start) {
542 		char *data = bvec_kmap_irq(&bv, &flags);
543 		memset(data, 0, bv.bv_len);
544 		flush_dcache_page(bv.bv_page);
545 		bvec_kunmap_irq(data, &flags);
546 	}
547 }
548 EXPORT_SYMBOL(zero_fill_bio_iter);
549 
550 /**
551  * bio_put - release a reference to a bio
552  * @bio:   bio to release reference to
553  *
554  * Description:
555  *   Put a reference to a &struct bio, either one you have gotten with
556  *   bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
557  **/
558 void bio_put(struct bio *bio)
559 {
560 	if (!bio_flagged(bio, BIO_REFFED))
561 		bio_free(bio);
562 	else {
563 		BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
564 
565 		/*
566 		 * last put frees it
567 		 */
568 		if (atomic_dec_and_test(&bio->__bi_cnt))
569 			bio_free(bio);
570 	}
571 }
572 EXPORT_SYMBOL(bio_put);
573 
574 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
575 {
576 	if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
577 		blk_recount_segments(q, bio);
578 
579 	return bio->bi_phys_segments;
580 }
581 EXPORT_SYMBOL(bio_phys_segments);
582 
583 /**
584  * 	__bio_clone_fast - clone a bio that shares the original bio's biovec
585  * 	@bio: destination bio
586  * 	@bio_src: bio to clone
587  *
588  *	Clone a &bio. Caller will own the returned bio, but not
589  *	the actual data it points to. Reference count of returned
590  * 	bio will be one.
591  *
592  * 	Caller must ensure that @bio_src is not freed before @bio.
593  */
594 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
595 {
596 	BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
597 
598 	/*
599 	 * most users will be overriding ->bi_disk with a new target,
600 	 * so we don't set nor calculate new physical/hw segment counts here
601 	 */
602 	bio->bi_disk = bio_src->bi_disk;
603 	bio->bi_partno = bio_src->bi_partno;
604 	bio_set_flag(bio, BIO_CLONED);
605 	if (bio_flagged(bio_src, BIO_THROTTLED))
606 		bio_set_flag(bio, BIO_THROTTLED);
607 	bio->bi_opf = bio_src->bi_opf;
608 	bio->bi_write_hint = bio_src->bi_write_hint;
609 	bio->bi_iter = bio_src->bi_iter;
610 	bio->bi_io_vec = bio_src->bi_io_vec;
611 
612 	bio_clone_blkcg_association(bio, bio_src);
613 }
614 EXPORT_SYMBOL(__bio_clone_fast);
615 
616 /**
617  *	bio_clone_fast - clone a bio that shares the original bio's biovec
618  *	@bio: bio to clone
619  *	@gfp_mask: allocation priority
620  *	@bs: bio_set to allocate from
621  *
622  * 	Like __bio_clone_fast, only also allocates the returned bio
623  */
624 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
625 {
626 	struct bio *b;
627 
628 	b = bio_alloc_bioset(gfp_mask, 0, bs);
629 	if (!b)
630 		return NULL;
631 
632 	__bio_clone_fast(b, bio);
633 
634 	if (bio_integrity(bio)) {
635 		int ret;
636 
637 		ret = bio_integrity_clone(b, bio, gfp_mask);
638 
639 		if (ret < 0) {
640 			bio_put(b);
641 			return NULL;
642 		}
643 	}
644 
645 	return b;
646 }
647 EXPORT_SYMBOL(bio_clone_fast);
648 
649 /**
650  * 	bio_clone_bioset - clone a bio
651  * 	@bio_src: bio to clone
652  *	@gfp_mask: allocation priority
653  *	@bs: bio_set to allocate from
654  *
655  *	Clone bio. Caller will own the returned bio, but not the actual data it
656  *	points to. Reference count of returned bio will be one.
657  */
658 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
659 			     struct bio_set *bs)
660 {
661 	struct bvec_iter iter;
662 	struct bio_vec bv;
663 	struct bio *bio;
664 
665 	/*
666 	 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
667 	 * bio_src->bi_io_vec to bio->bi_io_vec.
668 	 *
669 	 * We can't do that anymore, because:
670 	 *
671 	 *  - The point of cloning the biovec is to produce a bio with a biovec
672 	 *    the caller can modify: bi_idx and bi_bvec_done should be 0.
673 	 *
674 	 *  - The original bio could've had more than BIO_MAX_PAGES biovecs; if
675 	 *    we tried to clone the whole thing bio_alloc_bioset() would fail.
676 	 *    But the clone should succeed as long as the number of biovecs we
677 	 *    actually need to allocate is fewer than BIO_MAX_PAGES.
678 	 *
679 	 *  - Lastly, bi_vcnt should not be looked at or relied upon by code
680 	 *    that does not own the bio - reason being drivers don't use it for
681 	 *    iterating over the biovec anymore, so expecting it to be kept up
682 	 *    to date (i.e. for clones that share the parent biovec) is just
683 	 *    asking for trouble and would force extra work on
684 	 *    __bio_clone_fast() anyways.
685 	 */
686 
687 	bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
688 	if (!bio)
689 		return NULL;
690 	bio->bi_disk		= bio_src->bi_disk;
691 	bio->bi_opf		= bio_src->bi_opf;
692 	bio->bi_write_hint	= bio_src->bi_write_hint;
693 	bio->bi_iter.bi_sector	= bio_src->bi_iter.bi_sector;
694 	bio->bi_iter.bi_size	= bio_src->bi_iter.bi_size;
695 
696 	switch (bio_op(bio)) {
697 	case REQ_OP_DISCARD:
698 	case REQ_OP_SECURE_ERASE:
699 	case REQ_OP_WRITE_ZEROES:
700 		break;
701 	case REQ_OP_WRITE_SAME:
702 		bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
703 		break;
704 	default:
705 		bio_for_each_segment(bv, bio_src, iter)
706 			bio->bi_io_vec[bio->bi_vcnt++] = bv;
707 		break;
708 	}
709 
710 	if (bio_integrity(bio_src)) {
711 		int ret;
712 
713 		ret = bio_integrity_clone(bio, bio_src, gfp_mask);
714 		if (ret < 0) {
715 			bio_put(bio);
716 			return NULL;
717 		}
718 	}
719 
720 	bio_clone_blkcg_association(bio, bio_src);
721 
722 	return bio;
723 }
724 EXPORT_SYMBOL(bio_clone_bioset);
725 
726 /**
727  *	bio_add_pc_page	-	attempt to add page to bio
728  *	@q: the target queue
729  *	@bio: destination bio
730  *	@page: page to add
731  *	@len: vec entry length
732  *	@offset: vec entry offset
733  *
734  *	Attempt to add a page to the bio_vec maplist. This can fail for a
735  *	number of reasons, such as the bio being full or target block device
736  *	limitations. The target block device must allow bio's up to PAGE_SIZE,
737  *	so it is always possible to add a single page to an empty bio.
