blob: 9298c65ad9c74bb1c4adde9ef9ffa07b397ca63f [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>
Tejun Heo852c7882012-03-05 13:15:27 -080022#include <linux/iocontext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/slab.h>
24#include <linux/init.h>
25#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mempool.h>
28#include <linux/workqueue.h>
Tejun Heo852c7882012-03-05 13:15:27 -080029#include <linux/cgroup.h>
James Bottomley f1970ba2005-06-20 14:06:52 +020030#include <scsi/sg.h> /* for struct sg_iovec */
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
Denis ChengRq6feef532008-10-09 08:57:05 +020040static mempool_t *bio_split_pool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * if you change this list, also change bvec_alloc or things will
44 * break badly! cannot be bigger than what you can fit into an
45 * unsigned short
46 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Martin K. Petersendf677142011-03-08 08:28:01 +010048static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
50};
51#undef BV
52
53/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
55 * IO code that does not need private memory pools.
56 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020057struct bio_set *fs_bio_set;
Kent Overstreet3f86a822012-09-06 15:35:01 -070058EXPORT_SYMBOL(fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Jens Axboebb799ca2008-12-10 15:35:05 +010060/*
61 * Our slab pool management
62 */
63struct bio_slab {
64 struct kmem_cache *slab;
65 unsigned int slab_ref;
66 unsigned int slab_size;
67 char name[8];
68};
69static DEFINE_MUTEX(bio_slab_lock);
70static struct bio_slab *bio_slabs;
71static unsigned int bio_slab_nr, bio_slab_max;
72
73static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
74{
75 unsigned int sz = sizeof(struct bio) + extra_size;
76 struct kmem_cache *slab = NULL;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020077 struct bio_slab *bslab, *new_bio_slabs;
Jens Axboebb799ca2008-12-10 15:35:05 +010078 unsigned int i, entry = -1;
79
80 mutex_lock(&bio_slab_lock);
81
82 i = 0;
83 while (i < bio_slab_nr) {
Thiago Farinaf06f1352010-01-19 14:07:09 +010084 bslab = &bio_slabs[i];
Jens Axboebb799ca2008-12-10 15:35:05 +010085
86 if (!bslab->slab && entry == -1)
87 entry = i;
88 else if (bslab->slab_size == sz) {
89 slab = bslab->slab;
90 bslab->slab_ref++;
91 break;
92 }
93 i++;
94 }
95
96 if (slab)
97 goto out_unlock;
98
99 if (bio_slab_nr == bio_slab_max && entry == -1) {
100 bio_slab_max <<= 1;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200101 new_bio_slabs = krealloc(bio_slabs,
102 bio_slab_max * sizeof(struct bio_slab),
103 GFP_KERNEL);
104 if (!new_bio_slabs)
Jens Axboebb799ca2008-12-10 15:35:05 +0100105 goto out_unlock;
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);
114 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
115 if (!slab)
116 goto out_unlock;
117
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -0700118 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
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
Jens Axboebb799ca2008-12-10 15:35:05 +0100161void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
162{
163 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
164
165 if (idx == BIOVEC_MAX_IDX)
166 mempool_free(bv, bs->bvec_pool);
167 else {
168 struct biovec_slab *bvs = bvec_slabs + idx;
169
170 kmem_cache_free(bvs->slab, bv);
171 }
172}
173
Jens Axboe7ff93452008-12-11 11:53:43 +0100174struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
175 struct bio_set *bs)
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:
211 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
212 } 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) {
253 if (bio_has_allocated_vec(bio))
254 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
255
256 /*
257 * If we have front padding, adjust the bio pointer before freeing
258 */
259 p = bio;
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 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);
294 bio->bi_flags = flags|(1 << BIO_UPTODATE);
295}
296EXPORT_SYMBOL(bio_reset);
297
298/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 * bio_alloc_bioset - allocate a bio for I/O
300 * @gfp_mask: the GFP_ mask given to the slab allocator
301 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200302 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 *
304 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700305 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
306 * backed by the @bs's mempool.
307 *
308 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
309 * able to allocate a bio. This is due to the mempool guarantees. To make this
310 * work, callers must never allocate more than 1 bio at a time from this pool.
311 * Callers that need to allocate more than 1 bio must always submit the
312 * previously allocated bio for IO before attempting to allocate a new one.
313 * Failure to do so can cause deadlocks under memory pressure.
314 *
315 * RETURNS:
316 * Pointer to new bio on success, NULL on failure.
