blob: b980ecde026a3a0e1fca405053572e93ba86c44d [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>
22#include <linux/slab.h>
23#include <linux/init.h>
24#include <linux/kernel.h>
25#include <linux/module.h>
26#include <linux/mempool.h>
27#include <linux/workqueue.h>
James Bottomley f1970ba2005-06-20 14:06:52 +020028#include <scsi/sg.h> /* for struct sg_iovec */
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Li Zefan55782132009-06-09 13:43:05 +080030#include <trace/events/block.h>
Ingo Molnar0bfc2452008-11-26 11:59:56 +010031
Jens Axboe392ddc32008-12-23 12:42:54 +010032/*
33 * Test patch to inline a certain number of bi_io_vec's inside the bio
34 * itself, to shrink a bio data allocation from two mempool calls to one
35 */
36#define BIO_INLINE_VECS 4
37
Denis ChengRq6feef532008-10-09 08:57:05 +020038static mempool_t *bio_split_pool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Jens Axboebb799ca2008-12-10 15:35:05 +010057/*
58 * Our slab pool management
59 */
60struct bio_slab {
61 struct kmem_cache *slab;
62 unsigned int slab_ref;
63 unsigned int slab_size;
64 char name[8];
65};
66static DEFINE_MUTEX(bio_slab_lock);
67static struct bio_slab *bio_slabs;
68static unsigned int bio_slab_nr, bio_slab_max;
69
70static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
71{
72 unsigned int sz = sizeof(struct bio) + extra_size;
73 struct kmem_cache *slab = NULL;
74 struct bio_slab *bslab;
75 unsigned int i, entry = -1;
76
77 mutex_lock(&bio_slab_lock);
78
79 i = 0;
80 while (i < bio_slab_nr) {
Thiago Farinaf06f1352010-01-19 14:07:09 +010081 bslab = &bio_slabs[i];
Jens Axboebb799ca2008-12-10 15:35:05 +010082
83 if (!bslab->slab && entry == -1)
84 entry = i;
85 else if (bslab->slab_size == sz) {
86 slab = bslab->slab;
87 bslab->slab_ref++;
88 break;
89 }
90 i++;
91 }
92
93 if (slab)
94 goto out_unlock;
95
96 if (bio_slab_nr == bio_slab_max && entry == -1) {
97 bio_slab_max <<= 1;
98 bio_slabs = krealloc(bio_slabs,
99 bio_slab_max * sizeof(struct bio_slab),
100 GFP_KERNEL);
101 if (!bio_slabs)
102 goto out_unlock;
103 }
104 if (entry == -1)
105 entry = bio_slab_nr++;
106
107 bslab = &bio_slabs[entry];
108
109 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
110 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
111 if (!slab)
112 goto out_unlock;
113
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -0700114 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
Jens Axboebb799ca2008-12-10 15:35:05 +0100115 bslab->slab = slab;
116 bslab->slab_ref = 1;
117 bslab->slab_size = sz;
118out_unlock:
119 mutex_unlock(&bio_slab_lock);
120 return slab;
121}
122
123static void bio_put_slab(struct bio_set *bs)
124{
125 struct bio_slab *bslab = NULL;
126 unsigned int i;
127
128 mutex_lock(&bio_slab_lock);
129
130 for (i = 0; i < bio_slab_nr; i++) {
131 if (bs->bio_slab == bio_slabs[i].slab) {
132 bslab = &bio_slabs[i];
133 break;
134 }
135 }
136
137 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
138 goto out;
139
140 WARN_ON(!bslab->slab_ref);
141
142 if (--bslab->slab_ref)
143 goto out;
144
145 kmem_cache_destroy(bslab->slab);
146 bslab->slab = NULL;
147
148out:
149 mutex_unlock(&bio_slab_lock);
150}
151
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200152unsigned int bvec_nr_vecs(unsigned short idx)
153{
154 return bvec_slabs[idx].nr_vecs;
155}
156
Jens Axboebb799ca2008-12-10 15:35:05 +0100157void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
158{
159 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
160
161 if (idx == BIOVEC_MAX_IDX)
162 mempool_free(bv, bs->bvec_pool);
163 else {
164 struct biovec_slab *bvs = bvec_slabs + idx;
165
166 kmem_cache_free(bvs->slab, bv);
167 }
168}
169
Jens Axboe7ff93452008-12-11 11:53:43 +0100170struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
171 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172{
173 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100176 * see comment near bvec_array define!
177 */
178 switch (nr) {
179 case 1:
180 *idx = 0;
181 break;
182 case 2 ... 4:
183 *idx = 1;
184 break;
185 case 5 ... 16:
186 *idx = 2;
187 break;
188 case 17 ... 64:
189 *idx = 3;
190 break;
191 case 65 ... 128:
192 *idx = 4;
193 break;
194 case 129 ... BIO_MAX_PAGES:
195 *idx = 5;
196 break;
197 default:
198 return NULL;
199 }
200
201 /*
202 * idx now points to the pool we want to allocate from. only the
203 * 1-vec entry pool is mempool backed.
