blob: cf7591551b1716b74fa3765cb1c271a554e8ef56 [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;
Mel Gormand0164ad2015-11-06 16:28:21 -0800214 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __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 /*
Mel Gormand0164ad2015-11-06 16:28:21 -0800224 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
Jens Axboe7ff93452008-12-11 11:53:43 +0100225 * is set, retry with the 1-entry mempool
226 */
227 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
Mel Gormand0164ad2015-11-06 16:28:21 -0800228 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
Jens Axboe7ff93452008-12-11 11:53:43 +0100229 *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));
Jens Axboec4cf5262015-04-17 16:15:18 -0600272 atomic_set(&bio->__bi_remaining, 1);
Jens Axboedac56212015-04-17 16:23:59 -0600273 atomic_set(&bio->__bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200275EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277/**
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700278 * bio_reset - reinitialize a bio
279 * @bio: bio to reset
280 *
281 * Description:
282 * After calling bio_reset(), @bio will be in the same state as a freshly
283 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
284 * preserved are the ones that are initialized by bio_alloc_bioset(). See
285 * comment in struct bio.
286 */
287void bio_reset(struct bio *bio)
288{
289 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
290
Kent Overstreet4254bba2012-09-06 15:35:00 -0700291 __bio_free(bio);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700292
293 memset(bio, 0, BIO_RESET_BYTES);
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200294 bio->bi_flags = flags;
Jens Axboec4cf5262015-04-17 16:15:18 -0600295 atomic_set(&bio->__bi_remaining, 1);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700296}
297EXPORT_SYMBOL(bio_reset);
298
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200299static void bio_chain_endio(struct bio *bio)
Kent Overstreet196d38b2013-11-23 18:34:15 -0800300{
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200301 struct bio *parent = bio->bi_private;
302
303 parent->bi_error = bio->bi_error;
304 bio_endio(parent);
Kent Overstreet196d38b2013-11-23 18:34:15 -0800305 bio_put(bio);
306}
307
Mike Snitzer326e1db2015-05-22 09:14:03 -0400308/*
309 * Increment chain count for the bio. Make sure the CHAIN flag update
310 * is visible before the raised count.
311 */
312static inline void bio_inc_remaining(struct bio *bio)
313{
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600314 bio_set_flag(bio, BIO_CHAIN);
Mike Snitzer326e1db2015-05-22 09:14:03 -0400315 smp_mb__before_atomic();
316 atomic_inc(&bio->__bi_remaining);
317}
318
Kent Overstreet196d38b2013-11-23 18:34:15 -0800319/**
320 * bio_chain - chain bio completions
Randy Dunlap1051a902014-04-20 16:03:31 -0700321 * @bio: the target bio
322 * @parent: the @bio's parent bio
Kent Overstreet196d38b2013-11-23 18:34:15 -0800323 *
324 * The caller won't have a bi_end_io called when @bio completes - instead,
325 * @parent's bi_end_io won't be called until both @parent and @bio have
326 * completed; the chained bio will also be freed when it completes.
327 *
328 * The caller must not set bi_private or bi_end_io in @bio.
329 */
330void bio_chain(struct bio *bio, struct bio *parent)
331{
332 BUG_ON(bio->bi_private || bio->bi_end_io);
333
334 bio->bi_private = parent;
335 bio->bi_end_io = bio_chain_endio;
Jens Axboec4cf5262015-04-17 16:15:18 -0600336 bio_inc_remaining(parent);
Kent Overstreet196d38b2013-11-23 18:34:15 -0800337}
338EXPORT_SYMBOL(bio_chain);
339
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700340static void bio_alloc_rescue(struct work_struct *work)
341{
342 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
343 struct bio *bio;
344
345 while (1) {
346 spin_lock(&bs->rescue_lock);
347 bio = bio_list_pop(&bs->rescue_list);
348 spin_unlock(&bs->rescue_lock);
349
350 if (!bio)
351 break;
352
353 generic_make_request(bio);
354 }
355}
356
357static void punt_bios_to_rescuer(struct bio_set *bs)
358{
359 struct bio_list punt, nopunt;
360 struct bio *bio;
361
362 /*
363 * In order to guarantee forward progress we must punt only bios that
364 * were allocated from this bio_set; otherwise, if there was a bio on
365 * there for a stacking driver higher up in the stack, processing it
366 * could require allocating bios from this bio_set, and doing that from
367 * our own rescuer would be bad.
368 *
369 * Since bio lists are singly linked, pop them all instead of trying to
370 * remove from the middle of the list:
371 */
372
373 bio_list_init(&punt);
374 bio_list_init(&nopunt);
375
376 while ((bio = bio_list_pop(current->bio_list)))
377 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
378
379 *current->bio_list = nopunt;
380
381 spin_lock(&bs->rescue_lock);
382 bio_list_merge(&bs->rescue_list, &punt);
383 spin_unlock(&bs->rescue_lock);
384
385 queue_work(bs->rescue_workqueue, &bs->rescue_work);
386}
387
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700388/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 * bio_alloc_bioset - allocate a bio for I/O
390 * @gfp_mask: the GFP_ mask given to the slab allocator
391 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200392 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 *
394 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700395 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
396 * backed by the @bs's mempool.
397 *
Mel Gormand0164ad2015-11-06 16:28:21 -0800398 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
399 * always be able to allocate a bio. This is due to the mempool guarantees.
400 * To make this work, callers must never allocate more than 1 bio at a time
401 * from this pool. Callers that need to allocate more than 1 bio must always
402 * submit the previously allocated bio for IO before attempting to allocate
403 * a new one. Failure to do so can cause deadlocks under memory pressure.
Kent Overstreet3f86a822012-09-06 15:35:01 -0700404 *
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700405 * Note that when running under generic_make_request() (i.e. any block
406 * driver), bios are not submitted until after you return - see the code in
407 * generic_make_request() that converts recursion into iteration, to prevent
408 * stack overflows.
409 *
410 * This would normally mean allocating multiple bios under
411 * generic_make_request() would be susceptible to deadlocks, but we have
412 * deadlock avoidance code that resubmits any blocked bios from a rescuer
413 * thread.
414 *
415 * However, we do not guarantee forward progress for allocations from other
416 * mempools. Doing multiple allocations from the same mempool under
417 * generic_make_request() should be avoided - instead, use bio_set's front_pad
418 * for per bio allocations.
419 *
Kent Overstreet3f86a822012-09-06 15:35:01 -0700420 * RETURNS:
421 * Pointer to new bio on success, NULL on failure.
422 */
Al Virodd0fc662005-10-07 07:46:04 +0100423struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700425 gfp_t saved_gfp = gfp_mask;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700426 unsigned front_pad;
427 unsigned inline_vecs;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200428 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100429 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200430 struct bio *bio;
431 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200432
Kent Overstreet3f86a822012-09-06 15:35:01 -0700433 if (!bs) {
434 if (nr_iovecs > UIO_MAXIOV)
435 return NULL;
436
437 p = kmalloc(sizeof(struct bio) +
438 nr_iovecs * sizeof(struct bio_vec),
439 gfp_mask);
440 front_pad = 0;
441 inline_vecs = nr_iovecs;
442 } else {
Junichi Nomurad8f429e2014-10-03 17:27:12 -0400443 /* should not use nobvec bioset for nr_iovecs > 0 */
444 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
445 return NULL;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700446 /*
447 * generic_make_request() converts recursion to iteration; this
448 * means if we're running beneath it, any bios we allocate and
449 * submit will not be submitted (and thus freed) until after we
450 * return.
451 *
452 * This exposes us to a potential deadlock if we allocate
453 * multiple bios from the same bio_set() while running
454 * underneath generic_make_request(). If we were to allocate
455 * multiple bios (say a stacking block driver that was splitting
456 * bios), we would deadlock if we exhausted the mempool's
457 * reserve.