317 */
Al Virodd0fc662005-10-07 07:46:04 +0100318struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Kent Overstreet3f86a822012-09-06 15:35:01 -0700320 unsigned front_pad;
321 unsigned inline_vecs;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200322 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100323 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200324 struct bio *bio;
325 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200326
Kent Overstreet3f86a822012-09-06 15:35:01 -0700327 if (!bs) {
328 if (nr_iovecs > UIO_MAXIOV)
329 return NULL;
330
331 p = kmalloc(sizeof(struct bio) +
332 nr_iovecs * sizeof(struct bio_vec),
333 gfp_mask);
334 front_pad = 0;
335 inline_vecs = nr_iovecs;
336 } else {
337 p = mempool_alloc(bs->bio_pool, gfp_mask);
338 front_pad = bs->front_pad;
339 inline_vecs = BIO_INLINE_VECS;
340 }
341
Tejun Heo451a9eb2009-04-15 19:50:51 +0200342 if (unlikely(!p))
343 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100344
Kent Overstreet3f86a822012-09-06 15:35:01 -0700345 bio = p + front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100346 bio_init(bio);
347
Kent Overstreet3f86a822012-09-06 15:35:01 -0700348 if (nr_iovecs > inline_vecs) {
Ingo Molnar34053972009-02-21 11:16:36 +0100349 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
350 if (unlikely(!bvl))
351 goto err_free;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700352 } else if (nr_iovecs) {
353 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100354 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700355
356 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100357 bio->bi_flags |= idx << BIO_POOL_OFFSET;
358 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100359 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100361
362err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200363 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100364 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200366EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368void zero_fill_bio(struct bio *bio)
369{
370 unsigned long flags;
371 struct bio_vec *bv;
372 int i;
373
374 bio_for_each_segment(bv, bio, i) {
375 char *data = bvec_kmap_irq(bv, &flags);
376 memset(data, 0, bv->bv_len);
377 flush_dcache_page(bv->bv_page);
378 bvec_kunmap_irq(data, &flags);
379 }
380}
381EXPORT_SYMBOL(zero_fill_bio);
382
383/**
384 * bio_put - release a reference to a bio
385 * @bio: bio to release reference to
386 *
387 * Description:
388 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100389 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 **/
391void bio_put(struct bio *bio)
392{
393 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
394
395 /*
396 * last put frees it
397 */
Kent Overstreet4254bba2012-09-06 15:35:00 -0700398 if (atomic_dec_and_test(&bio->bi_cnt))
399 bio_free(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200401EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Jens Axboe165125e2007-07-24 09:28:11 +0200403inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
405 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
406 blk_recount_segments(q, bio);
407
408 return bio->bi_phys_segments;
409}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200410EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412/**
413 * __bio_clone - clone a bio
414 * @bio: destination bio
415 * @bio_src: bio to clone
416 *
417 * Clone a &bio. Caller will own the returned bio, but not
418 * the actual data it points to. Reference count of returned
419 * bio will be one.
420 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800421void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
Andrew Mortone525e152005-08-07 09:42:12 -0700423 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
424 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Jens Axboe5d840702008-01-25 12:44:44 +0100426 /*
427 * most users will be overriding ->bi_bdev with a new target,
428 * so we don't set nor calculate new physical/hw segment counts here
429 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 bio->bi_sector = bio_src->bi_sector;
431 bio->bi_bdev = bio_src->bi_bdev;
432 bio->bi_flags |= 1 << BIO_CLONED;
433 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 bio->bi_vcnt = bio_src->bi_vcnt;
435 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700436 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200438EXPORT_SYMBOL(__bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440/**
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700441 * bio_clone_bioset - clone a bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * @bio: bio to clone
443 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700444 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 *
446 * Like __bio_clone, only also allocates the returned bio
447 */
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700448struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask,
449 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700451 struct bio *b;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700453 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200454 if (!b)
455 return NULL;
456
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200457 __bio_clone(b, bio);
458
459 if (bio_integrity(bio)) {
460 int ret;
461
Kent Overstreet1e2a410f2012-09-06 15:34:56 -0700462 ret = bio_integrity_clone(b, bio, gfp_mask);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200463
Li Zefan059ea332009-03-09 10:42:45 +0100464 if (ret < 0) {
465 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200466 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100467 }
Peter Osterlund36763472005-09-06 15:16:42 -0700468 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 return b;
471}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700472EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474/**
475 * bio_get_nr_vecs - return approx number of vecs
476 * @bdev: I/O target
477 *
478 * Return the approximate number of pages we can send to this target.
479 * There's no guarantee that you will be able to fit this number of pages
480 * into a bio, it does not account for dynamic restrictions that vary
481 * on offset.
482 */
483int bio_get_nr_vecs(struct block_device *bdev)
484{
Jens Axboe165125e2007-07-24 09:28:11 +0200485 struct request_queue *q = bdev_get_queue(bdev);
Bernd Schubertf908ee92012-05-11 16:36:44 +0200486 int nr_pages;
487
488 nr_pages = min_t(unsigned,
Kent Overstreet5abebfd2012-02-08 22:07:18 +0100489 queue_max_segments(q),
490 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
Bernd Schubertf908ee92012-05-11 16:36:44 +0200491
492 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
493
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200495EXPORT_SYMBOL(bio_get_nr_vecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Jens Axboe165125e2007-07-24 09:28:11 +0200497static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600498 *page, unsigned int len, unsigned int offset,
499 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
501 int retried_segments = 0;
502 struct bio_vec *bvec;
503
504 /*
505 * cloned bio must not modify vec list
506 */
507 if (unlikely(bio_flagged(bio, BIO_CLONED)))
508 return 0;
509
Jens Axboe80cfd542006-01-06 09:43:28 +0100510 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 return 0;
512
Jens Axboe80cfd542006-01-06 09:43:28 +0100513 /*
514 * For filesystems with a blocksize smaller than the pagesize
515 * we will often be called with the same page as last time and
516 * a consecutive offset. Optimize this special case.
