1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4  */
5 #include <linux/mm.h>
6 #include <linux/swap.h>
7 #include <linux/bio.h>
8 #include <linux/blkdev.h>
9 #include <linux/uio.h>
10 #include <linux/iocontext.h>
11 #include <linux/slab.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/export.h>
15 #include <linux/mempool.h>
16 #include <linux/workqueue.h>
17 #include <linux/cgroup.h>
18 #include <linux/blk-cgroup.h>
19 #include <linux/highmem.h>
20 #include <linux/sched/sysctl.h>
21 #include <linux/blk-crypto.h>
22 #include <linux/xarray.h>
23 
24 #include <trace/events/block.h>
25 #include "blk.h"
26 #include "blk-rq-qos.h"
27 
28 struct bio_alloc_cache {
29 	struct bio_list		free_list;
30 	unsigned int		nr;
31 };
32 
33 static struct biovec_slab {
34 	int nr_vecs;
35 	char *name;
36 	struct kmem_cache *slab;
37 } bvec_slabs[] __read_mostly = {
38 	{ .nr_vecs = 16, .name = "biovec-16" },
39 	{ .nr_vecs = 64, .name = "biovec-64" },
40 	{ .nr_vecs = 128, .name = "biovec-128" },
41 	{ .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
42 };
43 
biovec_slab(unsigned short nr_vecs)44 static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
45 {
46 	switch (nr_vecs) {
47 	/* smaller bios use inline vecs */
48 	case 5 ... 16:
49 		return &bvec_slabs[0];
50 	case 17 ... 64:
51 		return &bvec_slabs[1];
52 	case 65 ... 128:
53 		return &bvec_slabs[2];
54 	case 129 ... BIO_MAX_VECS:
55 		return &bvec_slabs[3];
56 	default:
57 		BUG();
58 		return NULL;
59 	}
60 }
61 
62 /*
63  * fs_bio_set is the bio_set containing bio and iovec memory pools used by
64  * IO code that does not need private memory pools.
65  */
66 struct bio_set fs_bio_set;
67 EXPORT_SYMBOL(fs_bio_set);
68 
69 /*
70  * Our slab pool management
71  */
72 struct bio_slab {
73 	struct kmem_cache *slab;
74 	unsigned int slab_ref;
75 	unsigned int slab_size;
76 	char name[8];
77 };
78 static DEFINE_MUTEX(bio_slab_lock);
79 static DEFINE_XARRAY(bio_slabs);
80 
create_bio_slab(unsigned int size)81 static struct bio_slab *create_bio_slab(unsigned int size)
82 {
83 	struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
84 
85 	if (!bslab)
86 		return NULL;
87 
88 	snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
89 	bslab->slab = kmem_cache_create(bslab->name, size,
90 			ARCH_KMALLOC_MINALIGN,
91 			SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
92 	if (!bslab->slab)
93 		goto fail_alloc_slab;
94 
95 	bslab->slab_ref = 1;
96 	bslab->slab_size = size;
97 
98 	if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
99 		return bslab;
100 
101 	kmem_cache_destroy(bslab->slab);
102 
103 fail_alloc_slab:
104 	kfree(bslab);
105 	return NULL;
106 }
107 
bs_bio_slab_size(struct bio_set * bs)108 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
109 {
110 	return bs->front_pad + sizeof(struct bio) + bs->back_pad;
111 }
112 
bio_find_or_create_slab(struct bio_set * bs)113 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
114 {
115 	unsigned int size = bs_bio_slab_size(bs);
116 	struct bio_slab *bslab;
117 
118 	mutex_lock(&bio_slab_lock);
119 	bslab = xa_load(&bio_slabs, size);
120 	if (bslab)
121 		bslab->slab_ref++;
122 	else
123 		bslab = create_bio_slab(size);
124 	mutex_unlock(&bio_slab_lock);
125 
126 	if (bslab)
127 		return bslab->slab;
128 	return NULL;
129 }
130 
bio_put_slab(struct bio_set * bs)131 static void bio_put_slab(struct bio_set *bs)
132 {
133 	struct bio_slab *bslab = NULL;
134 	unsigned int slab_size = bs_bio_slab_size(bs);
135 
136 	mutex_lock(&bio_slab_lock);
137 
138 	bslab = xa_load(&bio_slabs, slab_size);
139 	if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
140 		goto out;
141 
142 	WARN_ON_ONCE(bslab->slab != bs->bio_slab);
143 
144 	WARN_ON(!bslab->slab_ref);
145 
146 	if (--bslab->slab_ref)
147 		goto out;
148 
149 	xa_erase(&bio_slabs, slab_size);
150 
151 	kmem_cache_destroy(bslab->slab);
152 	kfree(bslab);
153 
154 out:
155 	mutex_unlock(&bio_slab_lock);
156 }
157 
bvec_free(mempool_t * pool,struct bio_vec * bv,unsigned short nr_vecs)158 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
159 {
160 	BUG_ON(nr_vecs > BIO_MAX_VECS);
161 
162 	if (nr_vecs == BIO_MAX_VECS)
163 		mempool_free(bv, pool);
164 	else if (nr_vecs > BIO_INLINE_VECS)
165 		kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
166 }
167 
168 /*
169  * Make the first allocation restricted and don't dump info on allocation
170  * failures, since we'll fall back to the mempool in case of failure.
171  */
bvec_alloc_gfp(gfp_t gfp)172 static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
173 {
174 	return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
175 		__GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
176 }
177 
bvec_alloc(mempool_t * pool,unsigned short * nr_vecs,gfp_t gfp_mask)178 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
179 		gfp_t gfp_mask)
180 {
181 	struct biovec_slab *bvs = biovec_slab(*nr_vecs);
182 
183 	if (WARN_ON_ONCE(!bvs))
184 		return NULL;
185 
186 	/*
187 	 * Upgrade the nr_vecs request to take full advantage of the allocation.
188 	 * We also rely on this in the bvec_free path.
189 	 */
190 	*nr_vecs = bvs->nr_vecs;
191 
192 	/*
193 	 * Try a slab allocation first for all smaller allocations.  If that
194 	 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
195 	 * The mempool is sized to handle up to BIO_MAX_VECS entries.
196 	 */
197 	if (*nr_vecs < BIO_MAX_VECS) {
198 		struct bio_vec *bvl;
199 
200 		bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
201 		if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
202 			return bvl;
203 		*nr_vecs = BIO_MAX_VECS;
204 	}
205 
206 	return mempool_alloc(pool, gfp_mask);
207 }
208 
bio_uninit(struct bio * bio)209 void bio_uninit(struct bio *bio)
210 {
211 #ifdef CONFIG_BLK_CGROUP
212 	if (bio->bi_blkg) {
213 		blkg_put(bio->bi_blkg);
214 		bio->bi_blkg = NULL;
215 	}
216 #endif
217 	if (bio_integrity(bio))
218 		bio_integrity_free(bio);
219 
220 	bio_crypt_free_ctx(bio);
221 }
222 EXPORT_SYMBOL(bio_uninit);
223 
bio_free(struct bio * bio)224 static void bio_free(struct bio *bio)
225 {
226 	struct bio_set *bs = bio->bi_pool;
227 	void *p;
228 
229 	bio_uninit(bio);
230 
231 	if (bs) {
232 		bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
233 
234 		/*
235 		 * If we have front padding, adjust the bio pointer before freeing
236 		 */
237 		p = bio;
238 		p -= bs->front_pad;
239 
240 		mempool_free(p, &bs->bio_pool);
241 	} else {
242 		/* Bio was allocated by bio_kmalloc() */
243 		kfree(bio);
244 	}
245 }
246 
247 /*
248  * Users of this function have their own bio allocation. Subsequently,
249  * they must remember to pair any call to bio_init() with bio_uninit()
250  * when IO has completed, or when the bio is released.
