blob: 9bfade8a609bfa33c9f78b9f2faa9d01dfea9176 [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;
Jens Axboec7c22e42008-09-13 20:26:01 +0200258 bio->bi_comp_cpu = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 atomic_set(&bio->bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200261EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263/**
264 * bio_alloc_bioset - allocate a bio for I/O
265 * @gfp_mask: the GFP_ mask given to the slab allocator
266 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200267 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 *
269 * Description:
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200270 * bio_alloc_bioset will try its own mempool to satisfy the allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 * If %__GFP_WAIT is set then we will block on the internal pool waiting
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200272 * for a &struct bio to become free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 *
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200274 * Note that the caller must set ->bi_destructor on successful return
Jens Axboebb799ca2008-12-10 15:35:05 +0100275 * of a bio, to do the appropriate freeing of the bio once the reference
276 * count drops to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 **/
Al Virodd0fc662005-10-07 07:46:04 +0100278struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200280 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100281 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200282 struct bio *bio;
283 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200284
Tejun Heo451a9eb2009-04-15 19:50:51 +0200285 p = mempool_alloc(bs->bio_pool, gfp_mask);
286 if (unlikely(!p))
287 return NULL;
288 bio = p + bs->front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100289
290 bio_init(bio);
291
292 if (unlikely(!nr_iovecs))
293 goto out_set;
294
295 if (nr_iovecs <= BIO_INLINE_VECS) {
296 bvl = bio->bi_inline_vecs;
297 nr_iovecs = BIO_INLINE_VECS;
298 } else {
299 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
300 if (unlikely(!bvl))
301 goto err_free;
302
303 nr_iovecs = bvec_nr_vecs(idx);
304 }
Tejun Heo451a9eb2009-04-15 19:50:51 +0200305out_set:
Ingo Molnar34053972009-02-21 11:16:36 +0100306 bio->bi_flags |= idx << BIO_POOL_OFFSET;
307 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100308 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100310
311err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200312 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100313 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200315EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Tejun Heo451a9eb2009-04-15 19:50:51 +0200317static void bio_fs_destructor(struct bio *bio)
318{
319 bio_free(bio, fs_bio_set);
320}
321
322/**
323 * bio_alloc - allocate a new bio, memory pool backed
324 * @gfp_mask: allocation mask to use
325 * @nr_iovecs: number of iovecs
326 *
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100327 * bio_alloc will allocate a bio and associated bio_vec array that can hold
328 * at least @nr_iovecs entries. Allocations will be done from the
329 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
330 *
331 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
332 * a bio. This is due to the mempool guarantees. To make this work, callers
333 * must never allocate more than 1 bio at a time from this pool. Callers
334 * that need to allocate more than 1 bio must always submit the previously
335 * allocated bio for IO before attempting to allocate a new one. Failure to
336 * do so can cause livelocks under memory pressure.
Tejun Heo451a9eb2009-04-15 19:50:51 +0200337 *
338 * RETURNS:
339 * Pointer to new bio on success, NULL on failure.
340 */
341struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
342{
343 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
344
345 if (bio)
346 bio->bi_destructor = bio_fs_destructor;
347
348 return bio;
349}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200350EXPORT_SYMBOL(bio_alloc);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200351
352static void bio_kmalloc_destructor(struct bio *bio)
353{
354 if (bio_integrity(bio))
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200355 bio_integrity_free(bio, fs_bio_set);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200356 kfree(bio);
357}
358
Jens Axboe86c824b2009-04-15 09:00:07 +0200359/**
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100360 * bio_kmalloc - allocate a bio for I/O using kmalloc()
Jens Axboe86c824b2009-04-15 09:00:07 +0200361 * @gfp_mask: the GFP_ mask given to the slab allocator
362 * @nr_iovecs: number of iovecs to pre-allocate
363 *
364 * Description:
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100365 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
366 * %__GFP_WAIT, the allocation is guaranteed to succeed.