738  *
739  *	This should only be used by REQ_PC bios.
740  */
741 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
742 		    *page, unsigned int len, unsigned int offset)
743 {
744 	int retried_segments = 0;
745 	struct bio_vec *bvec;
746 
747 	/*
748 	 * cloned bio must not modify vec list
749 	 */
750 	if (unlikely(bio_flagged(bio, BIO_CLONED)))
751 		return 0;
752 
753 	if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
754 		return 0;
755 
756 	/*
757 	 * For filesystems with a blocksize smaller than the pagesize
758 	 * we will often be called with the same page as last time and
759 	 * a consecutive offset.  Optimize this special case.
760 	 */
761 	if (bio->bi_vcnt > 0) {
762 		struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
763 
764 		if (page == prev->bv_page &&
765 		    offset == prev->bv_offset + prev->bv_len) {
766 			prev->bv_len += len;
767 			bio->bi_iter.bi_size += len;
768 			goto done;
769 		}
770 
771 		/*
772 		 * If the queue doesn't support SG gaps and adding this
773 		 * offset would create a gap, disallow it.
774 		 */
775 		if (bvec_gap_to_prev(q, prev, offset))
776 			return 0;
777 	}
778 
779 	if (bio_full(bio))
780 		return 0;
781 
782 	/*
783 	 * setup the new entry, we might clear it again later if we
784 	 * cannot add the page
785 	 */
786 	bvec = &bio->bi_io_vec[bio->bi_vcnt];
787 	bvec->bv_page = page;
788 	bvec->bv_len = len;
789 	bvec->bv_offset = offset;
790 	bio->bi_vcnt++;
791 	bio->bi_phys_segments++;
792 	bio->bi_iter.bi_size += len;
793 
794 	/*
795 	 * Perform a recount if the number of segments is greater
796 	 * than queue_max_segments(q).
797 	 */
798 
799 	while (bio->bi_phys_segments > queue_max_segments(q)) {
800 
801 		if (retried_segments)
802 			goto failed;
803 
804 		retried_segments = 1;
805 		blk_recount_segments(q, bio);
806 	}
807 
808 	/* If we may be able to merge these biovecs, force a recount */
809 	if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
810 		bio_clear_flag(bio, BIO_SEG_VALID);
811 
812  done:
813 	return len;
814 
815  failed:
816 	bvec->bv_page = NULL;
817 	bvec->bv_len = 0;
818 	bvec->bv_offset = 0;
819 	bio->bi_vcnt--;
820 	bio->bi_iter.bi_size -= len;
821 	blk_recount_segments(q, bio);
822 	return 0;
823 }
824 EXPORT_SYMBOL(bio_add_pc_page);
825 
826 /**
827  * __bio_try_merge_page - try appending data to an existing bvec.
828  * @bio: destination bio
829  * @page: page to add
830  * @len: length of the data to add
831  * @off: offset of the data in @page
832  *
833  * Try to add the data at @page + @off to the last bvec of @bio.  This is a
834  * a useful optimisation for file systems with a block size smaller than the
835  * page size.
836  *
837  * Return %true on success or %false on failure.
838  */
839 bool __bio_try_merge_page(struct bio *bio, struct page *page,
840 		unsigned int len, unsigned int off)
841 {
842 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
843 		return false;
844 
845 	if (bio->bi_vcnt > 0) {
846 		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
847 
848 		if (page == bv->bv_page && off == bv->bv_offset + bv->bv_len) {
849 			bv->bv_len += len;
850 			bio->bi_iter.bi_size += len;
851 			return true;
852 		}
853 	}
854 	return false;
855 }
856 EXPORT_SYMBOL_GPL(__bio_try_merge_page);
857 
858 /**
859  * __bio_add_page - add page to a bio in a new segment
860  * @bio: destination bio
861  * @page: page to add
862  * @len: length of the data to add
863  * @off: offset of the data in @page
864  *
865  * Add the data at @page + @off to @bio as a new bvec.  The caller must ensure
866  * that @bio has space for another bvec.
867  */
868 void __bio_add_page(struct bio *bio, struct page *page,
869 		unsigned int len, unsigned int off)
870 {
871 	struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
872 
873 	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
874 	WARN_ON_ONCE(bio_full(bio));
875 
876 	bv->bv_page = page;
877 	bv->bv_offset = off;
878 	bv->bv_len = len;
879 
880 	bio->bi_iter.bi_size += len;
881 	bio->bi_vcnt++;
882 }
883 EXPORT_SYMBOL_GPL(__bio_add_page);
884 
885 /**
886  *	bio_add_page	-	attempt to add page to bio
887  *	@bio: destination bio
888  *	@page: page to add
889  *	@len: vec entry length
890  *	@offset: vec entry offset
891  *
892  *	Attempt to add a page to the bio_vec maplist. This will only fail
893  *	if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
894  */
895 int bio_add_page(struct bio *bio, struct page *page,
896 		 unsigned int len, unsigned int offset)
897 {
898 	if (!__bio_try_merge_page(bio, page, len, offset)) {
899 		if (bio_full(bio))
900 			return 0;
901 		__bio_add_page(bio, page, len, offset);
902 	}
903 	return len;
904 }
905 EXPORT_SYMBOL(bio_add_page);
906 
907 /**
908  * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
909  * @bio: bio to add pages to
910  * @iter: iov iterator describing the region to be mapped
911  *
912  * Pins as many pages from *iter and appends them to @bio's bvec array. The
913  * pages will have to be released using put_page() when done.
914  */
915 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
916 {
917 	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
918 	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
919 	struct page **pages = (struct page **)bv;
920 	size_t offset, diff;
921 	ssize_t size;
922 
923 	size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
924 	if (unlikely(size <= 0))
925 		return size ? size : -EFAULT;
926 	nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE;
927 
928 	/*
929 	 * Deep magic below:  We need to walk the pinned pages backwards
930 	 * because we are abusing the space allocated for the bio_vecs
931 	 * for the page array.  Because the bio_vecs are larger than the
932 	 * page pointers by definition this will always work.  But it also
933 	 * means we can't use bio_add_page, so any changes to it's semantics
934 	 * need to be reflected here as well.