458 *
459 * We solve this, and guarantee forward progress, with a rescuer
460 * workqueue per bio_set. If we go to allocate and there are
461 * bios on current->bio_list, we first try the allocation
Mel Gormand0164ad2015-11-06 16:28:21 -0800462 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
463 * bios we would be blocking to the rescuer workqueue before
464 * we retry with the original gfp_flags.
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700465 */
466
467 if (current->bio_list && !bio_list_empty(current->bio_list))
Mel Gormand0164ad2015-11-06 16:28:21 -0800468 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700469
Kent Overstreet3f86a822012-09-06 15:35:01 -0700470 p = mempool_alloc(bs->bio_pool, gfp_mask);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700471 if (!p && gfp_mask != saved_gfp) {
472 punt_bios_to_rescuer(bs);
473 gfp_mask = saved_gfp;
474 p = mempool_alloc(bs->bio_pool, gfp_mask);
475 }
476
Kent Overstreet3f86a822012-09-06 15:35:01 -0700477 front_pad = bs->front_pad;
478 inline_vecs = BIO_INLINE_VECS;
479 }
480
Tejun Heo451a9eb2009-04-15 19:50:51 +0200481 if (unlikely(!p))
482 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100483
Kent Overstreet3f86a822012-09-06 15:35:01 -0700484 bio = p + front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100485 bio_init(bio);
486
Kent Overstreet3f86a822012-09-06 15:35:01 -0700487 if (nr_iovecs > inline_vecs) {
Kent Overstreet9f060e22012-10-12 15:29:33 -0700488 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700489 if (!bvl && gfp_mask != saved_gfp) {
490 punt_bios_to_rescuer(bs);
491 gfp_mask = saved_gfp;
Kent Overstreet9f060e22012-10-12 15:29:33 -0700492 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -0700493 }
494
Ingo Molnar34053972009-02-21 11:16:36 +0100495 if (unlikely(!bvl))
496 goto err_free;
Kent Overstreeta38352e2012-05-25 13:03:11 -0700497
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600498 bio_set_flag(bio, BIO_OWNS_VEC);
Kent Overstreet3f86a822012-09-06 15:35:01 -0700499 } else if (nr_iovecs) {
500 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100501 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700502
503 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100504 bio->bi_flags |= idx << BIO_POOL_OFFSET;
505 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100506 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100508
509err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200510 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100511 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200513EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515void zero_fill_bio(struct bio *bio)
516{
517 unsigned long flags;
Kent Overstreet79886132013-11-23 17:19:00 -0800518 struct bio_vec bv;
519 struct bvec_iter iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Kent Overstreet79886132013-11-23 17:19:00 -0800521 bio_for_each_segment(bv, bio, iter) {
522 char *data = bvec_kmap_irq(&bv, &flags);
523 memset(data, 0, bv.bv_len);
524 flush_dcache_page(bv.bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 bvec_kunmap_irq(data, &flags);
526 }
527}
528EXPORT_SYMBOL(zero_fill_bio);
529
530/**
531 * bio_put - release a reference to a bio
532 * @bio: bio to release reference to
533 *
534 * Description:
535 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100536 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 **/
538void bio_put(struct bio *bio)
539{
Jens Axboedac56212015-04-17 16:23:59 -0600540 if (!bio_flagged(bio, BIO_REFFED))
Kent Overstreet4254bba2012-09-06 15:35:00 -0700541 bio_free(bio);
Jens Axboedac56212015-04-17 16:23:59 -0600542 else {
543 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
544
545 /*
546 * last put frees it
547 */
548 if (atomic_dec_and_test(&bio->__bi_cnt))
549 bio_free(bio);
550 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200552EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Jens Axboe165125e2007-07-24 09:28:11 +0200554inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
556 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
557 blk_recount_segments(q, bio);
558
559 return bio->bi_phys_segments;
560}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200561EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/**
Kent Overstreet59d276f2013-11-23 18:19:27 -0800564 * __bio_clone_fast - clone a bio that shares the original bio's biovec
565 * @bio: destination bio
566 * @bio_src: bio to clone
567 *
568 * Clone a &bio. Caller will own the returned bio, but not
569 * the actual data it points to. Reference count of returned
570 * bio will be one.
571 *
572 * Caller must ensure that @bio_src is not freed before @bio.
573 */
574void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
575{
576 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
577
578 /*
579 * most users will be overriding ->bi_bdev with a new target,
580 * so we don't set nor calculate new physical/hw segment counts here
581 */
582 bio->bi_bdev = bio_src->bi_bdev;
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600583 bio_set_flag(bio, BIO_CLONED);
Kent Overstreet59d276f2013-11-23 18:19:27 -0800584 bio->bi_rw = bio_src->bi_rw;
585 bio->bi_iter = bio_src->bi_iter;
586 bio->bi_io_vec = bio_src->bi_io_vec;
587}
588EXPORT_SYMBOL(__bio_clone_fast);
589
590/**
591 * bio_clone_fast - clone a bio that shares the original bio's biovec
592 * @bio: bio to clone
593 * @gfp_mask: allocation priority
594 * @bs: bio_set to allocate from
595 *
596 * Like __bio_clone_fast, only also allocates the returned bio
597 */
598struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
599{
600 struct bio *b;
601
602 b = bio_alloc_bioset(gfp_mask, 0, bs);
603 if (!b)
604 return NULL;
605
606 __bio_clone_fast(b, bio);
607
608 if (bio_integrity(bio)) {
609 int ret;
610
611 ret = bio_integrity_clone(b, bio, gfp_mask);
612
613 if (ret < 0) {
614 bio_put(b);
615 return NULL;
616 }
617 }
618
619 return b;
620}
621EXPORT_SYMBOL(bio_clone_fast);
622
623/**
Kent Overstreetbdb53202013-11-23 17:26:46 -0800624 * bio_clone_bioset - clone a bio
625 * @bio_src: bio to clone
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700627 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 *
Kent Overstreetbdb53202013-11-23 17:26:46 -0800629 * Clone bio. Caller will own the returned bio, but not the actual data it
630 * points to. Reference count of returned bio will be one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 */
Kent Overstreetbdb53202013-11-23 17:26:46 -0800632struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700633 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
Kent Overstreetbdb53202013-11-23 17:26:46 -0800635 struct bvec_iter iter;
636 struct bio_vec bv;
637 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Kent Overstreetbdb53202013-11-23 17:26:46 -0800639 /*
640 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
641 * bio_src->bi_io_vec to bio->bi_io_vec.
642 *
643 * We can't do that anymore, because:
644 *
645 * - The point of cloning the biovec is to produce a bio with a biovec
646 * the caller can modify: bi_idx and bi_bvec_done should be 0.
647 *
648 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
649 * we tried to clone the whole thing bio_alloc_bioset() would fail.
650 * But the clone should succeed as long as the number of biovecs we
651 * actually need to allocate is fewer than BIO_MAX_PAGES.
652 *
653 * - Lastly, bi_vcnt should not be looked at or relied upon by code
654 * that does not own the bio - reason being drivers don't use it for
655 * iterating over the biovec anymore, so expecting it to be kept up
656 * to date (i.e. for clones that share the parent biovec) is just
657 * asking for trouble and would force extra work on
658 * __bio_clone_fast() anyways.