204 */
205 if (*idx == BIOVEC_MAX_IDX) {
206fallback:
207 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
208 } else {
209 struct biovec_slab *bvs = bvec_slabs + *idx;
210 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200212 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100213 * Make this allocation restricted and don't dump info on
214 * allocation failures, since we'll fallback to the mempool
215 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200216 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100217 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
218
219 /*
220 * Try a slab allocation. If this fails and __GFP_WAIT
221 * is set, retry with the 1-entry mempool
222 */
223 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
224 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
225 *idx = BIOVEC_MAX_IDX;
226 goto fallback;
227 }
228 }
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return bvl;
231}
232
Jens Axboe7ff93452008-12-11 11:53:43 +0100233void bio_free(struct bio *bio, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
Jens Axboebb799ca2008-12-10 15:35:05 +0100235 void *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Jens Axboe392ddc32008-12-23 12:42:54 +0100237 if (bio_has_allocated_vec(bio))
Jens Axboebb799ca2008-12-10 15:35:05 +0100238 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
Jens Axboe992c5dd2007-07-18 13:18:08 +0200239
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200240 if (bio_integrity(bio))
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200241 bio_integrity_free(bio, bs);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200242
Jens Axboebb799ca2008-12-10 15:35:05 +0100243 /*
244 * If we have front padding, adjust the bio pointer before freeing
245 */
246 p = bio;
247 if (bs->front_pad)
248 p -= bs->front_pad;
249
250 mempool_free(p, bs->bio_pool);
Peter Osterlund36763472005-09-06 15:16:42 -0700251}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200252EXPORT_SYMBOL(bio_free);
Peter Osterlund36763472005-09-06 15:16:42 -0700253
Arjan van de Ven858119e2006-01-14 13:20:43 -0800254void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
Jens Axboe2b94de52007-07-18 13:14:03 +0200256 memset(bio, 0, sizeof(*bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 bio->bi_flags = 1 << BIO_UPTODATE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 atomic_set(&bio->bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200260EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262/**
263 * bio_alloc_bioset - allocate a bio for I/O
264 * @gfp_mask: the GFP_ mask given to the slab allocator
265 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200266 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 *
268 * Description:
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200269 * bio_alloc_bioset will try its own mempool to satisfy the allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 * If %__GFP_WAIT is set then we will block on the internal pool waiting
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200271 * for a &struct bio to become free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 *
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200273 * Note that the caller must set ->bi_destructor on successful return
Jens Axboebb799ca2008-12-10 15:35:05 +0100274 * of a bio, to do the appropriate freeing of the bio once the reference
275 * count drops to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 **/
Al Virodd0fc662005-10-07 07:46:04 +0100277struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200279 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100280 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200281 struct bio *bio;
282 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200283
Tejun Heo451a9eb2009-04-15 19:50:51 +0200284 p = mempool_alloc(bs->bio_pool, gfp_mask);
285 if (unlikely(!p))
286 return NULL;
287 bio = p + bs->front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100288
289 bio_init(bio);
290
291 if (unlikely(!nr_iovecs))
292 goto out_set;
293
294 if (nr_iovecs <= BIO_INLINE_VECS) {
295 bvl = bio->bi_inline_vecs;
296 nr_iovecs = BIO_INLINE_VECS;
297 } else {
298 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
299 if (unlikely(!bvl))
300 goto err_free;
301
302 nr_iovecs = bvec_nr_vecs(idx);
303 }
Tejun Heo451a9eb2009-04-15 19:50:51 +0200304out_set:
Ingo Molnar34053972009-02-21 11:16:36 +0100305 bio->bi_flags |= idx << BIO_POOL_OFFSET;
306 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100307 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100309
310err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200311 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100312 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200314EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
Tejun Heo451a9eb2009-04-15 19:50:51 +0200316static void bio_fs_destructor(struct bio *bio)
317{
318 bio_free(bio, fs_bio_set);
319}
320
321/**
322 * bio_alloc - allocate a new bio, memory pool backed
323 * @gfp_mask: allocation mask to use
324 * @nr_iovecs: number of iovecs
325 *
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100326 * bio_alloc will allocate a bio and associated bio_vec array that can hold
327 * at least @nr_iovecs entries. Allocations will be done from the
328 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
329 *
330 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
331 * a bio. This is due to the mempool guarantees. To make this work, callers
332 * must never allocate more than 1 bio at a time from this pool. Callers
333 * that need to allocate more than 1 bio must always submit the previously
334 * allocated bio for IO before attempting to allocate a new one. Failure to
335 * do so can cause livelocks under memory pressure.
Tejun Heo451a9eb2009-04-15 19:50:51 +0200336 *
337 * RETURNS:
338 * Pointer to new bio on success, NULL on failure.
339 */
Dan Carpenter121f0992011-11-16 09:21:50 +0100340struct bio *bio_alloc(gfp_t gfp_mask, unsigned int nr_iovecs)
Tejun Heo451a9eb2009-04-15 19:50:51 +0200341{
342 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
343
344 if (bio)
345 bio->bi_destructor = bio_fs_destructor;
346
347 return bio;
348}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200349EXPORT_SYMBOL(bio_alloc);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200350
351static void bio_kmalloc_destructor(struct bio *bio)
352{
353 if (bio_integrity(bio))
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200354 bio_integrity_free(bio, fs_bio_set);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200355 kfree(bio);
356}
357
Jens Axboe86c824b2009-04-15 09:00:07 +0200358/**
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100359 * bio_kmalloc - allocate a bio for I/O using kmalloc()
Jens Axboe86c824b2009-04-15 09:00:07 +0200360 * @gfp_mask: the GFP_ mask given to the slab allocator
361 * @nr_iovecs: number of iovecs to pre-allocate
362 *
363 * Description:
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100364 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
365 * %__GFP_WAIT, the allocation is guaranteed to succeed.
