blob: dc17afd672e354810d4b7b2d54e21d466bf8cca8 [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) }
Jens Axboebb799ca2008-12-10 15:35:05 +010046struct 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
114 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
115 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
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200267 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 *
269 * Description:
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200270 * bio_alloc_bioset will first 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
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200272 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
273 * fall back to just using @kmalloc to allocate the required memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 *
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200275 * Note that the caller must set ->bi_destructor on successful return
Jens Axboebb799ca2008-12-10 15:35:05 +0100276 * of a bio, to do the appropriate freeing of the bio once the reference
277 * count drops to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 **/
Al Virodd0fc662005-10-07 07:46:04 +0100279struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200281 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100282 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200283 struct bio *bio;
284 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200285
Tejun Heo451a9eb2009-04-15 19:50:51 +0200286 p = mempool_alloc(bs->bio_pool, gfp_mask);
287 if (unlikely(!p))
288 return NULL;
289 bio = p + bs->front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100290
291 bio_init(bio);
292
293 if (unlikely(!nr_iovecs))
294 goto out_set;
295
296 if (nr_iovecs <= BIO_INLINE_VECS) {
297 bvl = bio->bi_inline_vecs;
298 nr_iovecs = BIO_INLINE_VECS;
299 } else {
300 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
301 if (unlikely(!bvl))
302 goto err_free;
303
304 nr_iovecs = bvec_nr_vecs(idx);
305 }
Tejun Heo451a9eb2009-04-15 19:50:51 +0200306out_set:
Ingo Molnar34053972009-02-21 11:16:36 +0100307 bio->bi_flags |= idx << BIO_POOL_OFFSET;
308 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100309 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100311
312err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200313 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100314 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200316EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Tejun Heo451a9eb2009-04-15 19:50:51 +0200318static void bio_fs_destructor(struct bio *bio)
319{
320 bio_free(bio, fs_bio_set);
321}
322
323/**
324 * bio_alloc - allocate a new bio, memory pool backed
325 * @gfp_mask: allocation mask to use
326 * @nr_iovecs: number of iovecs
327 *
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100328 * bio_alloc will allocate a bio and associated bio_vec array that can hold
329 * at least @nr_iovecs entries. Allocations will be done from the
330 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
331 *
332 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
333 * a bio. This is due to the mempool guarantees. To make this work, callers
334 * must never allocate more than 1 bio at a time from this pool. Callers
335 * that need to allocate more than 1 bio must always submit the previously
336 * allocated bio for IO before attempting to allocate a new one. Failure to
337 * do so can cause livelocks under memory pressure.
Tejun Heo451a9eb2009-04-15 19:50:51 +0200338 *
339 * RETURNS:
340 * Pointer to new bio on success, NULL on failure.
341 */
342struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
343{
344 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
345
346 if (bio)
347 bio->bi_destructor = bio_fs_destructor;
348
349 return bio;
350}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200351EXPORT_SYMBOL(bio_alloc);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200352
353static void bio_kmalloc_destructor(struct bio *bio)
354{
355 if (bio_integrity(bio))
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200356 bio_integrity_free(bio, fs_bio_set);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200357 kfree(bio);
358}
359
Jens Axboe86c824b2009-04-15 09:00:07 +0200360/**
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100361 * bio_kmalloc - allocate a bio for I/O using kmalloc()
Jens Axboe86c824b2009-04-15 09:00:07 +0200362 * @gfp_mask: the GFP_ mask given to the slab allocator
363 * @nr_iovecs: number of iovecs to pre-allocate
364 *
365 * Description:
Alberto Bertogli5f04eeb2009-11-02 11:39:42 +0100366 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
367 * %__GFP_WAIT, the allocation is guaranteed to succeed.
Jens Axboe86c824b2009-04-15 09:00:07 +0200368 *
369 **/
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200370struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
371{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200372 struct bio *bio;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200373
Tejun Heo451a9eb2009-04-15 19:50:51 +0200374 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
375 gfp_mask);
376 if (unlikely(!bio))
377 return NULL;
378
379 bio_init(bio);
380 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
381 bio->bi_max_vecs = nr_iovecs;
382 bio->bi_io_vec = bio->bi_inline_vecs;
383 bio->bi_destructor = bio_kmalloc_destructor;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200384
385 return bio;
386}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200387EXPORT_SYMBOL(bio_kmalloc);
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200388
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389void zero_fill_bio(struct bio *bio)
390{
391 unsigned long flags;
392 struct bio_vec *bv;
393 int i;
394
395 bio_for_each_segment(bv, bio, i) {
396 char *data = bvec_kmap_irq(bv, &flags);
397 memset(data, 0, bv->bv_len);
398 flush_dcache_page(bv->bv_page);
399 bvec_kunmap_irq(data, &flags);
400 }
401}
402EXPORT_SYMBOL(zero_fill_bio);
403
404/**
405 * bio_put - release a reference to a bio
406 * @bio: bio to release reference to
407 *
408 * Description:
409 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100410 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 **/
412void bio_put(struct bio *bio)
413{
414 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
415
416 /*
417 * last put frees it
418 */
419 if (atomic_dec_and_test(&bio->bi_cnt)) {
420 bio->bi_next = NULL;
421 bio->bi_destructor(bio);
422 }
423}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200424EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Jens Axboe165125e2007-07-24 09:28:11 +0200426inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427{
428 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
429 blk_recount_segments(q, bio);
430
431 return bio->bi_phys_segments;
432}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200433EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435/**
436 * __bio_clone - clone a bio
437 * @bio: destination bio
438 * @bio_src: bio to clone
439 *
440 * Clone a &bio. Caller will own the returned bio, but not
441 * the actual data it points to. Reference count of returned
442 * bio will be one.