Jens Axboe86c824b2009-04-15 09:00:07 +0200367 *
368 **/
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200369struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
370{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200371 struct bio *bio;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200372
Jens Axboef3f63c12010-10-29 11:46:56 -0600373 if (nr_iovecs > UIO_MAXIOV)
374 return NULL;
375
Tejun Heo451a9eb2009-04-15 19:50:51 +0200376 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
377 gfp_mask);
378 if (unlikely(!bio))
379 return NULL;
380
381 bio_init(bio);
382 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
383 bio->bi_max_vecs = nr_iovecs;
384 bio->bi_io_vec = bio->bi_inline_vecs;
385 bio->bi_destructor = bio_kmalloc_destructor;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200386
387 return bio;
388}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200389EXPORT_SYMBOL(bio_kmalloc);
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391void zero_fill_bio(struct bio *bio)
392{
393 unsigned long flags;
394 struct bio_vec *bv;
395 int i;
396
397 bio_for_each_segment(bv, bio, i) {
398 char *data = bvec_kmap_irq(bv, &flags);
399 memset(data, 0, bv->bv_len);
400 flush_dcache_page(bv->bv_page);
401 bvec_kunmap_irq(data, &flags);
402 }
403}
404EXPORT_SYMBOL(zero_fill_bio);
405
406/**
407 * bio_put - release a reference to a bio
408 * @bio: bio to release reference to
409 *
410 * Description:
411 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100412 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 **/
414void bio_put(struct bio *bio)
415{
416 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
417
418 /*
419 * last put frees it
420 */
421 if (atomic_dec_and_test(&bio->bi_cnt)) {
422 bio->bi_next = NULL;
423 bio->bi_destructor(bio);
424 }
425}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200426EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Jens Axboe165125e2007-07-24 09:28:11 +0200428inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
430 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
431 blk_recount_segments(q, bio);
432
433 return bio->bi_phys_segments;
434}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200435EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437/**
438 * __bio_clone - clone a bio
439 * @bio: destination bio
440 * @bio_src: bio to clone
441 *
442 * Clone a &bio. Caller will own the returned bio, but not
443 * the actual data it points to. Reference count of returned
444 * bio will be one.
445 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800446void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
Andrew Mortone525e152005-08-07 09:42:12 -0700448 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
449 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Jens Axboe5d840702008-01-25 12:44:44 +0100451 /*
452 * most users will be overriding ->bi_bdev with a new target,
453 * so we don't set nor calculate new physical/hw segment counts here
454 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 bio->bi_sector = bio_src->bi_sector;
456 bio->bi_bdev = bio_src->bi_bdev;
457 bio->bi_flags |= 1 << BIO_CLONED;
458 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 bio->bi_vcnt = bio_src->bi_vcnt;
460 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700461 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200463EXPORT_SYMBOL(__bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465/**
466 * bio_clone - clone a bio
467 * @bio: bio to clone
468 * @gfp_mask: allocation priority
469 *
470 * Like __bio_clone, only also allocates the returned bio
471 */
Al Virodd0fc662005-10-07 07:46:04 +0100472struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473{
474 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
475
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200476 if (!b)
477 return NULL;
478
479 b->bi_destructor = bio_fs_destructor;
480 __bio_clone(b, bio);
481
482 if (bio_integrity(bio)) {
483 int ret;
484
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200485 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200486
Li Zefan059ea332009-03-09 10:42:45 +0100487 if (ret < 0) {
488 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200489 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100490 }
Peter Osterlund36763472005-09-06 15:16:42 -0700491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493 return b;
494}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200495EXPORT_SYMBOL(bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497/**
498 * bio_get_nr_vecs - return approx number of vecs
499 * @bdev: I/O target
500 *
501 * Return the approximate number of pages we can send to this target.
502 * There's no guarantee that you will be able to fit this number of pages
503 * into a bio, it does not account for dynamic restrictions that vary
504 * on offset.
505 */
506int bio_get_nr_vecs(struct block_device *bdev)
507{
Jens Axboe165125e2007-07-24 09:28:11 +0200508 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 int nr_pages;
510
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400511 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
Martin K. Petersen8a783622010-02-26 00:20:39 -0500512 if (nr_pages > queue_max_segments(q))
513 nr_pages = queue_max_segments(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
515 return nr_pages;
516}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200517EXPORT_SYMBOL(bio_get_nr_vecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Jens Axboe165125e2007-07-24 09:28:11 +0200519static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600520 *page, unsigned int len, unsigned int offset,
521 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522{
523 int retried_segments = 0;
524 struct bio_vec *bvec;
525
526 /*
527 * cloned bio must not modify vec list
528 */
529 if (unlikely(bio_flagged(bio, BIO_CLONED)))
530 return 0;
531
Jens Axboe80cfd542006-01-06 09:43:28 +0100532 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 return 0;
534
Jens Axboe80cfd542006-01-06 09:43:28 +0100535 /*
536 * For filesystems with a blocksize smaller than the pagesize
537 * we will often be called with the same page as last time and
538 * a consecutive offset. Optimize this special case.
539 */
540 if (bio->bi_vcnt > 0) {
541 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
542
543 if (page == prev->bv_page &&
544 offset == prev->bv_offset + prev->bv_len) {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300545 unsigned int prev_bv_len = prev->bv_len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100546 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200547
548 if (q->merge_bvec_fn) {
549 struct bvec_merge_data bvm = {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300550 /* prev_bvec is already charged in
551 bi_size, discharge it in order to
552 simulate merging updated prev_bvec
553 as new bvec. */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200554 .bi_bdev = bio->bi_bdev,
555 .bi_sector = bio->bi_sector,
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300556 .bi_size = bio->bi_size - prev_bv_len,
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200557 .bi_rw = bio->bi_rw,
558 };
559
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300560 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200561 prev->bv_len -= len;
562 return 0;
563 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100564 }
565
566 goto done;
567 }
568 }
569
570 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 return 0;
572
573 /*
574 * we might lose a segment or two here, but rather that than
575 * make this too complex.