935 	 */
936 	bio->bi_iter.bi_size += size;
937 	bio->bi_vcnt += nr_pages;
938 
939 	diff = (nr_pages * PAGE_SIZE - offset) - size;
940 	while (nr_pages--) {
941 		bv[nr_pages].bv_page = pages[nr_pages];
942 		bv[nr_pages].bv_len = PAGE_SIZE;
943 		bv[nr_pages].bv_offset = 0;
944 	}
945 
946 	bv[0].bv_offset += offset;
947 	bv[0].bv_len -= offset;
948 	if (diff)
949 		bv[bio->bi_vcnt - 1].bv_len -= diff;
950 
951 	iov_iter_advance(iter, size);
952 	return 0;
953 }
954 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
955 
956 static void submit_bio_wait_endio(struct bio *bio)
957 {
958 	complete(bio->bi_private);
959 }
960 
961 /**
962  * submit_bio_wait - submit a bio, and wait until it completes
963  * @bio: The &struct bio which describes the I/O
964  *
965  * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
966  * bio_endio() on failure.
967  *
968  * WARNING: Unlike to how submit_bio() is usually used, this function does not
969  * result in bio reference to be consumed. The caller must drop the reference
970  * on his own.
971  */
972 int submit_bio_wait(struct bio *bio)
973 {
974 	DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
975 
976 	bio->bi_private = &done;
977 	bio->bi_end_io = submit_bio_wait_endio;
978 	bio->bi_opf |= REQ_SYNC;
979 	submit_bio(bio);
980 	wait_for_completion_io(&done);
981 
982 	return blk_status_to_errno(bio->bi_status);
983 }
984 EXPORT_SYMBOL(submit_bio_wait);
985 
986 /**
987  * bio_advance - increment/complete a bio by some number of bytes
988  * @bio:	bio to advance
989  * @bytes:	number of bytes to complete
990  *
991  * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
992  * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
993  * be updated on the last bvec as well.
994  *
995  * @bio will then represent the remaining, uncompleted portion of the io.
996  */
997 void bio_advance(struct bio *bio, unsigned bytes)
998 {
999 	if (bio_integrity(bio))
1000 		bio_integrity_advance(bio, bytes);
1001 
1002 	bio_advance_iter(bio, &bio->bi_iter, bytes);
1003 }
1004 EXPORT_SYMBOL(bio_advance);
1005 
1006 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1007 			struct bio *src, struct bvec_iter *src_iter)
1008 {
1009 	struct bio_vec src_bv, dst_bv;
1010 	void *src_p, *dst_p;
1011 	unsigned bytes;
1012 
1013 	while (src_iter->bi_size && dst_iter->bi_size) {
1014 		src_bv = bio_iter_iovec(src, *src_iter);
1015 		dst_bv = bio_iter_iovec(dst, *dst_iter);
1016 
1017 		bytes = min(src_bv.bv_len, dst_bv.bv_len);
1018 
1019 		src_p = kmap_atomic(src_bv.bv_page);
1020 		dst_p = kmap_atomic(dst_bv.bv_page);
1021 
1022 		memcpy(dst_p + dst_bv.bv_offset,
1023 		       src_p + src_bv.bv_offset,
1024 		       bytes);
1025 
1026 		kunmap_atomic(dst_p);
1027 		kunmap_atomic(src_p);
1028 
1029 		flush_dcache_page(dst_bv.bv_page);
1030 
1031 		bio_advance_iter(src, src_iter, bytes);
1032 		bio_advance_iter(dst, dst_iter, bytes);
1033 	}
1034 }
1035 EXPORT_SYMBOL(bio_copy_data_iter);
1036 
1037 /**
1038  * bio_copy_data - copy contents of data buffers from one bio to another
1039  * @src: source bio
1040  * @dst: destination bio
1041  *
1042  * Stops when it reaches the end of either @src or @dst - that is, copies
1043  * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1044  */
1045 void bio_copy_data(struct bio *dst, struct bio *src)
1046 {
1047 	struct bvec_iter src_iter = src->bi_iter;
1048 	struct bvec_iter dst_iter = dst->bi_iter;
1049 
1050 	bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1051 }
1052 EXPORT_SYMBOL(bio_copy_data);
1053 
1054 /**
1055  * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1056  * another
1057  * @src: source bio list
1058  * @dst: destination bio list
1059  *
1060  * Stops when it reaches the end of either the @src list or @dst list - that is,
1061  * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1062  * bios).
1063  */
1064 void bio_list_copy_data(struct bio *dst, struct bio *src)
1065 {
1066 	struct bvec_iter src_iter = src->bi_iter;
1067 	struct bvec_iter dst_iter = dst->bi_iter;
1068 
1069 	while (1) {
1070 		if (!src_iter.bi_size) {
1071 			src = src->bi_next;
1072 			if (!src)
1073 				break;
1074 
1075 			src_iter = src->bi_iter;
1076 		}
1077 
1078 		if (!dst_iter.bi_size) {
1079 			dst = dst->bi_next;
1080 			if (!dst)
1081 				break;
1082 
1083 			dst_iter = dst->bi_iter;
1084 		}
1085 
1086 		bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1087 	}
1088 }
1089 EXPORT_SYMBOL(bio_list_copy_data);
1090 
1091 struct bio_map_data {
1092 	int is_our_pages;
1093 	struct iov_iter iter;
1094 	struct iovec iov[];
1095 };
1096 
1097 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
1098 					       gfp_t gfp_mask)
1099 {
1100 	struct bio_map_data *bmd;
1101 	if (data->nr_segs > UIO_MAXIOV)
1102 		return NULL;
1103 
1104 	bmd = kmalloc(sizeof(struct bio_map_data) +
1105 		       sizeof(struct iovec) * data->nr_segs, gfp_mask);
1106 	if (!bmd)
1107 		return NULL;
1108 	memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1109 	bmd->iter = *data;
1110 	bmd->iter.iov = bmd->iov;
1111 	return bmd;
1112 }
1113 
1114 /**
1115  * bio_copy_from_iter - copy all pages from iov_iter to bio
1116  * @bio: The &struct bio which describes the I/O as destination
1117  * @iter: iov_iter as source
1118  *
1119  * Copy all pages from iov_iter to bio.
1120  * Returns 0 on success, or error on failure.
1121  */
1122 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
1123 {
1124 	int i;
1125 	struct bio_vec *bvec;
1126 
1127 	bio_for_each_segment_all(bvec, bio, i) {
1128 		ssize_t ret;
1129 
1130 		ret = copy_page_from_iter(bvec->bv_page,
1131 					  bvec->bv_offset,
1132 					  bvec->bv_len,
1133 					  iter);
1134 
1135 		if (!iov_iter_count(iter))
1136 			break;
1137 
1138 		if (ret < bvec->bv_len)
1139 			return -EFAULT;
1140 	}
1141 
1142 	return 0;
1143 }
1144 
1145 /**
1146  * bio_copy_to_iter - copy all pages from bio to iov_iter
1147  * @bio: The &struct bio which describes the I/O as source
1148  * @iter: iov_iter as destination
1149  *
1150  * Copy all pages from bio to iov_iter.