659 */
660
Kent Overstreet8423ae32014-02-10 17:45:50 -0800661 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
Kent Overstreetbdb53202013-11-23 17:26:46 -0800662 if (!bio)
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200663 return NULL;
664
Kent Overstreetbdb53202013-11-23 17:26:46 -0800665 bio->bi_bdev = bio_src->bi_bdev;
666 bio->bi_rw = bio_src->bi_rw;
667 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
668 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200669
Kent Overstreet8423ae32014-02-10 17:45:50 -0800670 if (bio->bi_rw & REQ_DISCARD)
671 goto integrity_clone;
672
673 if (bio->bi_rw & REQ_WRITE_SAME) {
674 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
675 goto integrity_clone;
676 }
677
Kent Overstreetbdb53202013-11-23 17:26:46 -0800678 bio_for_each_segment(bv, bio_src, iter)
679 bio->bi_io_vec[bio->bi_vcnt++] = bv;
680
Kent Overstreet8423ae32014-02-10 17:45:50 -0800681integrity_clone:
Kent Overstreetbdb53202013-11-23 17:26:46 -0800682 if (bio_integrity(bio_src)) {
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200683 int ret;
684
Kent Overstreetbdb53202013-11-23 17:26:46 -0800685 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
Li Zefan059ea332009-03-09 10:42:45 +0100686 if (ret < 0) {
Kent Overstreetbdb53202013-11-23 17:26:46 -0800687 bio_put(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200688 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100689 }
Peter Osterlund36763472005-09-06 15:16:42 -0700690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Kent Overstreetbdb53202013-11-23 17:26:46 -0800692 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700694EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
696/**
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800697 * bio_add_pc_page - attempt to add page to bio
698 * @q: the target queue
699 * @bio: destination bio
700 * @page: page to add
701 * @len: vec entry length
702 * @offset: vec entry offset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800704 * Attempt to add a page to the bio_vec maplist. This can fail for a
705 * number of reasons, such as the bio being full or target block device
706 * limitations. The target block device must allow bio's up to PAGE_SIZE,
707 * so it is always possible to add a single page to an empty bio.
708 *
709 * This should only be used by REQ_PC bios.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800711int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
712 *page, unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
714 int retried_segments = 0;
715 struct bio_vec *bvec;
716
717 /*
718 * cloned bio must not modify vec list
719 */
720 if (unlikely(bio_flagged(bio, BIO_CLONED)))
721 return 0;
722
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800723 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 return 0;
725
Jens Axboe80cfd542006-01-06 09:43:28 +0100726 /*
727 * For filesystems with a blocksize smaller than the pagesize
728 * we will often be called with the same page as last time and
729 * a consecutive offset. Optimize this special case.
730 */
731 if (bio->bi_vcnt > 0) {
732 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
733
734 if (page == prev->bv_page &&
735 offset == prev->bv_offset + prev->bv_len) {
736 prev->bv_len += len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800737 bio->bi_iter.bi_size += len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100738 goto done;
739 }
Jens Axboe66cb45a2014-06-24 16:22:24 -0600740
741 /*
742 * If the queue doesn't support SG gaps and adding this
743 * offset would create a gap, disallow it.
744 */
Keith Busch03100aa2015-08-19 14:24:05 -0700745 if (bvec_gap_to_prev(q, prev, offset))
Jens Axboe66cb45a2014-06-24 16:22:24 -0600746 return 0;
Jens Axboe80cfd542006-01-06 09:43:28 +0100747 }
748
749 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return 0;
751
752 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 * setup the new entry, we might clear it again later if we
754 * cannot add the page
755 */
756 bvec = &bio->bi_io_vec[bio->bi_vcnt];
757 bvec->bv_page = page;
758 bvec->bv_len = len;
759 bvec->bv_offset = offset;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800760 bio->bi_vcnt++;
761 bio->bi_phys_segments++;
762 bio->bi_iter.bi_size += len;
763
764 /*
765 * Perform a recount if the number of segments is greater
766 * than queue_max_segments(q).
767 */
768
769 while (bio->bi_phys_segments > queue_max_segments(q)) {
770
771 if (retried_segments)
772 goto failed;
773
774 retried_segments = 1;
775 blk_recount_segments(q, bio);
776 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 /* If we may be able to merge these biovecs, force a recount */
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800779 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600780 bio_clear_flag(bio, BIO_SEG_VALID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Jens Axboe80cfd542006-01-06 09:43:28 +0100782 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 return len;
Maurizio Lombardifcbf6a02014-12-10 14:16:53 -0800784
785 failed:
786 bvec->bv_page = NULL;
787 bvec->bv_len = 0;
788 bvec->bv_offset = 0;
789 bio->bi_vcnt--;
790 bio->bi_iter.bi_size -= len;
791 blk_recount_segments(q, bio);
792 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200794EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600795
796/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 * bio_add_page - attempt to add page to bio
798 * @bio: destination bio
799 * @page: page to add
800 * @len: vec entry length
801 * @offset: vec entry offset
802 *
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800803 * Attempt to add a page to the bio_vec maplist. This will only fail
804 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 */
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800806int bio_add_page(struct bio *bio, struct page *page,
807 unsigned int len, unsigned int offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808{
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800809 struct bio_vec *bv;
Jens Axboe762380a2014-06-05 13:38:39 -0600810
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800811 /*
812 * cloned bio must not modify vec list
813 */
814 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
815 return 0;
Jens Axboe58a49152014-06-10 12:53:56 -0600816
Kent Overstreetc66a14d2013-11-23 22:30:22 -0800817 /*
818 * For filesystems with a blocksize smaller than the pagesize
819 * we will often be called with the same page as last time and
820 * a consecutive offset. Optimize this special case.
821 */
822 if (bio->bi_vcnt > 0) {
823 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
824
825 if (page == bv->bv_page &&
826 offset == bv->bv_offset + bv->bv_len) {
827 bv->bv_len += len;
828 goto done;
829 }
830 }
831
832 if (bio->bi_vcnt >= bio->bi_max_vecs)
833 return 0;
834
835 bv = &bio->bi_io_vec[bio->bi_vcnt];
836 bv->bv_page = page;
837 bv->bv_len = len;
838 bv->bv_offset = offset;
839
840 bio->bi_vcnt++;
841done:
842 bio->bi_iter.bi_size += len;
843 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200845EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Kent Overstreet9e882242012-09-10 14:41:12 -0700847struct submit_bio_ret {
848 struct completion event;
849 int error;
850};
851
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200852static void submit_bio_wait_endio(struct bio *bio)
Kent Overstreet9e882242012-09-10 14:41:12 -0700853{
854 struct submit_bio_ret *ret = bio->bi_private;
855
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200856 ret->error = bio->bi_error;
Kent Overstreet9e882242012-09-10 14:41:12 -0700857 complete(&ret->event);
858}
859
860/**
861 * submit_bio_wait - submit a bio, and wait until it completes
862 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
863 * @bio: The &struct bio which describes the I/O
864 *
865 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
866 * bio_endio() on failure.
867 */
868int submit_bio_wait(int rw, struct bio *bio)
869{
870 struct submit_bio_ret ret;
871
872 rw |= REQ_SYNC;
873 init_completion(&ret.event);
874 bio->bi_private = &ret;
875 bio->bi_end_io = submit_bio_wait_endio;
876 submit_bio(rw, bio);
Stephane Gasparinid57d6112016-02-09 17:07:38 +0100877 wait_for_completion_io(&ret.event);
Kent Overstreet9e882242012-09-10 14:41:12 -0700878
879 return ret.error;
880}
881EXPORT_SYMBOL(submit_bio_wait);
882
Kent Overstreet054bdf62012-09-28 13:17:55 -0700883/**
884 * bio_advance - increment/complete a bio by some number of bytes
885 * @bio: bio to advance
886 * @bytes: number of bytes to complete
887 *
888 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
889 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
890 * be updated on the last bvec as well.
891 *
892 * @bio will then represent the remaining, uncompleted portion of the io.