517 */
518 if (bio->bi_vcnt > 0) {
519 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
520
521 if (page == prev->bv_page &&
522 offset == prev->bv_offset + prev->bv_len) {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300523 unsigned int prev_bv_len = prev->bv_len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100524 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200525
526 if (q->merge_bvec_fn) {
527 struct bvec_merge_data bvm = {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300528 /* prev_bvec is already charged in
529 bi_size, discharge it in order to
530 simulate merging updated prev_bvec
531 as new bvec. */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200532 .bi_bdev = bio->bi_bdev,
533 .bi_sector = bio->bi_sector,
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300534 .bi_size = bio->bi_size - prev_bv_len,
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200535 .bi_rw = bio->bi_rw,
536 };
537
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300538 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200539 prev->bv_len -= len;
540 return 0;
541 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100542 }
543
544 goto done;
545 }
546 }
547
548 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 return 0;
550
551 /*
552 * we might lose a segment or two here, but rather that than
553 * make this too complex.
554 */
555
Martin K. Petersen8a783622010-02-26 00:20:39 -0500556 while (bio->bi_phys_segments >= queue_max_segments(q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 if (retried_segments)
559 return 0;
560
561 retried_segments = 1;
562 blk_recount_segments(q, bio);
563 }
564
565 /*
566 * setup the new entry, we might clear it again later if we
567 * cannot add the page
568 */
569 bvec = &bio->bi_io_vec[bio->bi_vcnt];
570 bvec->bv_page = page;
571 bvec->bv_len = len;
572 bvec->bv_offset = offset;
573
574 /*
575 * if queue has other restrictions (eg varying max sector size
576 * depending on offset), it can specify a merge_bvec_fn in the
577 * queue to get further control
578 */
579 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200580 struct bvec_merge_data bvm = {
581 .bi_bdev = bio->bi_bdev,
582 .bi_sector = bio->bi_sector,
583 .bi_size = bio->bi_size,
584 .bi_rw = bio->bi_rw,
585 };
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 /*
588 * merge_bvec_fn() returns number of bytes it can accept
589 * at this offset
590 */
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300591 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 bvec->bv_page = NULL;
593 bvec->bv_len = 0;
594 bvec->bv_offset = 0;
595 return 0;
596 }
597 }
598
599 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200600 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
602
603 bio->bi_vcnt++;
604 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100605 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 bio->bi_size += len;
607 return len;
608}
609
610/**
Mike Christie6e68af62005-11-11 05:30:27 -0600611 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100612 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600613 * @bio: destination bio
614 * @page: page to add
615 * @len: vec entry length
616 * @offset: vec entry offset
617 *
618 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200619 * number of reasons, such as the bio being full or target block device
620 * limitations. The target block device must allow bio's up to PAGE_SIZE,
621 * so it is always possible to add a single page to an empty bio.
622 *
623 * This should only be used by REQ_PC bios.
Mike Christie6e68af62005-11-11 05:30:27 -0600624 */
Jens Axboe165125e2007-07-24 09:28:11 +0200625int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600626 unsigned int len, unsigned int offset)
627{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400628 return __bio_add_page(q, bio, page, len, offset,
629 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600630}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200631EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600632
633/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 * bio_add_page - attempt to add page to bio
635 * @bio: destination bio
636 * @page: page to add
637 * @len: vec entry length
638 * @offset: vec entry offset
639 *
640 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200641 * number of reasons, such as the bio being full or target block device
642 * limitations. The target block device must allow bio's up to PAGE_SIZE,
643 * so it is always possible to add a single page to an empty bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 */
645int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
646 unsigned int offset)
647{
Mike Christiedefd94b2005-12-05 02:37:06 -0600648 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400649 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200651EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653struct bio_map_data {
654 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200655 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900656 int nr_sgvecs;
657 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658};
659
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200660static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900661 struct sg_iovec *iov, int iov_count,
662 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663{
664 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200665 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
666 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900667 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 bio->bi_private = bmd;
669}
670
671static void bio_free_map_data(struct bio_map_data *bmd)
672{
673 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200674 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 kfree(bmd);
676}
677
Dan Carpenter121f0992011-11-16 09:21:50 +0100678static struct bio_map_data *bio_alloc_map_data(int nr_segs,
679 unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200680 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Jens Axboef3f63c12010-10-29 11:46:56 -0600682 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Jens Axboef3f63c12010-10-29 11:46:56 -0600684 if (iov_count > UIO_MAXIOV)
685 return NULL;
686
687 bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (!bmd)
689 return NULL;
690
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200691 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200692 if (!bmd->iovecs) {
693 kfree(bmd);
694 return NULL;
695 }
696
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200697 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200698 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 return bmd;
700
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200701 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 kfree(bmd);
703 return NULL;
704}
705
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200706static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200707 struct sg_iovec *iov, int iov_count,
708 int to_user, int from_user, int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200709{
710 int ret = 0, i;
711 struct bio_vec *bvec;
712 int iov_idx = 0;
713 unsigned int iov_off = 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200714
715 __bio_for_each_segment(bvec, bio, i, 0) {
716 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200717 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200718
719 while (bv_len && iov_idx < iov_count) {
720 unsigned int bytes;
Michal Simek0e0c6212009-06-10 12:57:07 -0700721 char __user *iov_addr;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200722
723 bytes = min_t(unsigned int,
724 iov[iov_idx].iov_len - iov_off, bv_len);
725 iov_addr = iov[iov_idx].iov_base + iov_off;
726
727 if (!ret) {
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200728 if (to_user)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200729 ret = copy_to_user(iov_addr, bv_addr,
730 bytes);
731
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200732 if (from_user)
733 ret = copy_from_user(bv_addr, iov_addr,
734 bytes);
735
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200736 if (ret)
737 ret = -EFAULT;
738 }
739
740 bv_len -= bytes;
741 bv_addr += bytes;
742 iov_addr += bytes;
743 iov_off += bytes;
744
745 if (iov[iov_idx].iov_len == iov_off) {
746 iov_idx++;
747 iov_off = 0;
748 }
749 }
750
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900751 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200752 __free_page(bvec->bv_page);
753 }
754
755 return ret;
756}
757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758/**
759 * bio_uncopy_user - finish previously mapped bio
760 * @bio: bio being terminated
761 *
762 * Free pages allocated from bio_copy_user() and write back data
763 * to user space in case of a read.