1151  * Returns 0 on success, or error on failure.
1152  */
1153 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1154 {
1155 	int i;
1156 	struct bio_vec *bvec;
1157 
1158 	bio_for_each_segment_all(bvec, bio, i) {
1159 		ssize_t ret;
1160 
1161 		ret = copy_page_to_iter(bvec->bv_page,
1162 					bvec->bv_offset,
1163 					bvec->bv_len,
1164 					&iter);
1165 
1166 		if (!iov_iter_count(&iter))
1167 			break;
1168 
1169 		if (ret < bvec->bv_len)
1170 			return -EFAULT;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 void bio_free_pages(struct bio *bio)
1177 {
1178 	struct bio_vec *bvec;
1179 	int i;
1180 
1181 	bio_for_each_segment_all(bvec, bio, i)
1182 		__free_page(bvec->bv_page);
1183 }
1184 EXPORT_SYMBOL(bio_free_pages);
1185 
1186 /**
1187  *	bio_uncopy_user	-	finish previously mapped bio
1188  *	@bio: bio being terminated
1189  *
1190  *	Free pages allocated from bio_copy_user_iov() and write back data
1191  *	to user space in case of a read.
1192  */
1193 int bio_uncopy_user(struct bio *bio)
1194 {
1195 	struct bio_map_data *bmd = bio->bi_private;
1196 	int ret = 0;
1197 
1198 	if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1199 		/*
1200 		 * if we're in a workqueue, the request is orphaned, so
1201 		 * don't copy into a random user address space, just free
1202 		 * and return -EINTR so user space doesn't expect any data.
1203 		 */
1204 		if (!current->mm)
1205 			ret = -EINTR;
1206 		else if (bio_data_dir(bio) == READ)
1207 			ret = bio_copy_to_iter(bio, bmd->iter);
1208 		if (bmd->is_our_pages)
1209 			bio_free_pages(bio);
1210 	}
1211 	kfree(bmd);
1212 	bio_put(bio);
1213 	return ret;
1214 }
1215 
1216 /**
1217  *	bio_copy_user_iov	-	copy user data to bio
1218  *	@q:		destination block queue
1219  *	@map_data:	pointer to the rq_map_data holding pages (if necessary)
1220  *	@iter:		iovec iterator
1221  *	@gfp_mask:	memory allocation flags
1222  *
1223  *	Prepares and returns a bio for indirect user io, bouncing data
1224  *	to/from kernel pages as necessary. Must be paired with
1225  *	call bio_uncopy_user() on io completion.
1226  */
1227 struct bio *bio_copy_user_iov(struct request_queue *q,
1228 			      struct rq_map_data *map_data,
1229 			      struct iov_iter *iter,
1230 			      gfp_t gfp_mask)
1231 {
1232 	struct bio_map_data *bmd;
1233 	struct page *page;
1234 	struct bio *bio;
1235 	int i = 0, ret;
1236 	int nr_pages;
1237 	unsigned int len = iter->count;
1238 	unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1239 
1240 	bmd = bio_alloc_map_data(iter, gfp_mask);
1241 	if (!bmd)
1242 		return ERR_PTR(-ENOMEM);
1243 
1244 	/*
1245 	 * We need to do a deep copy of the iov_iter including the iovecs.
1246 	 * The caller provided iov might point to an on-stack or otherwise
1247 	 * shortlived one.
1248 	 */
1249 	bmd->is_our_pages = map_data ? 0 : 1;
1250 
1251 	nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1252 	if (nr_pages > BIO_MAX_PAGES)
1253 		nr_pages = BIO_MAX_PAGES;
1254 
1255 	ret = -ENOMEM;
1256 	bio = bio_kmalloc(gfp_mask, nr_pages);
1257 	if (!bio)
1258 		goto out_bmd;
1259 
1260 	ret = 0;
1261 
1262 	if (map_data) {
1263 		nr_pages = 1 << map_data->page_order;
1264 		i = map_data->offset / PAGE_SIZE;
1265 	}
1266 	while (len) {
1267 		unsigned int bytes = PAGE_SIZE;
1268 
1269 		bytes -= offset;
1270 
1271 		if (bytes > len)
1272 			bytes = len;
1273 
1274 		if (map_data) {
1275 			if (i == map_data->nr_entries * nr_pages) {
1276 				ret = -ENOMEM;
1277 				break;
1278 			}
1279 
1280 			page = map_data->pages[i / nr_pages];
1281 			page += (i % nr_pages);
1282 
1283 			i++;
1284 		} else {
1285 			page = alloc_page(q->bounce_gfp | gfp_mask);
1286 			if (!page) {
1287 				ret = -ENOMEM;
1288 				break;
1289 			}
1290 		}
1291 
1292 		if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1293 			break;
1294 
1295 		len -= bytes;
1296 		offset = 0;
1297 	}
1298 
1299 	if (ret)
1300 		goto cleanup;
1301 
1302 	if (map_data)
1303 		map_data->offset += bio->bi_iter.bi_size;
1304 
1305 	/*
1306 	 * success
1307 	 */
1308 	if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1309 	    (map_data && map_data->from_user)) {
1310 		ret = bio_copy_from_iter(bio, iter);
1311 		if (ret)
1312 			goto cleanup;
1313 	} else {
1314 		iov_iter_advance(iter, bio->bi_iter.bi_size);
1315 	}
1316 
1317 	bio->bi_private = bmd;
1318 	if (map_data && map_data->null_mapped)
1319 		bio_set_flag(bio, BIO_NULL_MAPPED);
1320 	return bio;
1321 cleanup:
1322 	if (!map_data)
1323 		bio_free_pages(bio);
1324 	bio_put(bio);
1325 out_bmd:
1326 	kfree(bmd);
1327 	return ERR_PTR(ret);
1328 }
1329 
1330 /**
1331  *	bio_map_user_iov - map user iovec into bio
1332  *	@q:		the struct request_queue for the bio
1333  *	@iter:		iovec iterator
1334  *	@gfp_mask:	memory allocation flags
1335  *
1336  *	Map the user space address into a bio suitable for io to a block
1337  *	device. Returns an error pointer in case of error.