Jens Axboe86c824b2009-04-15 09:00:07 +0200366 *
367 **/
Dan Carpenter121f0992011-11-16 09:21:50 +0100368struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned int nr_iovecs)
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200369{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200370 struct bio *bio;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200371
Jens Axboef3f63c12010-10-29 11:46:56 -0600372 if (nr_iovecs > UIO_MAXIOV)
373 return NULL;
374
Tejun Heo451a9eb2009-04-15 19:50:51 +0200375 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
376 gfp_mask);
377 if (unlikely(!bio))
378 return NULL;
379
380 bio_init(bio);
381 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
382 bio->bi_max_vecs = nr_iovecs;
383 bio->bi_io_vec = bio->bi_inline_vecs;
384 bio->bi_destructor = bio_kmalloc_destructor;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200385
386 return bio;
387}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200388EXPORT_SYMBOL(bio_kmalloc);
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390void zero_fill_bio(struct bio *bio)
391{
392 unsigned long flags;
393 struct bio_vec *bv;
394 int i;
395
396 bio_for_each_segment(bv, bio, i) {
397 char *data = bvec_kmap_irq(bv, &flags);
398 memset(data, 0, bv->bv_len);
399 flush_dcache_page(bv->bv_page);
400 bvec_kunmap_irq(data, &flags);
401 }
402}
403EXPORT_SYMBOL(zero_fill_bio);
404
405/**
406 * bio_put - release a reference to a bio
407 * @bio: bio to release reference to
408 *
409 * Description:
410 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100411 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 **/
413void bio_put(struct bio *bio)
414{
415 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
416
417 /*
418 * last put frees it
419 */
420 if (atomic_dec_and_test(&bio->bi_cnt)) {
421 bio->bi_next = NULL;
422 bio->bi_destructor(bio);
423 }
424}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200425EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
Jens Axboe165125e2007-07-24 09:28:11 +0200427inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
429 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
430 blk_recount_segments(q, bio);
431
432 return bio->bi_phys_segments;
433}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200434EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/**
437 * __bio_clone - clone a bio
438 * @bio: destination bio
439 * @bio_src: bio to clone
440 *
441 * Clone a &bio. Caller will own the returned bio, but not
442 * the actual data it points to. Reference count of returned
443 * bio will be one.
444 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800445void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
Andrew Mortone525e152005-08-07 09:42:12 -0700447 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
448 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Jens Axboe5d840702008-01-25 12:44:44 +0100450 /*
451 * most users will be overriding ->bi_bdev with a new target,
452 * so we don't set nor calculate new physical/hw segment counts here
453 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 bio->bi_sector = bio_src->bi_sector;
455 bio->bi_bdev = bio_src->bi_bdev;
456 bio->bi_flags |= 1 << BIO_CLONED;
457 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 bio->bi_vcnt = bio_src->bi_vcnt;
459 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700460 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200462EXPORT_SYMBOL(__bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464/**
465 * bio_clone - clone a bio
466 * @bio: bio to clone
467 * @gfp_mask: allocation priority
468 *
469 * Like __bio_clone, only also allocates the returned bio
470 */
Al Virodd0fc662005-10-07 07:46:04 +0100471struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
473 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
474
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200475 if (!b)
476 return NULL;
477
478 b->bi_destructor = bio_fs_destructor;
479 __bio_clone(b, bio);
480
481 if (bio_integrity(bio)) {
482 int ret;
483
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200484 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200485
Li Zefan059ea332009-03-09 10:42:45 +0100486 if (ret < 0) {
487 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200488 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100489 }
Peter Osterlund36763472005-09-06 15:16:42 -0700490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
492 return b;
493}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200494EXPORT_SYMBOL(bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496/**
497 * bio_get_nr_vecs - return approx number of vecs
498 * @bdev: I/O target
499 *
500 * Return the approximate number of pages we can send to this target.
501 * There's no guarantee that you will be able to fit this number of pages
502 * into a bio, it does not account for dynamic restrictions that vary
503 * on offset.
504 */
505int bio_get_nr_vecs(struct block_device *bdev)
506{
Jens Axboe165125e2007-07-24 09:28:11 +0200507 struct request_queue *q = bdev_get_queue(bdev);
Kent Overstreet5abebfd2012-02-08 22:07:18 +0100508 return min_t(unsigned,
509 queue_max_segments(q),
510 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200512EXPORT_SYMBOL(bio_get_nr_vecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Jens Axboe165125e2007-07-24 09:28:11 +0200514static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600515 *page, unsigned int len, unsigned int offset,
516 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
518 int retried_segments = 0;
519 struct bio_vec *bvec;
520
521 /*
522 * cloned bio must not modify vec list
523 */
524 if (unlikely(bio_flagged(bio, BIO_CLONED)))
525 return 0;
526
Jens Axboe80cfd542006-01-06 09:43:28 +0100527 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 return 0;
529
Jens Axboe80cfd542006-01-06 09:43:28 +0100530 /*
531 * For filesystems with a blocksize smaller than the pagesize
532 * we will often be called with the same page as last time and
533 * a consecutive offset. Optimize this special case.
534 */
535 if (bio->bi_vcnt > 0) {
536 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
537
538 if (page == prev->bv_page &&
539 offset == prev->bv_offset + prev->bv_len) {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300540 unsigned int prev_bv_len = prev->bv_len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100541 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200542
543 if (q->merge_bvec_fn) {
544 struct bvec_merge_data bvm = {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300545 /* prev_bvec is already charged in
546 bi_size, discharge it in order to
547 simulate merging updated prev_bvec
548 as new bvec. */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200549 .bi_bdev = bio->bi_bdev,
550 .bi_sector = bio->bi_sector,
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300551 .bi_size = bio->bi_size - prev_bv_len,
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200552 .bi_rw = bio->bi_rw,
553 };
554
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300555 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200556 prev->bv_len -= len;
557 return 0;
558 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100559 }
560
561 goto done;
562 }
563 }
564
565 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 return 0;
567
568 /*
569 * we might lose a segment or two here, but rather that than
570 * make this too complex.