251  */
bio_init(struct bio * bio,struct bio_vec * table,unsigned short max_vecs)252 void bio_init(struct bio *bio, struct bio_vec *table,
253 	      unsigned short max_vecs)
254 {
255 	bio->bi_next = NULL;
256 	bio->bi_bdev = NULL;
257 	bio->bi_opf = 0;
258 	bio->bi_flags = 0;
259 	bio->bi_ioprio = 0;
260 	bio->bi_write_hint = 0;
261 	bio->bi_status = 0;
262 	bio->bi_iter.bi_sector = 0;
263 	bio->bi_iter.bi_size = 0;
264 	bio->bi_iter.bi_idx = 0;
265 	bio->bi_iter.bi_bvec_done = 0;
266 	bio->bi_end_io = NULL;
267 	bio->bi_private = NULL;
268 #ifdef CONFIG_BLK_CGROUP
269 	bio->bi_blkg = NULL;
270 	bio->bi_issue.value = 0;
271 #ifdef CONFIG_BLK_CGROUP_IOCOST
272 	bio->bi_iocost_cost = 0;
273 #endif
274 #endif
275 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
276 	bio->bi_crypt_context = NULL;
277 #endif
278 #ifdef CONFIG_BLK_DEV_INTEGRITY
279 	bio->bi_integrity = NULL;
280 #endif
281 	bio->bi_vcnt = 0;
282 
283 	atomic_set(&bio->__bi_remaining, 1);
284 	atomic_set(&bio->__bi_cnt, 1);
285 	bio->bi_cookie = BLK_QC_T_NONE;
286 
287 	bio->bi_max_vecs = max_vecs;
288 	bio->bi_io_vec = table;
289 	bio->bi_pool = NULL;
290 }
291 EXPORT_SYMBOL(bio_init);
292 
293 /**
294  * bio_reset - reinitialize a bio
295  * @bio:	bio to reset
296  *
297  * Description:
298  *   After calling bio_reset(), @bio will be in the same state as a freshly
299  *   allocated bio returned bio bio_alloc_bioset() - the only fields that are
300  *   preserved are the ones that are initialized by bio_alloc_bioset(). See
301  *   comment in struct bio.
302  */
bio_reset(struct bio * bio)303 void bio_reset(struct bio *bio)
304 {
305 	bio_uninit(bio);
306 	memset(bio, 0, BIO_RESET_BYTES);
307 	atomic_set(&bio->__bi_remaining, 1);
308 }
309 EXPORT_SYMBOL(bio_reset);
310 
__bio_chain_endio(struct bio * bio)311 static struct bio *__bio_chain_endio(struct bio *bio)
312 {
313 	struct bio *parent = bio->bi_private;
314 
315 	if (bio->bi_status && !parent->bi_status)
316 		parent->bi_status = bio->bi_status;
317 	bio_put(bio);
318 	return parent;
319 }
320 
bio_chain_endio(struct bio * bio)321 static void bio_chain_endio(struct bio *bio)
322 {
323 	bio_endio(__bio_chain_endio(bio));
324 }
325 
326 /**
327  * bio_chain - chain bio completions
328  * @bio: the target bio
329  * @parent: the parent bio of @bio
330  *
331  * The caller won't have a bi_end_io called when @bio completes - instead,
332  * @parent's bi_end_io won't be called until both @parent and @bio have
333  * completed; the chained bio will also be freed when it completes.
334  *
335  * The caller must not set bi_private or bi_end_io in @bio.
336  */
bio_chain(struct bio * bio,struct bio * parent)337 void bio_chain(struct bio *bio, struct bio *parent)
338 {
339 	BUG_ON(bio->bi_private || bio->bi_end_io);
340 
341 	bio->bi_private = parent;
342 	bio->bi_end_io	= bio_chain_endio;
343 	bio_inc_remaining(parent);
344 }
345 EXPORT_SYMBOL(bio_chain);
346 
bio_alloc_rescue(struct work_struct * work)347 static void bio_alloc_rescue(struct work_struct *work)
348 {
349 	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
350 	struct bio *bio;
351 
352 	while (1) {
353 		spin_lock(&bs->rescue_lock);
354 		bio = bio_list_pop(&bs->rescue_list);
355 		spin_unlock(&bs->rescue_lock);
356 
357 		if (!bio)
358 			break;
359 
360 		submit_bio_noacct(bio);
361 	}
362 }
363 
punt_bios_to_rescuer(struct bio_set * bs)364 static void punt_bios_to_rescuer(struct bio_set *bs)
365 {
366 	struct bio_list punt, nopunt;
367 	struct bio *bio;
368 
369 	if (WARN_ON_ONCE(!bs->rescue_workqueue))
370 		return;
371 	/*
372 	 * In order to guarantee forward progress we must punt only bios that
373 	 * were allocated from this bio_set; otherwise, if there was a bio on
374 	 * there for a stacking driver higher up in the stack, processing it
375 	 * could require allocating bios from this bio_set, and doing that from
376 	 * our own rescuer would be bad.
377 	 *
378 	 * Since bio lists are singly linked, pop them all instead of trying to
379 	 * remove from the middle of the list:
380 	 */
381 
382 	bio_list_init(&punt);
383 	bio_list_init(&nopunt);
384 
385 	while ((bio = bio_list_pop(&current->bio_list[0])))
386 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
387 	current->bio_list[0] = nopunt;
388 
389 	bio_list_init(&nopunt);
390 	while ((bio = bio_list_pop(&current->bio_list[1])))
391 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
392 	current->bio_list[1] = nopunt;
393 
394 	spin_lock(&bs->rescue_lock);
395 	bio_list_merge(&bs->rescue_list, &punt);
396 	spin_unlock(&bs->rescue_lock);
397 
398 	queue_work(bs->rescue_workqueue, &bs->rescue_work);
399 }
400 
401 /**
402  * bio_alloc_bioset - allocate a bio for I/O
403  * @gfp_mask:   the GFP_* mask given to the slab allocator
404  * @nr_iovecs:	number of iovecs to pre-allocate
405  * @bs:		the bio_set to allocate from.
406  *
407  * Allocate a bio from the mempools in @bs.
408  *
409  * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
410  * allocate a bio.  This is due to the mempool guarantees.  To make this work,
411  * callers must never allocate more than 1 bio at a time from the general pool.
412  * Callers that need to allocate more than 1 bio must always submit the
413  * previously allocated bio for IO before attempting to allocate a new one.
414  * Failure to do so can cause deadlocks under memory pressure.
415  *
416  * Note that when running under submit_bio_noacct() (i.e. any block driver),
417  * bios are not submitted until after you return - see the code in
418  * submit_bio_noacct() that converts recursion into iteration, to prevent
419  * stack overflows.
420  *
421  * This would normally mean allocating multiple bios under submit_bio_noacct()
422  * would be susceptible to deadlocks, but we have
423  * deadlock avoidance code that resubmits any blocked bios from a rescuer
424  * thread.
425  *
426  * However, we do not guarantee forward progress for allocations from other
427  * mempools. Doing multiple allocations from the same mempool under
428  * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
429  * for per bio allocations.
430  *
431  * Returns: Pointer to new bio on success, NULL on failure.
432  */
bio_alloc_bioset(gfp_t gfp_mask,unsigned short nr_iovecs,struct bio_set * bs)433 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs,
434 			     struct bio_set *bs)
435 {
436 	gfp_t saved_gfp = gfp_mask;
437 	struct bio *bio;
438 	void *p;
439 
440 	/* should not use nobvec bioset for nr_iovecs > 0 */
441 	if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0))
442 		return NULL;
443 
444 	/*
445 	 * submit_bio_noacct() converts recursion to iteration; this means if
446 	 * we're running beneath it, any bios we allocate and submit will not be
447 	 * submitted (and thus freed) until after we return.