1338  */
1339 struct bio *bio_map_user_iov(struct request_queue *q,
1340 			     struct iov_iter *iter,
1341 			     gfp_t gfp_mask)
1342 {
1343 	int j;
1344 	struct bio *bio;
1345 	int ret;
1346 	struct bio_vec *bvec;
1347 
1348 	if (!iov_iter_count(iter))
1349 		return ERR_PTR(-EINVAL);
1350 
1351 	bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
1352 	if (!bio)
1353 		return ERR_PTR(-ENOMEM);
1354 
1355 	while (iov_iter_count(iter)) {
1356 		struct page **pages;
1357 		ssize_t bytes;
1358 		size_t offs, added = 0;
1359 		int npages;
1360 
1361 		bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
1362 		if (unlikely(bytes <= 0)) {
1363 			ret = bytes ? bytes : -EFAULT;
1364 			goto out_unmap;
1365 		}
1366 
1367 		npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
1368 
1369 		if (unlikely(offs & queue_dma_alignment(q))) {
1370 			ret = -EINVAL;
1371 			j = 0;
1372 		} else {
1373 			for (j = 0; j < npages; j++) {
1374 				struct page *page = pages[j];
1375 				unsigned int n = PAGE_SIZE - offs;
1376 				unsigned short prev_bi_vcnt = bio->bi_vcnt;
1377 
1378 				if (n > bytes)
1379 					n = bytes;
1380 
1381 				if (!bio_add_pc_page(q, bio, page, n, offs))
1382 					break;
1383 
1384 				/*
1385 				 * check if vector was merged with previous
1386 				 * drop page reference if needed
1387 				 */
1388 				if (bio->bi_vcnt == prev_bi_vcnt)
1389 					put_page(page);
1390 
1391 				added += n;
1392 				bytes -= n;
1393 				offs = 0;
1394 			}
1395 			iov_iter_advance(iter, added);
1396 		}
1397 		/*
1398 		 * release the pages we didn't map into the bio, if any
1399 		 */
1400 		while (j < npages)
1401 			put_page(pages[j++]);
1402 		kvfree(pages);
1403 		/* couldn't stuff something into bio? */
1404 		if (bytes)
1405 			break;
1406 	}
1407 
1408 	bio_set_flag(bio, BIO_USER_MAPPED);
1409 
1410 	/*
1411 	 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1412 	 * it would normally disappear when its bi_end_io is run.
1413 	 * however, we need it for the unmap, so grab an extra
1414 	 * reference to it
1415 	 */
1416 	bio_get(bio);
1417 	return bio;
1418 
1419  out_unmap:
1420 	bio_for_each_segment_all(bvec, bio, j) {
1421 		put_page(bvec->bv_page);
1422 	}
1423 	bio_put(bio);
1424 	return ERR_PTR(ret);
1425 }
1426 
1427 static void __bio_unmap_user(struct bio *bio)
1428 {
1429 	struct bio_vec *bvec;
1430 	int i;
1431 
1432 	/*
1433 	 * make sure we dirty pages we wrote to
1434 	 */
1435 	bio_for_each_segment_all(bvec, bio, i) {
1436 		if (bio_data_dir(bio) == READ)
1437 			set_page_dirty_lock(bvec->bv_page);
1438 
1439 		put_page(bvec->bv_page);
1440 	}
1441 
1442 	bio_put(bio);
1443 }
1444 
1445 /**
1446  *	bio_unmap_user	-	unmap a bio
1447  *	@bio:		the bio being unmapped
1448  *
1449  *	Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1450  *	process context.
1451  *
1452  *	bio_unmap_user() may sleep.
1453  */
1454 void bio_unmap_user(struct bio *bio)
1455 {
1456 	__bio_unmap_user(bio);
1457 	bio_put(bio);
1458 }
1459 
1460 static void bio_map_kern_endio(struct bio *bio)
1461 {
1462 	bio_put(bio);
1463 }
1464 
1465 /**
1466  *	bio_map_kern	-	map kernel address into bio
1467  *	@q: the struct request_queue for the bio
1468  *	@data: pointer to buffer to map
1469  *	@len: length in bytes
1470  *	@gfp_mask: allocation flags for bio allocation
1471  *
1472  *	Map the kernel address into a bio suitable for io to a block
1473  *	device. Returns an error pointer in case of error.
1474  */
1475 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1476 			 gfp_t gfp_mask)
1477 {
1478 	unsigned long kaddr = (unsigned long)data;
1479 	unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1480 	unsigned long start = kaddr >> PAGE_SHIFT;
1481 	const int nr_pages = end - start;
1482 	int offset, i;
1483 	struct bio *bio;
1484 
1485 	bio = bio_kmalloc(gfp_mask, nr_pages);
1486 	if (!bio)
1487 		return ERR_PTR(-ENOMEM);
1488 
1489 	offset = offset_in_page(kaddr);
1490 	for (i = 0; i < nr_pages; i++) {
1491 		unsigned int bytes = PAGE_SIZE - offset;
1492 
1493 		if (len <= 0)
1494 			break;
1495 
1496 		if (bytes > len)
1497 			bytes = len;
1498 
1499 		if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1500 				    offset) < bytes) {
1501 			/* we don't support partial mappings */
1502 			bio_put(bio);
1503 			return ERR_PTR(-EINVAL);
1504 		}
1505 
1506 		data += bytes;
1507 		len -= bytes;
1508 		offset = 0;
1509 	}
1510 
1511 	bio->bi_end_io = bio_map_kern_endio;
1512 	return bio;
1513 }
1514 EXPORT_SYMBOL(bio_map_kern);
1515 
1516 static void bio_copy_kern_endio(struct bio *bio)
1517 {
1518 	bio_free_pages(bio);
1519 	bio_put(bio);
1520 }
1521 
1522 static void bio_copy_kern_endio_read(struct bio *bio)
1523 {
1524 	char *p = bio->bi_private;
1525 	struct bio_vec *bvec;
1526 	int i;
1527 
1528 	bio_for_each_segment_all(bvec, bio, i) {
1529 		memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1530 		p += bvec->bv_len;
1531 	}
1532 
1533 	bio_copy_kern_endio(bio);
1534 }
1535 
1536 /**
1537  *	bio_copy_kern	-	copy kernel address into bio
1538  *	@q: the struct request_queue for the bio
1539  *	@data: pointer to buffer to copy
1540  *	@len: length in bytes
1541  *	@gfp_mask: allocation flags for bio and page allocation
1542  *	@reading: data direction is READ
1543  *
1544  *	copy the kernel address into a bio suitable for io to a block
1545  *	device. Returns an error pointer in case of error.