576 */
577
Martin K. Petersen8a783622010-02-26 00:20:39 -0500578 while (bio->bi_phys_segments >= queue_max_segments(q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 if (retried_segments)
581 return 0;
582
583 retried_segments = 1;
584 blk_recount_segments(q, bio);
585 }
586
587 /*
588 * setup the new entry, we might clear it again later if we
589 * cannot add the page
590 */
591 bvec = &bio->bi_io_vec[bio->bi_vcnt];
592 bvec->bv_page = page;
593 bvec->bv_len = len;
594 bvec->bv_offset = offset;
595
596 /*
597 * if queue has other restrictions (eg varying max sector size
598 * depending on offset), it can specify a merge_bvec_fn in the
599 * queue to get further control
600 */
601 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200602 struct bvec_merge_data bvm = {
603 .bi_bdev = bio->bi_bdev,
604 .bi_sector = bio->bi_sector,
605 .bi_size = bio->bi_size,
606 .bi_rw = bio->bi_rw,
607 };
608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 /*
610 * merge_bvec_fn() returns number of bytes it can accept
611 * at this offset
612 */
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300613 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 bvec->bv_page = NULL;
615 bvec->bv_len = 0;
616 bvec->bv_offset = 0;
617 return 0;
618 }
619 }
620
621 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200622 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
624
625 bio->bi_vcnt++;
626 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100627 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 bio->bi_size += len;
629 return len;
630}
631
632/**
Mike Christie6e68af62005-11-11 05:30:27 -0600633 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100634 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600635 * @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.
644 *
645 * This should only be used by REQ_PC bios.
Mike Christie6e68af62005-11-11 05:30:27 -0600646 */
Jens Axboe165125e2007-07-24 09:28:11 +0200647int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600648 unsigned int len, unsigned int offset)
649{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400650 return __bio_add_page(q, bio, page, len, offset,
651 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600652}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200653EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600654
655/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 * bio_add_page - attempt to add page to bio
657 * @bio: destination bio
658 * @page: page to add
659 * @len: vec entry length
660 * @offset: vec entry offset
661 *
662 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200663 * number of reasons, such as the bio being full or target block device
664 * limitations. The target block device must allow bio's up to PAGE_SIZE,
665 * so it is always possible to add a single page to an empty bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 */
667int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
668 unsigned int offset)
669{
Mike Christiedefd94b2005-12-05 02:37:06 -0600670 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400671 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200673EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675struct bio_map_data {
676 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200677 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900678 int nr_sgvecs;
679 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680};
681
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200682static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900683 struct sg_iovec *iov, int iov_count,
684 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
686 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200687 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
688 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900689 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 bio->bi_private = bmd;
691}
692
693static void bio_free_map_data(struct bio_map_data *bmd)
694{
695 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200696 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 kfree(bmd);
698}
699
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200700static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
701 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
Jens Axboef3f63c12010-10-29 11:46:56 -0600703 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
Jens Axboef3f63c12010-10-29 11:46:56 -0600705 if (iov_count > UIO_MAXIOV)
706 return NULL;
707
708 bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 if (!bmd)
710 return NULL;
711
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200712 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200713 if (!bmd->iovecs) {
714 kfree(bmd);
715 return NULL;
716 }
717
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200718 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200719 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 return bmd;
721
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200722 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 kfree(bmd);
724 return NULL;
725}
726
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200727static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200728 struct sg_iovec *iov, int iov_count,
729 int to_user, int from_user, int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200730{
731 int ret = 0, i;
732 struct bio_vec *bvec;
733 int iov_idx = 0;
734 unsigned int iov_off = 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200735
736 __bio_for_each_segment(bvec, bio, i, 0) {
737 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200738 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200739
740 while (bv_len && iov_idx < iov_count) {
741 unsigned int bytes;
Michal Simek0e0c6212009-06-10 12:57:07 -0700742 char __user *iov_addr;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200743
744 bytes = min_t(unsigned int,
745 iov[iov_idx].iov_len - iov_off, bv_len);
746 iov_addr = iov[iov_idx].iov_base + iov_off;
747
748 if (!ret) {
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200749 if (to_user)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200750 ret = copy_to_user(iov_addr, bv_addr,
751 bytes);
752
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200753 if (from_user)
754 ret = copy_from_user(bv_addr, iov_addr,
755 bytes);
756
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200757 if (ret)
758 ret = -EFAULT;
759 }
760
761 bv_len -= bytes;
762 bv_addr += bytes;
763 iov_addr += bytes;
764 iov_off += bytes;
765
766 if (iov[iov_idx].iov_len == iov_off) {
767 iov_idx++;
768 iov_off = 0;
769 }
770 }
771
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900772 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200773 __free_page(bvec->bv_page);
774 }
775
776 return ret;
777}
778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779/**
780 * bio_uncopy_user - finish previously mapped bio
781 * @bio: bio being terminated
782 *
783 * Free pages allocated from bio_copy_user() and write back data
784 * to user space in case of a read.