764 */
765int bio_uncopy_user(struct bio *bio)
766{
767 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900768 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
FUJITA Tomonori81882762008-09-02 16:20:19 +0900770 if (!bio_flagged(bio, BIO_NULL_MAPPED))
771 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200772 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
773 0, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 bio_free_map_data(bmd);
775 bio_put(bio);
776 return ret;
777}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200778EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200781 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900783 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200784 * @iov: the iovec.
785 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900787 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 *
789 * Prepares and returns a bio for indirect user io, bouncing data
790 * to/from kernel pages as necessary. Must be paired with
791 * call bio_uncopy_user() on io completion.
792 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900793struct bio *bio_copy_user_iov(struct request_queue *q,
794 struct rq_map_data *map_data,
795 struct sg_iovec *iov, int iov_count,
796 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 struct bio_map_data *bmd;
799 struct bio_vec *bvec;
800 struct page *page;
801 struct bio *bio;
802 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200803 int nr_pages = 0;
804 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900805 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200807 for (i = 0; i < iov_count; i++) {
808 unsigned long uaddr;
809 unsigned long end;
810 unsigned long start;
811
812 uaddr = (unsigned long)iov[i].iov_base;
813 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
814 start = uaddr >> PAGE_SHIFT;
815
Jens Axboecb4644c2010-11-10 14:36:25 +0100816 /*
817 * Overflow, abort
818 */
819 if (end < start)
820 return ERR_PTR(-EINVAL);
821
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200822 nr_pages += end - start;
823 len += iov[i].iov_len;
824 }
825
FUJITA Tomonori69838722009-04-28 20:24:29 +0200826 if (offset)
827 nr_pages++;
828
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900829 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (!bmd)
831 return ERR_PTR(-ENOMEM);
832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900834 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 if (!bio)
836 goto out_bmd;
837
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200838 if (!write_to_vm)
839 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900842
843 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900844 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900845 i = map_data->offset / PAGE_SIZE;
846 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900848 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900850 bytes -= offset;
851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (bytes > len)
853 bytes = len;
854
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900855 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900856 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900857 ret = -ENOMEM;
858 break;
859 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900860
861 page = map_data->pages[i / nr_pages];
862 page += (i % nr_pages);
863
864 i++;
865 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900866 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900867 if (!page) {
868 ret = -ENOMEM;
869 break;
870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 }
872
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900873 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
876 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900877 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879
880 if (ret)
881 goto cleanup;
882
883 /*
884 * success
885 */
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200886 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
887 (map_data && map_data->from_user)) {
888 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200889 if (ret)
890 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 }
892
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900893 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 return bio;
895cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900896 if (!map_data)
897 bio_for_each_segment(bvec, bio, i)
898 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
900 bio_put(bio);
901out_bmd:
902 bio_free_map_data(bmd);
903 return ERR_PTR(ret);
904}
905
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200906/**
907 * bio_copy_user - copy user data to bio
908 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900909 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200910 * @uaddr: start of user address
911 * @len: length in bytes
912 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900913 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200914 *
915 * Prepares and returns a bio for indirect user io, bouncing data
916 * to/from kernel pages as necessary. Must be paired with
917 * call bio_uncopy_user() on io completion.
918 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900919struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
920 unsigned long uaddr, unsigned int len,
921 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200922{
923 struct sg_iovec iov;
924
925 iov.iov_base = (void __user *)uaddr;
926 iov.iov_len = len;
927
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900928 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200929}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200930EXPORT_SYMBOL(bio_copy_user);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200931
Jens Axboe165125e2007-07-24 09:28:11 +0200932static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200933 struct block_device *bdev,
934 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900935 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936{
James Bottomley f1970ba2005-06-20 14:06:52 +0200937 int i, j;
938 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 struct page **pages;
940 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200941 int cur_page = 0;
942 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
James Bottomley f1970ba2005-06-20 14:06:52 +0200944 for (i = 0; i < iov_count; i++) {
945 unsigned long uaddr = (unsigned long)iov[i].iov_base;
946 unsigned long len = iov[i].iov_len;
947 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
948 unsigned long start = uaddr >> PAGE_SHIFT;
949
Jens Axboecb4644c2010-11-10 14:36:25 +0100950 /*
951 * Overflow, abort
952 */
953 if (end < start)
954 return ERR_PTR(-EINVAL);
955
James Bottomley f1970ba2005-06-20 14:06:52 +0200956 nr_pages += end - start;
957 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100958 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200959 */
Mike Christiead2d7222006-12-01 10:40:20 +0100960 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200961 return ERR_PTR(-EINVAL);
962 }
963
964 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 return ERR_PTR(-EINVAL);
966
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900967 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 if (!bio)
969 return ERR_PTR(-ENOMEM);
970
971 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900972 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 if (!pages)
974 goto out;
975
James Bottomley f1970ba2005-06-20 14:06:52 +0200976 for (i = 0; i < iov_count; i++) {
977 unsigned long uaddr = (unsigned long)iov[i].iov_base;
978 unsigned long len = iov[i].iov_len;
979 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
980 unsigned long start = uaddr >> PAGE_SHIFT;
981 const int local_nr_pages = end - start;
982 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +0100983
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700984 ret = get_user_pages_fast(uaddr, local_nr_pages,
985 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200986 if (ret < local_nr_pages) {
987 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200988 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
James Bottomley f1970ba2005-06-20 14:06:52 +0200991 offset = uaddr & ~PAGE_MASK;
992 for (j = cur_page; j < page_limit; j++) {
993 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
James Bottomley f1970ba2005-06-20 14:06:52 +0200995 if (len <= 0)
996 break;
997
998 if (bytes > len)
999 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
James Bottomley f1970ba2005-06-20 14:06:52 +02001001 /*
1002 * sorry...