1546  */
1547 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1548 			  gfp_t gfp_mask, int reading)
1549 {
1550 	unsigned long kaddr = (unsigned long)data;
1551 	unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1552 	unsigned long start = kaddr >> PAGE_SHIFT;
1553 	struct bio *bio;
1554 	void *p = data;
1555 	int nr_pages = 0;
1556 
1557 	/*
1558 	 * Overflow, abort
1559 	 */
1560 	if (end < start)
1561 		return ERR_PTR(-EINVAL);
1562 
1563 	nr_pages = end - start;
1564 	bio = bio_kmalloc(gfp_mask, nr_pages);
1565 	if (!bio)
1566 		return ERR_PTR(-ENOMEM);
1567 
1568 	while (len) {
1569 		struct page *page;
1570 		unsigned int bytes = PAGE_SIZE;
1571 
1572 		if (bytes > len)
1573 			bytes = len;
1574 
1575 		page = alloc_page(q->bounce_gfp | gfp_mask);
1576 		if (!page)
1577 			goto cleanup;
1578 
1579 		if (!reading)
1580 			memcpy(page_address(page), p, bytes);
1581 
1582 		if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1583 			break;
1584 
1585 		len -= bytes;
1586 		p += bytes;
1587 	}
1588 
1589 	if (reading) {
1590 		bio->bi_end_io = bio_copy_kern_endio_read;
1591 		bio->bi_private = data;
1592 	} else {
1593 		bio->bi_end_io = bio_copy_kern_endio;
1594 	}
1595 
1596 	return bio;
1597 
1598 cleanup:
1599 	bio_free_pages(bio);
1600 	bio_put(bio);
1601 	return ERR_PTR(-ENOMEM);
1602 }
1603 
1604 /*
1605  * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1606  * for performing direct-IO in BIOs.
1607  *
1608  * The problem is that we cannot run set_page_dirty() from interrupt context
1609  * because the required locks are not interrupt-safe.  So what we can do is to
1610  * mark the pages dirty _before_ performing IO.  And in interrupt context,
1611  * check that the pages are still dirty.   If so, fine.  If not, redirty them
1612  * in process context.
1613  *
1614  * We special-case compound pages here: normally this means reads into hugetlb
1615  * pages.  The logic in here doesn't really work right for compound pages
1616  * because the VM does not uniformly chase down the head page in all cases.
1617  * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1618  * handle them at all.  So we skip compound pages here at an early stage.
1619  *
1620  * Note that this code is very hard to test under normal circumstances because
1621  * direct-io pins the pages with get_user_pages().  This makes
1622  * is_page_cache_freeable return false, and the VM will not clean the pages.
1623  * But other code (eg, flusher threads) could clean the pages if they are mapped
1624  * pagecache.
1625  *
1626  * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1627  * deferred bio dirtying paths.
1628  */
1629 
1630 /*
1631  * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1632  */
1633 void bio_set_pages_dirty(struct bio *bio)
1634 {
1635 	struct bio_vec *bvec;
1636 	int i;
1637 
1638 	bio_for_each_segment_all(bvec, bio, i) {
1639 		struct page *page = bvec->bv_page;
1640 
1641 		if (page && !PageCompound(page))
1642 			set_page_dirty_lock(page);
1643 	}
1644 }
1645 EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1646 
1647 static void bio_release_pages(struct bio *bio)
1648 {
1649 	struct bio_vec *bvec;
1650 	int i;
1651 
1652 	bio_for_each_segment_all(bvec, bio, i) {
1653 		struct page *page = bvec->bv_page;
1654 
1655 		if (page)
1656 			put_page(page);
1657 	}
1658 }
1659 
1660 /*
1661  * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1662  * If they are, then fine.  If, however, some pages are clean then they must
1663  * have been written out during the direct-IO read.  So we take another ref on
1664  * the BIO and the offending pages and re-dirty the pages in process context.
1665  *
1666  * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1667  * here on.  It will run one put_page() against each page and will run one
1668  * bio_put() against the BIO.
1669  */
1670 
1671 static void bio_dirty_fn(struct work_struct *work);
1672 
1673 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1674 static DEFINE_SPINLOCK(bio_dirty_lock);
1675 static struct bio *bio_dirty_list;
1676 
1677 /*
1678  * This runs in process context
1679  */
1680 static void bio_dirty_fn(struct work_struct *work)
1681 {
1682 	unsigned long flags;
1683 	struct bio *bio;
1684 
1685 	spin_lock_irqsave(&bio_dirty_lock, flags);
1686 	bio = bio_dirty_list;
1687 	bio_dirty_list = NULL;
1688 	spin_unlock_irqrestore(&bio_dirty_lock, flags);
1689 
1690 	while (bio) {
1691 		struct bio *next = bio->bi_private;
1692 
1693 		bio_set_pages_dirty(bio);
1694 		bio_release_pages(bio);
1695 		bio_put(bio);
1696 		bio = next;
1697 	}
1698 }
1699 
1700 void bio_check_pages_dirty(struct bio *bio)
1701 {
1702 	struct bio_vec *bvec;
1703 	int nr_clean_pages = 0;
1704 	int i;
1705 
1706 	bio_for_each_segment_all(bvec, bio, i) {
1707 		struct page *page = bvec->bv_page;
1708 
1709 		if (PageDirty(page) || PageCompound(page)) {
1710 			put_page(page);
1711 			bvec->bv_page = NULL;
1712 		} else {
1713 			nr_clean_pages++;
1714 		}
1715 	}
1716 
1717 	if (nr_clean_pages) {
1718 		unsigned long flags;
1719 
1720 		spin_lock_irqsave(&bio_dirty_lock, flags);
1721 		bio->bi_private = bio_dirty_list;
1722 		bio_dirty_list = bio;
1723 		spin_unlock_irqrestore(&bio_dirty_lock, flags);
1724 		schedule_work(&bio_dirty_work);
1725 	} else {
1726 		bio_put(bio);
1727 	}
1728 }
1729 EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1730 
1731 void generic_start_io_acct(struct request_queue *q, int op,
1732 			   unsigned long sectors, struct hd_struct *part)
1733 {
1734 	const int sgrp = op_stat_group(op);
1735 	int cpu = part_stat_lock();
1736 
1737 	part_round_stats(q, cpu, part);
1738 	part_stat_inc(cpu, part, ios[sgrp]);
1739 	part_stat_add(cpu, part, sectors[sgrp], sectors);
1740 	part_inc_in_flight(q, part, op_is_write(op));
1741 
1742 	part_stat_unlock();
1743 }
1744 EXPORT_SYMBOL(generic_start_io_acct);
1745 
1746 void generic_end_io_acct(struct request_queue *q, int req_op,
1747 			 struct hd_struct *part, unsigned long start_time)
1748 {
1749 	unsigned long duration = jiffies - start_time;
1750 	const int sgrp = op_stat_group(req_op);
1751 	int cpu = part_stat_lock();
1752 
1753 	part_stat_add(cpu, part, ticks[sgrp], duration);
1754 	part_round_stats(q, cpu, part);
1755 	part_dec_in_flight(q, part, op_is_write(req_op));
1756 
1757 	part_stat_unlock();
1758 }
1759 EXPORT_SYMBOL(generic_end_io_acct);
1760 
1761 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1762 void bio_flush_dcache_pages(struct bio *bi)
1763 {
1764 	struct bio_vec bvec;
1765 	struct bvec_iter iter;
1766 
1767 	bio_for_each_segment(bvec, bi, iter)
1768 		flush_dcache_page(bvec.bv_page);
1769 }
1770 EXPORT_SYMBOL(bio_flush_dcache_pages);
1771 #endif
1772 
1773 static inline bool bio_remaining_done(struct bio *bio)
1774 {
1775 	/*
1776 	 * If we're not chaining, then ->__bi_remaining is always 1 and
1777 	 * we always end io on the first invocation.