785 */
786int bio_uncopy_user(struct bio *bio)
787{
788 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900789 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
FUJITA Tomonori81882762008-09-02 16:20:19 +0900791 if (!bio_flagged(bio, BIO_NULL_MAPPED))
792 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200793 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
794 0, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 bio_free_map_data(bmd);
796 bio_put(bio);
797 return ret;
798}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200799EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200802 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900804 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200805 * @iov: the iovec.
806 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900808 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 *
810 * Prepares and returns a bio for indirect user io, bouncing data
811 * to/from kernel pages as necessary. Must be paired with
812 * call bio_uncopy_user() on io completion.
813 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900814struct bio *bio_copy_user_iov(struct request_queue *q,
815 struct rq_map_data *map_data,
816 struct sg_iovec *iov, int iov_count,
817 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 struct bio_map_data *bmd;
820 struct bio_vec *bvec;
821 struct page *page;
822 struct bio *bio;
823 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200824 int nr_pages = 0;
825 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900826 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200828 for (i = 0; i < iov_count; i++) {
829 unsigned long uaddr;
830 unsigned long end;
831 unsigned long start;
832
833 uaddr = (unsigned long)iov[i].iov_base;
834 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
835 start = uaddr >> PAGE_SHIFT;
836
Jens Axboecb4644c2010-11-10 14:36:25 +0100837 /*
838 * Overflow, abort
839 */
840 if (end < start)
841 return ERR_PTR(-EINVAL);
842
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200843 nr_pages += end - start;
844 len += iov[i].iov_len;
845 }
846
FUJITA Tomonori69838722009-04-28 20:24:29 +0200847 if (offset)
848 nr_pages++;
849
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900850 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 if (!bmd)
852 return ERR_PTR(-ENOMEM);
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900855 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (!bio)
857 goto out_bmd;
858
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200859 if (!write_to_vm)
860 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
862 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900863
864 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900865 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900866 i = map_data->offset / PAGE_SIZE;
867 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900869 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900871 bytes -= offset;
872
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (bytes > len)
874 bytes = len;
875
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900876 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900877 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900878 ret = -ENOMEM;
879 break;
880 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900881
882 page = map_data->pages[i / nr_pages];
883 page += (i % nr_pages);
884
885 i++;
886 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900887 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900888 if (!page) {
889 ret = -ENOMEM;
890 break;
891 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 }
893
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900894 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900898 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 }
900
901 if (ret)
902 goto cleanup;
903
904 /*
905 * success
906 */
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200907 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
908 (map_data && map_data->from_user)) {
909 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200910 if (ret)
911 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 }
913
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900914 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 return bio;
916cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900917 if (!map_data)
918 bio_for_each_segment(bvec, bio, i)
919 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 bio_put(bio);
922out_bmd:
923 bio_free_map_data(bmd);
924 return ERR_PTR(ret);
925}
926
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200927/**
928 * bio_copy_user - copy user data to bio
929 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900930 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200931 * @uaddr: start of user address
932 * @len: length in bytes
933 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900934 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200935 *
936 * Prepares and returns a bio for indirect user io, bouncing data
937 * to/from kernel pages as necessary. Must be paired with
938 * call bio_uncopy_user() on io completion.
939 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900940struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
941 unsigned long uaddr, unsigned int len,
942 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200943{
944 struct sg_iovec iov;
945
946 iov.iov_base = (void __user *)uaddr;
947 iov.iov_len = len;
948
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900949 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200950}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200951EXPORT_SYMBOL(bio_copy_user);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200952
Jens Axboe165125e2007-07-24 09:28:11 +0200953static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200954 struct block_device *bdev,
955 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900956 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957{
James Bottomley f1970ba2005-06-20 14:06:52 +0200958 int i, j;
959 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 struct page **pages;
961 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200962 int cur_page = 0;
963 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
James Bottomley f1970ba2005-06-20 14:06:52 +0200965 for (i = 0; i < iov_count; i++) {
966 unsigned long uaddr = (unsigned long)iov[i].iov_base;
967 unsigned long len = iov[i].iov_len;
968 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
969 unsigned long start = uaddr >> PAGE_SHIFT;
970
Jens Axboecb4644c2010-11-10 14:36:25 +0100971 /*
972 * Overflow, abort
973 */
974 if (end < start)
975 return ERR_PTR(-EINVAL);
976
James Bottomley f1970ba2005-06-20 14:06:52 +0200977 nr_pages += end - start;
978 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100979 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200980 */
Mike Christiead2d7222006-12-01 10:40:20 +0100981 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200982 return ERR_PTR(-EINVAL);
983 }
984
985 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 return ERR_PTR(-EINVAL);
987
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900988 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (!bio)
990 return ERR_PTR(-ENOMEM);
991
992 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900993 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (!pages)
995 goto out;
996
James Bottomley f1970ba2005-06-20 14:06:52 +0200997 for (i = 0; i < iov_count; i++) {
998 unsigned long uaddr = (unsigned long)iov[i].iov_base;
999 unsigned long len = iov[i].iov_len;
1000 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1001 unsigned long start = uaddr >> PAGE_SHIFT;
1002 const int local_nr_pages = end - start;
1003 const int page_limit = cur_page + local_nr_pages;
Jens Axboecb4644c2010-11-10 14:36:25 +01001004
Nick Pigginf5dd33c2008-07-25 19:45:25 -07001005 ret = get_user_pages_fast(uaddr, local_nr_pages,
1006 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +02001007 if (ret < local_nr_pages) {
1008 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +02001009 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +02001010 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
James Bottomley f1970ba2005-06-20 14:06:52 +02001012 offset = uaddr & ~PAGE_MASK;
1013 for (j = cur_page; j < page_limit; j++) {
1014 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
James Bottomley f1970ba2005-06-20 14:06:52 +02001016 if (len <= 0)
1017 break;
1018
1019 if (bytes > len)
1020 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
James Bottomley f1970ba2005-06-20 14:06:52 +02001022 /*
1023 * sorry...