571 */
572
Martin K. Petersen8a783622010-02-26 00:20:39 -0500573 while (bio->bi_phys_segments >= queue_max_segments(q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 if (retried_segments)
576 return 0;
577
578 retried_segments = 1;
579 blk_recount_segments(q, bio);
580 }
581
582 /*
583 * setup the new entry, we might clear it again later if we
584 * cannot add the page
585 */
586 bvec = &bio->bi_io_vec[bio->bi_vcnt];
587 bvec->bv_page = page;
588 bvec->bv_len = len;
589 bvec->bv_offset = offset;
590
591 /*
592 * if queue has other restrictions (eg varying max sector size
593 * depending on offset), it can specify a merge_bvec_fn in the
594 * queue to get further control
595 */
596 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200597 struct bvec_merge_data bvm = {
598 .bi_bdev = bio->bi_bdev,
599 .bi_sector = bio->bi_sector,
600 .bi_size = bio->bi_size,
601 .bi_rw = bio->bi_rw,
602 };
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 /*
605 * merge_bvec_fn() returns number of bytes it can accept
606 * at this offset
607 */
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300608 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 bvec->bv_page = NULL;
610 bvec->bv_len = 0;
611 bvec->bv_offset = 0;
612 return 0;
613 }
614 }
615
616 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200617 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
619
620 bio->bi_vcnt++;
621 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100622 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 bio->bi_size += len;
624 return len;
625}
626
627/**
Mike Christie6e68af62005-11-11 05:30:27 -0600628 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100629 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600630 * @bio: destination bio
631 * @page: page to add
632 * @len: vec entry length
633 * @offset: vec entry offset
634 *
635 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200636 * number of reasons, such as the bio being full or target block device
637 * limitations. The target block device must allow bio's up to PAGE_SIZE,
638 * so it is always possible to add a single page to an empty bio.
639 *
640 * This should only be used by REQ_PC bios.
Mike Christie6e68af62005-11-11 05:30:27 -0600641 */
Jens Axboe165125e2007-07-24 09:28:11 +0200642int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600643 unsigned int len, unsigned int offset)
644{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400645 return __bio_add_page(q, bio, page, len, offset,
646 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600647}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200648EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600649
650/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 * bio_add_page - attempt to add page to bio
652 * @bio: destination bio
653 * @page: page to add
654 * @len: vec entry length
655 * @offset: vec entry offset
656 *
657 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200658 * number of reasons, such as the bio being full or target block device
659 * limitations. The target block device must allow bio's up to PAGE_SIZE,
660 * so it is always possible to add a single page to an empty bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 */
662int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
663 unsigned int offset)
664{
Mike Christiedefd94b2005-12-05 02:37:06 -0600665 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400666 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200668EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
670struct bio_map_data {
671 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200672 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900673 int nr_sgvecs;
674 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675};
676
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200677static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900678 struct sg_iovec *iov, int iov_count,
679 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200682 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
683 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900684 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 bio->bi_private = bmd;
686}
687
688static void bio_free_map_data(struct bio_map_data *bmd)
689{
690 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200691 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 kfree(bmd);
693}
694
Dan Carpenter121f0992011-11-16 09:21:50 +0100695static struct bio_map_data *bio_alloc_map_data(int nr_segs,
696 unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200697 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Jens Axboef3f63c12010-10-29 11:46:56 -0600699 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Jens Axboef3f63c12010-10-29 11:46:56 -0600701 if (iov_count > UIO_MAXIOV)
702 return NULL;
703
704 bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 if (!bmd)
706 return NULL;
707
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200708 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200709 if (!bmd->iovecs) {
710 kfree(bmd);
711 return NULL;
712 }
713
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200714 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200715 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 return bmd;
717
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200718 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 kfree(bmd);
720 return NULL;
721}
722
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200723static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200724 struct sg_iovec *iov, int iov_count,
725 int to_user, int from_user, int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200726{
727 int ret = 0, i;
728 struct bio_vec *bvec;
729 int iov_idx = 0;
730 unsigned int iov_off = 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200731
732 __bio_for_each_segment(bvec, bio, i, 0) {
733 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200734 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200735
736 while (bv_len && iov_idx < iov_count) {
737 unsigned int bytes;
Michal Simek0e0c6212009-06-10 12:57:07 -0700738 char __user *iov_addr;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200739
740 bytes = min_t(unsigned int,
741 iov[iov_idx].iov_len - iov_off, bv_len);
742 iov_addr = iov[iov_idx].iov_base + iov_off;
743
744 if (!ret) {
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200745 if (to_user)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200746 ret = copy_to_user(iov_addr, bv_addr,
747 bytes);
748
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200749 if (from_user)
750 ret = copy_from_user(bv_addr, iov_addr,
751 bytes);
752
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200753 if (ret)
754 ret = -EFAULT;
755 }
756
757 bv_len -= bytes;
758 bv_addr += bytes;
759 iov_addr += bytes;
760 iov_off += bytes;
761
762 if (iov[iov_idx].iov_len == iov_off) {
763 iov_idx++;
764 iov_off = 0;
765 }
766 }
767
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900768 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200769 __free_page(bvec->bv_page);
770 }
771
772 return ret;
773}
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775/**
776 * bio_uncopy_user - finish previously mapped bio
777 * @bio: bio being terminated
778 *
779 * Free pages allocated from bio_copy_user() and write back data
780 * to user space in case of a read.
781 */
782int bio_uncopy_user(struct bio *bio)
783{
784 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900785 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
FUJITA Tomonori81882762008-09-02 16:20:19 +0900787 if (!bio_flagged(bio, BIO_NULL_MAPPED))
788 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200789 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
790 0, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 bio_free_map_data(bmd);
792 bio_put(bio);
793 return ret;
794}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200795EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
797/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200798 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900800 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200801 * @iov: the iovec.
802 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900804 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 *
806 * Prepares and returns a bio for indirect user io, bouncing data
807 * to/from kernel pages as necessary. Must be paired with
808 * call bio_uncopy_user() on io completion.