448 	 *
449 	 * This exposes us to a potential deadlock if we allocate multiple bios
450 	 * from the same bio_set() while running underneath submit_bio_noacct().
451 	 * If we were to allocate multiple bios (say a stacking block driver
452 	 * that was splitting bios), we would deadlock if we exhausted the
453 	 * mempool's reserve.
454 	 *
455 	 * We solve this, and guarantee forward progress, with a rescuer
456 	 * workqueue per bio_set. If we go to allocate and there are bios on
457 	 * current->bio_list, we first try the allocation without
458 	 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
459 	 * blocking to the rescuer workqueue before we retry with the original
460 	 * gfp_flags.
461 	 */
462 	if (current->bio_list &&
463 	    (!bio_list_empty(&current->bio_list[0]) ||
464 	     !bio_list_empty(&current->bio_list[1])) &&
465 	    bs->rescue_workqueue)
466 		gfp_mask &= ~__GFP_DIRECT_RECLAIM;
467 
468 	p = mempool_alloc(&bs->bio_pool, gfp_mask);
469 	if (!p && gfp_mask != saved_gfp) {
470 		punt_bios_to_rescuer(bs);
471 		gfp_mask = saved_gfp;
472 		p = mempool_alloc(&bs->bio_pool, gfp_mask);
473 	}
474 	if (unlikely(!p))
475 		return NULL;
476 
477 	bio = p + bs->front_pad;
478 	if (nr_iovecs > BIO_INLINE_VECS) {
479 		struct bio_vec *bvl = NULL;
480 
481 		bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
482 		if (!bvl && gfp_mask != saved_gfp) {
483 			punt_bios_to_rescuer(bs);
484 			gfp_mask = saved_gfp;
485 			bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
486 		}
487 		if (unlikely(!bvl))
488 			goto err_free;
489 
490 		bio_init(bio, bvl, nr_iovecs);
491 	} else if (nr_iovecs) {
492 		bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS);
493 	} else {
494 		bio_init(bio, NULL, 0);
495 	}
496 
497 	bio->bi_pool = bs;
498 	return bio;
499 
500 err_free:
501 	mempool_free(p, &bs->bio_pool);
502 	return NULL;
503 }
504 EXPORT_SYMBOL(bio_alloc_bioset);
505 
506 /**
507  * bio_kmalloc - kmalloc a bio for I/O
508  * @gfp_mask:   the GFP_* mask given to the slab allocator
509  * @nr_iovecs:	number of iovecs to pre-allocate
510  *
511  * Use kmalloc to allocate and initialize a bio.
512  *
513  * Returns: Pointer to new bio on success, NULL on failure.
514  */
bio_kmalloc(gfp_t gfp_mask,unsigned short nr_iovecs)515 struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs)
516 {
517 	struct bio *bio;
518 
519 	if (nr_iovecs > UIO_MAXIOV)
520 		return NULL;
521 
522 	bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask);
523 	if (unlikely(!bio))
524 		return NULL;
525 	bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs);
526 	bio->bi_pool = NULL;
527 	return bio;
528 }
529 EXPORT_SYMBOL(bio_kmalloc);
530 
zero_fill_bio(struct bio * bio)531 void zero_fill_bio(struct bio *bio)
532 {
533 	struct bio_vec bv;
534 	struct bvec_iter iter;
535 
536 	bio_for_each_segment(bv, bio, iter)
537 		memzero_bvec(&bv);
538 }
539 EXPORT_SYMBOL(zero_fill_bio);
540 
541 /**
542  * bio_truncate - truncate the bio to small size of @new_size
543  * @bio:	the bio to be truncated
544  * @new_size:	new size for truncating the bio
545  *
546  * Description:
547  *   Truncate the bio to new size of @new_size. If bio_op(bio) is
548  *   REQ_OP_READ, zero the truncated part. This function should only
549  *   be used for handling corner cases, such as bio eod.
550  */
bio_truncate(struct bio * bio,unsigned new_size)551 static void bio_truncate(struct bio *bio, unsigned new_size)
552 {
553 	struct bio_vec bv;
554 	struct bvec_iter iter;
555 	unsigned int done = 0;
556 	bool truncated = false;
557 
558 	if (new_size >= bio->bi_iter.bi_size)
559 		return;
560 
561 	if (bio_op(bio) != REQ_OP_READ)
562 		goto exit;
563 
564 	bio_for_each_segment(bv, bio, iter) {
565 		if (done + bv.bv_len > new_size) {
566 			unsigned offset;
567 
568 			if (!truncated)
569 				offset = new_size - done;
570 			else
571 				offset = 0;
572 			zero_user(bv.bv_page, offset, bv.bv_len - offset);
573 			truncated = true;
574 		}
575 		done += bv.bv_len;
576 	}
577 
578  exit:
579 	/*
580 	 * Don't touch bvec table here and make it really immutable, since
581 	 * fs bio user has to retrieve all pages via bio_for_each_segment_all
582 	 * in its .end_bio() callback.
583 	 *
584 	 * It is enough to truncate bio by updating .bi_size since we can make
585 	 * correct bvec with the updated .bi_size for drivers.
586 	 */
587 	bio->bi_iter.bi_size = new_size;
588 }
589 
590 /**
591  * guard_bio_eod - truncate a BIO to fit the block device
592  * @bio:	bio to truncate
593  *
594  * This allows us to do IO even on the odd last sectors of a device, even if the
595  * block size is some multiple of the physical sector size.
596  *
597  * We'll just truncate the bio to the size of the device, and clear the end of
598  * the buffer head manually.  Truly out-of-range accesses will turn into actual
599  * I/O errors, this only handles the "we need to be able to do I/O at the final
600  * sector" case.
601  */
guard_bio_eod(struct bio * bio)602 void guard_bio_eod(struct bio *bio)
603 {
604 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
605 
606 	if (!maxsector)
607 		return;
608 
609 	/*
610 	 * If the *whole* IO is past the end of the device,
611 	 * let it through, and the IO layer will turn it into
612 	 * an EIO.
613 	 */
614 	if (unlikely(bio->bi_iter.bi_sector >= maxsector))
615 		return;
616 
617 	maxsector -= bio->bi_iter.bi_sector;
618 	if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
619 		return;
620 
621 	bio_truncate(bio, maxsector << 9);
622 }
623 
624 #define ALLOC_CACHE_MAX		512
625 #define ALLOC_CACHE_SLACK	 64
626 
bio_alloc_cache_prune(struct bio_alloc_cache * cache,unsigned int nr)627 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
628 				  unsigned int nr)
629 {
630 	unsigned int i = 0;
631 	struct bio *bio;
632 
633 	while ((bio = bio_list_pop(&cache->free_list)) != NULL) {
634 		cache->nr--;
635 		bio_free(bio);
636 		if (++i == nr)
637 			break;
638 	}
639 }
640 
bio_cpu_dead(unsigned int cpu,struct hlist_node * node)641 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
642 {
643 	struct bio_set *bs;
644 
645 	bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
646 	if (bs->cache) {
647 		struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
648 
649 		bio_alloc_cache_prune(cache, -1U);
650 	}
651 	return 0;
652 }
653 
bio_alloc_cache_destroy(struct bio_set * bs)654 static void bio_alloc_cache_destroy(struct bio_set *bs)
655 {
656 	int cpu;
657 
658 	if (!bs->cache)
659 		return;
660 
661 	cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
662 	for_each_possible_cpu(cpu) {
663 		struct bio_alloc_cache *cache;
664 
665 		cache = per_cpu_ptr(bs->cache, cpu);
666 		bio_alloc_cache_prune(cache, -1U);
667 	}
668 	free_percpu(bs->cache);
669 }
670 
671 /**
672  * bio_put - release a reference to a bio
673  * @bio:   bio to release reference to
674  *
675  * Description:
676  *   Put a reference to a &struct bio, either one you have gotten with
677  *   bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
678  **/
bio_put(struct bio * bio)679 void bio_put(struct bio *bio)
680 {
681 	if (unlikely(bio_flagged(bio, BIO_REFFED))) {
682 		BUG_ON(!atomic_read(&bio->__bi_cnt));
683 		if (!atomic_dec_and_test(&bio->__bi_cnt))
684 			return;
685 	}
686 
687 	if (bio_flagged(bio, BIO_PERCPU_CACHE)) {
688 		struct bio_alloc_cache *cache;
689 
690 		bio_uninit(bio);
691 		cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
692 		bio_list_add_head(&cache->free_list, bio);
693 		if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK)
694 			bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK);
695 		put_cpu();
696 	} else {
697 		bio_free(bio);
698 	}
699 }
700 EXPORT_SYMBOL(bio_put);
701 
702 /**
703  * 	__bio_clone_fast - clone a bio that shares the original bio's biovec
704  * 	@bio: destination bio
705  * 	@bio_src: bio to clone
706  *
707  *	Clone a &bio. Caller will own the returned bio, but not
708  *	the actual data it points to. Reference count of returned
709  * 	bio will be one.