1024 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001025 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1026 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001027 break;
1028
1029 len -= bytes;
1030 offset = 0;
1031 }
1032
1033 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001035 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001037 while (j < page_limit)
1038 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 kfree(pages);
1042
1043 /*
1044 * set data direction, and check if mapped pages need bouncing
1045 */
1046 if (!write_to_vm)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001047 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
James Bottomley f1970ba2005-06-20 14:06:52 +02001049 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1051 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001052
1053 out_unmap:
1054 for (i = 0; i < nr_pages; i++) {
1055 if(!pages[i])
1056 break;
1057 page_cache_release(pages[i]);
1058 }
1059 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 kfree(pages);
1061 bio_put(bio);
1062 return ERR_PTR(ret);
1063}
1064
1065/**
1066 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001067 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 * @bdev: destination block device
1069 * @uaddr: start of user address
1070 * @len: length in bytes
1071 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001072 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 *
1074 * Map the user space address into a bio suitable for io to a block
1075 * device. Returns an error pointer in case of error.
1076 */
Jens Axboe165125e2007-07-24 09:28:11 +02001077struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001078 unsigned long uaddr, unsigned int len, int write_to_vm,
1079 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
James Bottomley f1970ba2005-06-20 14:06:52 +02001081 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001083 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001084 iov.iov_len = len;
1085
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001086 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001087}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001088EXPORT_SYMBOL(bio_map_user);
James Bottomley f1970ba2005-06-20 14:06:52 +02001089
1090/**
1091 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001092 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001093 * @bdev: destination block device
1094 * @iov: the iovec.
1095 * @iov_count: number of elements in the iovec
1096 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001097 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001098 *
1099 * Map the user space address into a bio suitable for io to a block
1100 * device. Returns an error pointer in case of error.
1101 */
Jens Axboe165125e2007-07-24 09:28:11 +02001102struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001103 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001104 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001105{
1106 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001107
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001108 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1109 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (IS_ERR(bio))
1111 return bio;
1112
1113 /*
1114 * subtle -- if __bio_map_user() ended up bouncing a bio,
1115 * it would normally disappear when its bi_end_io is run.
1116 * however, we need it for the unmap, so grab an extra
1117 * reference to it
1118 */
1119 bio_get(bio);
1120
Mike Christie0e75f902006-12-01 10:40:55 +01001121 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122}
1123
1124static void __bio_unmap_user(struct bio *bio)
1125{
1126 struct bio_vec *bvec;
1127 int i;
1128
1129 /*
1130 * make sure we dirty pages we wrote to
1131 */
1132 __bio_for_each_segment(bvec, bio, i, 0) {
1133 if (bio_data_dir(bio) == READ)
1134 set_page_dirty_lock(bvec->bv_page);
1135
1136 page_cache_release(bvec->bv_page);
1137 }
1138
1139 bio_put(bio);
1140}
1141
1142/**
1143 * bio_unmap_user - unmap a bio
1144 * @bio: the bio being unmapped
1145 *
1146 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1147 * a process context.
1148 *
1149 * bio_unmap_user() may sleep.