443 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800444void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
Andrew Mortone525e152005-08-07 09:42:12 -0700446 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
447 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Jens Axboe5d840702008-01-25 12:44:44 +0100449 /*
450 * most users will be overriding ->bi_bdev with a new target,
451 * so we don't set nor calculate new physical/hw segment counts here
452 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 bio->bi_sector = bio_src->bi_sector;
454 bio->bi_bdev = bio_src->bi_bdev;
455 bio->bi_flags |= 1 << BIO_CLONED;
456 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 bio->bi_vcnt = bio_src->bi_vcnt;
458 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700459 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200461EXPORT_SYMBOL(__bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463/**
464 * bio_clone - clone a bio
465 * @bio: bio to clone
466 * @gfp_mask: allocation priority
467 *
468 * Like __bio_clone, only also allocates the returned bio
469 */
Al Virodd0fc662005-10-07 07:46:04 +0100470struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
472 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
473
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200474 if (!b)
475 return NULL;
476
477 b->bi_destructor = bio_fs_destructor;
478 __bio_clone(b, bio);
479
480 if (bio_integrity(bio)) {
481 int ret;
482
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200483 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200484
Li Zefan059ea332009-03-09 10:42:45 +0100485 if (ret < 0) {
486 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200487 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100488 }
Peter Osterlund36763472005-09-06 15:16:42 -0700489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491 return b;
492}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200493EXPORT_SYMBOL(bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
495/**
496 * bio_get_nr_vecs - return approx number of vecs
497 * @bdev: I/O target
498 *
499 * Return the approximate number of pages we can send to this target.
500 * There's no guarantee that you will be able to fit this number of pages
501 * into a bio, it does not account for dynamic restrictions that vary
502 * on offset.
503 */
504int bio_get_nr_vecs(struct block_device *bdev)
505{
Jens Axboe165125e2007-07-24 09:28:11 +0200506 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 int nr_pages;
508
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400509 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
Martin K. Petersen8a783622010-02-26 00:20:39 -0500510 if (nr_pages > queue_max_segments(q))
511 nr_pages = queue_max_segments(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513 return nr_pages;
514}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200515EXPORT_SYMBOL(bio_get_nr_vecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Jens Axboe165125e2007-07-24 09:28:11 +0200517static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600518 *page, unsigned int len, unsigned int offset,
519 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
521 int retried_segments = 0;
522 struct bio_vec *bvec;
523
524 /*
525 * cloned bio must not modify vec list
526 */
527 if (unlikely(bio_flagged(bio, BIO_CLONED)))
528 return 0;
529
Jens Axboe80cfd542006-01-06 09:43:28 +0100530 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return 0;
532
Jens Axboe80cfd542006-01-06 09:43:28 +0100533 /*
534 * For filesystems with a blocksize smaller than the pagesize
535 * we will often be called with the same page as last time and
536 * a consecutive offset. Optimize this special case.
537 */
538 if (bio->bi_vcnt > 0) {
539 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
540
541 if (page == prev->bv_page &&
542 offset == prev->bv_offset + prev->bv_len) {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300543 unsigned int prev_bv_len = prev->bv_len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100544 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200545
546 if (q->merge_bvec_fn) {
547 struct bvec_merge_data bvm = {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300548 /* prev_bvec is already charged in
549 bi_size, discharge it in order to
550 simulate merging updated prev_bvec
551 as new bvec. */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200552 .bi_bdev = bio->bi_bdev,
553 .bi_sector = bio->bi_sector,
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300554 .bi_size = bio->bi_size - prev_bv_len,
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200555 .bi_rw = bio->bi_rw,
556 };
557
Jens Axboe95999452010-03-02 19:17:34 +0100558 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200559 prev->bv_len -= len;
560 return 0;
561 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100562 }
563
564 goto done;
565 }
566 }
567
568 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 return 0;
570
571 /*
572 * we might lose a segment or two here, but rather that than
573 * make this too complex.