710  *
711  * 	Caller must ensure that @bio_src is not freed before @bio.
712  */
__bio_clone_fast(struct bio * bio,struct bio * bio_src)713 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
714 {
715 	WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs);
716 
717 	/*
718 	 * most users will be overriding ->bi_bdev with a new target,
719 	 * so we don't set nor calculate new physical/hw segment counts here
720 	 */
721 	bio->bi_bdev = bio_src->bi_bdev;
722 	bio_set_flag(bio, BIO_CLONED);
723 	if (bio_flagged(bio_src, BIO_THROTTLED))
724 		bio_set_flag(bio, BIO_THROTTLED);
725 	if (bio_flagged(bio_src, BIO_REMAPPED))
726 		bio_set_flag(bio, BIO_REMAPPED);
727 	bio->bi_opf = bio_src->bi_opf;
728 	bio->bi_ioprio = bio_src->bi_ioprio;
729 	bio->bi_write_hint = bio_src->bi_write_hint;
730 	bio->bi_iter = bio_src->bi_iter;
731 	bio->bi_io_vec = bio_src->bi_io_vec;
732 
733 	bio_clone_blkg_association(bio, bio_src);
734 	blkcg_bio_issue_init(bio);
735 }
736 EXPORT_SYMBOL(__bio_clone_fast);
737 
738 /**
739  *	bio_clone_fast - clone a bio that shares the original bio's biovec
740  *	@bio: bio to clone
741  *	@gfp_mask: allocation priority
742  *	@bs: bio_set to allocate from
743  *
744  * 	Like __bio_clone_fast, only also allocates the returned bio
745  */
bio_clone_fast(struct bio * bio,gfp_t gfp_mask,struct bio_set * bs)746 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
747 {
748 	struct bio *b;
749 
750 	b = bio_alloc_bioset(gfp_mask, 0, bs);
751 	if (!b)
752 		return NULL;
753 
754 	__bio_clone_fast(b, bio);
755 
756 	if (bio_crypt_clone(b, bio, gfp_mask) < 0)
757 		goto err_put;
758 
759 	if (bio_integrity(bio) &&
760 	    bio_integrity_clone(b, bio, gfp_mask) < 0)
761 		goto err_put;
762 
763 	return b;
764 
765 err_put:
766 	bio_put(b);
767 	return NULL;
768 }
769 EXPORT_SYMBOL(bio_clone_fast);
770 
bio_devname(struct bio * bio,char * buf)771 const char *bio_devname(struct bio *bio, char *buf)
772 {
773 	return bdevname(bio->bi_bdev, buf);
774 }
775 EXPORT_SYMBOL(bio_devname);
776 
777 /**
778  * bio_full - check if the bio is full
779  * @bio:	bio to check
780  * @len:	length of one segment to be added
781  *
782  * Return true if @bio is full and one segment with @len bytes can't be
783  * added to the bio, otherwise return false
784  */
bio_full(struct bio * bio,unsigned len)785 static inline bool bio_full(struct bio *bio, unsigned len)
786 {
787 	if (bio->bi_vcnt >= bio->bi_max_vecs)
788 		return true;
789 	if (bio->bi_iter.bi_size > UINT_MAX - len)
790 		return true;
791 	return false;
792 }
793 
page_is_mergeable(const struct bio_vec * bv,struct page * page,unsigned int len,unsigned int off,bool * same_page)794 static inline bool page_is_mergeable(const struct bio_vec *bv,
795 		struct page *page, unsigned int len, unsigned int off,
796 		bool *same_page)
797 {
798 	size_t bv_end = bv->bv_offset + bv->bv_len;
799 	phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
800 	phys_addr_t page_addr = page_to_phys(page);
801 
802 	if (vec_end_addr + 1 != page_addr + off)
803 		return false;
804 	if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
805 		return false;
806 
807 	*same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
808 	if (*same_page)
809 		return true;
810 	return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
811 }
812 
813 /**
814  * __bio_try_merge_page - try appending data to an existing bvec.
815  * @bio: destination bio
816  * @page: start page to add
817  * @len: length of the data to add
818  * @off: offset of the data relative to @page
819  * @same_page: return if the segment has been merged inside the same page
820  *
821  * Try to add the data at @page + @off to the last bvec of @bio.  This is a
822  * useful optimisation for file systems with a block size smaller than the
823  * page size.
824  *
825  * Warn if (@len, @off) crosses pages in case that @same_page is true.
826  *
827  * Return %true on success or %false on failure.
828  */
__bio_try_merge_page(struct bio * bio,struct page * page,unsigned int len,unsigned int off,bool * same_page)829 static bool __bio_try_merge_page(struct bio *bio, struct page *page,
830 		unsigned int len, unsigned int off, bool *same_page)
831 {
832 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
833 		return false;
834 
835 	if (bio->bi_vcnt > 0) {
836 		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
837 
838 		if (page_is_mergeable(bv, page, len, off, same_page)) {
839 			if (bio->bi_iter.bi_size > UINT_MAX - len) {
840 				*same_page = false;
841 				return false;
842 			}
843 			bv->bv_len += len;
844 			bio->bi_iter.bi_size += len;
845 			return true;
846 		}
847 	}
848 	return false;
849 }
850 
851 /*
852  * Try to merge a page into a segment, while obeying the hardware segment
853  * size limit.  This is not for normal read/write bios, but for passthrough
854  * or Zone Append operations that we can't split.
855  */
bio_try_merge_hw_seg(struct request_queue * q,struct bio * bio,struct page * page,unsigned len,unsigned offset,bool * same_page)856 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
857 				 struct page *page, unsigned len,
858 				 unsigned offset, bool *same_page)
859 {
860 	struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
861 	unsigned long mask = queue_segment_boundary(q);
862 	phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
863 	phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
864 
865 	if ((addr1 | mask) != (addr2 | mask))
866 		return false;
867 	if (bv->bv_len + len > queue_max_segment_size(q))
868 		return false;
869 	return __bio_try_merge_page(bio, page, len, offset, same_page);
870 }
871 
872 /**
873  * bio_add_hw_page - attempt to add a page to a bio with hw constraints
874  * @q: the target queue
875  * @bio: destination bio
876  * @page: page to add
877  * @len: vec entry length
878  * @offset: vec entry offset
879  * @max_sectors: maximum number of sectors that can be added
880  * @same_page: return if the segment has been merged inside the same page
881  *
882  * Add a page to a bio while respecting the hardware max_sectors, max_segment
883  * and gap limitations.