1150 */
1151void bio_unmap_user(struct bio *bio)
1152{
1153 __bio_unmap_user(bio);
1154 bio_put(bio);
1155}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001156EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
NeilBrown6712ecf2007-09-27 12:47:43 +02001158static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001159{
Jens Axboeb8238252005-06-20 14:05:27 +02001160 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001161}
1162
Jens Axboe165125e2007-07-24 09:28:11 +02001163static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001164 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001165{
1166 unsigned long kaddr = (unsigned long)data;
1167 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1168 unsigned long start = kaddr >> PAGE_SHIFT;
1169 const int nr_pages = end - start;
1170 int offset, i;
1171 struct bio *bio;
1172
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001173 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001174 if (!bio)
1175 return ERR_PTR(-ENOMEM);
1176
1177 offset = offset_in_page(kaddr);
1178 for (i = 0; i < nr_pages; i++) {
1179 unsigned int bytes = PAGE_SIZE - offset;
1180
1181 if (len <= 0)
1182 break;
1183
1184 if (bytes > len)
1185 bytes = len;
1186
Mike Christiedefd94b2005-12-05 02:37:06 -06001187 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1188 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001189 break;
1190
1191 data += bytes;
1192 len -= bytes;
1193 offset = 0;
1194 }
1195
Jens Axboeb8238252005-06-20 14:05:27 +02001196 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001197 return bio;
1198}
1199
1200/**
1201 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001202 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001203 * @data: pointer to buffer to map
1204 * @len: length in bytes
1205 * @gfp_mask: allocation flags for bio allocation
1206 *
1207 * Map the kernel address into a bio suitable for io to a block
1208 * device. Returns an error pointer in case of error.
1209 */
Jens Axboe165125e2007-07-24 09:28:11 +02001210struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001211 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001212{
1213 struct bio *bio;
1214
1215 bio = __bio_map_kern(q, data, len, gfp_mask);
1216 if (IS_ERR(bio))
1217 return bio;
1218
1219 if (bio->bi_size == len)
1220 return bio;
1221
1222 /*
1223 * Don't support partial mappings.
1224 */
1225 bio_put(bio);
1226 return ERR_PTR(-EINVAL);
1227}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001228EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001229
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001230static void bio_copy_kern_endio(struct bio *bio, int err)
1231{
1232 struct bio_vec *bvec;
1233 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001234 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001235 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001236 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001237
1238 __bio_for_each_segment(bvec, bio, i, 0) {
1239 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001240 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001241
Tejun Heo4fc981e2009-05-19 18:33:06 +09001242 if (read)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001243 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001244
1245 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001246 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001247 }
1248
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001249 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001250 bio_put(bio);
1251}
1252
1253/**
1254 * bio_copy_kern - copy kernel address into bio
1255 * @q: the struct request_queue for the bio
1256 * @data: pointer to buffer to copy
1257 * @len: length in bytes
1258 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001259 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001260 *
1261 * copy the kernel address into a bio suitable for io to a block
1262 * device. Returns an error pointer in case of error.
1263 */
1264struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1265 gfp_t gfp_mask, int reading)
1266{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001267 struct bio *bio;
1268 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001269 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001270
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001271 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1272 if (IS_ERR(bio))
1273 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001274
1275 if (!reading) {
1276 void *p = data;
1277
1278 bio_for_each_segment(bvec, bio, i) {
1279 char *addr = page_address(bvec->bv_page);
1280
1281 memcpy(addr, p, bvec->bv_len);
1282 p += bvec->bv_len;
1283 }
1284 }
1285
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001286 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001287
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001288 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001289}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001290EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292/*
1293 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1294 * for performing direct-IO in BIOs.
1295 *
1296 * The problem is that we cannot run set_page_dirty() from interrupt context
1297 * because the required locks are not interrupt-safe. So what we can do is to
1298 * mark the pages dirty _before_ performing IO. And in interrupt context,
1299 * check that the pages are still dirty. If so, fine. If not, redirty them
1300 * in process context.
1301 *
1302 * We special-case compound pages here: normally this means reads into hugetlb
1303 * pages. The logic in here doesn't really work right for compound pages
1304 * because the VM does not uniformly chase down the head page in all cases.
1305 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1306 * handle them at all. So we skip compound pages here at an early stage.
1307 *
1308 * Note that this code is very hard to test under normal circumstances because
1309 * direct-io pins the pages with get_user_pages(). This makes
1310 * is_page_cache_freeable return false, and the VM will not clean the pages.
1311 * But other code (eg, pdflush) could clean the pages if they are mapped
1312 * pagecache.
1313 *
1314 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1315 * deferred bio dirtying paths.
1316 */
1317
1318/*
1319 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1320 */
1321void bio_set_pages_dirty(struct bio *bio)
1322{
1323 struct bio_vec *bvec = bio->bi_io_vec;
1324 int i;
1325
1326 for (i = 0; i < bio->bi_vcnt; i++) {
1327 struct page *page = bvec[i].bv_page;
1328
1329 if (page && !PageCompound(page))
1330 set_page_dirty_lock(page);
1331 }
1332}
1333
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001334static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335{
1336 struct bio_vec *bvec = bio->bi_io_vec;
1337 int i;
1338
1339 for (i = 0; i < bio->bi_vcnt; i++) {
1340 struct page *page = bvec[i].bv_page;
1341
1342 if (page)
1343 put_page(page);
1344 }
1345}
1346
1347/*
1348 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1349 * If they are, then fine. If, however, some pages are clean then they must
1350 * have been written out during the direct-IO read. So we take another ref on
1351 * the BIO and the offending pages and re-dirty the pages in process context.
1352 *
1353 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1354 * here on. It will run one page_cache_release() against each page and will
1355 * run one bio_put() against the BIO.