1778 	 */
1779 	if (!bio_flagged(bio, BIO_CHAIN))
1780 		return true;
1781 
1782 	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1783 
1784 	if (atomic_dec_and_test(&bio->__bi_remaining)) {
1785 		bio_clear_flag(bio, BIO_CHAIN);
1786 		return true;
1787 	}
1788 
1789 	return false;
1790 }
1791 
1792 /**
1793  * bio_endio - end I/O on a bio
1794  * @bio:	bio
1795  *
1796  * Description:
1797  *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1798  *   way to end I/O on a bio. No one should call bi_end_io() directly on a
1799  *   bio unless they own it and thus know that it has an end_io function.
1800  *
1801  *   bio_endio() can be called several times on a bio that has been chained
1802  *   using bio_chain().  The ->bi_end_io() function will only be called the
1803  *   last time.  At this point the BLK_TA_COMPLETE tracing event will be
1804  *   generated if BIO_TRACE_COMPLETION is set.
1805  **/
1806 void bio_endio(struct bio *bio)
1807 {
1808 again:
1809 	if (!bio_remaining_done(bio))
1810 		return;
1811 	if (!bio_integrity_endio(bio))
1812 		return;
1813 
1814 	if (bio->bi_disk)
1815 		rq_qos_done_bio(bio->bi_disk->queue, bio);
1816 
1817 	/*
1818 	 * Need to have a real endio function for chained bios, otherwise
1819 	 * various corner cases will break (like stacking block devices that
1820 	 * save/restore bi_end_io) - however, we want to avoid unbounded
1821 	 * recursion and blowing the stack. Tail call optimization would
1822 	 * handle this, but compiling with frame pointers also disables
1823 	 * gcc's sibling call optimization.
1824 	 */
1825 	if (bio->bi_end_io == bio_chain_endio) {
1826 		bio = __bio_chain_endio(bio);
1827 		goto again;
1828 	}
1829 
1830 	if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1831 		trace_block_bio_complete(bio->bi_disk->queue, bio,
1832 					 blk_status_to_errno(bio->bi_status));
1833 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1834 	}
1835 
1836 	blk_throtl_bio_endio(bio);
1837 	/* release cgroup info */
1838 	bio_uninit(bio);
1839 	if (bio->bi_end_io)
1840 		bio->bi_end_io(bio);
1841 }
1842 EXPORT_SYMBOL(bio_endio);
1843 
1844 /**
1845  * bio_split - split a bio
1846  * @bio:	bio to split
1847  * @sectors:	number of sectors to split from the front of @bio
1848  * @gfp:	gfp mask
1849  * @bs:		bio set to allocate from
1850  *
1851  * Allocates and returns a new bio which represents @sectors from the start of
1852  * @bio, and updates @bio to represent the remaining sectors.
1853  *
1854  * Unless this is a discard request the newly allocated bio will point
1855  * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1856  * @bio is not freed before the split.
1857  */
1858 struct bio *bio_split(struct bio *bio, int sectors,
1859 		      gfp_t gfp, struct bio_set *bs)
1860 {
1861 	struct bio *split;
1862 
1863 	BUG_ON(sectors <= 0);
1864 	BUG_ON(sectors >= bio_sectors(bio));
1865 
1866 	split = bio_clone_fast(bio, gfp, bs);
1867 	if (!split)
1868 		return NULL;
1869 
1870 	split->bi_iter.bi_size = sectors << 9;
1871 
1872 	if (bio_integrity(split))
1873 		bio_integrity_trim(split);
1874 
1875 	bio_advance(bio, split->bi_iter.bi_size);
1876 
1877 	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1878 		bio_set_flag(split, BIO_TRACE_COMPLETION);
1879 
1880 	return split;
1881 }
1882 EXPORT_SYMBOL(bio_split);
1883 
1884 /**
1885  * bio_trim - trim a bio
1886  * @bio:	bio to trim
1887  * @offset:	number of sectors to trim from the front of @bio
1888  * @size:	size we want to trim @bio to, in sectors
1889  */
1890 void bio_trim(struct bio *bio, int offset, int size)
1891 {
1892 	/* 'bio' is a cloned bio which we need to trim to match
1893 	 * the given offset and size.
1894 	 */
1895 
1896 	size <<= 9;
1897 	if (offset == 0 && size == bio->bi_iter.bi_size)
1898 		return;
1899 
1900 	bio_clear_flag(bio, BIO_SEG_VALID);
1901 
1902 	bio_advance(bio, offset << 9);
1903 
1904 	bio->bi_iter.bi_size = size;
1905 
1906 	if (bio_integrity(bio))
1907 		bio_integrity_trim(bio);
1908 
1909 }
1910 EXPORT_SYMBOL_GPL(bio_trim);
1911 
1912 /*
1913  * create memory pools for biovec's in a bio_set.
1914  * use the global biovec slabs created for general use.
1915  */
1916 int biovec_init_pool(mempool_t *pool, int pool_entries)
1917 {
1918 	struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1919 
1920 	return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1921 }
1922 
1923 /*
1924  * bioset_exit - exit a bioset initialized with bioset_init()
1925  *
1926  * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1927  * kzalloc()).
1928  */
1929 void bioset_exit(struct bio_set *bs)
1930 {
1931 	if (bs->rescue_workqueue)
1932 		destroy_workqueue(bs->rescue_workqueue);
1933 	bs->rescue_workqueue = NULL;
1934 
1935 	mempool_exit(&bs->bio_pool);
1936 	mempool_exit(&bs->bvec_pool);
1937 
1938 	bioset_integrity_free(bs);
1939 	if (bs->bio_slab)
1940 		bio_put_slab(bs);
1941 	bs->bio_slab = NULL;
1942 }
1943 EXPORT_SYMBOL(bioset_exit);
1944 
1945 /**
1946  * bioset_init - Initialize a bio_set
1947  * @bs:		pool to initialize
1948  * @pool_size:	Number of bio and bio_vecs to cache in the mempool
1949  * @front_pad:	Number of bytes to allocate in front of the returned bio
1950  * @flags:	Flags to modify behavior, currently %BIOSET_NEED_BVECS
1951  *              and %BIOSET_NEED_RESCUER
1952  *
1953  * Description:
1954  *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1955  *    to ask for a number of bytes to be allocated in front of the bio.
1956  *    Front pad allocation is useful for embedding the bio inside
1957  *    another structure, to avoid allocating extra data to go with the bio.
