blob: d6e5ba3399f0ae151ea040e2ec1fd1df1c3dba6a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Jens Axboe0fe23472006-09-04 15:41:16 +02002 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070022#include <linux/uio.h>
Tejun Heo852c7882012-03-05 13:15:27 -080023#include <linux/iocontext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
25#include <linux/init.h>
26#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050027#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/mempool.h>
29#include <linux/workqueue.h>
Tejun Heo852c7882012-03-05 13:15:27 -080030#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Li Zefan55782132009-06-09 13:43:05 +080032#include <trace/events/block.h>
Ingo Molnar0bfc2452008-11-26 11:59:56 +010033
Jens Axboe392ddc32008-12-23 12:42:54 +010034/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/*
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
43 * unsigned short
44 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Martin K. Petersendf677142011-03-08 08:28:01 +010046static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48};
49#undef BV
50
51/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
54 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020055struct bio_set *fs_bio_set;
Kent Overstreet3f86a822012-09-06 15:35:01 -070056EXPORT_SYMBOL(fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Jens Axboebb799ca2008-12-10 15:35:05 +010058/*
59 * Our slab pool management
60 */
61struct bio_slab {
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
65 char name[8];
66};
67static DEFINE_MUTEX(bio_slab_lock);
68static struct bio_slab *bio_slabs;
69static unsigned int bio_slab_nr, bio_slab_max;
70
71static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72{
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020075 struct bio_slab *bslab, *new_bio_slabs;
Anna Leuschner386bc352012-10-22 21:53:36 +020076 unsigned int new_bio_slab_max;
Jens Axboebb799ca2008-12-10 15:35:05 +010077 unsigned int i, entry = -1;
78
79 mutex_lock(&bio_slab_lock);
80
81 i = 0;
82 while (i < bio_slab_nr) {
Thiago Farinaf06f1352010-01-19 14:07:09 +010083 bslab = &bio_slabs[i];
Jens Axboebb799ca2008-12-10 15:35:05 +010084
85 if (!bslab->slab && entry == -1)
86 entry = i;
87 else if (bslab->slab_size == sz) {
88 slab = bslab->slab;
89 bslab->slab_ref++;
90 break;
91 }
92 i++;
93 }
94
95 if (slab)
96 goto out_unlock;
97
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
Anna Leuschner386bc352012-10-22 21:53:36 +020099 new_bio_slab_max = bio_slab_max << 1;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200100 new_bio_slabs = krealloc(bio_slabs,
Anna Leuschner386bc352012-10-22 21:53:36 +0200101 new_bio_slab_max * sizeof(struct bio_slab),
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200102 GFP_KERNEL);
103 if (!new_bio_slabs)
Jens Axboebb799ca2008-12-10 15:35:05 +0100104 goto out_unlock;
Anna Leuschner386bc352012-10-22 21:53:36 +0200105 bio_slab_max = new_bio_slab_max;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200106 bio_slabs = new_bio_slabs;
Jens Axboebb799ca2008-12-10 15:35:05 +0100107 }
108 if (entry == -1)
109 entry = bio_slab_nr++;
110
111 bslab = &bio_slabs[entry];
112
113 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
Mikulas Patocka6a241482014-03-28 15:51:55 -0400114 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
115 SLAB_HWCACHE_ALIGN, NULL);
Jens Axboebb799ca2008-12-10 15:35:05 +0100116 if (!slab)
117 goto out_unlock;
118
Jens Axboebb799ca2008-12-10 15:35:05 +0100119 bslab->slab = slab;
120 bslab->slab_ref = 1;
121 bslab->slab_size = sz;
122out_unlock:
123 mutex_unlock(&bio_slab_lock);
124 return slab;
125}
126
127static void bio_put_slab(struct bio_set *bs)
128{
129 struct bio_slab *bslab = NULL;
130 unsigned int i;
131
132 mutex_lock(&bio_slab_lock);
133
134 for (i = 0; i < bio_slab_nr; i++) {
135 if (bs->bio_slab == bio_slabs[i].slab) {
136 bslab = &bio_slabs[i];
137 break;
138 }
139 }
140
141 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 goto out;
143
144 WARN_ON(!bslab->slab_ref);
145
146 if (--bslab->slab_ref)
147 goto out;
148
149 kmem_cache_destroy(bslab->slab);
150 bslab->slab = NULL;
151
152out:
153 mutex_unlock(&bio_slab_lock);
154}
155
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200156unsigned int bvec_nr_vecs(unsigned short idx)
157{
158 return bvec_slabs[idx].nr_vecs;
159}
160
Kent Overstreet9f060e22012-10-12 15:29:33 -0700161void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
Jens Axboebb799ca2008-12-10 15:35:05 +0100162{
163 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
164
165 if (idx == BIOVEC_MAX_IDX)
Kent Overstreet9f060e22012-10-12 15:29:33 -0700166 mempool_free(bv, pool);
Jens Axboebb799ca2008-12-10 15:35:05 +0100167 else {
168 struct biovec_slab *bvs = bvec_slabs + idx;
169
170 kmem_cache_free(bvs->slab, bv);
171 }
172}
173
Kent Overstreet9f060e22012-10-12 15:29:33 -0700174struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
175 mempool_t *pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
177 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100180 * see comment near bvec_array define!
181 */
182 switch (nr) {
183 case 1:
184 *idx = 0;
185 break;
186 case 2 ... 4:
187 *idx = 1;
188 break;
189 case 5 ... 16:
190 *idx = 2;
191 break;
192 case 17 ... 64:
193 *idx = 3;
194 break;
195 case 65 ... 128:
196 *idx = 4;
197 break;
198 case 129 ... BIO_MAX_PAGES:
199 *idx = 5;
200 break;
201 default:
202 return NULL;
203 }
204
205 /*
206 * idx now points to the pool we want to allocate from. only the
207 * 1-vec entry pool is mempool backed.
208 */
209 if (*idx == BIOVEC_MAX_IDX) {
210fallback:
Kent Overstreet9f060e22012-10-12 15:29:33 -0700211 bvl = mempool_alloc(pool, gfp_mask);
Jens Axboe7ff93452008-12-11 11:53:43 +0100212 } else {
213 struct biovec_slab *bvs = bvec_slabs + *idx;
214 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200216 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100217 * Make this allocation restricted and don't dump info on
218 * allocation failures, since we'll fallback to the mempool
219 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200220 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100221 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
222
223 /*
224 * Try a slab allocation. If this fails and __GFP_WAIT
225 * is set, retry with the 1-entry mempool
226 */
227 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
228 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
229 *idx = BIOVEC_MAX_IDX;
230 goto fallback;
231 }
232 }
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return bvl;
235}
236
Kent Overstreet4254bba2012-09-06 15:35:00 -0700237static void __bio_free(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Kent Overstreet4254bba2012-09-06 15:35:00 -0700239 bio_disassociate_task(bio);
Jens Axboe992c5dd2007-07-18 13:18:08 +0200240
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200241 if (bio_integrity(bio))
Kent Overstreet1e2a410f2012-09-06 15:34:56 -0700242 bio_integrity_free(bio);
Kent Overstreet4254bba2012-09-06 15:35:00 -0700243}
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200244
Kent Overstreet4254bba2012-09-06 15:35:00 -0700245static void bio_free(struct bio *bio)
246{
247 struct bio_set *bs = bio->bi_pool;
248 void *p;
249
250 __bio_free(bio);
251
252 if (bs) {
Kent Overstreeta38352e2012-05-25 13:03:11 -0700253 if (bio_flagged(bio, BIO_OWNS_VEC))
Kent Overstreet9f060e22012-10-12 15:29:33 -0700254 bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
Kent Overstreet4254bba2012-09-06 15:35:00 -0700255
256 /*
257 * If we have front padding, adjust the bio pointer before freeing
258 */
259 p = bio;
Jens Axboebb799ca2008-12-10 15:35:05 +0100260 p -= bs->front_pad;
261
Kent Overstreet4254bba2012-09-06 15:35:00 -0700262 mempool_free(p, bs->bio_pool);
263 } else {
264 /* Bio was allocated by bio_kmalloc() */
265 kfree(bio);
266 }
Peter Osterlund36763472005-09-06 15:16:42 -0700267}
268
Arjan van de Ven858119e2006-01-14 13:20:43 -0800269void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Jens Axboe2b94de52007-07-18 13:14:03 +0200271 memset(bio, 0, sizeof(*bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 bio->bi_flags = 1 << BIO_UPTODATE;
Jens Axboec4cf5262015-04-17 16:15:18 -0600273 atomic_set(&bio->__bi_remaining, 1);
Jens Axboedac56212015-04-17 16:23:59 -0600274 atomic_set(&bio->__bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200276EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
278/**
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700279 * bio_reset - reinitialize a bio
280 * @bio: bio to reset
281 *
282 * Description:
283 * After calling bio_reset(), @bio will be in the same state as a freshly
284 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
285 * preserved are the ones that are initialized by bio_alloc_bioset(). See
286 * comment in struct bio.