574 */
575
Martin K. Petersen8a783622010-02-26 00:20:39 -0500576 while (bio->bi_phys_segments >= queue_max_segments(q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 if (retried_segments)
579 return 0;
580
581 retried_segments = 1;
582 blk_recount_segments(q, bio);
583 }
584
585 /*
586 * setup the new entry, we might clear it again later if we
587 * cannot add the page
588 */
589 bvec = &bio->bi_io_vec[bio->bi_vcnt];
590 bvec->bv_page = page;
591 bvec->bv_len = len;
592 bvec->bv_offset = offset;
593
594 /*
595 * if queue has other restrictions (eg varying max sector size
596 * depending on offset), it can specify a merge_bvec_fn in the
597 * queue to get further control
598 */
599 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200600 struct bvec_merge_data bvm = {
601 .bi_bdev = bio->bi_bdev,
602 .bi_sector = bio->bi_sector,
603 .bi_size = bio->bi_size,
604 .bi_rw = bio->bi_rw,
605 };
606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /*
608 * merge_bvec_fn() returns number of bytes it can accept
609 * at this offset
610 */
Jens Axboe95999452010-03-02 19:17:34 +0100611 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 bvec->bv_page = NULL;
613 bvec->bv_len = 0;
614 bvec->bv_offset = 0;
615 return 0;
616 }
617 }
618
619 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200620 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
622
623 bio->bi_vcnt++;
624 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100625 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 bio->bi_size += len;
627 return len;
628}
629
630/**
Mike Christie6e68af62005-11-11 05:30:27 -0600631 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100632 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600633 * @bio: destination bio
634 * @page: page to add
635 * @len: vec entry length
636 * @offset: vec entry offset
637 *
638 * Attempt to add a page to the bio_vec maplist. This can fail for a
639 * number of reasons, such as the bio being full or target block
640 * device limitations. The target block device must allow bio's
641 * smaller than PAGE_SIZE, so it is always possible to add a single
642 * page to an empty bio. This should only be used by REQ_PC bios.
643 */
Jens Axboe165125e2007-07-24 09:28:11 +0200644int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600645 unsigned int len, unsigned int offset)
646{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400647 return __bio_add_page(q, bio, page, len, offset,
648 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600649}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200650EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600651
652/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 * bio_add_page - attempt to add page to bio
654 * @bio: destination bio
655 * @page: page to add
656 * @len: vec entry length
657 * @offset: vec entry offset
658 *
659 * Attempt to add a page to the bio_vec maplist. This can fail for a
660 * number of reasons, such as the bio being full or target block
661 * device limitations. The target block device must allow bio's
662 * smaller than PAGE_SIZE, so it is always possible to add a single
663 * page to an empty bio.
664 */
665int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
666 unsigned int offset)
667{
Mike Christiedefd94b2005-12-05 02:37:06 -0600668 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400669 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200671EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673struct bio_map_data {
674 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200675 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900676 int nr_sgvecs;
677 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678};
679
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200680static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900681 struct sg_iovec *iov, int iov_count,
682 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200685 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
686 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900687 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 bio->bi_private = bmd;
689}
690
691static void bio_free_map_data(struct bio_map_data *bmd)
692{
693 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200694 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 kfree(bmd);
696}
697
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200698static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
699 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200701 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
703 if (!bmd)
704 return NULL;
705
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200706 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200707 if (!bmd->iovecs) {
708 kfree(bmd);
709 return NULL;
710 }
711
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200712 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200713 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 return bmd;
715
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200716 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 kfree(bmd);
718 return NULL;
719}
720
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200721static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200722 struct sg_iovec *iov, int iov_count,
723 int to_user, int from_user, int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200724{
725 int ret = 0, i;
726 struct bio_vec *bvec;
727 int iov_idx = 0;
728 unsigned int iov_off = 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200729
730 __bio_for_each_segment(bvec, bio, i, 0) {
731 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200732 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200733
734 while (bv_len && iov_idx < iov_count) {
735 unsigned int bytes;
Michal Simek0e0c6212009-06-10 12:57:07 -0700736 char __user *iov_addr;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200737
738 bytes = min_t(unsigned int,
739 iov[iov_idx].iov_len - iov_off, bv_len);
740 iov_addr = iov[iov_idx].iov_base + iov_off;
741
742 if (!ret) {
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200743 if (to_user)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200744 ret = copy_to_user(iov_addr, bv_addr,
745 bytes);
746
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200747 if (from_user)
748 ret = copy_from_user(bv_addr, iov_addr,
749 bytes);
750
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200751 if (ret)
752 ret = -EFAULT;
753 }
754
755 bv_len -= bytes;
756 bv_addr += bytes;
757 iov_addr += bytes;
758 iov_off += bytes;
759
760 if (iov[iov_idx].iov_len == iov_off) {
761 iov_idx++;
762 iov_off = 0;
763 }
764 }
765
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900766 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200767 __free_page(bvec->bv_page);
768 }
769
770 return ret;
771}
772
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773/**
774 * bio_uncopy_user - finish previously mapped bio
775 * @bio: bio being terminated
776 *
777 * Free pages allocated from bio_copy_user() and write back data
778 * to user space in case of a read.