884  */
bio_add_hw_page(struct request_queue * q,struct bio * bio,struct page * page,unsigned int len,unsigned int offset,unsigned int max_sectors,bool * same_page)885 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
886 		struct page *page, unsigned int len, unsigned int offset,
887 		unsigned int max_sectors, bool *same_page)
888 {
889 	struct bio_vec *bvec;
890 
891 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
892 		return 0;
893 
894 	if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
895 		return 0;
896 
897 	if (bio->bi_vcnt > 0) {
898 		if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
899 			return len;
900 
901 		/*
902 		 * If the queue doesn't support SG gaps and adding this segment
903 		 * would create a gap, disallow it.
904 		 */
905 		bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
906 		if (bvec_gap_to_prev(q, bvec, offset))
907 			return 0;
908 	}
909 
910 	if (bio_full(bio, len))
911 		return 0;
912 
913 	if (bio->bi_vcnt >= queue_max_segments(q))
914 		return 0;
915 
916 	bvec = &bio->bi_io_vec[bio->bi_vcnt];
917 	bvec->bv_page = page;
918 	bvec->bv_len = len;
919 	bvec->bv_offset = offset;
920 	bio->bi_vcnt++;
921 	bio->bi_iter.bi_size += len;
922 	return len;
923 }
924 
925 /**
926  * bio_add_pc_page	- attempt to add page to passthrough bio
927  * @q: the target queue
928  * @bio: destination bio
929  * @page: page to add
930  * @len: vec entry length
931  * @offset: vec entry offset
932  *
933  * Attempt to add a page to the bio_vec maplist. This can fail for a
934  * number of reasons, such as the bio being full or target block device
935  * limitations. The target block device must allow bio's up to PAGE_SIZE,
936  * so it is always possible to add a single page to an empty bio.
937  *
938  * This should only be used by passthrough bios.
939  */
bio_add_pc_page(struct request_queue * q,struct bio * bio,struct page * page,unsigned int len,unsigned int offset)940 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
941 		struct page *page, unsigned int len, unsigned int offset)
942 {
943 	bool same_page = false;
944 	return bio_add_hw_page(q, bio, page, len, offset,
945 			queue_max_hw_sectors(q), &same_page);
946 }
947 EXPORT_SYMBOL(bio_add_pc_page);
948 
949 /**
950  * bio_add_zone_append_page - attempt to add page to zone-append bio
951  * @bio: destination bio
952  * @page: page to add
953  * @len: vec entry length
954  * @offset: vec entry offset
955  *
956  * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
957  * for a zone-append request. This can fail for a number of reasons, such as the
958  * bio being full or the target block device is not a zoned block device or
959  * other limitations of the target block device. The target block device must
960  * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
961  * to an empty bio.
962  *
963  * Returns: number of bytes added to the bio, or 0 in case of a failure.
964  */
bio_add_zone_append_page(struct bio * bio,struct page * page,unsigned int len,unsigned int offset)965 int bio_add_zone_append_page(struct bio *bio, struct page *page,
966 			     unsigned int len, unsigned int offset)
967 {
968 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
969 	bool same_page = false;
970 
971 	if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
972 		return 0;
973 
974 	if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
975 		return 0;
976 
977 	return bio_add_hw_page(q, bio, page, len, offset,
978 			       queue_max_zone_append_sectors(q), &same_page);
979 }
980 EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
981 
982 /**
983  * __bio_add_page - add page(s) to a bio in a new segment
984  * @bio: destination bio
985  * @page: start page to add
986  * @len: length of the data to add, may cross pages
987  * @off: offset of the data relative to @page, may cross pages
988  *
989  * Add the data at @page + @off to @bio as a new bvec.  The caller must ensure
990  * that @bio has space for another bvec.
991  */
__bio_add_page(struct bio * bio,struct page * page,unsigned int len,unsigned int off)992 void __bio_add_page(struct bio *bio, struct page *page,
993 		unsigned int len, unsigned int off)
994 {
995 	struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
996 
997 	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
998 	WARN_ON_ONCE(bio_full(bio, len));
999 
1000 	bv->bv_page = page;
1001 	bv->bv_offset = off;
1002 	bv->bv_len = len;
1003 
1004 	bio->bi_iter.bi_size += len;
1005 	bio->bi_vcnt++;
1006 
1007 	if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
1008 		bio_set_flag(bio, BIO_WORKINGSET);
1009 }
1010 EXPORT_SYMBOL_GPL(__bio_add_page);
1011 
1012 /**
1013  *	bio_add_page	-	attempt to add page(s) to bio
1014  *	@bio: destination bio
1015  *	@page: start page to add
1016  *	@len: vec entry length, may cross pages
1017  *	@offset: vec entry offset relative to @page, may cross pages
1018  *
1019  *	Attempt to add page(s) to the bio_vec maplist. This will only fail
1020  *	if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1021  */
bio_add_page(struct bio * bio,struct page * page,unsigned int len,unsigned int offset)1022 int bio_add_page(struct bio *bio, struct page *page,
1023 		 unsigned int len, unsigned int offset)
1024 {
1025 	bool same_page = false;
1026 
1027 	if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1028 		if (bio_full(bio, len))
1029 			return 0;
1030 		__bio_add_page(bio, page, len, offset);
1031 	}
1032 	return len;
1033 }
1034 EXPORT_SYMBOL(bio_add_page);
1035 
__bio_release_pages(struct bio * bio,bool mark_dirty)1036 void __bio_release_pages(struct bio *bio, bool mark_dirty)
1037 {
1038 	struct bvec_iter_all iter_all;
1039 	struct bio_vec *bvec;
1040 
1041 	bio_for_each_segment_all(bvec, bio, iter_all) {
1042 		if (mark_dirty && !PageCompound(bvec->bv_page))
1043 			set_page_dirty_lock(bvec->bv_page);
1044 		put_page(bvec->bv_page);
1045 	}
1046 }
1047 EXPORT_SYMBOL_GPL(__bio_release_pages);
1048 
bio_iov_bvec_set(struct bio * bio,struct iov_iter * iter)1049 void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1050 {
1051 	size_t size = iov_iter_count(iter);
1052 
1053 	WARN_ON_ONCE(bio->bi_max_vecs);
1054 
1055 	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1056 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1057 		size_t max_sectors = queue_max_zone_append_sectors(q);
1058 
1059 		size = min(size, max_sectors << SECTOR_SHIFT);
1060 	}
1061 
1062 	bio->bi_vcnt = iter->nr_segs;
1063 	bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1064 	bio->bi_iter.bi_bvec_done = iter->iov_offset;
1065 	bio->bi_iter.bi_size = size;
1066 	bio_set_flag(bio, BIO_NO_PAGE_REF);
1067 	bio_set_flag(bio, BIO_CLONED);
1068 }
1069 
bio_put_pages(struct page ** pages,size_t size,size_t off)1070 static void bio_put_pages(struct page **pages, size_t size, size_t off)
1071 {
1072 	size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE);
1073 
1074 	for (i = 0; i < nr; i++)
1075 		put_page(pages[i]);
1076 }
1077 
1078 #define PAGE_PTRS_PER_BVEC     (sizeof(struct bio_vec) / sizeof(struct page *))
1079 
1080 /**
1081  * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1082  * @bio: bio to add pages to
1083  * @iter: iov iterator describing the region to be mapped
1084  *
1085  * Pins pages from *iter and appends them to @bio's bvec array. The
1086  * pages will have to be released using put_page() when done.
1087  * For multi-segment *iter, this function only adds pages from the
1088  * next non-empty segment of the iov iterator.
1089  */
__bio_iov_iter_get_pages(struct bio * bio,struct iov_iter * iter)1090 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1091 {
1092 	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1093 	unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1094 	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1095 	struct page **pages = (struct page **)bv;
1096 	bool same_page = false;
1097 	ssize_t size, left;
1098 	unsigned len, i;
1099 	size_t offset;
1100 
1101 	/*
1102 	 * Move page array up in the allocated memory for the bio vecs as far as
1103 	 * possible so that we can start filling biovecs from the beginning
1104 	 * without overwriting the temporary page array.