809 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900810struct bio *bio_copy_user_iov(struct request_queue *q,
811 struct rq_map_data *map_data,
812 struct sg_iovec *iov, int iov_count,
813 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 struct bio_map_data *bmd;
816 struct bio_vec *bvec;
817 struct page *page;
818 struct bio *bio;
819 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200820 int nr_pages = 0;
821 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900822 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200824 for (i = 0; i < iov_count; i++) {
825 unsigned long uaddr;
826 unsigned long end;
827 unsigned long start;
828
829 uaddr = (unsigned long)iov[i].iov_base;
830 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
831 start = uaddr >> PAGE_SHIFT;
832
Jens Axboecb4644c2010-11-10 14:36:25 +0100833 /*
834 * Overflow, abort
835 */
836 if (end < start)
837 return ERR_PTR(-EINVAL);
838
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200839 nr_pages += end - start;
840 len += iov[i].iov_len;
841 }
842
FUJITA Tomonori69838722009-04-28 20:24:29 +0200843 if (offset)
844 nr_pages++;
845
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900846 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 if (!bmd)
848 return ERR_PTR(-ENOMEM);
849
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900851 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (!bio)
853 goto out_bmd;
854
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200855 if (!write_to_vm)
856 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900859
860 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900861 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900862 i = map_data->offset / PAGE_SIZE;
863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900865 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900867 bytes -= offset;
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 if (bytes > len)
870 bytes = len;
871
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900872 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900873 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900874 ret = -ENOMEM;
875 break;
876 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900877
878 page = map_data->pages[i / nr_pages];
879 page += (i % nr_pages);
880
881 i++;
882 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900883 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900884 if (!page) {
885 ret = -ENOMEM;
886 break;
887 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 }
889
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900890 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
893 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900894 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 }
896
897 if (ret)
898 goto cleanup;
899
900 /*
901 * success
902 */
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200903 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
904 (map_data && map_data->from_user)) {
905 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200906 if (ret)
907 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 }
909
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900910 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 return bio;
912cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900913 if (!map_data)
914 bio_for_each_segment(bvec, bio, i)
915 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 bio_put(bio);
918out_bmd:
919 bio_free_map_data(bmd);
920 return ERR_PTR(ret);
921}
922
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200923/**
924 * bio_copy_user - copy user data to bio
925 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900926 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200927 * @uaddr: start of user address
928 * @len: length in bytes
929 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900930 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200931 *
932 * Prepares and returns a bio for indirect user io, bouncing data
933 * to/from kernel pages as necessary. Must be paired with
934 * call bio_uncopy_user() on io completion.
935 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900936struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
937 unsigned long uaddr, unsigned int len,
938 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200939{
940 struct sg_iovec iov;
941
942 iov.iov_base = (void __user *)uaddr;
943 iov.iov_len = len;
944
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900945 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200946}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200947EXPORT_SYMBOL(bio_copy_user);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200948
Jens Axboe165125e2007-07-24 09:28:11 +0200949static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200950 struct block_device *bdev,
951 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900952 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
James Bottomley f1970ba2005-06-20 14:06:52 +0200954 int i, j;
955 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 struct page **pages;
957 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200958 int cur_page = 0;
959 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
James Bottomley f1970ba2005-06-20 14:06:52 +0200961 for (i = 0; i < iov_count; i++) {
962 unsigned long uaddr = (unsigned long)iov[i].iov_base;
963 unsigned long len = iov[i].iov_len;
964 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
965 unsigned long start = uaddr >> PAGE_SHIFT;
966
Jens Axboecb4644c2010-11-10 14:36:25 +0100967 /*
968 * Overflow, abort
969 */
970 if (end < start)
971 return ERR_PTR(-EINVAL);
972
James Bottomley f1970ba2005-06-20 14:06:52 +0200973 nr_pages += end - start;
974 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100975 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200976 */
Mike Christiead2d7222006-12-01 10:40:20 +0100977 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200978 return ERR_PTR(-EINVAL);
979 }
980
981 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 return ERR_PTR(-EINVAL);
983
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900984 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 if (!bio)
986 return ERR_PTR(-ENOMEM);
987
988 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900989 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 if (!pages)
991 goto out;
992
James Bottomley f1970ba2005-06-20 14:06:52 +0200993 for (i = 0; i < iov_count; i++) {
994 unsigned long uaddr = (unsigned long)iov[i].iov_base;
995 unsigned long len = iov[i].iov_len;
996 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
997 unsigned long start = uaddr >> PAGE_SHIFT;
998 const int local_nr_pages = end - start;
999 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001000
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001001 ret = get_user_pages_fast(uaddr, local_nr_pages,
1002 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +02001003 if (ret < local_nr_pages) {
1004 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001005 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
James Bottomley f1970ba2005-06-20 14:06:52 +02001008 offset = uaddr & ~PAGE_MASK;
1009 for (j = cur_page; j < page_limit; j++) {
1010 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
James Bottomley f1970ba2005-06-20 14:06:52 +02001012 if (len <= 0)
1013 break;
1014
1015 if (bytes > len)
1016 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
James Bottomley f1970ba2005-06-20 14:06:52 +02001018 /*
1019 * sorry...
1020 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001021 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1022 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001023 break;
1024
1025 len -= bytes;
1026 offset = 0;
1027 }
1028
1029 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001031 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001033 while (j < page_limit)
1034 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
1036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 kfree(pages);
1038
1039 /*
1040 * set data direction, and check if mapped pages need bouncing
1041 */
1042 if (!write_to_vm)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001043 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
James Bottomley f1970ba2005-06-20 14:06:52 +02001045 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1047 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001048
1049 out_unmap:
1050 for (i = 0; i < nr_pages; i++) {
1051 if(!pages[i])
1052 break;
1053 page_cache_release(pages[i]);
1054 }
1055 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 kfree(pages);
1057 bio_put(bio);
1058 return ERR_PTR(ret);
1059}
1060
1061/**
1062 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001063 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 * @bdev: destination block device
1065 * @uaddr: start of user address
1066 * @len: length in bytes
1067 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001068 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 *
1070 * Map the user space address into a bio suitable for io to a block
1071 * device. Returns an error pointer in case of error.