893 */
894void bio_advance(struct bio *bio, unsigned bytes)
895{
896 if (bio_integrity(bio))
897 bio_integrity_advance(bio, bytes);
898
Kent Overstreet4550dd62013-08-07 14:26:21 -0700899 bio_advance_iter(bio, &bio->bi_iter, bytes);
Kent Overstreet054bdf62012-09-28 13:17:55 -0700900}
901EXPORT_SYMBOL(bio_advance);
902
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700903/**
Kent Overstreeta0787602012-09-10 14:03:28 -0700904 * bio_alloc_pages - allocates a single page for each bvec in a bio
905 * @bio: bio to allocate pages for
906 * @gfp_mask: flags for allocation
907 *
908 * Allocates pages up to @bio->bi_vcnt.
909 *
910 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
911 * freed.
912 */
913int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
914{
915 int i;
916 struct bio_vec *bv;
917
918 bio_for_each_segment_all(bv, bio, i) {
919 bv->bv_page = alloc_page(gfp_mask);
920 if (!bv->bv_page) {
921 while (--bv >= bio->bi_io_vec)
922 __free_page(bv->bv_page);
923 return -ENOMEM;
924 }
925 }
926
927 return 0;
928}
929EXPORT_SYMBOL(bio_alloc_pages);
930
931/**
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700932 * bio_copy_data - copy contents of data buffers from one chain of bios to
933 * another
934 * @src: source bio list
935 * @dst: destination bio list
936 *
937 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
938 * @src and @dst as linked lists of bios.
939 *
940 * Stops when it reaches the end of either @src or @dst - that is, copies
941 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
942 */
943void bio_copy_data(struct bio *dst, struct bio *src)
944{
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700945 struct bvec_iter src_iter, dst_iter;
946 struct bio_vec src_bv, dst_bv;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700947 void *src_p, *dst_p;
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700948 unsigned bytes;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700949
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700950 src_iter = src->bi_iter;
951 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700952
953 while (1) {
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700954 if (!src_iter.bi_size) {
955 src = src->bi_next;
956 if (!src)
957 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700958
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700959 src_iter = src->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700960 }
961
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700962 if (!dst_iter.bi_size) {
963 dst = dst->bi_next;
964 if (!dst)
965 break;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700966
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700967 dst_iter = dst->bi_iter;
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700968 }
969
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700970 src_bv = bio_iter_iovec(src, src_iter);
971 dst_bv = bio_iter_iovec(dst, dst_iter);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700972
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700973 bytes = min(src_bv.bv_len, dst_bv.bv_len);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700974
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700975 src_p = kmap_atomic(src_bv.bv_page);
976 dst_p = kmap_atomic(dst_bv.bv_page);
977
978 memcpy(dst_p + dst_bv.bv_offset,
979 src_p + src_bv.bv_offset,
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700980 bytes);
981
982 kunmap_atomic(dst_p);
983 kunmap_atomic(src_p);
984
Kent Overstreet1cb9dda2013-08-07 14:26:39 -0700985 bio_advance_iter(src, &src_iter, bytes);
986 bio_advance_iter(dst, &dst_iter, bytes);
Kent Overstreet16ac3d62012-09-10 13:57:51 -0700987 }
988}
989EXPORT_SYMBOL(bio_copy_data);
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991struct bio_map_data {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900992 int is_our_pages;
Kent Overstreet26e49cf2015-01-18 16:16:31 +0100993 struct iov_iter iter;
994 struct iovec iov[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995};
996
Fabian Frederick7410b3c2014-04-22 15:09:07 -0600997static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200998 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
Jens Axboef3f63c12010-10-29 11:46:56 -06001000 if (iov_count > UIO_MAXIOV)
1001 return NULL;
1002
Kent Overstreetc8db4442013-11-22 19:39:06 -08001003 return kmalloc(sizeof(struct bio_map_data) +
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001004 sizeof(struct iovec) * iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
1006
Dongsu Park9124d3f2015-01-18 16:16:34 +01001007/**
1008 * bio_copy_from_iter - copy all pages from iov_iter to bio
1009 * @bio: The &struct bio which describes the I/O as destination
1010 * @iter: iov_iter as source
1011 *
1012 * Copy all pages from iov_iter to bio.
1013 * Returns 0 on success, or error on failure.
1014 */
1015static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001016{
Dongsu Park9124d3f2015-01-18 16:16:34 +01001017 int i;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001018 struct bio_vec *bvec;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001019
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001020 bio_for_each_segment_all(bvec, bio, i) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001021 ssize_t ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001022
Dongsu Park9124d3f2015-01-18 16:16:34 +01001023 ret = copy_page_from_iter(bvec->bv_page,
1024 bvec->bv_offset,
1025 bvec->bv_len,
1026 &iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001027
Dongsu Park9124d3f2015-01-18 16:16:34 +01001028 if (!iov_iter_count(&iter))
1029 break;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001030
Dongsu Park9124d3f2015-01-18 16:16:34 +01001031 if (ret < bvec->bv_len)
1032 return -EFAULT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001033 }
1034
Dongsu Park9124d3f2015-01-18 16:16:34 +01001035 return 0;
1036}
1037
1038/**
1039 * bio_copy_to_iter - copy all pages from bio to iov_iter
1040 * @bio: The &struct bio which describes the I/O as source
1041 * @iter: iov_iter as destination
1042 *
1043 * Copy all pages from bio to iov_iter.
1044 * Returns 0 on success, or error on failure.
1045 */
1046static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1047{
1048 int i;
1049 struct bio_vec *bvec;
1050
1051 bio_for_each_segment_all(bvec, bio, i) {
1052 ssize_t ret;
1053
1054 ret = copy_page_to_iter(bvec->bv_page,
1055 bvec->bv_offset,
1056 bvec->bv_len,
1057 &iter);
1058
1059 if (!iov_iter_count(&iter))
1060 break;
1061
1062 if (ret < bvec->bv_len)
1063 return -EFAULT;
1064 }
1065
1066 return 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001067}
1068
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001069static void bio_free_pages(struct bio *bio)
1070{
1071 struct bio_vec *bvec;
1072 int i;
1073
1074 bio_for_each_segment_all(bvec, bio, i)
1075 __free_page(bvec->bv_page);
1076}
1077
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078/**
1079 * bio_uncopy_user - finish previously mapped bio
1080 * @bio: bio being terminated
1081 *
Christoph Hellwigddad8dd2015-01-18 16:16:29 +01001082 * Free pages allocated from bio_copy_user_iov() and write back data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 * to user space in case of a read.
1084 */
1085int bio_uncopy_user(struct bio *bio)
1086{
1087 struct bio_map_data *bmd = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001088 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Roland Dreier35dc2482013-08-05 17:55:01 -07001090 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1091 /*
1092 * if we're in a workqueue, the request is orphaned, so
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001093 * don't copy into a random user address space, just free
1094 * and return -EINTR so user space doesn't expect any data.
Roland Dreier35dc2482013-08-05 17:55:01 -07001095 */
Hannes Reinecke2d99b552016-02-12 09:39:15 +01001096 if (!current->mm)
1097 ret = -EINTR;
1098 else if (bio_data_dir(bio) == READ)
Dongsu Park9124d3f2015-01-18 16:16:34 +01001099 ret = bio_copy_to_iter(bio, bmd->iter);
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001100 if (bmd->is_our_pages)
1101 bio_free_pages(bio);
Roland Dreier35dc2482013-08-05 17:55:01 -07001102 }
Kent Overstreetc8db4442013-11-22 19:39:06 -08001103 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 bio_put(bio);
1105 return ret;
1106}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001107EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001110 * bio_copy_user_iov - copy user data to bio
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001111 * @q: destination block queue
1112 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1113 * @iter: iovec iterator
1114 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 *
1116 * Prepares and returns a bio for indirect user io, bouncing data
1117 * to/from kernel pages as necessary. Must be paired with
1118 * call bio_uncopy_user() on io completion.