1003 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001004 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1005 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001006 break;
1007
1008 len -= bytes;
1009 offset = 0;
1010 }
1011
1012 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001014 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001016 while (j < page_limit)
1017 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 kfree(pages);
1021
1022 /*
1023 * set data direction, and check if mapped pages need bouncing
1024 */
1025 if (!write_to_vm)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001026 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
James Bottomley f1970ba2005-06-20 14:06:52 +02001028 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1030 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001031
1032 out_unmap:
1033 for (i = 0; i < nr_pages; i++) {
1034 if(!pages[i])
1035 break;
1036 page_cache_release(pages[i]);
1037 }
1038 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 kfree(pages);
1040 bio_put(bio);
1041 return ERR_PTR(ret);
1042}
1043
1044/**
1045 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001046 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 * @bdev: destination block device
1048 * @uaddr: start of user address
1049 * @len: length in bytes
1050 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001051 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 *
1053 * Map the user space address into a bio suitable for io to a block
1054 * device. Returns an error pointer in case of error.
1055 */
Jens Axboe165125e2007-07-24 09:28:11 +02001056struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001057 unsigned long uaddr, unsigned int len, int write_to_vm,
1058 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
James Bottomley f1970ba2005-06-20 14:06:52 +02001060 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001062 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001063 iov.iov_len = len;
1064
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001065 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001066}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001067EXPORT_SYMBOL(bio_map_user);
James Bottomley f1970ba2005-06-20 14:06:52 +02001068
1069/**
1070 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001071 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001072 * @bdev: destination block device
1073 * @iov: the iovec.
1074 * @iov_count: number of elements in the iovec
1075 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001076 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001077 *
1078 * Map the user space address into a bio suitable for io to a block
1079 * device. Returns an error pointer in case of error.
1080 */
Jens Axboe165125e2007-07-24 09:28:11 +02001081struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001082 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001083 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001084{
1085 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001086
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001087 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1088 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (IS_ERR(bio))
1090 return bio;
1091
1092 /*
1093 * subtle -- if __bio_map_user() ended up bouncing a bio,
1094 * it would normally disappear when its bi_end_io is run.
1095 * however, we need it for the unmap, so grab an extra
1096 * reference to it
1097 */
1098 bio_get(bio);
1099
Mike Christie0e75f902006-12-01 10:40:55 +01001100 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101}
1102
1103static void __bio_unmap_user(struct bio *bio)
1104{
1105 struct bio_vec *bvec;
1106 int i;
1107
1108 /*
1109 * make sure we dirty pages we wrote to
1110 */
1111 __bio_for_each_segment(bvec, bio, i, 0) {
1112 if (bio_data_dir(bio) == READ)
1113 set_page_dirty_lock(bvec->bv_page);
1114
1115 page_cache_release(bvec->bv_page);
1116 }
1117
1118 bio_put(bio);
1119}
1120
1121/**
1122 * bio_unmap_user - unmap a bio
1123 * @bio: the bio being unmapped
1124 *
1125 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1126 * a process context.
1127 *
1128 * bio_unmap_user() may sleep.
1129 */
1130void bio_unmap_user(struct bio *bio)
1131{
1132 __bio_unmap_user(bio);
1133 bio_put(bio);
1134}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001135EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
NeilBrown6712ecf2007-09-27 12:47:43 +02001137static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001138{
Jens Axboeb8238252005-06-20 14:05:27 +02001139 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001140}
1141
Jens Axboe165125e2007-07-24 09:28:11 +02001142static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001143 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001144{
1145 unsigned long kaddr = (unsigned long)data;
1146 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1147 unsigned long start = kaddr >> PAGE_SHIFT;
1148 const int nr_pages = end - start;
1149 int offset, i;
1150 struct bio *bio;
1151
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001152 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001153 if (!bio)
1154 return ERR_PTR(-ENOMEM);
1155
1156 offset = offset_in_page(kaddr);
1157 for (i = 0; i < nr_pages; i++) {
1158 unsigned int bytes = PAGE_SIZE - offset;
1159
1160 if (len <= 0)
1161 break;
1162
1163 if (bytes > len)
1164 bytes = len;
1165
Mike Christiedefd94b2005-12-05 02:37:06 -06001166 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1167 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001168 break;
1169
1170 data += bytes;
1171 len -= bytes;
1172 offset = 0;
1173 }
1174
Jens Axboeb8238252005-06-20 14:05:27 +02001175 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001176 return bio;
1177}
1178
1179/**
1180 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001181 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001182 * @data: pointer to buffer to map
1183 * @len: length in bytes
1184 * @gfp_mask: allocation flags for bio allocation
1185 *
1186 * Map the kernel address into a bio suitable for io to a block
1187 * device. Returns an error pointer in case of error.