1356 */
1357
David Howells65f27f32006-11-22 14:55:48 +00001358static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
David Howells65f27f32006-11-22 14:55:48 +00001360static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361static DEFINE_SPINLOCK(bio_dirty_lock);
1362static struct bio *bio_dirty_list;
1363
1364/*
1365 * This runs in process context
1366 */
David Howells65f27f32006-11-22 14:55:48 +00001367static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368{
1369 unsigned long flags;
1370 struct bio *bio;
1371
1372 spin_lock_irqsave(&bio_dirty_lock, flags);
1373 bio = bio_dirty_list;
1374 bio_dirty_list = NULL;
1375 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1376
1377 while (bio) {
1378 struct bio *next = bio->bi_private;
1379
1380 bio_set_pages_dirty(bio);
1381 bio_release_pages(bio);
1382 bio_put(bio);
1383 bio = next;
1384 }
1385}
1386
1387void bio_check_pages_dirty(struct bio *bio)
1388{
1389 struct bio_vec *bvec = bio->bi_io_vec;
1390 int nr_clean_pages = 0;
1391 int i;
1392
1393 for (i = 0; i < bio->bi_vcnt; i++) {
1394 struct page *page = bvec[i].bv_page;
1395
1396 if (PageDirty(page) || PageCompound(page)) {
1397 page_cache_release(page);
1398 bvec[i].bv_page = NULL;
1399 } else {
1400 nr_clean_pages++;
1401 }
1402 }
1403
1404 if (nr_clean_pages) {
1405 unsigned long flags;
1406
1407 spin_lock_irqsave(&bio_dirty_lock, flags);
1408 bio->bi_private = bio_dirty_list;
1409 bio_dirty_list = bio;
1410 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1411 schedule_work(&bio_dirty_work);
1412 } else {
1413 bio_put(bio);
1414 }
1415}
1416
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001417#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1418void bio_flush_dcache_pages(struct bio *bi)
1419{
1420 int i;
1421 struct bio_vec *bvec;
1422
1423 bio_for_each_segment(bvec, bi, i)
1424 flush_dcache_page(bvec->bv_page);
1425}
1426EXPORT_SYMBOL(bio_flush_dcache_pages);
1427#endif
1428
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429/**
1430 * bio_endio - end I/O on a bio
1431 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 * @error: error, if any
1433 *
1434 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001435 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001436 * preferred way to end I/O on a bio, it takes care of clearing
1437 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1438 * established -Exxxx (-EIO, for instance) error values in case
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001439 * something went wrong. No one should call bi_end_io() directly on a
NeilBrown5bb23a62007-09-27 12:46:13 +02001440 * bio unless they own it and thus know that it has an end_io
1441 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001443void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
1445 if (error)
1446 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001447 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1448 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
NeilBrown5bb23a62007-09-27 12:46:13 +02001450 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001451 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001453EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
1455void bio_pair_release(struct bio_pair *bp)
1456{
1457 if (atomic_dec_and_test(&bp->cnt)) {
1458 struct bio *master = bp->bio1.bi_private;
1459
NeilBrown6712ecf2007-09-27 12:47:43 +02001460 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 mempool_free(bp, bp->bio2.bi_private);
1462 }
1463}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001464EXPORT_SYMBOL(bio_pair_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
NeilBrown6712ecf2007-09-27 12:47:43 +02001466static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467{
1468 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1469
1470 if (err)
1471 bp->error = err;
1472
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
NeilBrown6712ecf2007-09-27 12:47:43 +02001476static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
1478 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1479
1480 if (err)
1481 bp->error = err;
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484}
1485
1486/*
Alberto Bertoglic7eee1b2009-01-25 23:36:14 -02001487 * split a bio - only worry about a bio with a single page in its iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001489struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
Denis ChengRq6feef532008-10-09 08:57:05 +02001491 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
1493 if (!bp)
1494 return bp;
1495
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001496 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001497 bi->bi_sector + first_sectors);
1498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 BUG_ON(bi->bi_vcnt != 1);
1500 BUG_ON(bi->bi_idx != 0);
1501 atomic_set(&bp->cnt, 3);
1502 bp->error = 0;
1503 bp->bio1 = *bi;
1504 bp->bio2 = *bi;
1505 bp->bio2.bi_sector += first_sectors;
1506 bp->bio2.bi_size -= first_sectors << 9;
1507 bp->bio1.bi_size = first_sectors << 9;
1508
1509 bp->bv1 = bi->bi_io_vec[0];
1510 bp->bv2 = bi->bi_io_vec[0];
1511 bp->bv2.bv_offset += first_sectors << 9;
1512 bp->bv2.bv_len -= first_sectors << 9;
1513 bp->bv1.bv_len = first_sectors << 9;
1514
1515 bp->bio1.bi_io_vec = &bp->bv1;
1516 bp->bio2.bi_io_vec = &bp->bv2;
1517
NeilBrowna2eb0c12006-05-22 22:35:27 -07001518 bp->bio1.bi_max_vecs = 1;
1519 bp->bio2.bi_max_vecs = 1;
1520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 bp->bio1.bi_end_io = bio_pair_end_1;
1522 bp->bio2.bi_end_io = bio_pair_end_2;
1523
1524 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001525 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001527 if (bio_integrity(bi))
1528 bio_integrity_split(bi, bp, first_sectors);
1529
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 return bp;
1531}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001532EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001534/**
1535 * bio_sector_offset - Find hardware sector offset in bio
1536 * @bio: bio to inspect
1537 * @index: bio_vec index
1538 * @offset: offset in bv_page
1539 *
1540 * Return the number of hardware sectors between beginning of bio
1541 * and an end point indicated by a bio_vec index and an offset
1542 * within that vector's page.