287 */
288void bio_reset(struct bio *bio)
289{
290 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
291
Kent Overstreet4254bba2012-09-06 15:35:00 -0700292 __bio_free(bio);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700293
294 memset(bio, 0, BIO_RESET_BYTES);
Jens Axboec4cf5262015-04-17 16:15:18 -0600295 bio->bi_flags = flags | (1 << BIO_UPTODATE);
296 atomic_set(&bio->__bi_remaining, 1);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700297}
298EXPORT_SYMBOL(bio_reset);
299
Kent Overstreet196d38b2013-11-23 18:34:15 -0800300static void bio_chain_endio(struct bio *bio, int error)
301{
302 bio_endio(bio->bi_private, error);
303 bio_put(bio);
304}
305
Mike Snitzer326e1db2015-05-22 09:14:03 -0400306/*
307 * Increment chain count for the bio. Make sure the CHAIN flag update
308 * is visible before the raised count.
309 */
310static inline void bio_inc_remaining(struct bio *bio)
311{
312 bio->bi_flags |= (1 << BIO_CHAIN);
313 smp_mb__before_atomic();
314 atomic_inc(&bio->__bi_remaining);
315}
316
Kent Overstreet196d38b2013-11-23 18:34:15 -0800317/**
318 * bio_chain - chain bio completions
Randy Dunlap1051a902014-04-20 16:03:31 -0700319 * @bio: the target bio
320 * @parent: the @bio's parent bio
Kent Overstreet196d38b2013-11-23 18:34:15 -0800321 *
322 * The caller won't have a bi_end_io called when @bio completes - instead,
323 * @parent's bi_end_io won't be called until both @parent and @bio have
324 * completed; the chained bio will also be freed when it completes.
325 *
326 * The caller must not set bi_private or bi_end_io in @bio.
327 */
328void bio_chain(struct bio *bio, struct bio *parent)
329{
330 BUG_ON(bio->bi_private || bio->bi_end_io);
331
332 bio->bi_private = parent;
333 bio->bi_end_io = bio_chain_endio;
Jens Axboec4cf5262015-04-17 16:15:18 -0600334 bio_inc_remaining(parent);
Kent Overstreet196d38b2013-11-23 18:34:15 -0800335}
336EXPORT_SYMBOL(bio_chain);
337
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700338static void bio_alloc_rescue(struct work_struct *work)
339{
340 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
341 struct bio *bio;
342
343 while (1) {
344 spin_lock(&bs->rescue_lock);
345 bio = bio_list_pop(&bs->rescue_list);
346 spin_unlock(&bs->rescue_lock);
347
348 if (!bio)
349 break;
350
351 generic_make_request(bio);
352 }
353}
354
355static void punt_bios_to_rescuer(struct bio_set *bs)
356{
357 struct bio_list punt, nopunt;
358 struct bio *bio;
359
360 /*
361 * In order to guarantee forward progress we must punt only bios that
362 * were allocated from this bio_set; otherwise, if there was a bio on
363 * there for a stacking driver higher up in the stack, processing it
364 * could require allocating bios from this bio_set, and doing that from
365 * our own rescuer would be bad.
366 *
367 * Since bio lists are singly linked, pop them all instead of trying to
368 * remove from the middle of the list:
369 */
370
371 bio_list_init(&punt);
372 bio_list_init(&nopunt);
373
374 while ((bio = bio_list_pop(current->bio_list)))
375 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
376
377 *current->bio_list = nopunt;
378
379 spin_lock(&bs->rescue_lock);
380 bio_list_merge(&bs->rescue_list, &punt);
381 spin_unlock(&bs->rescue_lock);
382
383 queue_work(bs->rescue_workqueue, &bs->rescue_work);
384}
385
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700386/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 * bio_alloc_bioset - allocate a bio for I/O
388 * @gfp_mask: the GFP_ mask given to the slab allocator
389 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200390 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 *
392 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700393 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
394 * backed by the @bs's mempool.
395 *
396 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
397 * able to allocate a bio. This is due to the mempool guarantees. To make this
398 * work, callers must never allocate more than 1 bio at a time from this pool.
399 * Callers that need to allocate more than 1 bio must always submit the
400 * previously allocated bio for IO before attempting to allocate a new one.
401 * Failure to do so can cause deadlocks under memory pressure.
402 *
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700403 * Note that when running under generic_make_request() (i.e. any block
404 * driver), bios are not submitted until after you return - see the code in
405 * generic_make_request() that converts recursion into iteration, to prevent
406 * stack overflows.
407 *
408 * This would normally mean allocating multiple bios under
409 * generic_make_request() would be susceptible to deadlocks, but we have
410 * deadlock avoidance code that resubmits any blocked bios from a rescuer
411 * thread.
412 *
413 * However, we do not guarantee forward progress for allocations from other
414 * mempools. Doing multiple allocations from the same mempool under
415 * generic_make_request() should be avoided - instead, use bio_set's front_pad
416 * for per bio allocations.
417 *
Kent Overstreet3f86a822012-09-06 15:35:01 -0700418 * RETURNS:
419 * Pointer to new bio on success, NULL on failure.
420 */
Al Virodd0fc662005-10-07 07:46:04 +0100421struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700423 gfp_t saved_gfp = gfp_mask;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700424 unsigned front_pad;
425 unsigned inline_vecs;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200426 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100427 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200428 struct bio *bio;
429 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200430
Kent Overstreet3f86a822012-09-06 15:35:01 -0700431 if (!bs) {
432 if (nr_iovecs > UIO_MAXIOV)
433 return NULL;
434
435 p = kmalloc(sizeof(struct bio) +
436 nr_iovecs * sizeof(struct bio_vec),
437 gfp_mask);
438 front_pad = 0;
439 inline_vecs = nr_iovecs;
440 } else {
Junichi Nomurad8f429e2014-10-03 17:27:12 -0400441 /* should not use nobvec bioset for nr_iovecs > 0 */
442 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
443 return NULL;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700444 /*
445 * generic_make_request() converts recursion to iteration; this
446 * means if we're running beneath it, any bios we allocate and
447 * submit will not be submitted (and thus freed) until after we
448 * return.
449 *
450 * This exposes us to a potential deadlock if we allocate
451 * multiple bios from the same bio_set() while running
452 * underneath generic_make_request(). If we were to allocate
453 * multiple bios (say a stacking block driver that was splitting
454 * bios), we would deadlock if we exhausted the mempool's
455 * reserve.
456 *
457 * We solve this, and guarantee forward progress, with a rescuer
458 * workqueue per bio_set. If we go to allocate and there are
459 * bios on current->bio_list, we first try the allocation
460 * without __GFP_WAIT; if that fails, we punt those bios we
461 * would be blocking to the rescuer workqueue before we retry
462 * with the original gfp_flags.