1188 */
Jens Axboe165125e2007-07-24 09:28:11 +02001189struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001190 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001191{
1192 struct bio *bio;
1193
1194 bio = __bio_map_kern(q, data, len, gfp_mask);
1195 if (IS_ERR(bio))
1196 return bio;
1197
1198 if (bio->bi_size == len)
1199 return bio;
1200
1201 /*
1202 * Don't support partial mappings.
1203 */
1204 bio_put(bio);
1205 return ERR_PTR(-EINVAL);
1206}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001207EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001208
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001209static void bio_copy_kern_endio(struct bio *bio, int err)
1210{
1211 struct bio_vec *bvec;
1212 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001213 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001214 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001215 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001216
1217 __bio_for_each_segment(bvec, bio, i, 0) {
1218 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001219 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001220
Tejun Heo4fc981e2009-05-19 18:33:06 +09001221 if (read)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001222 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001223
1224 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001225 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001226 }
1227
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001228 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001229 bio_put(bio);
1230}
1231
1232/**
1233 * bio_copy_kern - copy kernel address into bio
1234 * @q: the struct request_queue for the bio
1235 * @data: pointer to buffer to copy
1236 * @len: length in bytes
1237 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001238 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001239 *
1240 * copy the kernel address into a bio suitable for io to a block
1241 * device. Returns an error pointer in case of error.
1242 */
1243struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1244 gfp_t gfp_mask, int reading)
1245{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001246 struct bio *bio;
1247 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001248 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001249
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001250 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1251 if (IS_ERR(bio))
1252 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001253
1254 if (!reading) {
1255 void *p = data;
1256
1257 bio_for_each_segment(bvec, bio, i) {
1258 char *addr = page_address(bvec->bv_page);
1259
1260 memcpy(addr, p, bvec->bv_len);
1261 p += bvec->bv_len;
1262 }
1263 }
1264
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001265 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001266
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001267 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001268}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001269EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271/*
1272 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1273 * for performing direct-IO in BIOs.
1274 *
1275 * The problem is that we cannot run set_page_dirty() from interrupt context
1276 * because the required locks are not interrupt-safe. So what we can do is to
1277 * mark the pages dirty _before_ performing IO. And in interrupt context,
1278 * check that the pages are still dirty. If so, fine. If not, redirty them
1279 * in process context.
1280 *
1281 * We special-case compound pages here: normally this means reads into hugetlb
1282 * pages. The logic in here doesn't really work right for compound pages
1283 * because the VM does not uniformly chase down the head page in all cases.
1284 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1285 * handle them at all. So we skip compound pages here at an early stage.
1286 *
1287 * Note that this code is very hard to test under normal circumstances because
1288 * direct-io pins the pages with get_user_pages(). This makes
1289 * is_page_cache_freeable return false, and the VM will not clean the pages.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001290 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 * pagecache.
1292 *
1293 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1294 * deferred bio dirtying paths.
1295 */
1296
1297/*
1298 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1299 */
1300void bio_set_pages_dirty(struct bio *bio)
1301{
1302 struct bio_vec *bvec = bio->bi_io_vec;
1303 int i;
1304
1305 for (i = 0; i < bio->bi_vcnt; i++) {
1306 struct page *page = bvec[i].bv_page;
1307
1308 if (page && !PageCompound(page))
1309 set_page_dirty_lock(page);
1310 }
1311}
1312
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001313static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314{
1315 struct bio_vec *bvec = bio->bi_io_vec;
1316 int i;
1317
1318 for (i = 0; i < bio->bi_vcnt; i++) {
1319 struct page *page = bvec[i].bv_page;
1320
1321 if (page)
1322 put_page(page);
1323 }
1324}
1325
1326/*
1327 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1328 * If they are, then fine. If, however, some pages are clean then they must
1329 * have been written out during the direct-IO read. So we take another ref on
1330 * the BIO and the offending pages and re-dirty the pages in process context.
1331 *
1332 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1333 * here on. It will run one page_cache_release() against each page and will
1334 * run one bio_put() against the BIO.