1119 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001120struct bio *bio_copy_user_iov(struct request_queue *q,
1121 struct rq_map_data *map_data,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001122 const struct iov_iter *iter,
1123 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 struct page *page;
1127 struct bio *bio;
1128 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001129 int nr_pages = 0;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001130 unsigned int len = iter->count;
Geliang Tangbd5cece2015-11-21 17:27:31 +08001131 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001133 for (i = 0; i < iter->nr_segs; i++) {
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001134 unsigned long uaddr;
1135 unsigned long end;
1136 unsigned long start;
1137
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001138 uaddr = (unsigned long) iter->iov[i].iov_base;
1139 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1140 >> PAGE_SHIFT;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001141 start = uaddr >> PAGE_SHIFT;
1142
Jens Axboecb4644c2010-11-10 14:36:25 +01001143 /*
1144 * Overflow, abort
1145 */
1146 if (end < start)
1147 return ERR_PTR(-EINVAL);
1148
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001149 nr_pages += end - start;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001150 }
1151
FUJITA Tomonori69838722009-04-28 20:24:29 +02001152 if (offset)
1153 nr_pages++;
1154
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001155 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (!bmd)
1157 return ERR_PTR(-ENOMEM);
1158
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001159 /*
1160 * We need to do a deep copy of the iov_iter including the iovecs.
1161 * The caller provided iov might point to an on-stack or otherwise
1162 * shortlived one.
1163 */
1164 bmd->is_our_pages = map_data ? 0 : 1;
1165 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1166 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1167 iter->nr_segs, iter->count);
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001170 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 if (!bio)
1172 goto out_bmd;
1173
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001174 if (iter->type & WRITE)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001175 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
1177 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001178
1179 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001180 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001181 i = map_data->offset / PAGE_SIZE;
1182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001184 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001186 bytes -= offset;
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 if (bytes > len)
1189 bytes = len;
1190
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001191 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001192 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001193 ret = -ENOMEM;
1194 break;
1195 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001196
1197 page = map_data->pages[i / nr_pages];
1198 page += (i % nr_pages);
1199
1200 i++;
1201 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001202 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +09001203 if (!page) {
1204 ret = -ENOMEM;
1205 break;
1206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 }
1208
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001209 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
1212 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +09001213 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215
1216 if (ret)
1217 goto cleanup;
1218
1219 /*
1220 * success
1221 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001222 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +02001223 (map_data && map_data->from_user)) {
Dongsu Park9124d3f2015-01-18 16:16:34 +01001224 ret = bio_copy_from_iter(bio, *iter);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +02001225 if (ret)
1226 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 }
1228
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001229 bio->bi_private = bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 return bio;
1231cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +09001232 if (!map_data)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001233 bio_free_pages(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 bio_put(bio);
1235out_bmd:
Kent Overstreetc8db4442013-11-22 19:39:06 -08001236 kfree(bmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 return ERR_PTR(ret);
1238}
1239
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001240/**
1241 * bio_map_user_iov - map user iovec into bio
1242 * @q: the struct request_queue for the bio
1243 * @iter: iovec iterator
1244 * @gfp_mask: memory allocation flags
1245 *
1246 * Map the user space address into a bio suitable for io to a block
1247 * device. Returns an error pointer in case of error.
1248 */
1249struct bio *bio_map_user_iov(struct request_queue *q,
1250 const struct iov_iter *iter,
1251 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252{
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001253 int j;
James Bottomley f1970ba2005-06-20 14:06:52 +02001254 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 struct page **pages;
1256 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001257 int cur_page = 0;
1258 int ret, offset;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001259 struct iov_iter i;
1260 struct iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001262 iov_for_each(iov, i, *iter) {
1263 unsigned long uaddr = (unsigned long) iov.iov_base;
1264 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001265 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1266 unsigned long start = uaddr >> PAGE_SHIFT;
1267
Jens Axboecb4644c2010-11-10 14:36:25 +01001268 /*
1269 * Overflow, abort
1270 */
1271 if (end < start)
1272 return ERR_PTR(-EINVAL);
1273
James Bottomley f1970ba2005-06-20 14:06:52 +02001274 nr_pages += end - start;
1275 /*
Mike Christiead2d7222006-12-01 10:40:20 +01001276 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +02001277 */
Mike Christiead2d7222006-12-01 10:40:20 +01001278 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +02001279 return ERR_PTR(-EINVAL);
1280 }
1281
1282 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 return ERR_PTR(-EINVAL);
1284
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001285 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if (!bio)
1287 return ERR_PTR(-ENOMEM);
1288
1289 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001290 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 if (!pages)
1292 goto out;
1293
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001294 iov_for_each(iov, i, *iter) {
1295 unsigned long uaddr = (unsigned long) iov.iov_base;
1296 unsigned long len = iov.iov_len;
James Bottomley f1970ba2005-06-20 14:06:52 +02001297 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1298 unsigned long start = uaddr >> PAGE_SHIFT;
1299 const int local_nr_pages = end - start;
1300 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001301
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001302 ret = get_user_pages_fast(uaddr, local_nr_pages,
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001303 (iter->type & WRITE) != WRITE,
1304 &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +02001305 if (ret < local_nr_pages) {
1306 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001307 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001308 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
Geliang Tangbd5cece2015-11-21 17:27:31 +08001310 offset = offset_in_page(uaddr);
James Bottomley f1970ba2005-06-20 14:06:52 +02001311 for (j = cur_page; j < page_limit; j++) {
1312 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
James Bottomley f1970ba2005-06-20 14:06:52 +02001314 if (len <= 0)
1315 break;
1316
1317 if (bytes > len)
1318 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
James Bottomley f1970ba2005-06-20 14:06:52 +02001320 /*
1321 * sorry...
1322 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001323 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1324 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001325 break;
1326
1327 len -= bytes;
1328 offset = 0;
1329 }
1330
1331 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001333 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001335 while (j < page_limit)
1336 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 kfree(pages);
1340
1341 /*
1342 * set data direction, and check if mapped pages need bouncing
1343 */
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001344 if (iter->type & WRITE)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001345 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001347 bio_set_flag(bio, BIO_USER_MAPPED);
Christoph Hellwig37f19e52015-01-18 16:16:33 +01001348
1349 /*
1350 * subtle -- if __bio_map_user() ended up bouncing a bio,
1351 * it would normally disappear when its bi_end_io is run.
1352 * however, we need it for the unmap, so grab an extra
1353 * reference to it
1354 */
1355 bio_get(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001357
1358 out_unmap:
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001359 for (j = 0; j < nr_pages; j++) {
1360 if (!pages[j])
James Bottomley f1970ba2005-06-20 14:06:52 +02001361 break;
Kent Overstreet26e49cf2015-01-18 16:16:31 +01001362 page_cache_release(pages[j]);
James Bottomley f1970ba2005-06-20 14:06:52 +02001363 }
1364 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 kfree(pages);
1366 bio_put(bio);
1367 return ERR_PTR(ret);
1368}
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370static void __bio_unmap_user(struct bio *bio)
1371{
1372 struct bio_vec *bvec;
1373 int i;
1374
1375 /*
1376 * make sure we dirty pages we wrote to
1377 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001378 bio_for_each_segment_all(bvec, bio, i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 if (bio_data_dir(bio) == READ)
1380 set_page_dirty_lock(bvec->bv_page);
1381
1382 page_cache_release(bvec->bv_page);
1383 }
1384
1385 bio_put(bio);
1386}
1387
1388/**
1389 * bio_unmap_user - unmap a bio
1390 * @bio: the bio being unmapped
1391 *
1392 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1393 * a process context.
1394 *
1395 * bio_unmap_user() may sleep.