1105 	*/
1106 	BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1107 	pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1108 
1109 	size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1110 	if (unlikely(size <= 0))
1111 		return size ? size : -EFAULT;
1112 
1113 	for (left = size, i = 0; left > 0; left -= len, i++) {
1114 		struct page *page = pages[i];
1115 
1116 		len = min_t(size_t, PAGE_SIZE - offset, left);
1117 
1118 		if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1119 			if (same_page)
1120 				put_page(page);
1121 		} else {
1122 			if (WARN_ON_ONCE(bio_full(bio, len))) {
1123 				bio_put_pages(pages + i, left, offset);
1124 				return -EINVAL;
1125 			}
1126 			__bio_add_page(bio, page, len, offset);
1127 		}
1128 		offset = 0;
1129 	}
1130 
1131 	iov_iter_advance(iter, size);
1132 	return 0;
1133 }
1134 
__bio_iov_append_get_pages(struct bio * bio,struct iov_iter * iter)1135 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
1136 {
1137 	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1138 	unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1139 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1140 	unsigned int max_append_sectors = queue_max_zone_append_sectors(q);
1141 	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1142 	struct page **pages = (struct page **)bv;
1143 	ssize_t size, left;
1144 	unsigned len, i;
1145 	size_t offset;
1146 	int ret = 0;
1147 
1148 	if (WARN_ON_ONCE(!max_append_sectors))
1149 		return 0;
1150 
1151 	/*
1152 	 * Move page array up in the allocated memory for the bio vecs as far as
1153 	 * possible so that we can start filling biovecs from the beginning
1154 	 * without overwriting the temporary page array.
1155 	 */
1156 	BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1157 	pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1158 
1159 	size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1160 	if (unlikely(size <= 0))
1161 		return size ? size : -EFAULT;
1162 
1163 	for (left = size, i = 0; left > 0; left -= len, i++) {
1164 		struct page *page = pages[i];
1165 		bool same_page = false;
1166 
1167 		len = min_t(size_t, PAGE_SIZE - offset, left);
1168 		if (bio_add_hw_page(q, bio, page, len, offset,
1169 				max_append_sectors, &same_page) != len) {
1170 			bio_put_pages(pages + i, left, offset);
1171 			ret = -EINVAL;
1172 			break;
1173 		}
1174 		if (same_page)
1175 			put_page(page);
1176 		offset = 0;
1177 	}
1178 
1179 	iov_iter_advance(iter, size - left);
1180 	return ret;
1181 }
1182 
1183 /**
1184  * bio_iov_iter_get_pages - add user or kernel pages to a bio
1185  * @bio: bio to add pages to
1186  * @iter: iov iterator describing the region to be added
1187  *
1188  * This takes either an iterator pointing to user memory, or one pointing to
1189  * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1190  * map them into the kernel. On IO completion, the caller should put those
1191  * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1192  * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1193  * to ensure the bvecs and pages stay referenced until the submitted I/O is
1194  * completed by a call to ->ki_complete() or returns with an error other than
1195  * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1196  * on IO completion. If it isn't, then pages should be released.
1197  *
1198  * The function tries, but does not guarantee, to pin as many pages as
1199  * fit into the bio, or are requested in @iter, whatever is smaller. If
1200  * MM encounters an error pinning the requested pages, it stops. Error
1201  * is returned only if 0 pages could be pinned.
1202  *
1203  * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1204  * responsible for setting BIO_WORKINGSET if necessary.
1205  */
bio_iov_iter_get_pages(struct bio * bio,struct iov_iter * iter)1206 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1207 {
1208 	int ret = 0;
1209 
1210 	if (iov_iter_is_bvec(iter)) {
1211 		bio_iov_bvec_set(bio, iter);
1212 		iov_iter_advance(iter, bio->bi_iter.bi_size);
1213 		return 0;
1214 	}
1215 
1216 	do {
1217 		if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1218 			ret = __bio_iov_append_get_pages(bio, iter);
1219 		else
1220 			ret = __bio_iov_iter_get_pages(bio, iter);
1221 	} while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1222 
1223 	/* don't account direct I/O as memory stall */
1224 	bio_clear_flag(bio, BIO_WORKINGSET);
1225 	return bio->bi_vcnt ? 0 : ret;
1226 }
1227 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1228 
submit_bio_wait_endio(struct bio * bio)1229 static void submit_bio_wait_endio(struct bio *bio)
1230 {
1231 	complete(bio->bi_private);
1232 }
1233 
1234 /**
1235  * submit_bio_wait - submit a bio, and wait until it completes
1236  * @bio: The &struct bio which describes the I/O
1237  *
1238  * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1239  * bio_endio() on failure.
1240  *
1241  * WARNING: Unlike to how submit_bio() is usually used, this function does not
1242  * result in bio reference to be consumed. The caller must drop the reference
1243  * on his own.
1244  */
submit_bio_wait(struct bio * bio)1245 int submit_bio_wait(struct bio *bio)
1246 {
1247 	DECLARE_COMPLETION_ONSTACK_MAP(done,
1248 			bio->bi_bdev->bd_disk->lockdep_map);
1249 	unsigned long hang_check;
1250 
1251 	bio->bi_private = &done;
1252 	bio->bi_end_io = submit_bio_wait_endio;
1253 	bio->bi_opf |= REQ_SYNC;
1254 	submit_bio(bio);
1255 
1256 	/* Prevent hang_check timer from firing at us during very long I/O */
1257 	hang_check = sysctl_hung_task_timeout_secs;
1258 	if (hang_check)
1259 		while (!wait_for_completion_io_timeout(&done,
1260 					hang_check * (HZ/2)))
1261 			;
1262 	else
1263 		wait_for_completion_io(&done);
1264 
1265 	return blk_status_to_errno(bio->bi_status);
1266 }
1267 EXPORT_SYMBOL(submit_bio_wait);
1268 
__bio_advance(struct bio * bio,unsigned bytes)1269 void __bio_advance(struct bio *bio, unsigned bytes)
1270 {
1271 	if (bio_integrity(bio))
1272 		bio_integrity_advance(bio, bytes);
1273 
1274 	bio_crypt_advance(bio, bytes);
1275 	bio_advance_iter(bio, &bio->bi_iter, bytes);
1276 }
1277 EXPORT_SYMBOL(__bio_advance);
1278 
bio_copy_data_iter(struct bio * dst,struct bvec_iter * dst_iter,struct bio * src,struct bvec_iter * src_iter)1279 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1280 			struct bio *src, struct bvec_iter *src_iter)
1281 {
1282 	while (src_iter->bi_size && dst_iter->bi_size) {
1283 		struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1284 		struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1285 		unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1286 		void *src_buf;
1287 
1288 		src_buf = bvec_kmap_local(&src_bv);
1289 		memcpy_to_bvec(&dst_bv, src_buf);
1290 		kunmap_local(src_buf);
1291 
1292 		bio_advance_iter_single(src, src_iter, bytes);
1293 		bio_advance_iter_single(dst, dst_iter, bytes);
1294 	}
1295 }
1296 EXPORT_SYMBOL(bio_copy_data_iter);
1297 
1298 /**
1299  * bio_copy_data - copy contents of data buffers from one bio to another
1300  * @src: source bio
1301  * @dst: destination bio
1302  *
1303  * Stops when it reaches the end of either @src or @dst - that is, copies
1304  * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1305  */
bio_copy_data(struct bio * dst,struct bio * src)1306 void bio_copy_data(struct bio *dst, struct bio *src)
1307 {
1308 	struct bvec_iter src_iter = src->bi_iter;
1309 	struct bvec_iter dst_iter = dst->bi_iter;
1310 
1311 	bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1312 }
1313 EXPORT_SYMBOL(bio_copy_data);
1314 
bio_free_pages(struct bio * bio)1315 void bio_free_pages(struct bio *bio)
1316 {
1317 	struct bio_vec *bvec;
1318 	struct bvec_iter_all iter_all;
1319 
1320 	bio_for_each_segment_all(bvec, bio, iter_all)
1321 		__free_page(bvec->bv_page);
1322 }
1323 EXPORT_SYMBOL(bio_free_pages);
1324 
1325 /*
1326  * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1327  * for performing direct-IO in BIOs.