1543 */
1544sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1545 unsigned int offset)
1546{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001547 unsigned int sector_sz;
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001548 struct bio_vec *bv;
1549 sector_t sectors;
1550 int i;
1551
Martin K. Petersene1defc42009-05-22 17:17:49 -04001552 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001553 sectors = 0;
1554
1555 if (index >= bio->bi_idx)
1556 index = bio->bi_vcnt - 1;
1557
1558 __bio_for_each_segment(bv, bio, i, 0) {
1559 if (i == index) {
1560 if (offset > bv->bv_offset)
1561 sectors += (offset - bv->bv_offset) / sector_sz;
1562 break;
1563 }
1564
1565 sectors += bv->bv_len / sector_sz;
1566 }
1567
1568 return sectors;
1569}
1570EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572/*
1573 * create memory pools for biovec's in a bio_set.
1574 * use the global biovec slabs created for general use.
1575 */
Jens Axboe59725112007-04-02 10:06:42 +02001576static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577{
Jens Axboe7ff93452008-12-11 11:53:43 +01001578 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Jens Axboe7ff93452008-12-11 11:53:43 +01001580 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1581 if (!bs->bvec_pool)
1582 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 return 0;
1585}
1586
1587static void biovec_free_pools(struct bio_set *bs)
1588{
Jens Axboe7ff93452008-12-11 11:53:43 +01001589 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590}
1591
1592void bioset_free(struct bio_set *bs)
1593{
1594 if (bs->bio_pool)
1595 mempool_destroy(bs->bio_pool);
1596
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001597 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001599 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 kfree(bs);
1602}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001603EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
Jens Axboebb799ca2008-12-10 15:35:05 +01001605/**
1606 * bioset_create - Create a bio_set
1607 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1608 * @front_pad: Number of bytes to allocate in front of the returned bio
1609 *
1610 * Description:
1611 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1612 * to ask for a number of bytes to be allocated in front of the bio.
1613 * Front pad allocation is useful for embedding the bio inside
1614 * another structure, to avoid allocating extra data to go with the bio.
1615 * Note that the bio must be embedded at the END of that structure always,
1616 * or things will break badly.
1617 */
1618struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619{
Jens Axboe392ddc32008-12-23 12:42:54 +01001620 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001621 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
Jens Axboe1b434492008-10-22 20:32:58 +02001623 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 if (!bs)
1625 return NULL;
1626
Jens Axboebb799ca2008-12-10 15:35:05 +01001627 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001628
Jens Axboe392ddc32008-12-23 12:42:54 +01001629 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001630 if (!bs->bio_slab) {
1631 kfree(bs);
1632 return NULL;
1633 }
1634
1635 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (!bs->bio_pool)
1637 goto bad;
1638
Jens Axboebb799ca2008-12-10 15:35:05 +01001639 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 return bs;
1641
1642bad:
1643 bioset_free(bs);
1644 return NULL;
1645}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001646EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
1648static void __init biovec_init_slabs(void)
1649{
1650 int i;
1651
1652 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1653 int size;
1654 struct biovec_slab *bvs = bvec_slabs + i;
1655
Jens Axboea7fcd372008-12-05 16:10:29 +01001656 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1657 bvs->slab = NULL;
1658 continue;
1659 }
Jens Axboea7fcd372008-12-05 16:10:29 +01001660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 size = bvs->nr_vecs * sizeof(struct bio_vec);
1662 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001663 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 }
1665}
1666
1667static int __init init_bio(void)
1668{
Jens Axboebb799ca2008-12-10 15:35:05 +01001669 bio_slab_max = 2;
1670 bio_slab_nr = 0;
1671 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1672 if (!bio_slabs)
1673 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001675 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 biovec_init_slabs();
1677
Jens Axboebb799ca2008-12-10 15:35:05 +01001678 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 if (!fs_bio_set)
1680 panic("bio: can't allocate bios\n");
1681
Martin K. Petersena91a2782011-03-17 11:11:05 +01001682 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1683 panic("bio: can't create integrity pool\n");
1684
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001685 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1686 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 if (!bio_split_pool)
1688 panic("bio: can't create split pool\n");
1689
1690 return 0;
1691}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692subsys_initcall(init_bio);