1072 */
Jens Axboe165125e2007-07-24 09:28:11 +02001073struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001074 unsigned long uaddr, unsigned int len, int write_to_vm,
1075 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076{
James Bottomley f1970ba2005-06-20 14:06:52 +02001077 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001079 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001080 iov.iov_len = len;
1081
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001082 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001083}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001084EXPORT_SYMBOL(bio_map_user);
James Bottomley f1970ba2005-06-20 14:06:52 +02001085
1086/**
1087 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001088 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001089 * @bdev: destination block device
1090 * @iov: the iovec.
1091 * @iov_count: number of elements in the iovec
1092 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001093 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001094 *
1095 * Map the user space address into a bio suitable for io to a block
1096 * device. Returns an error pointer in case of error.
1097 */
Jens Axboe165125e2007-07-24 09:28:11 +02001098struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001099 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001100 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001101{
1102 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001103
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001104 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1105 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 if (IS_ERR(bio))
1107 return bio;
1108
1109 /*
1110 * subtle -- if __bio_map_user() ended up bouncing a bio,
1111 * it would normally disappear when its bi_end_io is run.
1112 * however, we need it for the unmap, so grab an extra
1113 * reference to it
1114 */
1115 bio_get(bio);
1116
Mike Christie0e75f902006-12-01 10:40:55 +01001117 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
1120static void __bio_unmap_user(struct bio *bio)
1121{
1122 struct bio_vec *bvec;
1123 int i;
1124
1125 /*
1126 * make sure we dirty pages we wrote to
1127 */
1128 __bio_for_each_segment(bvec, bio, i, 0) {
1129 if (bio_data_dir(bio) == READ)
1130 set_page_dirty_lock(bvec->bv_page);
1131
1132 page_cache_release(bvec->bv_page);
1133 }
1134
1135 bio_put(bio);
1136}
1137
1138/**
1139 * bio_unmap_user - unmap a bio
1140 * @bio: the bio being unmapped
1141 *
1142 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1143 * a process context.
1144 *
1145 * bio_unmap_user() may sleep.
1146 */
1147void bio_unmap_user(struct bio *bio)
1148{
1149 __bio_unmap_user(bio);
1150 bio_put(bio);
1151}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001152EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
NeilBrown6712ecf2007-09-27 12:47:43 +02001154static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001155{
Jens Axboeb8238252005-06-20 14:05:27 +02001156 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001157}
1158
Jens Axboe165125e2007-07-24 09:28:11 +02001159static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001160 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001161{
1162 unsigned long kaddr = (unsigned long)data;
1163 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1164 unsigned long start = kaddr >> PAGE_SHIFT;
1165 const int nr_pages = end - start;
1166 int offset, i;
1167 struct bio *bio;
1168
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001169 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001170 if (!bio)
1171 return ERR_PTR(-ENOMEM);
1172
1173 offset = offset_in_page(kaddr);
1174 for (i = 0; i < nr_pages; i++) {
1175 unsigned int bytes = PAGE_SIZE - offset;
1176
1177 if (len <= 0)
1178 break;
1179
1180 if (bytes > len)
1181 bytes = len;
1182
Mike Christiedefd94b2005-12-05 02:37:06 -06001183 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1184 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001185 break;
1186
1187 data += bytes;
1188 len -= bytes;
1189 offset = 0;
1190 }
1191
Jens Axboeb8238252005-06-20 14:05:27 +02001192 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001193 return bio;
1194}
1195
1196/**
1197 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001198 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001199 * @data: pointer to buffer to map
1200 * @len: length in bytes
1201 * @gfp_mask: allocation flags for bio allocation
1202 *
1203 * Map the kernel address into a bio suitable for io to a block
1204 * device. Returns an error pointer in case of error.
1205 */
Jens Axboe165125e2007-07-24 09:28:11 +02001206struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001207 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001208{
1209 struct bio *bio;
1210
1211 bio = __bio_map_kern(q, data, len, gfp_mask);
1212 if (IS_ERR(bio))
1213 return bio;
1214
1215 if (bio->bi_size == len)
1216 return bio;
1217
1218 /*
1219 * Don't support partial mappings.
1220 */
1221 bio_put(bio);
1222 return ERR_PTR(-EINVAL);
1223}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001224EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001225
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001226static void bio_copy_kern_endio(struct bio *bio, int err)
1227{
1228 struct bio_vec *bvec;
1229 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001230 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001231 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001232 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001233
1234 __bio_for_each_segment(bvec, bio, i, 0) {
1235 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001236 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001237
Tejun Heo4fc981e2009-05-19 18:33:06 +09001238 if (read)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001239 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001240
1241 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001242 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001243 }
1244
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001245 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001246 bio_put(bio);
1247}
1248
1249/**
1250 * bio_copy_kern - copy kernel address into bio
1251 * @q: the struct request_queue for the bio
1252 * @data: pointer to buffer to copy
1253 * @len: length in bytes
1254 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001255 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001256 *
1257 * copy the kernel address into a bio suitable for io to a block
1258 * device. Returns an error pointer in case of error.
1259 */
1260struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1261 gfp_t gfp_mask, int reading)
1262{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001263 struct bio *bio;
1264 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001265 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001266
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001267 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1268 if (IS_ERR(bio))
1269 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001270
1271 if (!reading) {
1272 void *p = data;
1273
1274 bio_for_each_segment(bvec, bio, i) {
1275 char *addr = page_address(bvec->bv_page);
1276
1277 memcpy(addr, p, bvec->bv_len);
1278 p += bvec->bv_len;
1279 }
1280 }
1281
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001282 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001283
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001284 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001285}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001286EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288/*
1289 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1290 * for performing direct-IO in BIOs.
1291 *
1292 * The problem is that we cannot run set_page_dirty() from interrupt context
1293 * because the required locks are not interrupt-safe. So what we can do is to
1294 * mark the pages dirty _before_ performing IO. And in interrupt context,
1295 * check that the pages are still dirty. If so, fine. If not, redirty them
1296 * in process context.