1958  *    Note that the bio must be embedded at the END of that structure always,
1959  *    or things will break badly.
1960  *    If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1961  *    for allocating iovecs.  This pool is not needed e.g. for bio_clone_fast().
1962  *    If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1963  *    dispatch queued requests when the mempool runs out of space.
1964  *
1965  */
1966 int bioset_init(struct bio_set *bs,
1967 		unsigned int pool_size,
1968 		unsigned int front_pad,
1969 		int flags)
1970 {
1971 	unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1972 
1973 	bs->front_pad = front_pad;
1974 
1975 	spin_lock_init(&bs->rescue_lock);
1976 	bio_list_init(&bs->rescue_list);
1977 	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1978 
1979 	bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1980 	if (!bs->bio_slab)
1981 		return -ENOMEM;
1982 
1983 	if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1984 		goto bad;
1985 
1986 	if ((flags & BIOSET_NEED_BVECS) &&
1987 	    biovec_init_pool(&bs->bvec_pool, pool_size))
1988 		goto bad;
1989 
1990 	if (!(flags & BIOSET_NEED_RESCUER))
1991 		return 0;
1992 
1993 	bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1994 	if (!bs->rescue_workqueue)
1995 		goto bad;
1996 
1997 	return 0;
1998 bad:
1999 	bioset_exit(bs);
2000 	return -ENOMEM;
2001 }
2002 EXPORT_SYMBOL(bioset_init);
2003 
2004 /*
2005  * Initialize and setup a new bio_set, based on the settings from
2006  * another bio_set.
2007  */
2008 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
2009 {
2010 	int flags;
2011 
2012 	flags = 0;
2013 	if (src->bvec_pool.min_nr)
2014 		flags |= BIOSET_NEED_BVECS;
2015 	if (src->rescue_workqueue)
2016 		flags |= BIOSET_NEED_RESCUER;
2017 
2018 	return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
2019 }
2020 EXPORT_SYMBOL(bioset_init_from_src);
2021 
2022 #ifdef CONFIG_BLK_CGROUP
2023 
2024 #ifdef CONFIG_MEMCG
2025 /**
2026  * bio_associate_blkcg_from_page - associate a bio with the page's blkcg
2027  * @bio: target bio
2028  * @page: the page to lookup the blkcg from
2029  *
2030  * Associate @bio with the blkcg from @page's owning memcg.  This works like
2031  * every other associate function wrt references.
2032  */
2033 int bio_associate_blkcg_from_page(struct bio *bio, struct page *page)
2034 {
2035 	struct cgroup_subsys_state *blkcg_css;
2036 
2037 	if (unlikely(bio->bi_css))
2038 		return -EBUSY;
2039 	if (!page->mem_cgroup)
2040 		return 0;
2041 	blkcg_css = cgroup_get_e_css(page->mem_cgroup->css.cgroup,
2042 				     &io_cgrp_subsys);
2043 	bio->bi_css = blkcg_css;
2044 	return 0;
2045 }
2046 #endif /* CONFIG_MEMCG */
2047 
2048 /**
2049  * bio_associate_blkcg - associate a bio with the specified blkcg
2050  * @bio: target bio
2051  * @blkcg_css: css of the blkcg to associate
2052  *
2053  * Associate @bio with the blkcg specified by @blkcg_css.  Block layer will
2054  * treat @bio as if it were issued by a task which belongs to the blkcg.
2055  *
2056  * This function takes an extra reference of @blkcg_css which will be put
2057  * when @bio is released.  The caller must own @bio and is responsible for
2058  * synchronizing calls to this function.
2059  */
2060 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
2061 {
2062 	if (unlikely(bio->bi_css))
2063 		return -EBUSY;
2064 	css_get(blkcg_css);
2065 	bio->bi_css = blkcg_css;
2066 	return 0;
2067 }
2068 EXPORT_SYMBOL_GPL(bio_associate_blkcg);
2069 
2070 /**
2071  * bio_associate_blkg - associate a bio with the specified blkg
2072  * @bio: target bio
2073  * @blkg: the blkg to associate
2074  *
2075  * Associate @bio with the blkg specified by @blkg.  This is the queue specific
2076  * blkcg information associated with the @bio, a reference will be taken on the
2077  * @blkg and will be freed when the bio is freed.
2078  */
2079 int bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg)
2080 {
2081 	if (unlikely(bio->bi_blkg))
2082 		return -EBUSY;
2083 	blkg_get(blkg);
2084 	bio->bi_blkg = blkg;
2085 	return 0;
2086 }
2087 
2088 /**
2089  * bio_disassociate_task - undo bio_associate_current()
2090  * @bio: target bio
2091  */
2092 void bio_disassociate_task(struct bio *bio)
2093 {
2094 	if (bio->bi_ioc) {
2095 		put_io_context(bio->bi_ioc);
2096 		bio->bi_ioc = NULL;
2097 	}
2098 	if (bio->bi_css) {
2099 		css_put(bio->bi_css);
2100 		bio->bi_css = NULL;
2101 	}
2102 	if (bio->bi_blkg) {
2103 		blkg_put(bio->bi_blkg);
2104 		bio->bi_blkg = NULL;
2105 	}
2106 }
2107 
2108 /**
2109  * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2110  * @dst: destination bio
2111  * @src: source bio
2112  */
2113 void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2114 {
2115 	if (src->bi_css)
2116 		WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2117 }
2118 EXPORT_SYMBOL_GPL(bio_clone_blkcg_association);
2119 #endif /* CONFIG_BLK_CGROUP */
2120 
2121 static void __init biovec_init_slabs(void)
2122 {
2123 	int i;
2124 
2125 	for (i = 0; i < BVEC_POOL_NR; i++) {
2126 		int size;
2127 		struct biovec_slab *bvs = bvec_slabs + i;
2128 
2129 		if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2130 			bvs->slab = NULL;
2131 			continue;
2132 		}
2133 
2134 		size = bvs->nr_vecs * sizeof(struct bio_vec);
2135 		bvs->slab = kmem_cache_create(bvs->name, size, 0,
2136                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2137 	}
2138 }
2139 
2140 static int __init init_bio(void)
2141 {
2142 	bio_slab_max = 2;
2143 	bio_slab_nr = 0;
2144 	bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab),
2145 			    GFP_KERNEL);
2146 	if (!bio_slabs)
2147 		panic("bio: can't allocate bios\n");
2148 
2149 	bio_integrity_init();
2150 	biovec_init_slabs();
2151 
2152 	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
2153 		panic("bio: can't allocate bios\n");
2154 
2155 	if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
2156 		panic("bio: can't create integrity pool\n");
2157 
2158 	return 0;
2159 }
2160 subsys_initcall(init_bio);
2161