463 */
464
465 if (current->bio_list && !bio_list_empty(current->bio_list))
466 gfp_mask &= ~__GFP_WAIT;
467
Kent Overstreet3f86a822012-09-06 15:35:01 -0700468 p = mempool_alloc(bs->bio_pool, gfp_mask);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700469 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
Kent Overstreet3f86a822012-09-06 15:35:01 -0700475 front_pad = bs->front_pad;
476 inline_vecs = BIO_INLINE_VECS;
477 }
478
Tejun Heo451a9eb2009-04-15 19:50:51 +0200479 if (unlikely(!p))
480 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100481
Kent Overstreet3f86a822012-09-06 15:35:01 -0700482 bio = p + front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100483 bio_init(bio);
484
Kent Overstreet3f86a822012-09-06 15:35:01 -0700485 if (nr_iovecs > inline_vecs) {
Kent Overstreet9f060e22012-10-12 15:29:33 -0700486 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700487 if (!bvl && gfp_mask != saved_gfp) {
488 punt_bios_to_rescuer(bs);
489 gfp_mask = saved_gfp;
Kent Overstreet9f060e22012-10-12 15:29:33 -0700490 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700491 }
492
Ingo Molnar34053972009-02-21 11:16:36 +0100493 if (unlikely(!bvl))
494 goto err_free;
Kent Overstreeta38352e2012-05-25 13:03:11 -0700495
496 bio->bi_flags |= 1 << BIO_OWNS_VEC;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700497 } else if (nr_iovecs) {
498 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100499 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700500
501 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100502 bio->bi_flags |= idx << BIO_POOL_OFFSET;
503 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100504 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100506
507err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200508 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100509 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200511EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513void zero_fill_bio(struct bio *bio)
514{
515 unsigned long flags;
Kent Overstreet79886132013-11-23 17:19:00 -0800516 struct bio_vec bv;
517 struct bvec_iter iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Kent Overstreet79886132013-11-23 17:19:00 -0800519 bio_for_each_segment(bv, bio, iter) {
520 char *data = bvec_kmap_irq(&bv, &flags);
521 memset(data, 0, bv.bv_len);
522 flush_dcache_page(bv.bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 bvec_kunmap_irq(data, &flags);
524 }
525}
526EXPORT_SYMBOL(zero_fill_bio);
527
528/**
529 * bio_put - release a reference to a bio
530 * @bio: bio to release reference to
531 *
532 * Description:
533 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100534 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 **/
536void bio_put(struct bio *bio)
537{
Jens Axboedac56212015-04-17 16:23:59 -0600538 if (!bio_flagged(bio, BIO_REFFED))
Kent Overstreet4254bba2012-09-06 15:35:00 -0700539 bio_free(bio);
Jens Axboedac56212015-04-17 16:23:59 -0600540 else {
541 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
542
543 /*
544 * last put frees it
545 */
546 if (atomic_dec_and_test(&bio->__bi_cnt))
547 bio_free(bio);
548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200550EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Jens Axboe165125e2007-07-24 09:28:11 +0200552inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553{
554 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
555 blk_recount_segments(q, bio);
556
557 return bio->bi_phys_segments;
558}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200559EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561/**
Kent Overstreet59d276f2013-11-23 18:19:27 -0800562 * __bio_clone_fast - clone a bio that shares the original bio's biovec
563 * @bio: destination bio
564 * @bio_src: bio to clone
565 *
566 * Clone a &bio. Caller will own the returned bio, but not
567 * the actual data it points to. Reference count of returned
568 * bio will be one.
569 *
570 * Caller must ensure that @bio_src is not freed before @bio.
571 */
572void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
573{
574 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
575
576 /*
577 * most users will be overriding ->bi_bdev with a new target,
578 * so we don't set nor calculate new physical/hw segment counts here
579 */
580 bio->bi_bdev = bio_src->bi_bdev;
581 bio->bi_flags |= 1 << BIO_CLONED;
582 bio->bi_rw = bio_src->bi_rw;
583 bio->bi_iter = bio_src->bi_iter;
584 bio->bi_io_vec = bio_src->bi_io_vec;
585}
586EXPORT_SYMBOL(__bio_clone_fast);
587
588/**
589 * bio_clone_fast - clone a bio that shares the original bio's biovec
590 * @bio: bio to clone
591 * @gfp_mask: allocation priority
592 * @bs: bio_set to allocate from
593 *
594 * Like __bio_clone_fast, only also allocates the returned bio
595 */
596struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
597{
598 struct bio *b;
599
600 b = bio_alloc_bioset(gfp_mask, 0, bs);
601 if (!b)
602 return NULL;
603
604 __bio_clone_fast(b, bio);
605
606 if (bio_integrity(bio)) {
607 int ret;
608
609 ret = bio_integrity_clone(b, bio, gfp_mask);
610
611 if (ret < 0) {
612 bio_put(b);
613 return NULL;
614 }
615 }
616
617 return b;
618}
619EXPORT_SYMBOL(bio_clone_fast);
620
621/**
Kent Overstreetbdb53202013-11-23 17:26:46 -0800622 * bio_clone_bioset - clone a bio
623 * @bio_src: bio to clone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700625 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 *
Kent Overstreetbdb53202013-11-23 17:26:46 -0800627 * Clone bio. Caller will own the returned bio, but not the actual data it
628 * points to. Reference count of returned bio will be one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 */
Kent Overstreetbdb53202013-11-23 17:26:46 -0800630struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700631 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Kent Overstreetbdb53202013-11-23 17:26:46 -0800633 struct bvec_iter iter;
634 struct bio_vec bv;
635 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Kent Overstreetbdb53202013-11-23 17:26:46 -0800637 /*
638 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
639 * bio_src->bi_io_vec to bio->bi_io_vec.
640 *
641 * We can't do that anymore, because:
642 *
643 * - The point of cloning the biovec is to produce a bio with a biovec
644 * the caller can modify: bi_idx and bi_bvec_done should be 0.
645 *
646 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
647 * we tried to clone the whole thing bio_alloc_bioset() would fail.
648 * But the clone should succeed as long as the number of biovecs we
649 * actually need to allocate is fewer than BIO_MAX_PAGES.
650 *
651 * - Lastly, bi_vcnt should not be looked at or relied upon by code
652 * that does not own the bio - reason being drivers don't use it for
653 * iterating over the biovec anymore, so expecting it to be kept up
654 * to date (i.e. for clones that share the parent biovec) is just
655 * asking for trouble and would force extra work on
656 * __bio_clone_fast() anyways.
657 */
658
Kent Overstreet8423ae32014-02-10 17:45:50 -0800659 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
Kent Overstreetbdb53202013-11-23 17:26:46 -0800660 if (!bio)
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200661 return NULL;
662
Kent Overstreetbdb53202013-11-23 17:26:46 -0800663 bio->bi_bdev = bio_src->bi_bdev;
664 bio->bi_rw = bio_src->bi_rw;
665 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
666 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200667
Kent Overstreet8423ae32014-02-10 17:45:50 -0800668 if (bio->bi_rw & REQ_DISCARD)
669 goto integrity_clone;
670
671 if (bio->bi_rw & REQ_WRITE_SAME) {
672 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
673 goto integrity_clone;
674 }
675
Kent Overstreetbdb53202013-11-23 17:26:46 -0800676 bio_for_each_segment(bv, bio_src, iter)
677 bio->bi_io_vec[bio->bi_vcnt++] = bv;
678
Kent Overstreet8423ae32014-02-10 17:45:50 -0800679integrity_clone:
Kent Overstreetbdb53202013-11-23 17:26:46 -0800680 if (bio_integrity(bio_src)) {
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200681 int ret;
682
Kent Overstreetbdb53202013-11-23 17:26:46 -0800683 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
Li Zefan059ea332009-03-09 10:42:45 +0100684 if (ret < 0) {
Kent Overstreetbdb53202013-11-23 17:26:46 -0800685 bio_put(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200686 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100687 }
Peter Osterlund36763472005-09-06 15:16:42 -0700688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Kent Overstreetbdb53202013-11-23 17:26:46 -0800690 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700692EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
694/**
695 * bio_get_nr_vecs - return approx number of vecs
696 * @bdev: I/O target
697 *
698 * Return the approximate number of pages we can send to this target.
699 * There's no guarantee that you will be able to fit this number of pages
700 * into a bio, it does not account for dynamic restrictions that vary
701 * on offset.
702 */
703int bio_get_nr_vecs(struct block_device *bdev)
704{
Jens Axboe165125e2007-07-24 09:28:11 +0200705 struct request_queue *q = bdev_get_queue(bdev);
Bernd Schubertf908ee92012-05-11 16:36:44 +0200706 int nr_pages;
707
708 nr_pages = min_t(unsigned,
Kent Overstreet5abebfd2012-02-08 22:07:18 +0100709 queue_max_segments(q),
710 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
Bernd Schubertf908ee92012-05-11 16:36:44 +0200711
712 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
713
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200715EXPORT_SYMBOL(bio_get_nr_vecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Jens Axboe165125e2007-07-24 09:28:11 +0200717static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600718 *page, unsigned int len, unsigned int offset,
Akinobu Mita34f2fd82013-11-18 22:11:42 +0900719 unsigned int max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720{
721 int retried_segments = 0;
722 struct bio_vec *bvec;
723
724 /*
725 * cloned bio must not modify vec list
726 */
727 if (unlikely(bio_flagged(bio, BIO_CLONED)))
728 return 0;
729
Kent Overstreet4f024f32013-10-11 15:44:27 -0700730 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 return 0;
732
Jens Axboe80cfd542006-01-06 09:43:28 +0100733 /*
734 * For filesystems with a blocksize smaller than the pagesize
735 * we will often be called with the same page as last time and
736 * a consecutive offset. Optimize this special case.