1297 *
1298 * We special-case compound pages here: normally this means reads into hugetlb
1299 * pages. The logic in here doesn't really work right for compound pages
1300 * because the VM does not uniformly chase down the head page in all cases.
1301 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1302 * handle them at all. So we skip compound pages here at an early stage.
1303 *
1304 * Note that this code is very hard to test under normal circumstances because
1305 * direct-io pins the pages with get_user_pages(). This makes
1306 * is_page_cache_freeable return false, and the VM will not clean the pages.
1307 * But other code (eg, pdflush) could clean the pages if they are mapped
1308 * pagecache.
1309 *
1310 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1311 * deferred bio dirtying paths.
1312 */
1313
1314/*
1315 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1316 */
1317void bio_set_pages_dirty(struct bio *bio)
1318{
1319 struct bio_vec *bvec = bio->bi_io_vec;
1320 int i;
1321
1322 for (i = 0; i < bio->bi_vcnt; i++) {
1323 struct page *page = bvec[i].bv_page;
1324
1325 if (page && !PageCompound(page))
1326 set_page_dirty_lock(page);
1327 }
1328}
1329
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001330static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
1332 struct bio_vec *bvec = bio->bi_io_vec;
1333 int i;
1334
1335 for (i = 0; i < bio->bi_vcnt; i++) {
1336 struct page *page = bvec[i].bv_page;
1337
1338 if (page)
1339 put_page(page);
1340 }
1341}
1342
1343/*
1344 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1345 * If they are, then fine. If, however, some pages are clean then they must
1346 * have been written out during the direct-IO read. So we take another ref on
1347 * the BIO and the offending pages and re-dirty the pages in process context.
1348 *
1349 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1350 * here on. It will run one page_cache_release() against each page and will
1351 * run one bio_put() against the BIO.
1352 */
1353
David Howells65f27f32006-11-22 14:55:48 +00001354static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
David Howells65f27f32006-11-22 14:55:48 +00001356static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357static DEFINE_SPINLOCK(bio_dirty_lock);
1358static struct bio *bio_dirty_list;
1359
1360/*
1361 * This runs in process context
1362 */
David Howells65f27f32006-11-22 14:55:48 +00001363static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
1365 unsigned long flags;
1366 struct bio *bio;
1367
1368 spin_lock_irqsave(&bio_dirty_lock, flags);
1369 bio = bio_dirty_list;
1370 bio_dirty_list = NULL;
1371 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1372
1373 while (bio) {
1374 struct bio *next = bio->bi_private;
1375
1376 bio_set_pages_dirty(bio);
1377 bio_release_pages(bio);
1378 bio_put(bio);
1379 bio = next;
1380 }
1381}
1382
1383void bio_check_pages_dirty(struct bio *bio)
1384{
1385 struct bio_vec *bvec = bio->bi_io_vec;
1386 int nr_clean_pages = 0;
1387 int i;
1388
1389 for (i = 0; i < bio->bi_vcnt; i++) {
1390 struct page *page = bvec[i].bv_page;
1391
1392 if (PageDirty(page) || PageCompound(page)) {
1393 page_cache_release(page);
1394 bvec[i].bv_page = NULL;
1395 } else {
1396 nr_clean_pages++;
1397 }
1398 }
1399
1400 if (nr_clean_pages) {
1401 unsigned long flags;
1402
1403 spin_lock_irqsave(&bio_dirty_lock, flags);
1404 bio->bi_private = bio_dirty_list;
1405 bio_dirty_list = bio;
1406 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1407 schedule_work(&bio_dirty_work);
1408 } else {
1409 bio_put(bio);
1410 }
1411}
1412
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001413#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1414void bio_flush_dcache_pages(struct bio *bi)
1415{
1416 int i;
1417 struct bio_vec *bvec;
1418
1419 bio_for_each_segment(bvec, bi, i)
1420 flush_dcache_page(bvec->bv_page);
1421}
1422EXPORT_SYMBOL(bio_flush_dcache_pages);
1423#endif
1424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425/**
1426 * bio_endio - end I/O on a bio
1427 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 * @error: error, if any
1429 *
1430 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001431 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001432 * preferred way to end I/O on a bio, it takes care of clearing
1433 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1434 * established -Exxxx (-EIO, for instance) error values in case
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001435 * something went wrong. No one should call bi_end_io() directly on a
NeilBrown5bb23a62007-09-27 12:46:13 +02001436 * bio unless they own it and thus know that it has an end_io
1437 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001439void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440{
1441 if (error)
1442 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001443 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1444 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
NeilBrown5bb23a62007-09-27 12:46:13 +02001446 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001447 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001449EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
1451void bio_pair_release(struct bio_pair *bp)
1452{
1453 if (atomic_dec_and_test(&bp->cnt)) {
1454 struct bio *master = bp->bio1.bi_private;
1455
NeilBrown6712ecf2007-09-27 12:47:43 +02001456 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 mempool_free(bp, bp->bio2.bi_private);
1458 }
1459}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001460EXPORT_SYMBOL(bio_pair_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
NeilBrown6712ecf2007-09-27 12:47:43 +02001462static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463{
1464 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1465
1466 if (err)
1467 bp->error = err;
1468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
NeilBrown6712ecf2007-09-27 12:47:43 +02001472static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
1474 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1475
1476 if (err)
1477 bp->error = err;
1478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480}
1481
1482/*
Alberto Bertoglic7eee1b2009-01-25 23:36:14 -02001483 * split a bio - only worry about a bio with a single page in its iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001485struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486{
Denis ChengRq6feef532008-10-09 08:57:05 +02001487 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
1489 if (!bp)
1490 return bp;
1491
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001492 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001493 bi->bi_sector + first_sectors);
1494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 BUG_ON(bi->bi_vcnt != 1);
1496 BUG_ON(bi->bi_idx != 0);
1497 atomic_set(&bp->cnt, 3);
1498 bp->error = 0;
1499 bp->bio1 = *bi;
1500 bp->bio2 = *bi;
1501 bp->bio2.bi_sector += first_sectors;
1502 bp->bio2.bi_size -= first_sectors << 9;
1503 bp->bio1.bi_size = first_sectors << 9;
1504
1505 bp->bv1 = bi->bi_io_vec[0];
1506 bp->bv2 = bi->bi_io_vec[0];
1507 bp->bv2.bv_offset += first_sectors << 9;
1508 bp->bv2.bv_len -= first_sectors << 9;
1509 bp->bv1.bv_len = first_sectors << 9;
1510
1511 bp->bio1.bi_io_vec = &bp->bv1;
1512 bp->bio2.bi_io_vec = &bp->bv2;
1513
NeilBrowna2eb0c12006-05-22 22:35:27 -07001514 bp->bio1.bi_max_vecs = 1;
1515 bp->bio2.bi_max_vecs = 1;
1516
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 bp->bio1.bi_end_io = bio_pair_end_1;
1518 bp->bio2.bi_end_io = bio_pair_end_2;
1519
1520 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001521 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001523 if (bio_integrity(bi))
1524 bio_integrity_split(bi, bp, first_sectors);
1525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 return bp;
1527}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001528EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001530/**
1531 * bio_sector_offset - Find hardware sector offset in bio
1532 * @bio: bio to inspect
1533 * @index: bio_vec index
1534 * @offset: offset in bv_page
1535 *
1536 * Return the number of hardware sectors between beginning of bio
1537 * and an end point indicated by a bio_vec index and an offset
1538 * within that vector's page.