779 */
780int bio_uncopy_user(struct bio *bio)
781{
782 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900783 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
FUJITA Tomonori81882762008-09-02 16:20:19 +0900785 if (!bio_flagged(bio, BIO_NULL_MAPPED))
786 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200787 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
788 0, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 bio_free_map_data(bmd);
790 bio_put(bio);
791 return ret;
792}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200793EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
795/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200796 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900798 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200799 * @iov: the iovec.
800 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900802 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 *
804 * Prepares and returns a bio for indirect user io, bouncing data
805 * to/from kernel pages as necessary. Must be paired with
806 * call bio_uncopy_user() on io completion.
807 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900808struct bio *bio_copy_user_iov(struct request_queue *q,
809 struct rq_map_data *map_data,
810 struct sg_iovec *iov, int iov_count,
811 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 struct bio_map_data *bmd;
814 struct bio_vec *bvec;
815 struct page *page;
816 struct bio *bio;
817 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200818 int nr_pages = 0;
819 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900820 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200822 for (i = 0; i < iov_count; i++) {
823 unsigned long uaddr;
824 unsigned long end;
825 unsigned long start;
826
827 uaddr = (unsigned long)iov[i].iov_base;
828 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
829 start = uaddr >> PAGE_SHIFT;
830
831 nr_pages += end - start;
832 len += iov[i].iov_len;
833 }
834
FUJITA Tomonori69838722009-04-28 20:24:29 +0200835 if (offset)
836 nr_pages++;
837
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900838 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (!bmd)
840 return ERR_PTR(-ENOMEM);
841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900843 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (!bio)
845 goto out_bmd;
846
847 bio->bi_rw |= (!write_to_vm << BIO_RW);
848
849 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900850
851 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900852 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900853 i = map_data->offset / PAGE_SIZE;
854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900856 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900858 bytes -= offset;
859
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 if (bytes > len)
861 bytes = len;
862
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900863 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900864 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900865 ret = -ENOMEM;
866 break;
867 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900868
869 page = map_data->pages[i / nr_pages];
870 page += (i % nr_pages);
871
872 i++;
873 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900874 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900875 if (!page) {
876 ret = -ENOMEM;
877 break;
878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
880
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900881 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900885 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
887
888 if (ret)
889 goto cleanup;
890
891 /*
892 * success
893 */
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200894 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
895 (map_data && map_data->from_user)) {
896 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200897 if (ret)
898 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 }
900
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900901 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 return bio;
903cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900904 if (!map_data)
905 bio_for_each_segment(bvec, bio, i)
906 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 bio_put(bio);
909out_bmd:
910 bio_free_map_data(bmd);
911 return ERR_PTR(ret);
912}
913
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200914/**
915 * bio_copy_user - copy user data to bio
916 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900917 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200918 * @uaddr: start of user address
919 * @len: length in bytes
920 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900921 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200922 *
923 * Prepares and returns a bio for indirect user io, bouncing data
924 * to/from kernel pages as necessary. Must be paired with
925 * call bio_uncopy_user() on io completion.
926 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900927struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
928 unsigned long uaddr, unsigned int len,
929 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200930{
931 struct sg_iovec iov;
932
933 iov.iov_base = (void __user *)uaddr;
934 iov.iov_len = len;
935
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900936 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200937}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200938EXPORT_SYMBOL(bio_copy_user);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200939
Jens Axboe165125e2007-07-24 09:28:11 +0200940static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200941 struct block_device *bdev,
942 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900943 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944{
James Bottomley f1970ba2005-06-20 14:06:52 +0200945 int i, j;
946 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 struct page **pages;
948 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200949 int cur_page = 0;
950 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
James Bottomley f1970ba2005-06-20 14:06:52 +0200952 for (i = 0; i < iov_count; i++) {
953 unsigned long uaddr = (unsigned long)iov[i].iov_base;
954 unsigned long len = iov[i].iov_len;
955 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
956 unsigned long start = uaddr >> PAGE_SHIFT;
957
958 nr_pages += end - start;
959 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100960 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200961 */
Mike Christiead2d7222006-12-01 10:40:20 +0100962 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200963 return ERR_PTR(-EINVAL);
964 }
965
966 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 return ERR_PTR(-EINVAL);
968
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900969 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (!bio)
971 return ERR_PTR(-ENOMEM);
972
973 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900974 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 if (!pages)
976 goto out;
977
James Bottomley f1970ba2005-06-20 14:06:52 +0200978 for (i = 0; i < iov_count; i++) {
979 unsigned long uaddr = (unsigned long)iov[i].iov_base;
980 unsigned long len = iov[i].iov_len;
981 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
982 unsigned long start = uaddr >> PAGE_SHIFT;
983 const int local_nr_pages = end - start;
984 const int page_limit = cur_page + local_nr_pages;
985
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700986 ret = get_user_pages_fast(uaddr, local_nr_pages,
987 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200988 if (ret < local_nr_pages) {
989 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200990 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
James Bottomley f1970ba2005-06-20 14:06:52 +0200993 offset = uaddr & ~PAGE_MASK;
994 for (j = cur_page; j < page_limit; j++) {
995 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
James Bottomley f1970ba2005-06-20 14:06:52 +0200997 if (len <= 0)
998 break;
999
1000 if (bytes > len)
1001 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
James Bottomley f1970ba2005-06-20 14:06:52 +02001003 /*
1004 * sorry...