737 */
738 if (bio->bi_vcnt > 0) {
739 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
740
741 if (page == prev->bv_page &&
742 offset == prev->bv_offset + prev->bv_len) {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300743 unsigned int prev_bv_len = prev->bv_len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100744 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200745
746 if (q->merge_bvec_fn) {
747 struct bvec_merge_data bvm = {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300748 /* prev_bvec is already charged in
749 bi_size, discharge it in order to
750 simulate merging updated prev_bvec
751 as new bvec. */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200752 .bi_bdev = bio->bi_bdev,
Kent Overstreet4f024f32013-10-11 15:44:27 -0700753 .bi_sector = bio->bi_iter.bi_sector,
754 .bi_size = bio->bi_iter.bi_size -
755 prev_bv_len,
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200756 .bi_rw = bio->bi_rw,
757 };
758
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300759 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200760 prev->bv_len -= len;
761 return 0;
762 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100763 }
764
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800765 bio->bi_iter.bi_size += len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100766 goto done;
767 }
Jens Axboe66cb45a2014-06-24 16:22:24 -0600768
769 /*
770 * If the queue doesn't support SG gaps and adding this
771 * offset would create a gap, disallow it.
772 */
773 if (q->queue_flags & (1 << QUEUE_FLAG_SG_GAPS) &&
774 bvec_gap_to_prev(prev, offset))
775 return 0;
Jens Axboe80cfd542006-01-06 09:43:28 +0100776 }
777
778 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return 0;
780
781 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * setup the new entry, we might clear it again later if we
783 * cannot add the page
784 */
785 bvec = &bio->bi_io_vec[bio->bi_vcnt];
786 bvec->bv_page = page;
787 bvec->bv_len = len;
788 bvec->bv_offset = offset;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800789 bio->bi_vcnt++;
790 bio->bi_phys_segments++;
791 bio->bi_iter.bi_size += len;
792
793 /*
794 * Perform a recount if the number of segments is greater
795 * than queue_max_segments(q).
796 */
797
798 while (bio->bi_phys_segments > queue_max_segments(q)) {
799
800 if (retried_segments)
801 goto failed;
802
803 retried_segments = 1;
804 blk_recount_segments(q, bio);
805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
807 /*
808 * if queue has other restrictions (eg varying max sector size
809 * depending on offset), it can specify a merge_bvec_fn in the
810 * queue to get further control
811 */
812 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200813 struct bvec_merge_data bvm = {
814 .bi_bdev = bio->bi_bdev,
Kent Overstreet4f024f32013-10-11 15:44:27 -0700815 .bi_sector = bio->bi_iter.bi_sector,
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800816 .bi_size = bio->bi_iter.bi_size - len,
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200817 .bi_rw = bio->bi_rw,
818 };
819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 /*
821 * merge_bvec_fn() returns number of bytes it can accept
822 * at this offset
823 */
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800824 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len)
825 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 }
827
828 /* If we may be able to merge these biovecs, force a recount */
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800829 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
831
Jens Axboe80cfd542006-01-06 09:43:28 +0100832 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800834
835 failed:
836 bvec->bv_page = NULL;
837 bvec->bv_len = 0;
838 bvec->bv_offset = 0;
839 bio->bi_vcnt--;
840 bio->bi_iter.bi_size -= len;
841 blk_recount_segments(q, bio);
842 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
845/**
Mike Christie6e68af62005-11-11 05:30:27 -0600846 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100847 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600848 * @bio: destination bio
849 * @page: page to add
850 * @len: vec entry length
851 * @offset: vec entry offset
852 *
853 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200854 * number of reasons, such as the bio being full or target block device
855 * limitations. The target block device must allow bio's up to PAGE_SIZE,
856 * so it is always possible to add a single page to an empty bio.
857 *
858 * This should only be used by REQ_PC bios.
Mike Christie6e68af62005-11-11 05:30:27 -0600859 */
Jens Axboe165125e2007-07-24 09:28:11 +0200860int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600861 unsigned int len, unsigned int offset)
862{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400863 return __bio_add_page(q, bio, page, len, offset,
864 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600865}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200866EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600867
868/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 * bio_add_page - attempt to add page to bio
870 * @bio: destination bio
871 * @page: page to add
872 * @len: vec entry length
873 * @offset: vec entry offset
874 *
875 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200876 * number of reasons, such as the bio being full or target block device
877 * limitations. The target block device must allow bio's up to PAGE_SIZE,
878 * so it is always possible to add a single page to an empty bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 */
880int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
881 unsigned int offset)
882{
Mike Christiedefd94b2005-12-05 02:37:06 -0600883 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Jens Axboe58a49152014-06-10 12:53:56 -0600884 unsigned int max_sectors;
Jens Axboe762380a2014-06-05 13:38:39 -0600885
Jens Axboe58a49152014-06-10 12:53:56 -0600886 max_sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
887 if ((max_sectors < (len >> 9)) && !bio->bi_iter.bi_size)
888 max_sectors = len >> 9;
889
890 return __bio_add_page(q, bio, page, len, offset, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200892EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893
Kent Overstreet9e882242012-09-10 14:41:12 -0700894struct submit_bio_ret {
895 struct completion event;
896 int error;
897};
898
899static void submit_bio_wait_endio(struct bio *bio, int error)
900{
901 struct submit_bio_ret *ret = bio->bi_private;
902
903 ret->error = error;
904 complete(&ret->event);
905}
906
907/**
908 * submit_bio_wait - submit a bio, and wait until it completes
909 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
910 * @bio: The &struct bio which describes the I/O
911 *
912 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
913 * bio_endio() on failure.
914 */
915int submit_bio_wait(int rw, struct bio *bio)
916{
917 struct submit_bio_ret ret;
918
919 rw |= REQ_SYNC;
920 init_completion(&ret.event);
921 bio->bi_private = &ret;
922 bio->bi_end_io = submit_bio_wait_endio;
923 submit_bio(rw, bio);
924 wait_for_completion(&ret.event);
925
926 return ret.error;
927}
928EXPORT_SYMBOL(submit_bio_wait);
929
Kent Overstreet054bdf62012-09-28 13:17:55 -0700930/**
931 * bio_advance - increment/complete a bio by some number of bytes
932 * @bio: bio to advance
933 * @bytes: number of bytes to complete
934 *
935 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
936 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
937 * be updated on the last bvec as well.
938 *
939 * @bio will then represent the remaining, uncompleted portion of the io.
940 */
941void bio_advance(struct bio *bio, unsigned bytes)
942{
943 if (bio_integrity(bio))
944 bio_integrity_advance(bio, bytes);
945
Kent Overstreet4550dd62013-08-07 14:26:21 -0700946 bio_advance_iter(bio, &bio->bi_iter, bytes);
Kent Overstreet054bdf62012-09-28 13:17:55 -0700947}
948EXPORT_SYMBOL(bio_advance);
949
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700950/**
Kent Overstreeta0787602012-09-10 14:03:28 -0700951 * bio_alloc_pages - allocates a single page for each bvec in a bio
952 * @bio: bio to allocate pages for
953 * @gfp_mask: flags for allocation
954 *
955 * Allocates pages up to @bio->bi_vcnt.
956 *
957 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
958 * freed.
959 */
960int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
961{
962 int i;
963 struct bio_vec *bv;
964
965 bio_for_each_segment_all(bv, bio, i) {
966 bv->bv_page = alloc_page(gfp_mask);
967 if (!bv->bv_page) {
968 while (--bv >= bio->bi_io_vec)
969 __free_page(bv->bv_page);
970 return -ENOMEM;
971 }
972 }
973
974 return 0;
975}
976EXPORT_SYMBOL(bio_alloc_pages);
977
978/**
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700979 * bio_copy_data - copy contents of data buffers from one chain of bios to
980 * another
981 * @src: source bio list
982 * @dst: destination bio list
983 *
984 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
985 * @src and @dst as linked lists of bios.
986 *
987 * Stops when it reaches the end of either @src or @dst - that is, copies
988 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
989 */
990void bio_copy_data(struct bio *dst, struct bio *src)
991{
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700992 struct bvec_iter src_iter, dst_iter;
993 struct bio_vec src_bv, dst_bv;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700994 void *src_p, *dst_p;
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700995 unsigned bytes;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700996
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700997 src_iter = src->bi_iter;
998 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700999
1000 while (1) {
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001001 if (!src_iter.bi_size) {
1002 src = src->bi_next;
1003 if (!src)
1004 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001005
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001006 src_iter = src->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001007 }
1008
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001009 if (!dst_iter.bi_size) {
1010 dst = dst->bi_next;
1011 if (!dst)
1012 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001013
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001014 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001015 }
1016
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001017 src_bv = bio_iter_iovec(src, src_iter);
1018 dst_bv = bio_iter_iovec(dst, dst_iter);
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001019
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001020 bytes = min(src_bv.bv_len, dst_bv.bv_len);
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001021
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001022 src_p = kmap_atomic(src_bv.bv_page);
1023 dst_p = kmap_atomic(dst_bv.bv_page);
1024
1025 memcpy(dst_p + dst_bv.bv_offset,
1026 src_p + src_bv.bv_offset,
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001027 bytes);
1028
1029 kunmap_atomic(dst_p);
1030 kunmap_atomic(src_p);
1031
Kent Overstreet1cb9dda2013-08-07 14:26:39 -07001032 bio_advance_iter(src, &src_iter, bytes);
1033 bio_advance_iter(dst, &dst_iter, bytes);
Kent Overstreet16ac3d62012-09-10 13:57:51 -07001034 }
1035}
1036EXPORT_SYMBOL(bio_copy_data);
1037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038struct bio_map_data {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001039 int is_our_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001040 struct iov_iter iter;
1041 struct iovec iov[];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042};
1043
Fabian Frederick7410b3c2014-04-22 15:09:07 -06001044static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001045 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
Jens Axboef3f63c12010-10-29 11:46:56 -06001047 if (iov_count > UIO_MAXIOV)
1048 return NULL;
1049
Kent Overstreetc8db4442013-11-22 19:39:06 -08001050 return kmalloc(sizeof(struct bio_map_data) +
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001051 sizeof(struct iovec) * iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052}
1053
Dongsu Park9124d3f2015-01-18 16:16:34 +01001054/**
1055 * bio_copy_from_iter - copy all pages from iov_iter to bio
1056 * @bio: The &struct bio which describes the I/O as destination
1057 * @iter: iov_iter as source
1058 *
1059 * Copy all pages from iov_iter to bio.