1328  *
1329  * The problem is that we cannot run set_page_dirty() from interrupt context
1330  * because the required locks are not interrupt-safe.  So what we can do is to
1331  * mark the pages dirty _before_ performing IO.  And in interrupt context,
1332  * check that the pages are still dirty.   If so, fine.  If not, redirty them
1333  * in process context.
1334  *
1335  * We special-case compound pages here: normally this means reads into hugetlb
1336  * pages.  The logic in here doesn't really work right for compound pages
1337  * because the VM does not uniformly chase down the head page in all cases.
1338  * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1339  * handle them at all.  So we skip compound pages here at an early stage.
1340  *
1341  * Note that this code is very hard to test under normal circumstances because
1342  * direct-io pins the pages with get_user_pages().  This makes
1343  * is_page_cache_freeable return false, and the VM will not clean the pages.
1344  * But other code (eg, flusher threads) could clean the pages if they are mapped
1345  * pagecache.
1346  *
1347  * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1348  * deferred bio dirtying paths.
1349  */
1350 
1351 /*
1352  * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1353  */
bio_set_pages_dirty(struct bio * bio)1354 void bio_set_pages_dirty(struct bio *bio)
1355 {
1356 	struct bio_vec *bvec;
1357 	struct bvec_iter_all iter_all;
1358 
1359 	bio_for_each_segment_all(bvec, bio, iter_all) {
1360 		if (!PageCompound(bvec->bv_page))
1361 			set_page_dirty_lock(bvec->bv_page);
1362 	}
1363 }
1364 
1365 /*
1366  * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1367  * If they are, then fine.  If, however, some pages are clean then they must
1368  * have been written out during the direct-IO read.  So we take another ref on
1369  * the BIO and re-dirty the pages in process context.
1370  *
1371  * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1372  * here on.  It will run one put_page() against each page and will run one
1373  * bio_put() against the BIO.
1374  */
1375 
1376 static void bio_dirty_fn(struct work_struct *work);
1377 
1378 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1379 static DEFINE_SPINLOCK(bio_dirty_lock);
1380 static struct bio *bio_dirty_list;
1381 
1382 /*
1383  * This runs in process context
1384  */
bio_dirty_fn(struct work_struct * work)1385 static void bio_dirty_fn(struct work_struct *work)
1386 {
1387 	struct bio *bio, *next;
1388 
1389 	spin_lock_irq(&bio_dirty_lock);
1390 	next = bio_dirty_list;
1391 	bio_dirty_list = NULL;
1392 	spin_unlock_irq(&bio_dirty_lock);
1393 
1394 	while ((bio = next) != NULL) {
1395 		next = bio->bi_private;
1396 
1397 		bio_release_pages(bio, true);
1398 		bio_put(bio);
1399 	}
1400 }
1401 
bio_check_pages_dirty(struct bio * bio)1402 void bio_check_pages_dirty(struct bio *bio)
1403 {
1404 	struct bio_vec *bvec;
1405 	unsigned long flags;
1406 	struct bvec_iter_all iter_all;
1407 
1408 	bio_for_each_segment_all(bvec, bio, iter_all) {
1409 		if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1410 			goto defer;
1411 	}
1412 
1413 	bio_release_pages(bio, false);
1414 	bio_put(bio);
1415 	return;
1416 defer:
1417 	spin_lock_irqsave(&bio_dirty_lock, flags);
1418 	bio->bi_private = bio_dirty_list;
1419 	bio_dirty_list = bio;
1420 	spin_unlock_irqrestore(&bio_dirty_lock, flags);
1421 	schedule_work(&bio_dirty_work);
1422 }
1423 
bio_remaining_done(struct bio * bio)1424 static inline bool bio_remaining_done(struct bio *bio)
1425 {
1426 	/*
1427 	 * If we're not chaining, then ->__bi_remaining is always 1 and
1428 	 * we always end io on the first invocation.
1429 	 */
1430 	if (!bio_flagged(bio, BIO_CHAIN))
1431 		return true;
1432 
1433 	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1434 
1435 	if (atomic_dec_and_test(&bio->__bi_remaining)) {
1436 		bio_clear_flag(bio, BIO_CHAIN);
1437 		return true;
1438 	}
1439 
1440 	return false;
1441 }
1442 
1443 /**
1444  * bio_endio - end I/O on a bio
1445  * @bio:	bio
1446  *
1447  * Description:
1448  *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1449  *   way to end I/O on a bio. No one should call bi_end_io() directly on a
1450  *   bio unless they own it and thus know that it has an end_io function.
1451  *
1452  *   bio_endio() can be called several times on a bio that has been chained
1453  *   using bio_chain().  The ->bi_end_io() function will only be called the
1454  *   last time.
1455  **/
bio_endio(struct bio * bio)1456 void bio_endio(struct bio *bio)
1457 {
1458 again:
1459 	if (!bio_remaining_done(bio))
1460 		return;
1461 	if (!bio_integrity_endio(bio))
1462 		return;
1463 
1464 	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED))
1465 		rq_qos_done_bio(bdev_get_queue(bio->bi_bdev), bio);
1466 
1467 	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1468 		trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1469 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1470 	}
1471 
1472 	/*
1473 	 * Need to have a real endio function for chained bios, otherwise
1474 	 * various corner cases will break (like stacking block devices that
1475 	 * save/restore bi_end_io) - however, we want to avoid unbounded
1476 	 * recursion and blowing the stack. Tail call optimization would
1477 	 * handle this, but compiling with frame pointers also disables
1478 	 * gcc's sibling call optimization.
1479 	 */
1480 	if (bio->bi_end_io == bio_chain_endio) {
1481 		bio = __bio_chain_endio(bio);
1482 		goto again;
1483 	}
1484 
1485 	blk_throtl_bio_endio(bio);
1486 	/* release cgroup info */
1487 	bio_uninit(bio);
1488 	if (bio->bi_end_io)
1489 		bio->bi_end_io(bio);
1490 }
1491 EXPORT_SYMBOL(bio_endio);
1492 
1493 /**
1494  * bio_split - split a bio
1495  * @bio:	bio to split
1496  * @sectors:	number of sectors to split from the front of @bio
1497  * @gfp:	gfp mask
1498  * @bs:		bio set to allocate from
1499  *
1500  * Allocates and returns a new bio which represents @sectors from the start of
1501  * @bio, and updates @bio to represent the remaining sectors.
1502  *
1503  * Unless this is a discard request the newly allocated bio will point
1504  * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1505  * neither @bio nor @bs are freed before the split bio.
1506  */
bio_split(struct bio * bio,int sectors,gfp_t gfp,struct bio_set * bs)1507 struct bio *bio_split(struct bio *bio, int sectors,
1508 		      gfp_t gfp, struct bio_set *bs)
1509 {
1510 	struct bio *split;
1511 
1512 	BUG_ON(sectors <= 0);
1513 	BUG_ON(sectors >= bio_sectors(bio));
1514 
1515 	/* Zone append commands cannot be split */
1516 	if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1517 		return NULL;
1518 
1519 	split = bio_clone_fast(bio, gfp, bs);
1520 	if (!split)
1521 		return NULL;
1522 
1523 	split->bi_iter.bi_size = sectors << 9;
1524 
1525 	if (bio_integrity(split))
1526 		bio_integrity_trim(split);
1527 
1528 	bio_advance(bio, split->bi_iter.bi_size);
1529 
1530 	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1531 		bio_set_flag(split, BIO_TRACE_COMPLETION);
1532 
1533 	return split;
1534 }
1535 EXPORT_SYMBOL(bio_split);
1536 
1537 /**
1538  * bio_trim - trim a bio
1539  * @bio:	bio to trim
1540  * @offset:	number of sectors to trim from the front of @bio
1541  * @size:	size we want to trim @bio to, in sectors
1542  *
1543  * This function is typically used for bios that are cloned and submitted
1544  * to the underlying device in parts.