1539 */
1540sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1541 unsigned int offset)
1542{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001543 unsigned int sector_sz;
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001544 struct bio_vec *bv;
1545 sector_t sectors;
1546 int i;
1547
Martin K. Petersene1defc42009-05-22 17:17:49 -04001548 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001549 sectors = 0;
1550
1551 if (index >= bio->bi_idx)
1552 index = bio->bi_vcnt - 1;
1553
1554 __bio_for_each_segment(bv, bio, i, 0) {
1555 if (i == index) {
1556 if (offset > bv->bv_offset)
1557 sectors += (offset - bv->bv_offset) / sector_sz;
1558 break;
1559 }
1560
1561 sectors += bv->bv_len / sector_sz;
1562 }
1563
1564 return sectors;
1565}
1566EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568/*
1569 * create memory pools for biovec's in a bio_set.
1570 * use the global biovec slabs created for general use.
1571 */
Jens Axboe59725112007-04-02 10:06:42 +02001572static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
Jens Axboe7ff93452008-12-11 11:53:43 +01001574 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Jens Axboe7ff93452008-12-11 11:53:43 +01001576 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1577 if (!bs->bvec_pool)
1578 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 return 0;
1581}
1582
1583static void biovec_free_pools(struct bio_set *bs)
1584{
Jens Axboe7ff93452008-12-11 11:53:43 +01001585 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586}
1587
1588void bioset_free(struct bio_set *bs)
1589{
1590 if (bs->bio_pool)
1591 mempool_destroy(bs->bio_pool);
1592
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001593 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001595 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597 kfree(bs);
1598}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001599EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Jens Axboebb799ca2008-12-10 15:35:05 +01001601/**
1602 * bioset_create - Create a bio_set
1603 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1604 * @front_pad: Number of bytes to allocate in front of the returned bio
1605 *
1606 * Description:
1607 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1608 * to ask for a number of bytes to be allocated in front of the bio.
1609 * Front pad allocation is useful for embedding the bio inside
1610 * another structure, to avoid allocating extra data to go with the bio.
1611 * Note that the bio must be embedded at the END of that structure always,
1612 * or things will break badly.
1613 */
1614struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615{
Jens Axboe392ddc32008-12-23 12:42:54 +01001616 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001617 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
Jens Axboe1b434492008-10-22 20:32:58 +02001619 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 if (!bs)
1621 return NULL;
1622
Jens Axboebb799ca2008-12-10 15:35:05 +01001623 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001624
Jens Axboe392ddc32008-12-23 12:42:54 +01001625 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001626 if (!bs->bio_slab) {
1627 kfree(bs);
1628 return NULL;
1629 }
1630
1631 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if (!bs->bio_pool)
1633 goto bad;
1634
Jens Axboebb799ca2008-12-10 15:35:05 +01001635 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 return bs;
1637
1638bad:
1639 bioset_free(bs);
1640 return NULL;
1641}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001642EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644static void __init biovec_init_slabs(void)
1645{
1646 int i;
1647
1648 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1649 int size;
1650 struct biovec_slab *bvs = bvec_slabs + i;
1651
Jens Axboea7fcd372008-12-05 16:10:29 +01001652 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1653 bvs->slab = NULL;
1654 continue;
1655 }
Jens Axboea7fcd372008-12-05 16:10:29 +01001656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 size = bvs->nr_vecs * sizeof(struct bio_vec);
1658 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001659 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 }
1661}
1662
1663static int __init init_bio(void)
1664{
Jens Axboebb799ca2008-12-10 15:35:05 +01001665 bio_slab_max = 2;
1666 bio_slab_nr = 0;
1667 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1668 if (!bio_slabs)
1669 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001671 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 biovec_init_slabs();
1673
Jens Axboebb799ca2008-12-10 15:35:05 +01001674 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 if (!fs_bio_set)
1676 panic("bio: can't allocate bios\n");
1677
Martin K. Petersena91a2782011-03-17 11:11:05 +01001678 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1679 panic("bio: can't create integrity pool\n");
1680
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001681 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1682 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 if (!bio_split_pool)
1684 panic("bio: can't create split pool\n");
1685
1686 return 0;
1687}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688subsys_initcall(init_bio);