1060 * Returns 0 on success, or error on failure.
1061 */
1062static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001063{
Dongsu Park9124d3f2015-01-18 16:16:34 +01001064 int i;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001065 struct bio_vec *bvec;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001066
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001067 bio_for_each_segment_all(bvec, bio, i) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001068 ssize_t ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001069
Dongsu Park9124d3f2015-01-18 16:16:34 +01001070 ret = copy_page_from_iter(bvec->bv_page,
1071 bvec->bv_offset,
1072 bvec->bv_len,
1073 &iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001074
Dongsu Park9124d3f2015-01-18 16:16:34 +01001075 if (!iov_iter_count(&iter))
1076 break;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001077
Dongsu Park9124d3f2015-01-18 16:16:34 +01001078 if (ret < bvec->bv_len)
1079 return -EFAULT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001080 }
1081
Dongsu Park9124d3f2015-01-18 16:16:34 +01001082 return 0;
1083}
1084
1085/**
1086 * bio_copy_to_iter - copy all pages from bio to iov_iter
1087 * @bio: The &struct bio which describes the I/O as source
1088 * @iter: iov_iter as destination
1089 *
1090 * Copy all pages from bio to iov_iter.
1091 * Returns 0 on success, or error on failure.
1092 */
1093static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1094{
1095 int i;
1096 struct bio_vec *bvec;
1097
1098 bio_for_each_segment_all(bvec, bio, i) {
1099 ssize_t ret;
1100
1101 ret = copy_page_to_iter(bvec->bv_page,
1102 bvec->bv_offset,
1103 bvec->bv_len,
1104 &iter);
1105
1106 if (!iov_iter_count(&iter))
1107 break;
1108
1109 if (ret < bvec->bv_len)
1110 return -EFAULT;
1111 }
1112
1113 return 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001114}
1115
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001116static void bio_free_pages(struct bio *bio)
1117{
1118 struct bio_vec *bvec;
1119 int i;
1120
1121 bio_for_each_segment_all(bvec, bio, i)
1122 __free_page(bvec->bv_page);
1123}
1124
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125/**
1126 * bio_uncopy_user - finish previously mapped bio
1127 * @bio: bio being terminated
1128 *
Christoph Hellwigddad8dd2015-01-18 16:16:29 +01001129 * Free pages allocated from bio_copy_user_iov() and write back data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 * to user space in case of a read.
1131 */
1132int bio_uncopy_user(struct bio *bio)
1133{
1134 struct bio_map_data *bmd = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001135 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
Roland Dreier35dc2482013-08-05 17:55:01 -07001137 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1138 /*
1139 * if we're in a workqueue, the request is orphaned, so
1140 * don't copy into a random user address space, just free.
1141 */
Dongsu Park9124d3f2015-01-18 16:16:34 +01001142 if (current->mm && bio_data_dir(bio) == READ)
1143 ret = bio_copy_to_iter(bio, bmd->iter);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001144 if (bmd->is_our_pages)
1145 bio_free_pages(bio);
Roland Dreier35dc2482013-08-05 17:55:01 -07001146 }
Kent Overstreetc8db4442013-11-22 19:39:06 -08001147 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 bio_put(bio);
1149 return ret;
1150}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001151EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
1153/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001154 * bio_copy_user_iov - copy user data to bio
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001155 * @q: destination block queue
1156 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1157 * @iter: iovec iterator
1158 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 *
1160 * Prepares and returns a bio for indirect user io, bouncing data
1161 * to/from kernel pages as necessary. Must be paired with
1162 * call bio_uncopy_user() on io completion.
1163 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001164struct bio *bio_copy_user_iov(struct request_queue *q,
1165 struct rq_map_data *map_data,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001166 const struct iov_iter *iter,
1167 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 struct page *page;
1171 struct bio *bio;
1172 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001173 int nr_pages = 0;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001174 unsigned int len = iter->count;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001175 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001177 for (i = 0; i < iter->nr_segs; i++) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001178 unsigned long uaddr;
1179 unsigned long end;
1180 unsigned long start;
1181
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001182 uaddr = (unsigned long) iter->iov[i].iov_base;
1183 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1184 >> PAGE_SHIFT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001185 start = uaddr >> PAGE_SHIFT;
1186
Jens Axboecb4644c2010-11-10 14:36:25 +01001187 /*
1188 * Overflow, abort
1189 */
1190 if (end < start)
1191 return ERR_PTR(-EINVAL);
1192
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001193 nr_pages += end - start;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001194 }
1195
FUJITA Tomonori69838722009-04-28 20:24:29 +02001196 if (offset)
1197 nr_pages++;
1198
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001199 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 if (!bmd)
1201 return ERR_PTR(-ENOMEM);
1202
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001203 /*
1204 * We need to do a deep copy of the iov_iter including the iovecs.
1205 * The caller provided iov might point to an on-stack or otherwise
1206 * shortlived one.
1207 */
1208 bmd->is_our_pages = map_data ? 0 : 1;
1209 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1210 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1211 iter->nr_segs, iter->count);
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001214 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (!bio)
1216 goto out_bmd;
1217
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001218 if (iter->type & WRITE)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001219 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
1221 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001222
1223 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001224 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001225 i = map_data->offset / PAGE_SIZE;
1226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001228 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001230 bytes -= offset;
1231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (bytes > len)
1233 bytes = len;
1234
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001235 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001236 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001237 ret = -ENOMEM;
1238 break;
1239 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001240
1241 page = map_data->pages[i / nr_pages];
1242 page += (i % nr_pages);
1243
1244 i++;
1245 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001246 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001247 if (!page) {
1248 ret = -ENOMEM;
1249 break;
1250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 }
1252
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001253 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
1256 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001257 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 }
1259
1260 if (ret)
1261 goto cleanup;
1262
1263 /*
1264 * success
1265 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001266 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +02001267 (map_data && map_data->from_user)) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001268 ret = bio_copy_from_iter(bio, *iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001269 if (ret)
1270 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 }
1272
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001273 bio->bi_private = bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 return bio;
1275cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001276 if (!map_data)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001277 bio_free_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 bio_put(bio);
1279out_bmd:
Kent Overstreetc8db4442013-11-22 19:39:06 -08001280 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return ERR_PTR(ret);
1282}
1283
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001284/**
1285 * bio_map_user_iov - map user iovec into bio
1286 * @q: the struct request_queue for the bio
1287 * @iter: iovec iterator
1288 * @gfp_mask: memory allocation flags
1289 *
1290 * Map the user space address into a bio suitable for io to a block
1291 * device. Returns an error pointer in case of error.