1545  */
bio_trim(struct bio * bio,sector_t offset,sector_t size)1546 void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1547 {
1548 	if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1549 			 offset + size > bio->bi_iter.bi_size))
1550 		return;
1551 
1552 	size <<= 9;
1553 	if (offset == 0 && size == bio->bi_iter.bi_size)
1554 		return;
1555 
1556 	bio_advance(bio, offset << 9);
1557 	bio->bi_iter.bi_size = size;
1558 
1559 	if (bio_integrity(bio))
1560 		bio_integrity_trim(bio);
1561 }
1562 EXPORT_SYMBOL_GPL(bio_trim);
1563 
1564 /*
1565  * create memory pools for biovec's in a bio_set.
1566  * use the global biovec slabs created for general use.
1567  */
biovec_init_pool(mempool_t * pool,int pool_entries)1568 int biovec_init_pool(mempool_t *pool, int pool_entries)
1569 {
1570 	struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1571 
1572 	return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1573 }
1574 
1575 /*
1576  * bioset_exit - exit a bioset initialized with bioset_init()
1577  *
1578  * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1579  * kzalloc()).
1580  */
bioset_exit(struct bio_set * bs)1581 void bioset_exit(struct bio_set *bs)
1582 {
1583 	bio_alloc_cache_destroy(bs);
1584 	if (bs->rescue_workqueue)
1585 		destroy_workqueue(bs->rescue_workqueue);
1586 	bs->rescue_workqueue = NULL;
1587 
1588 	mempool_exit(&bs->bio_pool);
1589 	mempool_exit(&bs->bvec_pool);
1590 
1591 	bioset_integrity_free(bs);
1592 	if (bs->bio_slab)
1593 		bio_put_slab(bs);
1594 	bs->bio_slab = NULL;
1595 }
1596 EXPORT_SYMBOL(bioset_exit);
1597 
1598 /**
1599  * bioset_init - Initialize a bio_set
1600  * @bs:		pool to initialize
1601  * @pool_size:	Number of bio and bio_vecs to cache in the mempool
1602  * @front_pad:	Number of bytes to allocate in front of the returned bio
1603  * @flags:	Flags to modify behavior, currently %BIOSET_NEED_BVECS
1604  *              and %BIOSET_NEED_RESCUER
1605  *
1606  * Description:
1607  *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1608  *    to ask for a number of bytes to be allocated in front of the bio.
1609  *    Front pad allocation is useful for embedding the bio inside
1610  *    another structure, to avoid allocating extra data to go with the bio.
1611  *    Note that the bio must be embedded at the END of that structure always,
1612  *    or things will break badly.
1613  *    If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1614  *    for allocating iovecs.  This pool is not needed e.g. for bio_clone_fast().
1615  *    If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1616  *    dispatch queued requests when the mempool runs out of space.
1617  *
1618  */
bioset_init(struct bio_set * bs,unsigned int pool_size,unsigned int front_pad,int flags)1619 int bioset_init(struct bio_set *bs,
1620 		unsigned int pool_size,
1621 		unsigned int front_pad,
1622 		int flags)
1623 {
1624 	bs->front_pad = front_pad;
1625 	if (flags & BIOSET_NEED_BVECS)
1626 		bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1627 	else
1628 		bs->back_pad = 0;
1629 
1630 	spin_lock_init(&bs->rescue_lock);
1631 	bio_list_init(&bs->rescue_list);
1632 	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1633 
1634 	bs->bio_slab = bio_find_or_create_slab(bs);
1635 	if (!bs->bio_slab)
1636 		return -ENOMEM;
1637 
1638 	if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1639 		goto bad;
1640 
1641 	if ((flags & BIOSET_NEED_BVECS) &&
1642 	    biovec_init_pool(&bs->bvec_pool, pool_size))
1643 		goto bad;
1644 
1645 	if (flags & BIOSET_NEED_RESCUER) {
1646 		bs->rescue_workqueue = alloc_workqueue("bioset",
1647 							WQ_MEM_RECLAIM, 0);
1648 		if (!bs->rescue_workqueue)
1649 			goto bad;
1650 	}
1651 	if (flags & BIOSET_PERCPU_CACHE) {
1652 		bs->cache = alloc_percpu(struct bio_alloc_cache);
1653 		if (!bs->cache)
1654 			goto bad;
1655 		cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1656 	}
1657 
1658 	return 0;
1659 bad:
1660 	bioset_exit(bs);
1661 	return -ENOMEM;
1662 }
1663 EXPORT_SYMBOL(bioset_init);
1664 
1665 /*
1666  * Initialize and setup a new bio_set, based on the settings from
1667  * another bio_set.
1668  */
bioset_init_from_src(struct bio_set * bs,struct bio_set * src)1669 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1670 {
1671 	int flags;
1672 
1673 	flags = 0;
1674 	if (src->bvec_pool.min_nr)
1675 		flags |= BIOSET_NEED_BVECS;
1676 	if (src->rescue_workqueue)
1677 		flags |= BIOSET_NEED_RESCUER;
1678 
1679 	return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1680 }
1681 EXPORT_SYMBOL(bioset_init_from_src);
1682 
1683 /**
1684  * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb
1685  * @kiocb:	kiocb describing the IO
1686  * @nr_vecs:	number of iovecs to pre-allocate
1687  * @bs:		bio_set to allocate from
1688  *
1689  * Description:
1690  *    Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only
1691  *    used to check if we should dip into the per-cpu bio_set allocation
1692  *    cache. The allocation uses GFP_KERNEL internally. On return, the
1693  *    bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio
1694  *    MUST be done from process context, not hard/soft IRQ.
1695  *
1696  */
bio_alloc_kiocb(struct kiocb * kiocb,unsigned short nr_vecs,struct bio_set * bs)1697 struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs,
1698 			    struct bio_set *bs)
1699 {
1700 	struct bio_alloc_cache *cache;
1701 	struct bio *bio;
1702 
1703 	if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS)
1704 		return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
1705 
1706 	cache = per_cpu_ptr(bs->cache, get_cpu());
1707 	bio = bio_list_pop(&cache->free_list);
1708 	if (bio) {
1709 		cache->nr--;
1710 		put_cpu();
1711 		bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs);
1712 		bio->bi_pool = bs;
1713 		bio_set_flag(bio, BIO_PERCPU_CACHE);
1714 		return bio;
1715 	}
1716 	put_cpu();
1717 	bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
1718 	bio_set_flag(bio, BIO_PERCPU_CACHE);
1719 	return bio;
1720 }
1721 EXPORT_SYMBOL_GPL(bio_alloc_kiocb);
1722 
init_bio(void)1723 static int __init init_bio(void)
1724 {
1725 	int i;
1726 
1727 	bio_integrity_init();
1728 
1729 	for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1730 		struct biovec_slab *bvs = bvec_slabs + i;
1731 
1732 		bvs->slab = kmem_cache_create(bvs->name,
1733 				bvs->nr_vecs * sizeof(struct bio_vec), 0,
1734 				SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1735 	}
1736 
1737 	cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1738 					bio_cpu_dead);
1739 
1740 	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
1741 		panic("bio: can't allocate bios\n");
1742 
1743 	if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1744 		panic("bio: can't create integrity pool\n");
1745 
1746 	return 0;
1747 }
1748 subsys_initcall(init_bio);
1749