1396 */
1397void bio_unmap_user(struct bio *bio)
1398{
1399 __bio_unmap_user(bio);
1400 bio_put(bio);
1401}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001402EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001404static void bio_map_kern_endio(struct bio *bio)
Jens Axboeb8238252005-06-20 14:05:27 +02001405{
Jens Axboeb8238252005-06-20 14:05:27 +02001406 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001407}
1408
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001409/**
1410 * bio_map_kern - map kernel address into bio
1411 * @q: the struct request_queue for the bio
1412 * @data: pointer to buffer to map
1413 * @len: length in bytes
1414 * @gfp_mask: allocation flags for bio allocation
1415 *
1416 * Map the kernel address into a bio suitable for io to a block
1417 * device. Returns an error pointer in case of error.
1418 */
1419struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1420 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001421{
1422 unsigned long kaddr = (unsigned long)data;
1423 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1424 unsigned long start = kaddr >> PAGE_SHIFT;
1425 const int nr_pages = end - start;
1426 int offset, i;
1427 struct bio *bio;
1428
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001429 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001430 if (!bio)
1431 return ERR_PTR(-ENOMEM);
1432
1433 offset = offset_in_page(kaddr);
1434 for (i = 0; i < nr_pages; i++) {
1435 unsigned int bytes = PAGE_SIZE - offset;
1436
1437 if (len <= 0)
1438 break;
1439
1440 if (bytes > len)
1441 bytes = len;
1442
Mike Christiedefd94b2005-12-05 02:37:06 -06001443 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
Christoph Hellwig75c72b82015-01-18 16:16:32 +01001444 offset) < bytes) {
1445 /* we don't support partial mappings */
1446 bio_put(bio);
1447 return ERR_PTR(-EINVAL);
1448 }
Mike Christie df46b9a2005-06-20 14:04:44 +02001449
1450 data += bytes;
1451 len -= bytes;
1452 offset = 0;
1453 }
1454
Jens Axboeb8238252005-06-20 14:05:27 +02001455 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001456 return bio;
1457}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001458EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001459
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001460static void bio_copy_kern_endio(struct bio *bio)
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001461{
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001462 bio_free_pages(bio);
1463 bio_put(bio);
1464}
1465
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001466static void bio_copy_kern_endio_read(struct bio *bio)
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001467{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001468 char *p = bio->bi_private;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001469 struct bio_vec *bvec;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001470 int i;
1471
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001472 bio_for_each_segment_all(bvec, bio, i) {
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001473 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
Kent Overstreetc8db4442013-11-22 19:39:06 -08001474 p += bvec->bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001475 }
1476
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001477 bio_copy_kern_endio(bio);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001478}
1479
1480/**
1481 * bio_copy_kern - copy kernel address into bio
1482 * @q: the struct request_queue for the bio
1483 * @data: pointer to buffer to copy
1484 * @len: length in bytes
1485 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001486 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001487 *
1488 * copy the kernel address into a bio suitable for io to a block
1489 * device. Returns an error pointer in case of error.
1490 */
1491struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1492 gfp_t gfp_mask, int reading)
1493{
Christoph Hellwig42d26832015-01-18 16:16:28 +01001494 unsigned long kaddr = (unsigned long)data;
1495 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1496 unsigned long start = kaddr >> PAGE_SHIFT;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001497 struct bio *bio;
1498 void *p = data;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001499 int nr_pages = 0;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001500
Christoph Hellwig42d26832015-01-18 16:16:28 +01001501 /*
1502 * Overflow, abort
1503 */
1504 if (end < start)
1505 return ERR_PTR(-EINVAL);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001506
Christoph Hellwig42d26832015-01-18 16:16:28 +01001507 nr_pages = end - start;
1508 bio = bio_kmalloc(gfp_mask, nr_pages);
1509 if (!bio)
1510 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001511
Christoph Hellwig42d26832015-01-18 16:16:28 +01001512 while (len) {
1513 struct page *page;
1514 unsigned int bytes = PAGE_SIZE;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001515
Christoph Hellwig42d26832015-01-18 16:16:28 +01001516 if (bytes > len)
1517 bytes = len;
1518
1519 page = alloc_page(q->bounce_gfp | gfp_mask);
1520 if (!page)
1521 goto cleanup;
1522
1523 if (!reading)
1524 memcpy(page_address(page), p, bytes);
1525
1526 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1527 break;
1528
1529 len -= bytes;
1530 p += bytes;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001531 }
1532
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001533 if (reading) {
1534 bio->bi_end_io = bio_copy_kern_endio_read;
1535 bio->bi_private = data;
1536 } else {
1537 bio->bi_end_io = bio_copy_kern_endio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001538 bio->bi_rw |= REQ_WRITE;
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001539 }
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001540
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001541 return bio;
Christoph Hellwig42d26832015-01-18 16:16:28 +01001542
1543cleanup:
Christoph Hellwig1dfa0f62015-01-18 16:16:30 +01001544 bio_free_pages(bio);
Christoph Hellwig42d26832015-01-18 16:16:28 +01001545 bio_put(bio);
1546 return ERR_PTR(-ENOMEM);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001547}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001548EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550/*
1551 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1552 * for performing direct-IO in BIOs.
1553 *
1554 * The problem is that we cannot run set_page_dirty() from interrupt context
1555 * because the required locks are not interrupt-safe. So what we can do is to
1556 * mark the pages dirty _before_ performing IO. And in interrupt context,
1557 * check that the pages are still dirty. If so, fine. If not, redirty them
1558 * in process context.
1559 *
1560 * We special-case compound pages here: normally this means reads into hugetlb
1561 * pages. The logic in here doesn't really work right for compound pages
1562 * because the VM does not uniformly chase down the head page in all cases.
1563 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1564 * handle them at all. So we skip compound pages here at an early stage.
1565 *
1566 * Note that this code is very hard to test under normal circumstances because
1567 * direct-io pins the pages with get_user_pages(). This makes
1568 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001569 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 * pagecache.
1571 *
1572 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1573 * deferred bio dirtying paths.
1574 */
1575
1576/*
1577 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1578 */
1579void bio_set_pages_dirty(struct bio *bio)
1580{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001581 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 int i;
1583
Kent Overstreetcb34e052012-09-05 15:22:02 -07001584 bio_for_each_segment_all(bvec, bio, i) {
1585 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 if (page && !PageCompound(page))
1588 set_page_dirty_lock(page);
1589 }
1590}
1591
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001592static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001594 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 int i;
1596
Kent Overstreetcb34e052012-09-05 15:22:02 -07001597 bio_for_each_segment_all(bvec, bio, i) {
1598 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600 if (page)
1601 put_page(page);
1602 }
1603}
1604
1605/*
1606 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1607 * If they are, then fine. If, however, some pages are clean then they must
1608 * have been written out during the direct-IO read. So we take another ref on
1609 * the BIO and the offending pages and re-dirty the pages in process context.
1610 *
1611 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1612 * here on. It will run one page_cache_release() against each page and will
1613 * run one bio_put() against the BIO.