1292 */
1293struct bio *bio_map_user_iov(struct request_queue *q,
1294 const struct iov_iter *iter,
1295 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296{
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001297 int j;
James Bottomley f1970ba2005-06-20 14:06:52 +02001298 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 struct page **pages;
1300 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001301 int cur_page = 0;
1302 int ret, offset;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001303 struct iov_iter i;
1304 struct iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001306 iov_for_each(iov, i, *iter) {
1307 unsigned long uaddr = (unsigned long) iov.iov_base;
1308 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001309 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1310 unsigned long start = uaddr >> PAGE_SHIFT;
1311
Jens Axboecb4644c2010-11-10 14:36:25 +01001312 /*
1313 * Overflow, abort
1314 */
1315 if (end < start)
1316 return ERR_PTR(-EINVAL);
1317
James Bottomley f1970ba2005-06-20 14:06:52 +02001318 nr_pages += end - start;
1319 /*
Mike Christiead2d7222006-12-01 10:40:20 +01001320 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +02001321 */
Mike Christiead2d7222006-12-01 10:40:20 +01001322 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +02001323 return ERR_PTR(-EINVAL);
1324 }
1325
1326 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 return ERR_PTR(-EINVAL);
1328
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001329 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 if (!bio)
1331 return ERR_PTR(-ENOMEM);
1332
1333 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001334 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 if (!pages)
1336 goto out;
1337
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001338 iov_for_each(iov, i, *iter) {
1339 unsigned long uaddr = (unsigned long) iov.iov_base;
1340 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001341 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1342 unsigned long start = uaddr >> PAGE_SHIFT;
1343 const int local_nr_pages = end - start;
1344 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001345
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001346 ret = get_user_pages_fast(uaddr, local_nr_pages,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001347 (iter->type & WRITE) != WRITE,
1348 &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +02001349 if (ret < local_nr_pages) {
1350 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001351 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
James Bottomley f1970ba2005-06-20 14:06:52 +02001354 offset = uaddr & ~PAGE_MASK;
1355 for (j = cur_page; j < page_limit; j++) {
1356 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
James Bottomley f1970ba2005-06-20 14:06:52 +02001358 if (len <= 0)
1359 break;
1360
1361 if (bytes > len)
1362 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
James Bottomley f1970ba2005-06-20 14:06:52 +02001364 /*
1365 * sorry...
1366 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001367 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1368 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001369 break;
1370
1371 len -= bytes;
1372 offset = 0;
1373 }
1374
1375 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001377 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001379 while (j < page_limit)
1380 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 kfree(pages);
1384
1385 /*
1386 * set data direction, and check if mapped pages need bouncing
1387 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001388 if (iter->type & WRITE)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001389 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391 bio->bi_flags |= (1 << BIO_USER_MAPPED);
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001392
1393 /*
1394 * subtle -- if __bio_map_user() ended up bouncing a bio,
1395 * it would normally disappear when its bi_end_io is run.
1396 * however, we need it for the unmap, so grab an extra
1397 * reference to it
1398 */
1399 bio_get(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001401
1402 out_unmap:
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001403 for (j = 0; j < nr_pages; j++) {
1404 if (!pages[j])
James Bottomley f1970ba2005-06-20 14:06:52 +02001405 break;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001406 page_cache_release(pages[j]);
James Bottomley f1970ba2005-06-20 14:06:52 +02001407 }
1408 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 kfree(pages);
1410 bio_put(bio);
1411 return ERR_PTR(ret);
1412}
1413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414static void __bio_unmap_user(struct bio *bio)
1415{
1416 struct bio_vec *bvec;
1417 int i;
1418
1419 /*
1420 * make sure we dirty pages we wrote to
1421 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001422 bio_for_each_segment_all(bvec, bio, i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (bio_data_dir(bio) == READ)
1424 set_page_dirty_lock(bvec->bv_page);
1425
1426 page_cache_release(bvec->bv_page);
1427 }
1428
1429 bio_put(bio);
1430}
1431
1432/**
1433 * bio_unmap_user - unmap a bio
1434 * @bio: the bio being unmapped
1435 *
1436 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1437 * a process context.
1438 *
1439 * bio_unmap_user() may sleep.
1440 */
1441void bio_unmap_user(struct bio *bio)
1442{
1443 __bio_unmap_user(bio);
1444 bio_put(bio);
1445}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001446EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
NeilBrown6712ecf2007-09-27 12:47:43 +02001448static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001449{
Jens Axboeb8238252005-06-20 14:05:27 +02001450 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001451}
1452
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001453/**
1454 * bio_map_kern - map kernel address into bio
1455 * @q: the struct request_queue for the bio
1456 * @data: pointer to buffer to map
1457 * @len: length in bytes
1458 * @gfp_mask: allocation flags for bio allocation
1459 *
1460 * Map the kernel address into a bio suitable for io to a block
1461 * device. Returns an error pointer in case of error.
1462 */
1463struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1464 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001465{
1466 unsigned long kaddr = (unsigned long)data;
1467 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1468 unsigned long start = kaddr >> PAGE_SHIFT;
1469 const int nr_pages = end - start;
1470 int offset, i;
1471 struct bio *bio;
1472
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001473 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001474 if (!bio)
1475 return ERR_PTR(-ENOMEM);
1476
1477 offset = offset_in_page(kaddr);
1478 for (i = 0; i < nr_pages; i++) {
1479 unsigned int bytes = PAGE_SIZE - offset;
1480
1481 if (len <= 0)
1482 break;
1483
1484 if (bytes > len)
1485 bytes = len;
1486
Mike Christiedefd94b2005-12-05 02:37:06 -06001487 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001488 offset) < bytes) {
1489 /* we don't support partial mappings */
1490 bio_put(bio);
1491 return ERR_PTR(-EINVAL);
1492 }
Mike Christie df46b9a2005-06-20 14:04:44 +02001493
1494 data += bytes;
1495 len -= bytes;
1496 offset = 0;
1497 }
1498
Jens Axboeb8238252005-06-20 14:05:27 +02001499 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001500 return bio;
1501}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001502EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001503
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001504static void bio_copy_kern_endio(struct bio *bio, int err)
1505{
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001506 bio_free_pages(bio);
1507 bio_put(bio);
1508}
1509
1510static void bio_copy_kern_endio_read(struct bio *bio, int err)
1511{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001512 char *p = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001513 struct bio_vec *bvec;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001514 int i;
1515
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001516 bio_for_each_segment_all(bvec, bio, i) {
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001517 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
Kent Overstreetc8db4442013-11-22 19:39:06 -08001518 p += bvec->bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001519 }
1520
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001521 bio_copy_kern_endio(bio, err);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001522}
1523
1524/**
1525 * bio_copy_kern - copy kernel address into bio
1526 * @q: the struct request_queue for the bio
1527 * @data: pointer to buffer to copy
1528 * @len: length in bytes
1529 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001530 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001531 *
1532 * copy the kernel address into a bio suitable for io to a block
1533 * device. Returns an error pointer in case of error.
1534 */
1535struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1536 gfp_t gfp_mask, int reading)
1537{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001538 unsigned long kaddr = (unsigned long)data;
1539 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1540 unsigned long start = kaddr >> PAGE_SHIFT;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001541 struct bio *bio;
1542 void *p = data;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001543 int nr_pages = 0;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001544
Christoph Hellwig42d26832015-01-18 16:16:28 +01001545 /*
1546 * Overflow, abort
1547 */
1548 if (end < start)
1549 return ERR_PTR(-EINVAL);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001550
Christoph Hellwig42d26832015-01-18 16:16:28 +01001551 nr_pages = end - start;
1552 bio = bio_kmalloc(gfp_mask, nr_pages);
1553 if (!bio)
1554 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001555
Christoph Hellwig42d26832015-01-18 16:16:28 +01001556 while (len) {
1557 struct page *page;
1558 unsigned int bytes = PAGE_SIZE;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001559
Christoph Hellwig42d26832015-01-18 16:16:28 +01001560 if (bytes > len)
1561 bytes = len;
1562
1563 page = alloc_page(q->bounce_gfp | gfp_mask);
1564 if (!page)
1565 goto cleanup;
1566
1567 if (!reading)
1568 memcpy(page_address(page), p, bytes);
1569
1570 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1571 break;
1572
1573 len -= bytes;
1574 p += bytes;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001575 }
1576
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001577 if (reading) {
1578 bio->bi_end_io = bio_copy_kern_endio_read;
1579 bio->bi_private = data;
1580 } else {
1581 bio->bi_end_io = bio_copy_kern_endio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001582 bio->bi_rw |= REQ_WRITE;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001583 }
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001584
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001585 return bio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001586
1587cleanup:
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001588 bio_free_pages(bio);
Christoph Hellwig42d26832015-01-18 16:16:28 +01001589 bio_put(bio);
1590 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001591}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001592EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594/*
1595 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1596 * for performing direct-IO in BIOs.
1597 *
1598 * The problem is that we cannot run set_page_dirty() from interrupt context
1599 * because the required locks are not interrupt-safe. So what we can do is to
1600 * mark the pages dirty _before_ performing IO. And in interrupt context,
1601 * check that the pages are still dirty. If so, fine. If not, redirty them
1602 * in process context.
1603 *
1604 * We special-case compound pages here: normally this means reads into hugetlb
1605 * pages. The logic in here doesn't really work right for compound pages
1606 * because the VM does not uniformly chase down the head page in all cases.
1607 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1608 * handle them at all. So we skip compound pages here at an early stage.