1335 */
1336
David Howells65f27f32006-11-22 14:55:48 +00001337static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
David Howells65f27f32006-11-22 14:55:48 +00001339static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340static DEFINE_SPINLOCK(bio_dirty_lock);
1341static struct bio *bio_dirty_list;
1342
1343/*
1344 * This runs in process context
1345 */
David Howells65f27f32006-11-22 14:55:48 +00001346static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 unsigned long flags;
1349 struct bio *bio;
1350
1351 spin_lock_irqsave(&bio_dirty_lock, flags);
1352 bio = bio_dirty_list;
1353 bio_dirty_list = NULL;
1354 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1355
1356 while (bio) {
1357 struct bio *next = bio->bi_private;
1358
1359 bio_set_pages_dirty(bio);
1360 bio_release_pages(bio);
1361 bio_put(bio);
1362 bio = next;
1363 }
1364}
1365
1366void bio_check_pages_dirty(struct bio *bio)
1367{
1368 struct bio_vec *bvec = bio->bi_io_vec;
1369 int nr_clean_pages = 0;
1370 int i;
1371
1372 for (i = 0; i < bio->bi_vcnt; i++) {
1373 struct page *page = bvec[i].bv_page;
1374
1375 if (PageDirty(page) || PageCompound(page)) {
1376 page_cache_release(page);
1377 bvec[i].bv_page = NULL;
1378 } else {
1379 nr_clean_pages++;
1380 }
1381 }
1382
1383 if (nr_clean_pages) {
1384 unsigned long flags;
1385
1386 spin_lock_irqsave(&bio_dirty_lock, flags);
1387 bio->bi_private = bio_dirty_list;
1388 bio_dirty_list = bio;
1389 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1390 schedule_work(&bio_dirty_work);
1391 } else {
1392 bio_put(bio);
1393 }
1394}
1395
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001396#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1397void bio_flush_dcache_pages(struct bio *bi)
1398{
1399 int i;
1400 struct bio_vec *bvec;
1401
1402 bio_for_each_segment(bvec, bi, i)
1403 flush_dcache_page(bvec->bv_page);
1404}
1405EXPORT_SYMBOL(bio_flush_dcache_pages);
1406#endif
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408/**
1409 * bio_endio - end I/O on a bio
1410 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 * @error: error, if any
1412 *
1413 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001414 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001415 * preferred way to end I/O on a bio, it takes care of clearing
1416 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1417 * established -Exxxx (-EIO, for instance) error values in case
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001418 * something went wrong. No one should call bi_end_io() directly on a
NeilBrown5bb23a62007-09-27 12:46:13 +02001419 * bio unless they own it and thus know that it has an end_io
1420 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001422void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423{
1424 if (error)
1425 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001426 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1427 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
NeilBrown5bb23a62007-09-27 12:46:13 +02001429 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001430 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001432EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
1434void bio_pair_release(struct bio_pair *bp)
1435{
1436 if (atomic_dec_and_test(&bp->cnt)) {
1437 struct bio *master = bp->bio1.bi_private;
1438
NeilBrown6712ecf2007-09-27 12:47:43 +02001439 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 mempool_free(bp, bp->bio2.bi_private);
1441 }
1442}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001443EXPORT_SYMBOL(bio_pair_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
NeilBrown6712ecf2007-09-27 12:47:43 +02001445static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
1447 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1448
1449 if (err)
1450 bp->error = err;
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453}
1454
NeilBrown6712ecf2007-09-27 12:47:43 +02001455static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
1457 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1458
1459 if (err)
1460 bp->error = err;
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464
1465/*
Alberto Bertoglic7eee1b2009-01-25 23:36:14 -02001466 * split a bio - only worry about a bio with a single page in its iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001468struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
Denis ChengRq6feef532008-10-09 08:57:05 +02001470 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
1472 if (!bp)
1473 return bp;
1474
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001475 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001476 bi->bi_sector + first_sectors);
1477
Shaohua Li02f39392012-09-28 10:38:48 +02001478 BUG_ON(bi->bi_vcnt != 1 && bi->bi_vcnt != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 BUG_ON(bi->bi_idx != 0);
1480 atomic_set(&bp->cnt, 3);
1481 bp->error = 0;
1482 bp->bio1 = *bi;
1483 bp->bio2 = *bi;
1484 bp->bio2.bi_sector += first_sectors;
1485 bp->bio2.bi_size -= first_sectors << 9;
1486 bp->bio1.bi_size = first_sectors << 9;
1487
Shaohua Li02f39392012-09-28 10:38:48 +02001488 if (bi->bi_vcnt != 0) {
1489 bp->bv1 = bi->bi_io_vec[0];
1490 bp->bv2 = bi->bi_io_vec[0];
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001491
Shaohua Li02f39392012-09-28 10:38:48 +02001492 if (bio_is_rw(bi)) {
1493 bp->bv2.bv_offset += first_sectors << 9;
1494 bp->bv2.bv_len -= first_sectors << 9;
1495 bp->bv1.bv_len = first_sectors << 9;
1496 }
1497
1498 bp->bio1.bi_io_vec = &bp->bv1;
1499 bp->bio2.bi_io_vec = &bp->bv2;
1500
1501 bp->bio1.bi_max_vecs = 1;
1502 bp->bio2.bi_max_vecs = 1;
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 bp->bio1.bi_end_io = bio_pair_end_1;
1506 bp->bio2.bi_end_io = bio_pair_end_2;
1507
1508 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001509 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001511 if (bio_integrity(bi))
1512 bio_integrity_split(bi, bp, first_sectors);
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 return bp;
1515}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001516EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001518/**
1519 * bio_sector_offset - Find hardware sector offset in bio
1520 * @bio: bio to inspect
1521 * @index: bio_vec index
1522 * @offset: offset in bv_page
1523 *
1524 * Return the number of hardware sectors between beginning of bio
1525 * and an end point indicated by a bio_vec index and an offset
1526 * within that vector's page.
1527 */
1528sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1529 unsigned int offset)
1530{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001531 unsigned int sector_sz;
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001532 struct bio_vec *bv;
1533 sector_t sectors;
1534 int i;
1535
Martin K. Petersene1defc42009-05-22 17:17:49 -04001536 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001537 sectors = 0;
1538
1539 if (index >= bio->bi_idx)
1540 index = bio->bi_vcnt - 1;
1541
1542 __bio_for_each_segment(bv, bio, i, 0) {
1543 if (i == index) {
1544 if (offset > bv->bv_offset)
1545 sectors += (offset - bv->bv_offset) / sector_sz;
1546 break;
1547 }
1548
1549 sectors += bv->bv_len / sector_sz;
1550 }
1551
1552 return sectors;
1553}
1554EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556/*
1557 * create memory pools for biovec's in a bio_set.