1005 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001006 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1007 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001008 break;
1009
1010 len -= bytes;
1011 offset = 0;
1012 }
1013
1014 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001016 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001018 while (j < page_limit)
1019 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 }
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 kfree(pages);
1023
1024 /*
1025 * set data direction, and check if mapped pages need bouncing
1026 */
1027 if (!write_to_vm)
1028 bio->bi_rw |= (1 << BIO_RW);
1029
James Bottomley f1970ba2005-06-20 14:06:52 +02001030 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1032 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001033
1034 out_unmap:
1035 for (i = 0; i < nr_pages; i++) {
1036 if(!pages[i])
1037 break;
1038 page_cache_release(pages[i]);
1039 }
1040 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 kfree(pages);
1042 bio_put(bio);
1043 return ERR_PTR(ret);
1044}
1045
1046/**
1047 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001048 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 * @bdev: destination block device
1050 * @uaddr: start of user address
1051 * @len: length in bytes
1052 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001053 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 *
1055 * Map the user space address into a bio suitable for io to a block
1056 * device. Returns an error pointer in case of error.
1057 */
Jens Axboe165125e2007-07-24 09:28:11 +02001058struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001059 unsigned long uaddr, unsigned int len, int write_to_vm,
1060 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
James Bottomley f1970ba2005-06-20 14:06:52 +02001062 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001064 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001065 iov.iov_len = len;
1066
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001067 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001068}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001069EXPORT_SYMBOL(bio_map_user);
James Bottomley f1970ba2005-06-20 14:06:52 +02001070
1071/**
1072 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001073 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001074 * @bdev: destination block device
1075 * @iov: the iovec.
1076 * @iov_count: number of elements in the iovec
1077 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001078 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001079 *
1080 * Map the user space address into a bio suitable for io to a block
1081 * device. Returns an error pointer in case of error.
1082 */
Jens Axboe165125e2007-07-24 09:28:11 +02001083struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001084 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001085 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001086{
1087 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001088
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001089 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1090 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 if (IS_ERR(bio))
1092 return bio;
1093
1094 /*
1095 * subtle -- if __bio_map_user() ended up bouncing a bio,
1096 * it would normally disappear when its bi_end_io is run.
1097 * however, we need it for the unmap, so grab an extra
1098 * reference to it
1099 */
1100 bio_get(bio);
1101
Mike Christie0e75f902006-12-01 10:40:55 +01001102 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103}
1104
1105static void __bio_unmap_user(struct bio *bio)
1106{
1107 struct bio_vec *bvec;
1108 int i;
1109
1110 /*
1111 * make sure we dirty pages we wrote to
1112 */
1113 __bio_for_each_segment(bvec, bio, i, 0) {
1114 if (bio_data_dir(bio) == READ)
1115 set_page_dirty_lock(bvec->bv_page);
1116
1117 page_cache_release(bvec->bv_page);
1118 }
1119
1120 bio_put(bio);
1121}
1122
1123/**
1124 * bio_unmap_user - unmap a bio
1125 * @bio: the bio being unmapped
1126 *
1127 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1128 * a process context.
1129 *
1130 * bio_unmap_user() may sleep.
1131 */
1132void bio_unmap_user(struct bio *bio)
1133{
1134 __bio_unmap_user(bio);
1135 bio_put(bio);
1136}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001137EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
NeilBrown6712ecf2007-09-27 12:47:43 +02001139static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001140{
Jens Axboeb8238252005-06-20 14:05:27 +02001141 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001142}
1143
Jens Axboe165125e2007-07-24 09:28:11 +02001144static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001145 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001146{
1147 unsigned long kaddr = (unsigned long)data;
1148 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1149 unsigned long start = kaddr >> PAGE_SHIFT;
1150 const int nr_pages = end - start;
1151 int offset, i;
1152 struct bio *bio;
1153
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001154 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001155 if (!bio)
1156 return ERR_PTR(-ENOMEM);
1157
1158 offset = offset_in_page(kaddr);
1159 for (i = 0; i < nr_pages; i++) {
1160 unsigned int bytes = PAGE_SIZE - offset;
1161
1162 if (len <= 0)
1163 break;
1164
1165 if (bytes > len)
1166 bytes = len;
1167
Mike Christiedefd94b2005-12-05 02:37:06 -06001168 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1169 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001170 break;
1171
1172 data += bytes;
1173 len -= bytes;
1174 offset = 0;
1175 }
1176
Jens Axboeb8238252005-06-20 14:05:27 +02001177 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001178 return bio;
1179}
1180
1181/**
1182 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001183 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001184 * @data: pointer to buffer to map
1185 * @len: length in bytes
1186 * @gfp_mask: allocation flags for bio allocation
1187 *
1188 * Map the kernel address into a bio suitable for io to a block
1189 * device. Returns an error pointer in case of error.