1614 */
1615
David Howells65f27f32006-11-22 14:55:48 +00001616static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
David Howells65f27f32006-11-22 14:55:48 +00001618static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619static DEFINE_SPINLOCK(bio_dirty_lock);
1620static struct bio *bio_dirty_list;
1621
1622/*
1623 * This runs in process context
1624 */
David Howells65f27f32006-11-22 14:55:48 +00001625static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
1627 unsigned long flags;
1628 struct bio *bio;
1629
1630 spin_lock_irqsave(&bio_dirty_lock, flags);
1631 bio = bio_dirty_list;
1632 bio_dirty_list = NULL;
1633 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1634
1635 while (bio) {
1636 struct bio *next = bio->bi_private;
1637
1638 bio_set_pages_dirty(bio);
1639 bio_release_pages(bio);
1640 bio_put(bio);
1641 bio = next;
1642 }
1643}
1644
1645void bio_check_pages_dirty(struct bio *bio)
1646{
Kent Overstreetcb34e052012-09-05 15:22:02 -07001647 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 int nr_clean_pages = 0;
1649 int i;
1650
Kent Overstreetcb34e052012-09-05 15:22:02 -07001651 bio_for_each_segment_all(bvec, bio, i) {
1652 struct page *page = bvec->bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
1654 if (PageDirty(page) || PageCompound(page)) {
1655 page_cache_release(page);
Kent Overstreetcb34e052012-09-05 15:22:02 -07001656 bvec->bv_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 } else {
1658 nr_clean_pages++;
1659 }
1660 }
1661
1662 if (nr_clean_pages) {
1663 unsigned long flags;
1664
1665 spin_lock_irqsave(&bio_dirty_lock, flags);
1666 bio->bi_private = bio_dirty_list;
1667 bio_dirty_list = bio;
1668 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1669 schedule_work(&bio_dirty_work);
1670 } else {
1671 bio_put(bio);
1672 }
1673}
1674
Gu Zheng394ffa52014-11-24 11:05:22 +08001675void generic_start_io_acct(int rw, unsigned long sectors,
1676 struct hd_struct *part)
1677{
1678 int cpu = part_stat_lock();
1679
1680 part_round_stats(cpu, part);
1681 part_stat_inc(cpu, part, ios[rw]);
1682 part_stat_add(cpu, part, sectors[rw], sectors);
1683 part_inc_in_flight(part, rw);
1684
1685 part_stat_unlock();
1686}
1687EXPORT_SYMBOL(generic_start_io_acct);
1688
1689void generic_end_io_acct(int rw, struct hd_struct *part,
1690 unsigned long start_time)
1691{
1692 unsigned long duration = jiffies - start_time;
1693 int cpu = part_stat_lock();
1694
1695 part_stat_add(cpu, part, ticks[rw], duration);
1696 part_round_stats(cpu, part);
1697 part_dec_in_flight(part, rw);
1698
1699 part_stat_unlock();
1700}
1701EXPORT_SYMBOL(generic_end_io_acct);
1702
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001703#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1704void bio_flush_dcache_pages(struct bio *bi)
1705{
Kent Overstreet79886132013-11-23 17:19:00 -08001706 struct bio_vec bvec;
1707 struct bvec_iter iter;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001708
Kent Overstreet79886132013-11-23 17:19:00 -08001709 bio_for_each_segment(bvec, bi, iter)
1710 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001711}
1712EXPORT_SYMBOL(bio_flush_dcache_pages);
1713#endif
1714
Jens Axboec4cf5262015-04-17 16:15:18 -06001715static inline bool bio_remaining_done(struct bio *bio)
1716{
1717 /*
1718 * If we're not chaining, then ->__bi_remaining is always 1 and
1719 * we always end io on the first invocation.
1720 */
1721 if (!bio_flagged(bio, BIO_CHAIN))
1722 return true;
1723
1724 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1725
Mike Snitzer326e1db2015-05-22 09:14:03 -04001726 if (atomic_dec_and_test(&bio->__bi_remaining)) {
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001727 bio_clear_flag(bio, BIO_CHAIN);
Jens Axboec4cf5262015-04-17 16:15:18 -06001728 return true;
Mike Snitzer326e1db2015-05-22 09:14:03 -04001729 }
Jens Axboec4cf5262015-04-17 16:15:18 -06001730
1731 return false;
1732}
1733
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734/**
1735 * bio_endio - end I/O on a bio
1736 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 *
1738 * Description:
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001739 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1740 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1741 * bio unless they own it and thus know that it has an end_io function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 **/
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001743void bio_endio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
Kent Overstreet196d38b2013-11-23 18:34:15 -08001745 while (bio) {
Jens Axboec4cf5262015-04-17 16:15:18 -06001746 if (unlikely(!bio_remaining_done(bio)))
1747 break;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001748
1749 /*
1750 * Need to have a real endio function for chained bios,
1751 * otherwise various corner cases will break (like stacking
1752 * block devices that save/restore bi_end_io) - however, we want
1753 * to avoid unbounded recursion and blowing the stack. Tail call
1754 * optimization would handle this, but compiling with frame
1755 * pointers also disables gcc's sibling call optimization.
1756 */
1757 if (bio->bi_end_io == bio_chain_endio) {
1758 struct bio *parent = bio->bi_private;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001759 parent->bi_error = bio->bi_error;
Kent Overstreet196d38b2013-11-23 18:34:15 -08001760 bio_put(bio);
1761 bio = parent;
1762 } else {
1763 if (bio->bi_end_io)
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001764 bio->bi_end_io(bio);
Kent Overstreet196d38b2013-11-23 18:34:15 -08001765 bio = NULL;
1766 }
1767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001769EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Kent Overstreet196d38b2013-11-23 18:34:15 -08001771/**
Kent Overstreet20d01892013-11-23 18:21:01 -08001772 * bio_split - split a bio
1773 * @bio: bio to split
1774 * @sectors: number of sectors to split from the front of @bio
1775 * @gfp: gfp mask
1776 * @bs: bio set to allocate from
1777 *
1778 * Allocates and returns a new bio which represents @sectors from the start of
1779 * @bio, and updates @bio to represent the remaining sectors.
1780 *
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001781 * Unless this is a discard request the newly allocated bio will point
1782 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1783 * @bio is not freed before the split.
Kent Overstreet20d01892013-11-23 18:21:01 -08001784 */
1785struct bio *bio_split(struct bio *bio, int sectors,
1786 gfp_t gfp, struct bio_set *bs)
1787{
1788 struct bio *split = NULL;
1789
1790 BUG_ON(sectors <= 0);
1791 BUG_ON(sectors >= bio_sectors(bio));
1792
Martin K. Petersenf3f5da62015-07-22 07:57:12 -04001793 /*
1794 * Discards need a mutable bio_vec to accommodate the payload
1795 * required by the DSM TRIM and UNMAP commands.
1796 */
1797 if (bio->bi_rw & REQ_DISCARD)
1798 split = bio_clone_bioset(bio, gfp, bs);
1799 else
1800 split = bio_clone_fast(bio, gfp, bs);
1801
Kent Overstreet20d01892013-11-23 18:21:01 -08001802 if (!split)
1803 return NULL;
1804
1805 split->bi_iter.bi_size = sectors << 9;
1806
1807 if (bio_integrity(split))
1808 bio_integrity_trim(split, 0, sectors);
1809
1810 bio_advance(bio, split->bi_iter.bi_size);
1811
1812 return split;
1813}
1814EXPORT_SYMBOL(bio_split);
1815
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001816/**
Kent Overstreet6678d832013-08-07 11:14:32 -07001817 * bio_trim - trim a bio
1818 * @bio: bio to trim
1819 * @offset: number of sectors to trim from the front of @bio
1820 * @size: size we want to trim @bio to, in sectors
1821 */
1822void bio_trim(struct bio *bio, int offset, int size)
1823{
1824 /* 'bio' is a cloned bio which we need to trim to match
1825 * the given offset and size.
Kent Overstreet6678d832013-08-07 11:14:32 -07001826 */
Kent Overstreet6678d832013-08-07 11:14:32 -07001827
1828 size <<= 9;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001829 if (offset == 0 && size == bio->bi_iter.bi_size)
Kent Overstreet6678d832013-08-07 11:14:32 -07001830 return;
1831
Jens Axboeb7c44ed2015-07-24 12:37:59 -06001832 bio_clear_flag(bio, BIO_SEG_VALID);
Kent Overstreet6678d832013-08-07 11:14:32 -07001833
1834 bio_advance(bio, offset << 9);
1835
Kent Overstreet4f024f32013-10-11 15:44:27 -07001836 bio->bi_iter.bi_size = size;
Kent Overstreet6678d832013-08-07 11:14:32 -07001837}
1838EXPORT_SYMBOL_GPL(bio_trim);
1839
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840/*
1841 * create memory pools for biovec's in a bio_set.