1609 *
1610 * Note that this code is very hard to test under normal circumstances because
1611 * direct-io pins the pages with get_user_pages(). This makes
1612 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001613 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 * pagecache.
1615 *
1616 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1617 * deferred bio dirtying paths.
1618 */
1619
1620/*
1621 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1622 */
1623void bio_set_pages_dirty(struct bio *bio)
1624{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001625 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 int i;
1627
Kent Overstreetcb34e052012-09-05 15:22:02 -07001628 bio_for_each_segment_all(bvec, bio, i) {
1629 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 if (page && !PageCompound(page))
1632 set_page_dirty_lock(page);
1633 }
1634}
1635
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001636static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001638 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 int i;
1640
Kent Overstreetcb34e052012-09-05 15:22:02 -07001641 bio_for_each_segment_all(bvec, bio, i) {
1642 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644 if (page)
1645 put_page(page);
1646 }
1647}
1648
1649/*
1650 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1651 * If they are, then fine. If, however, some pages are clean then they must
1652 * have been written out during the direct-IO read. So we take another ref on
1653 * the BIO and the offending pages and re-dirty the pages in process context.
1654 *
1655 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1656 * here on. It will run one page_cache_release() against each page and will
1657 * run one bio_put() against the BIO.
1658 */
1659
David Howells65f27f32006-11-22 14:55:48 +00001660static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
David Howells65f27f32006-11-22 14:55:48 +00001662static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663static DEFINE_SPINLOCK(bio_dirty_lock);
1664static struct bio *bio_dirty_list;
1665
1666/*
1667 * This runs in process context
1668 */
David Howells65f27f32006-11-22 14:55:48 +00001669static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
1671 unsigned long flags;
1672 struct bio *bio;
1673
1674 spin_lock_irqsave(&bio_dirty_lock, flags);
1675 bio = bio_dirty_list;
1676 bio_dirty_list = NULL;
1677 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1678
1679 while (bio) {
1680 struct bio *next = bio->bi_private;
1681
1682 bio_set_pages_dirty(bio);
1683 bio_release_pages(bio);
1684 bio_put(bio);
1685 bio = next;
1686 }
1687}
1688
1689void bio_check_pages_dirty(struct bio *bio)
1690{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001691 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 int nr_clean_pages = 0;
1693 int i;
1694
Kent Overstreetcb34e052012-09-05 15:22:02 -07001695 bio_for_each_segment_all(bvec, bio, i) {
1696 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
1698 if (PageDirty(page) || PageCompound(page)) {
1699 page_cache_release(page);
Kent Overstreetcb34e052012-09-05 15:22:02 -07001700 bvec->bv_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 } else {
1702 nr_clean_pages++;
1703 }
1704 }
1705
1706 if (nr_clean_pages) {
1707 unsigned long flags;
1708
1709 spin_lock_irqsave(&bio_dirty_lock, flags);
1710 bio->bi_private = bio_dirty_list;
1711 bio_dirty_list = bio;
1712 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1713 schedule_work(&bio_dirty_work);
1714 } else {
1715 bio_put(bio);
1716 }
1717}
1718
Gu Zheng394ffa52014-11-24 11:05:22 +08001719void generic_start_io_acct(int rw, unsigned long sectors,
1720 struct hd_struct *part)
1721{
1722 int cpu = part_stat_lock();
1723
1724 part_round_stats(cpu, part);
1725 part_stat_inc(cpu, part, ios[rw]);
1726 part_stat_add(cpu, part, sectors[rw], sectors);
1727 part_inc_in_flight(part, rw);
1728
1729 part_stat_unlock();
1730}
1731EXPORT_SYMBOL(generic_start_io_acct);
1732
1733void generic_end_io_acct(int rw, struct hd_struct *part,
1734 unsigned long start_time)
1735{
1736 unsigned long duration = jiffies - start_time;
1737 int cpu = part_stat_lock();
1738
1739 part_stat_add(cpu, part, ticks[rw], duration);
1740 part_round_stats(cpu, part);
1741 part_dec_in_flight(part, rw);
1742
1743 part_stat_unlock();
1744}
1745EXPORT_SYMBOL(generic_end_io_acct);
1746
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001747#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1748void bio_flush_dcache_pages(struct bio *bi)
1749{
Kent Overstreet79886132013-11-23 17:19:00 -08001750 struct bio_vec bvec;
1751 struct bvec_iter iter;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001752
Kent Overstreet79886132013-11-23 17:19:00 -08001753 bio_for_each_segment(bvec, bi, iter)
1754 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001755}
1756EXPORT_SYMBOL(bio_flush_dcache_pages);
1757#endif
1758
Jens Axboec4cf5262015-04-17 16:15:18 -06001759static inline bool bio_remaining_done(struct bio *bio)
1760{
1761 /*
1762 * If we're not chaining, then ->__bi_remaining is always 1 and
1763 * we always end io on the first invocation.
1764 */
1765 if (!bio_flagged(bio, BIO_CHAIN))
1766 return true;
1767
1768 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1769
Mike Snitzer326e1db2015-05-22 09:14:03 -04001770 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1771 clear_bit(BIO_CHAIN, &bio->bi_flags);
Jens Axboec4cf5262015-04-17 16:15:18 -06001772 return true;
Mike Snitzer326e1db2015-05-22 09:14:03 -04001773 }
Jens Axboec4cf5262015-04-17 16:15:18 -06001774
1775 return false;
1776}
1777
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778/**
1779 * bio_endio - end I/O on a bio
1780 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 * @error: error, if any
1782 *
1783 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001784 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001785 * preferred way to end I/O on a bio, it takes care of clearing
1786 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1787 * established -Exxxx (-EIO, for instance) error values in case
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001788 * something went wrong. No one should call bi_end_io() directly on a
NeilBrown5bb23a62007-09-27 12:46:13 +02001789 * bio unless they own it and thus know that it has an end_io
1790 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001792void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793{
Kent Overstreet196d38b2013-11-23 18:34:15 -08001794 while (bio) {
Kent Overstreet196d38b2013-11-23 18:34:15 -08001795 if (error)
1796 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1797 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1798 error = -EIO;
1799
Jens Axboec4cf5262015-04-17 16:15:18 -06001800 if (unlikely(!bio_remaining_done(bio)))
1801 break;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001802
1803 /*
1804 * Need to have a real endio function for chained bios,
1805 * otherwise various corner cases will break (like stacking
1806 * block devices that save/restore bi_end_io) - however, we want
1807 * to avoid unbounded recursion and blowing the stack. Tail call
1808 * optimization would handle this, but compiling with frame
1809 * pointers also disables gcc's sibling call optimization.
1810 */
1811 if (bio->bi_end_io == bio_chain_endio) {
1812 struct bio *parent = bio->bi_private;
1813 bio_put(bio);
1814 bio = parent;
1815 } else {
1816 if (bio->bi_end_io)
1817 bio->bi_end_io(bio, error);
1818 bio = NULL;
1819 }
1820 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001822EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Kent Overstreet196d38b2013-11-23 18:34:15 -08001824/**
Kent Overstreet20d01892013-11-23 18:21:01 -08001825 * bio_split - split a bio
1826 * @bio: bio to split
1827 * @sectors: number of sectors to split from the front of @bio
1828 * @gfp: gfp mask
1829 * @bs: bio set to allocate from
1830 *
1831 * Allocates and returns a new bio which represents @sectors from the start of
1832 * @bio, and updates @bio to represent the remaining sectors.
1833 *
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001834 * Unless this is a discard request the newly allocated bio will point
1835 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1836 * @bio is not freed before the split.
Kent Overstreet20d01892013-11-23 18:21:01 -08001837 */
1838struct bio *bio_split(struct bio *bio, int sectors,
1839 gfp_t gfp, struct bio_set *bs)
1840{
1841 struct bio *split = NULL;
1842
1843 BUG_ON(sectors <= 0);
1844 BUG_ON(sectors >= bio_sectors(bio));
1845
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001846 /*
1847 * Discards need a mutable bio_vec to accommodate the payload
1848 * required by the DSM TRIM and UNMAP commands.
1849 */
1850 if (bio->bi_rw & REQ_DISCARD)
1851 split = bio_clone_bioset(bio, gfp, bs);
1852 else
1853 split = bio_clone_fast(bio, gfp, bs);
1854
Kent Overstreet20d01892013-11-23 18:21:01 -08001855 if (!split)
1856 return NULL;
1857
1858 split->bi_iter.bi_size = sectors << 9;
1859
1860 if (bio_integrity(split))
1861 bio_integrity_trim(split, 0, sectors);
1862
1863 bio_advance(bio, split->bi_iter.bi_size);
1864
1865 return split;
1866}
1867EXPORT_SYMBOL(bio_split);
1868
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001869/**
Kent Overstreet6678d832013-08-07 11:14:32 -07001870 * bio_trim - trim a bio
1871 * @bio: bio to trim
1872 * @offset: number of sectors to trim from the front of @bio
1873 * @size: size we want to trim @bio to, in sectors
1874 */
1875void bio_trim(struct bio *bio, int offset, int size)
1876{
1877 /* 'bio' is a cloned bio which we need to trim to match
1878 * the given offset and size.