1190 */
Jens Axboe165125e2007-07-24 09:28:11 +02001191struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001192 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001193{
1194 struct bio *bio;
1195
1196 bio = __bio_map_kern(q, data, len, gfp_mask);
1197 if (IS_ERR(bio))
1198 return bio;
1199
1200 if (bio->bi_size == len)
1201 return bio;
1202
1203 /*
1204 * Don't support partial mappings.
1205 */
1206 bio_put(bio);
1207 return ERR_PTR(-EINVAL);
1208}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001209EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001210
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001211static void bio_copy_kern_endio(struct bio *bio, int err)
1212{
1213 struct bio_vec *bvec;
1214 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001215 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001216 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001217 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001218
1219 __bio_for_each_segment(bvec, bio, i, 0) {
1220 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001221 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001222
Tejun Heo4fc981e2009-05-19 18:33:06 +09001223 if (read)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001224 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001225
1226 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001227 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001228 }
1229
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001230 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001231 bio_put(bio);
1232}
1233
1234/**
1235 * bio_copy_kern - copy kernel address into bio
1236 * @q: the struct request_queue for the bio
1237 * @data: pointer to buffer to copy
1238 * @len: length in bytes
1239 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001240 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001241 *
1242 * copy the kernel address into a bio suitable for io to a block
1243 * device. Returns an error pointer in case of error.
1244 */
1245struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1246 gfp_t gfp_mask, int reading)
1247{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001248 struct bio *bio;
1249 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001250 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001251
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001252 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1253 if (IS_ERR(bio))
1254 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001255
1256 if (!reading) {
1257 void *p = data;
1258
1259 bio_for_each_segment(bvec, bio, i) {
1260 char *addr = page_address(bvec->bv_page);
1261
1262 memcpy(addr, p, bvec->bv_len);
1263 p += bvec->bv_len;
1264 }
1265 }
1266
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001267 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001268
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001269 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001270}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001271EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001272
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273/*
1274 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1275 * for performing direct-IO in BIOs.
1276 *
1277 * The problem is that we cannot run set_page_dirty() from interrupt context
1278 * because the required locks are not interrupt-safe. So what we can do is to
1279 * mark the pages dirty _before_ performing IO. And in interrupt context,
1280 * check that the pages are still dirty. If so, fine. If not, redirty them
1281 * in process context.
1282 *
1283 * We special-case compound pages here: normally this means reads into hugetlb
1284 * pages. The logic in here doesn't really work right for compound pages
1285 * because the VM does not uniformly chase down the head page in all cases.
1286 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1287 * handle them at all. So we skip compound pages here at an early stage.
1288 *
1289 * Note that this code is very hard to test under normal circumstances because
1290 * direct-io pins the pages with get_user_pages(). This makes
1291 * is_page_cache_freeable return false, and the VM will not clean the pages.
1292 * But other code (eg, pdflush) could clean the pages if they are mapped
1293 * pagecache.
1294 *
1295 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1296 * deferred bio dirtying paths.
1297 */
1298
1299/*
1300 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1301 */
1302void bio_set_pages_dirty(struct bio *bio)
1303{
1304 struct bio_vec *bvec = bio->bi_io_vec;
1305 int i;
1306
1307 for (i = 0; i < bio->bi_vcnt; i++) {
1308 struct page *page = bvec[i].bv_page;
1309
1310 if (page && !PageCompound(page))
1311 set_page_dirty_lock(page);
1312 }
1313}
1314
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001315static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
1317 struct bio_vec *bvec = bio->bi_io_vec;
1318 int i;
1319
1320 for (i = 0; i < bio->bi_vcnt; i++) {
1321 struct page *page = bvec[i].bv_page;
1322
1323 if (page)
1324 put_page(page);
1325 }
1326}
1327
1328/*
1329 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1330 * If they are, then fine. If, however, some pages are clean then they must
1331 * have been written out during the direct-IO read. So we take another ref on
1332 * the BIO and the offending pages and re-dirty the pages in process context.
1333 *
1334 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1335 * here on. It will run one page_cache_release() against each page and will
1336 * run one bio_put() against the BIO.