1842 * use the global biovec slabs created for general use.
1843 */
Fabian Fredericka6c39cb4f2014-04-22 15:09:05 -06001844mempool_t *biovec_create_pool(int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845{
Jens Axboe7ff93452008-12-11 11:53:43 +01001846 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Kent Overstreet9f060e22012-10-12 15:29:33 -07001848 return mempool_create_slab_pool(pool_entries, bp->slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849}
1850
1851void bioset_free(struct bio_set *bs)
1852{
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001853 if (bs->rescue_workqueue)
1854 destroy_workqueue(bs->rescue_workqueue);
1855
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 if (bs->bio_pool)
1857 mempool_destroy(bs->bio_pool);
1858
Kent Overstreet9f060e22012-10-12 15:29:33 -07001859 if (bs->bvec_pool)
1860 mempool_destroy(bs->bvec_pool);
1861
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001862 bioset_integrity_free(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001863 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
1865 kfree(bs);
1866}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001867EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001869static struct bio_set *__bioset_create(unsigned int pool_size,
1870 unsigned int front_pad,
1871 bool create_bvec_pool)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
Jens Axboe392ddc32008-12-23 12:42:54 +01001873 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001874 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Jens Axboe1b434492008-10-22 20:32:58 +02001876 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if (!bs)
1878 return NULL;
1879
Jens Axboebb799ca2008-12-10 15:35:05 +01001880 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001881
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001882 spin_lock_init(&bs->rescue_lock);
1883 bio_list_init(&bs->rescue_list);
1884 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1885
Jens Axboe392ddc32008-12-23 12:42:54 +01001886 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001887 if (!bs->bio_slab) {
1888 kfree(bs);
1889 return NULL;
1890 }
1891
1892 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 if (!bs->bio_pool)
1894 goto bad;
1895
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001896 if (create_bvec_pool) {
1897 bs->bvec_pool = biovec_create_pool(pool_size);
1898 if (!bs->bvec_pool)
1899 goto bad;
1900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
Kent Overstreetdf2cb6d2012-09-10 14:33:46 -07001902 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1903 if (!bs->rescue_workqueue)
1904 goto bad;
1905
1906 return bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907bad:
1908 bioset_free(bs);
1909 return NULL;
1910}
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001911
1912/**
1913 * bioset_create - Create a bio_set
1914 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1915 * @front_pad: Number of bytes to allocate in front of the returned bio
1916 *
1917 * Description:
1918 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1919 * to ask for a number of bytes to be allocated in front of the bio.
1920 * Front pad allocation is useful for embedding the bio inside
1921 * another structure, to avoid allocating extra data to go with the bio.
1922 * Note that the bio must be embedded at the END of that structure always,
1923 * or things will break badly.
1924 */
1925struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1926{
1927 return __bioset_create(pool_size, front_pad, true);
1928}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001929EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
Junichi Nomurad8f429e2014-10-03 17:27:12 -04001931/**
1932 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1933 * @pool_size: Number of bio to cache in the mempool
1934 * @front_pad: Number of bytes to allocate in front of the returned bio
1935 *
1936 * Description:
1937 * Same functionality as bioset_create() except that mempool is not
1938 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1939 */
1940struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1941{
1942 return __bioset_create(pool_size, front_pad, false);
1943}
1944EXPORT_SYMBOL(bioset_create_nobvec);
1945
Tejun Heo852c7882012-03-05 13:15:27 -08001946#ifdef CONFIG_BLK_CGROUP
Tejun Heo1d933cf2015-05-22 17:13:24 -04001947
1948/**
1949 * bio_associate_blkcg - associate a bio with the specified blkcg
1950 * @bio: target bio
1951 * @blkcg_css: css of the blkcg to associate
1952 *
1953 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1954 * treat @bio as if it were issued by a task which belongs to the blkcg.
1955 *
1956 * This function takes an extra reference of @blkcg_css which will be put
1957 * when @bio is released. The caller must own @bio and is responsible for
1958 * synchronizing calls to this function.
1959 */
1960int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1961{
1962 if (unlikely(bio->bi_css))
1963 return -EBUSY;
1964 css_get(blkcg_css);
1965 bio->bi_css = blkcg_css;
1966 return 0;
1967}
Tejun Heo5aa2a962015-07-23 14:27:09 -04001968EXPORT_SYMBOL_GPL(bio_associate_blkcg);
Tejun Heo1d933cf2015-05-22 17:13:24 -04001969
Tejun Heo852c7882012-03-05 13:15:27 -08001970/**
1971 * bio_associate_current - associate a bio with %current
1972 * @bio: target bio
1973 *
1974 * Associate @bio with %current if it hasn't been associated yet. Block
1975 * layer will treat @bio as if it were issued by %current no matter which
1976 * task actually issues it.
1977 *
1978 * This function takes an extra reference of @task's io_context and blkcg
1979 * which will be put when @bio is released. The caller must own @bio,
1980 * ensure %current->io_context exists, and is responsible for synchronizing
1981 * calls to this function.
1982 */
1983int bio_associate_current(struct bio *bio)
1984{
1985 struct io_context *ioc;
Tejun Heo852c7882012-03-05 13:15:27 -08001986
Tejun Heo1d933cf2015-05-22 17:13:24 -04001987 if (bio->bi_css)
Tejun Heo852c7882012-03-05 13:15:27 -08001988 return -EBUSY;
1989
1990 ioc = current->io_context;
1991 if (!ioc)
1992 return -ENOENT;
1993
Tejun Heo852c7882012-03-05 13:15:27 -08001994 get_io_context_active(ioc);
1995 bio->bi_ioc = ioc;
Tejun Heoc165b3e2015-08-18 14:55:29 -07001996 bio->bi_css = task_get_css(current, io_cgrp_id);
Tejun Heo852c7882012-03-05 13:15:27 -08001997 return 0;
1998}
Tejun Heo5aa2a962015-07-23 14:27:09 -04001999EXPORT_SYMBOL_GPL(bio_associate_current);
Tejun Heo852c7882012-03-05 13:15:27 -08002000
2001/**
2002 * bio_disassociate_task - undo bio_associate_current()
2003 * @bio: target bio
2004 */
2005void bio_disassociate_task(struct bio *bio)
2006{
2007 if (bio->bi_ioc) {
2008 put_io_context(bio->bi_ioc);
2009 bio->bi_ioc = NULL;
2010 }
2011 if (bio->bi_css) {
2012 css_put(bio->bi_css);
2013 bio->bi_css = NULL;
2014 }
2015}
2016
2017#endif /* CONFIG_BLK_CGROUP */
2018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019static void __init biovec_init_slabs(void)
2020{
2021 int i;
2022
2023 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2024 int size;
2025 struct biovec_slab *bvs = bvec_slabs + i;
2026
Jens Axboea7fcd372008-12-05 16:10:29 +01002027 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2028 bvs->slab = NULL;
2029 continue;
2030 }
Jens Axboea7fcd372008-12-05 16:10:29 +01002031
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 size = bvs->nr_vecs * sizeof(struct bio_vec);
2033 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002034 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 }
2036}
2037
2038static int __init init_bio(void)
2039{
Jens Axboebb799ca2008-12-10 15:35:05 +01002040 bio_slab_max = 2;
2041 bio_slab_nr = 0;
2042 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2043 if (!bio_slabs)
2044 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
Martin K. Petersen7878cba2009-06-26 15:37:49 +02002046 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 biovec_init_slabs();
2048
Jens Axboebb799ca2008-12-10 15:35:05 +01002049 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (!fs_bio_set)
2051 panic("bio: can't allocate bios\n");
2052
Martin K. Petersena91a2782011-03-17 11:11:05 +01002053 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2054 panic("bio: can't create integrity pool\n");
2055
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 return 0;
2057}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058subsys_initcall(init_bio);