Kent Overstreet6678d832013-08-07 11:14:32 -07001879 */
Kent Overstreet6678d832013-08-07 11:14:32 -07001880
1881 size <<= 9;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001882 if (offset == 0 && size == bio->bi_iter.bi_size)
Kent Overstreet6678d832013-08-07 11:14:32 -07001883 return;
1884
1885 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1886
1887 bio_advance(bio, offset << 9);
1888
Kent Overstreet4f024f32013-10-11 15:44:27 -07001889 bio->bi_iter.bi_size = size;
Kent Overstreet6678d832013-08-07 11:14:32 -07001890}
1891EXPORT_SYMBOL_GPL(bio_trim);
1892
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893/*
1894 * create memory pools for biovec's in a bio_set.
1895 * use the global biovec slabs created for general use.
1896 */
Fabian Fredericka6c39cb4f2014-04-22 15:09:05 -06001897mempool_t *biovec_create_pool(int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
Jens Axboe7ff93452008-12-11 11:53:43 +01001899 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Kent Overstreet9f060e22012-10-12 15:29:33 -07001901 return mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902}
1903
1904void bioset_free(struct bio_set *bs)
1905{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001906 if (bs->rescue_workqueue)
1907 destroy_workqueue(bs->rescue_workqueue);
1908
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (bs->bio_pool)
1910 mempool_destroy(bs->bio_pool);
1911
Kent Overstreet9f060e22012-10-12 15:29:33 -07001912 if (bs->bvec_pool)
1913 mempool_destroy(bs->bvec_pool);
1914
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001915 bioset_integrity_free(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001916 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
1918 kfree(bs);
1919}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001920EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001922static struct bio_set *__bioset_create(unsigned int pool_size,
1923 unsigned int front_pad,
1924 bool create_bvec_pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925{
Jens Axboe392ddc32008-12-23 12:42:54 +01001926 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001927 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Jens Axboe1b434492008-10-22 20:32:58 +02001929 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 if (!bs)
1931 return NULL;
1932
Jens Axboebb799ca2008-12-10 15:35:05 +01001933 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001934
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001935 spin_lock_init(&bs->rescue_lock);
1936 bio_list_init(&bs->rescue_list);
1937 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1938
Jens Axboe392ddc32008-12-23 12:42:54 +01001939 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001940 if (!bs->bio_slab) {
1941 kfree(bs);
1942 return NULL;
1943 }
1944
1945 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 if (!bs->bio_pool)
1947 goto bad;
1948
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001949 if (create_bvec_pool) {
1950 bs->bvec_pool = biovec_create_pool(pool_size);
1951 if (!bs->bvec_pool)
1952 goto bad;
1953 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001955 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1956 if (!bs->rescue_workqueue)
1957 goto bad;
1958
1959 return bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960bad:
1961 bioset_free(bs);
1962 return NULL;
1963}
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001964
1965/**
1966 * bioset_create - Create a bio_set
1967 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1968 * @front_pad: Number of bytes to allocate in front of the returned bio
1969 *
1970 * Description:
1971 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1972 * to ask for a number of bytes to be allocated in front of the bio.
1973 * Front pad allocation is useful for embedding the bio inside
1974 * another structure, to avoid allocating extra data to go with the bio.
1975 * Note that the bio must be embedded at the END of that structure always,
1976 * or things will break badly.
1977 */
1978struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1979{
1980 return __bioset_create(pool_size, front_pad, true);
1981}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001982EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001984/**
1985 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1986 * @pool_size: Number of bio to cache in the mempool
1987 * @front_pad: Number of bytes to allocate in front of the returned bio
1988 *
1989 * Description:
1990 * Same functionality as bioset_create() except that mempool is not
1991 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1992 */
1993struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1994{
1995 return __bioset_create(pool_size, front_pad, false);
1996}
1997EXPORT_SYMBOL(bioset_create_nobvec);
1998
Tejun Heo852c7882012-03-05 13:15:27 -08001999#ifdef CONFIG_BLK_CGROUP
Tejun Heo1d933cf2015-05-22 17:13:24 -04002000
2001/**
2002 * bio_associate_blkcg - associate a bio with the specified blkcg
2003 * @bio: target bio
2004 * @blkcg_css: css of the blkcg to associate
2005 *
2006 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
2007 * treat @bio as if it were issued by a task which belongs to the blkcg.
2008 *
2009 * This function takes an extra reference of @blkcg_css which will be put
2010 * when @bio is released. The caller must own @bio and is responsible for
2011 * synchronizing calls to this function.
2012 */
2013int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
2014{
2015 if (unlikely(bio->bi_css))
2016 return -EBUSY;
2017 css_get(blkcg_css);
2018 bio->bi_css = blkcg_css;
2019 return 0;
2020}
Tejun Heo5aa2a962015-07-23 14:27:09 -04002021EXPORT_SYMBOL_GPL(bio_associate_blkcg);
Tejun Heo1d933cf2015-05-22 17:13:24 -04002022
Tejun Heo852c7882012-03-05 13:15:27 -08002023/**
2024 * bio_associate_current - associate a bio with %current
2025 * @bio: target bio
2026 *
2027 * Associate @bio with %current if it hasn't been associated yet. Block
2028 * layer will treat @bio as if it were issued by %current no matter which
2029 * task actually issues it.
2030 *
2031 * This function takes an extra reference of @task's io_context and blkcg
2032 * which will be put when @bio is released. The caller must own @bio,
2033 * ensure %current->io_context exists, and is responsible for synchronizing
2034 * calls to this function.
2035 */
2036int bio_associate_current(struct bio *bio)
2037{
2038 struct io_context *ioc;
Tejun Heo852c7882012-03-05 13:15:27 -08002039
Tejun Heo1d933cf2015-05-22 17:13:24 -04002040 if (bio->bi_css)
Tejun Heo852c7882012-03-05 13:15:27 -08002041 return -EBUSY;
2042
2043 ioc = current->io_context;
2044 if (!ioc)
2045 return -ENOENT;
2046
Tejun Heo852c7882012-03-05 13:15:27 -08002047 get_io_context_active(ioc);
2048 bio->bi_ioc = ioc;
Tejun Heoec438692015-05-22 17:13:22 -04002049 bio->bi_css = task_get_css(current, blkio_cgrp_id);
Tejun Heo852c7882012-03-05 13:15:27 -08002050 return 0;
2051}
Tejun Heo5aa2a962015-07-23 14:27:09 -04002052EXPORT_SYMBOL_GPL(bio_associate_current);
Tejun Heo852c7882012-03-05 13:15:27 -08002053
2054/**
2055 * bio_disassociate_task - undo bio_associate_current()
2056 * @bio: target bio
2057 */
2058void bio_disassociate_task(struct bio *bio)
2059{
2060 if (bio->bi_ioc) {
2061 put_io_context(bio->bi_ioc);
2062 bio->bi_ioc = NULL;
2063 }
2064 if (bio->bi_css) {
2065 css_put(bio->bi_css);
2066 bio->bi_css = NULL;
2067 }
2068}
2069
2070#endif /* CONFIG_BLK_CGROUP */
2071
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072static void __init biovec_init_slabs(void)
2073{
2074 int i;
2075
2076 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2077 int size;
2078 struct biovec_slab *bvs = bvec_slabs + i;
2079
Jens Axboea7fcd372008-12-05 16:10:29 +01002080 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2081 bvs->slab = NULL;
2082 continue;
2083 }
Jens Axboea7fcd372008-12-05 16:10:29 +01002084
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 size = bvs->nr_vecs * sizeof(struct bio_vec);
2086 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002087 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 }
2089}
2090
2091static int __init init_bio(void)
2092{
Jens Axboebb799ca2008-12-10 15:35:05 +01002093 bio_slab_max = 2;
2094 bio_slab_nr = 0;
2095 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2096 if (!bio_slabs)
2097 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002099 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 biovec_init_slabs();
2101
Jens Axboebb799ca2008-12-10 15:35:05 +01002102 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 if (!fs_bio_set)
2104 panic("bio: can't allocate bios\n");
2105
Martin K. Petersena91a2782011-03-17 11:11:05 +01002106 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2107 panic("bio: can't create integrity pool\n");
2108
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 return 0;
2110}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111subsys_initcall(init_bio);