1337 */
1338
David Howells65f27f32006-11-22 14:55:48 +00001339static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
David Howells65f27f32006-11-22 14:55:48 +00001341static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342static DEFINE_SPINLOCK(bio_dirty_lock);
1343static struct bio *bio_dirty_list;
1344
1345/*
1346 * This runs in process context
1347 */
David Howells65f27f32006-11-22 14:55:48 +00001348static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
1350 unsigned long flags;
1351 struct bio *bio;
1352
1353 spin_lock_irqsave(&bio_dirty_lock, flags);
1354 bio = bio_dirty_list;
1355 bio_dirty_list = NULL;
1356 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1357
1358 while (bio) {
1359 struct bio *next = bio->bi_private;
1360
1361 bio_set_pages_dirty(bio);
1362 bio_release_pages(bio);
1363 bio_put(bio);
1364 bio = next;
1365 }
1366}
1367
1368void bio_check_pages_dirty(struct bio *bio)
1369{
1370 struct bio_vec *bvec = bio->bi_io_vec;
1371 int nr_clean_pages = 0;
1372 int i;
1373
1374 for (i = 0; i < bio->bi_vcnt; i++) {
1375 struct page *page = bvec[i].bv_page;
1376
1377 if (PageDirty(page) || PageCompound(page)) {
1378 page_cache_release(page);
1379 bvec[i].bv_page = NULL;
1380 } else {
1381 nr_clean_pages++;
1382 }
1383 }
1384
1385 if (nr_clean_pages) {
1386 unsigned long flags;
1387
1388 spin_lock_irqsave(&bio_dirty_lock, flags);
1389 bio->bi_private = bio_dirty_list;
1390 bio_dirty_list = bio;
1391 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1392 schedule_work(&bio_dirty_work);
1393 } else {
1394 bio_put(bio);
1395 }
1396}
1397
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001398#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1399void bio_flush_dcache_pages(struct bio *bi)
1400{
1401 int i;
1402 struct bio_vec *bvec;
1403
1404 bio_for_each_segment(bvec, bi, i)
1405 flush_dcache_page(bvec->bv_page);
1406}
1407EXPORT_SYMBOL(bio_flush_dcache_pages);
1408#endif
1409
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410/**
1411 * bio_endio - end I/O on a bio
1412 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 * @error: error, if any
1414 *
1415 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001416 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001417 * preferred way to end I/O on a bio, it takes care of clearing
1418 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1419 * established -Exxxx (-EIO, for instance) error values in case
1420 * something went wrong. Noone should call bi_end_io() directly on a
1421 * bio unless they own it and thus know that it has an end_io
1422 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001424void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 if (error)
1427 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001428 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1429 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
NeilBrown5bb23a62007-09-27 12:46:13 +02001431 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001432 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001434EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436void bio_pair_release(struct bio_pair *bp)
1437{
1438 if (atomic_dec_and_test(&bp->cnt)) {
1439 struct bio *master = bp->bio1.bi_private;
1440
NeilBrown6712ecf2007-09-27 12:47:43 +02001441 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 mempool_free(bp, bp->bio2.bi_private);
1443 }
1444}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001445EXPORT_SYMBOL(bio_pair_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
NeilBrown6712ecf2007-09-27 12:47:43 +02001447static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448{
1449 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1450
1451 if (err)
1452 bp->error = err;
1453
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455}
1456
NeilBrown6712ecf2007-09-27 12:47:43 +02001457static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
1459 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1460
1461 if (err)
1462 bp->error = err;
1463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465}
1466
1467/*
Alberto Bertoglic7eee1b2009-01-25 23:36:14 -02001468 * split a bio - only worry about a bio with a single page in its iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001470struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Denis ChengRq6feef532008-10-09 08:57:05 +02001472 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
1474 if (!bp)
1475 return bp;
1476
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001477 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001478 bi->bi_sector + first_sectors);
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 BUG_ON(bi->bi_vcnt != 1);
1481 BUG_ON(bi->bi_idx != 0);
1482 atomic_set(&bp->cnt, 3);
1483 bp->error = 0;
1484 bp->bio1 = *bi;
1485 bp->bio2 = *bi;
1486 bp->bio2.bi_sector += first_sectors;
1487 bp->bio2.bi_size -= first_sectors << 9;
1488 bp->bio1.bi_size = first_sectors << 9;
1489
1490 bp->bv1 = bi->bi_io_vec[0];
1491 bp->bv2 = bi->bi_io_vec[0];
1492 bp->bv2.bv_offset += first_sectors << 9;
1493 bp->bv2.bv_len -= first_sectors << 9;
1494 bp->bv1.bv_len = first_sectors << 9;
1495
1496 bp->bio1.bi_io_vec = &bp->bv1;
1497 bp->bio2.bi_io_vec = &bp->bv2;
1498
NeilBrowna2eb0c12006-05-22 22:35:27 -07001499 bp->bio1.bi_max_vecs = 1;
1500 bp->bio2.bi_max_vecs = 1;
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 bp->bio1.bi_end_io = bio_pair_end_1;
1503 bp->bio2.bi_end_io = bio_pair_end_2;
1504
1505 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001506 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001508 if (bio_integrity(bi))
1509 bio_integrity_split(bi, bp, first_sectors);
1510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 return bp;
1512}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001513EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001515/**
1516 * bio_sector_offset - Find hardware sector offset in bio
1517 * @bio: bio to inspect
1518 * @index: bio_vec index
1519 * @offset: offset in bv_page
1520 *
1521 * Return the number of hardware sectors between beginning of bio
1522 * and an end point indicated by a bio_vec index and an offset
1523 * within that vector's page.