1558 * use the global biovec slabs created for general use.
1559 */
Jens Axboe59725112007-04-02 10:06:42 +02001560static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561{
Jens Axboe7ff93452008-12-11 11:53:43 +01001562 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
Jens Axboe7ff93452008-12-11 11:53:43 +01001564 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1565 if (!bs->bvec_pool)
1566 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 return 0;
1569}
1570
1571static void biovec_free_pools(struct bio_set *bs)
1572{
Jens Axboe7ff93452008-12-11 11:53:43 +01001573 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574}
1575
1576void bioset_free(struct bio_set *bs)
1577{
1578 if (bs->bio_pool)
1579 mempool_destroy(bs->bio_pool);
1580
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001581 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001583 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 kfree(bs);
1586}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001587EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
Jens Axboebb799ca2008-12-10 15:35:05 +01001589/**
1590 * bioset_create - Create a bio_set
1591 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1592 * @front_pad: Number of bytes to allocate in front of the returned bio
1593 *
1594 * Description:
1595 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1596 * to ask for a number of bytes to be allocated in front of the bio.
1597 * Front pad allocation is useful for embedding the bio inside
1598 * another structure, to avoid allocating extra data to go with the bio.
1599 * Note that the bio must be embedded at the END of that structure always,
1600 * or things will break badly.
1601 */
1602struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603{
Jens Axboe392ddc32008-12-23 12:42:54 +01001604 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001605 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
Jens Axboe1b434492008-10-22 20:32:58 +02001607 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 if (!bs)
1609 return NULL;
1610
Jens Axboebb799ca2008-12-10 15:35:05 +01001611 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001612
Jens Axboe392ddc32008-12-23 12:42:54 +01001613 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001614 if (!bs->bio_slab) {
1615 kfree(bs);
1616 return NULL;
1617 }
1618
1619 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 if (!bs->bio_pool)
1621 goto bad;
1622
Jens Axboebb799ca2008-12-10 15:35:05 +01001623 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 return bs;
1625
1626bad:
1627 bioset_free(bs);
1628 return NULL;
1629}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001630EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Tejun Heo852c7882012-03-05 13:15:27 -08001632#ifdef CONFIG_BLK_CGROUP
1633/**
1634 * bio_associate_current - associate a bio with %current
1635 * @bio: target bio
1636 *
1637 * Associate @bio with %current if it hasn't been associated yet. Block
1638 * layer will treat @bio as if it were issued by %current no matter which
1639 * task actually issues it.
1640 *
1641 * This function takes an extra reference of @task's io_context and blkcg
1642 * which will be put when @bio is released. The caller must own @bio,
1643 * ensure %current->io_context exists, and is responsible for synchronizing
1644 * calls to this function.
1645 */
1646int bio_associate_current(struct bio *bio)
1647{
1648 struct io_context *ioc;
1649 struct cgroup_subsys_state *css;
1650
1651 if (bio->bi_ioc)
1652 return -EBUSY;
1653
1654 ioc = current->io_context;
1655 if (!ioc)
1656 return -ENOENT;
1657
1658 /* acquire active ref on @ioc and associate */
1659 get_io_context_active(ioc);
1660 bio->bi_ioc = ioc;
1661
1662 /* associate blkcg if exists */
1663 rcu_read_lock();
1664 css = task_subsys_state(current, blkio_subsys_id);
1665 if (css && css_tryget(css))
1666 bio->bi_css = css;
1667 rcu_read_unlock();
1668
1669 return 0;
1670}
1671
1672/**
1673 * bio_disassociate_task - undo bio_associate_current()
1674 * @bio: target bio
1675 */
1676void bio_disassociate_task(struct bio *bio)
1677{
1678 if (bio->bi_ioc) {
1679 put_io_context(bio->bi_ioc);
1680 bio->bi_ioc = NULL;
1681 }
1682 if (bio->bi_css) {
1683 css_put(bio->bi_css);
1684 bio->bi_css = NULL;
1685 }
1686}
1687
1688#endif /* CONFIG_BLK_CGROUP */
1689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690static void __init biovec_init_slabs(void)
1691{
1692 int i;
1693
1694 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1695 int size;
1696 struct biovec_slab *bvs = bvec_slabs + i;
1697
Jens Axboea7fcd372008-12-05 16:10:29 +01001698 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1699 bvs->slab = NULL;
1700 continue;
1701 }
Jens Axboea7fcd372008-12-05 16:10:29 +01001702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 size = bvs->nr_vecs * sizeof(struct bio_vec);
1704 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001705 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 }
1707}
1708
1709static int __init init_bio(void)
1710{
Jens Axboebb799ca2008-12-10 15:35:05 +01001711 bio_slab_max = 2;
1712 bio_slab_nr = 0;
1713 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1714 if (!bio_slabs)
1715 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001717 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 biovec_init_slabs();
1719
Jens Axboebb799ca2008-12-10 15:35:05 +01001720 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 if (!fs_bio_set)
1722 panic("bio: can't allocate bios\n");
1723
Martin K. Petersena91a2782011-03-17 11:11:05 +01001724 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1725 panic("bio: can't create integrity pool\n");
1726
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001727 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1728 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 if (!bio_split_pool)
1730 panic("bio: can't create split pool\n");
1731
1732 return 0;
1733}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734subsys_initcall(init_bio);