1524 */
1525sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1526 unsigned int offset)
1527{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001528 unsigned int sector_sz;
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001529 struct bio_vec *bv;
1530 sector_t sectors;
1531 int i;
1532
Martin K. Petersene1defc42009-05-22 17:17:49 -04001533 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001534 sectors = 0;
1535
1536 if (index >= bio->bi_idx)
1537 index = bio->bi_vcnt - 1;
1538
1539 __bio_for_each_segment(bv, bio, i, 0) {
1540 if (i == index) {
1541 if (offset > bv->bv_offset)
1542 sectors += (offset - bv->bv_offset) / sector_sz;
1543 break;
1544 }
1545
1546 sectors += bv->bv_len / sector_sz;
1547 }
1548
1549 return sectors;
1550}
1551EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
1553/*
1554 * create memory pools for biovec's in a bio_set.
1555 * use the global biovec slabs created for general use.
1556 */
Jens Axboe59725112007-04-02 10:06:42 +02001557static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558{
Jens Axboe7ff93452008-12-11 11:53:43 +01001559 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
Jens Axboe7ff93452008-12-11 11:53:43 +01001561 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1562 if (!bs->bvec_pool)
1563 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 return 0;
1566}
1567
1568static void biovec_free_pools(struct bio_set *bs)
1569{
Jens Axboe7ff93452008-12-11 11:53:43 +01001570 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571}
1572
1573void bioset_free(struct bio_set *bs)
1574{
1575 if (bs->bio_pool)
1576 mempool_destroy(bs->bio_pool);
1577
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001578 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001580 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 kfree(bs);
1583}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001584EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
Jens Axboebb799ca2008-12-10 15:35:05 +01001586/**
1587 * bioset_create - Create a bio_set
1588 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1589 * @front_pad: Number of bytes to allocate in front of the returned bio
1590 *
1591 * Description:
1592 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1593 * to ask for a number of bytes to be allocated in front of the bio.
1594 * Front pad allocation is useful for embedding the bio inside
1595 * another structure, to avoid allocating extra data to go with the bio.
1596 * Note that the bio must be embedded at the END of that structure always,
1597 * or things will break badly.
1598 */
1599struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600{
Jens Axboe392ddc32008-12-23 12:42:54 +01001601 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001602 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
Jens Axboe1b434492008-10-22 20:32:58 +02001604 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 if (!bs)
1606 return NULL;
1607
Jens Axboebb799ca2008-12-10 15:35:05 +01001608 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001609
Jens Axboe392ddc32008-12-23 12:42:54 +01001610 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001611 if (!bs->bio_slab) {
1612 kfree(bs);
1613 return NULL;
1614 }
1615
1616 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 if (!bs->bio_pool)
1618 goto bad;
1619
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001620 if (bioset_integrity_create(bs, pool_size))
1621 goto bad;
1622
Jens Axboebb799ca2008-12-10 15:35:05 +01001623 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 return bs;
1625
1626bad:
1627 bioset_free(bs);
1628 return NULL;
1629}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001630EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632static void __init biovec_init_slabs(void)
1633{
1634 int i;
1635
1636 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1637 int size;
1638 struct biovec_slab *bvs = bvec_slabs + i;
1639
Jens Axboea7fcd372008-12-05 16:10:29 +01001640#ifndef CONFIG_BLK_DEV_INTEGRITY
1641 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1642 bvs->slab = NULL;
1643 continue;
1644 }
1645#endif
1646
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 size = bvs->nr_vecs * sizeof(struct bio_vec);
1648 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001649 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 }
1651}
1652
1653static int __init init_bio(void)
1654{
Jens Axboebb799ca2008-12-10 15:35:05 +01001655 bio_slab_max = 2;
1656 bio_slab_nr = 0;
1657 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1658 if (!bio_slabs)
1659 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001661 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 biovec_init_slabs();
1663
Jens Axboebb799ca2008-12-10 15:35:05 +01001664 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 if (!fs_bio_set)
1666 panic("bio: can't allocate bios\n");
1667
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001668 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1669 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 if (!bio_split_pool)
1671 panic("bio: can't create split pool\n");
1672
1673 return 0;
1674}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675subsys_initcall(init_bio);