blob: bb5768f59b32e22fdcde250f0ae8b01eb6b69b48 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Jens Axboe0fe23472006-09-04 15:41:16 +02002 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
16 *
17 */
18#include <linux/mm.h>
19#include <linux/swap.h>
20#include <linux/bio.h>
21#include <linux/blkdev.h>
Tejun Heo852c7882012-03-05 13:15:27 -080022#include <linux/iocontext.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/slab.h>
24#include <linux/init.h>
25#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mempool.h>
28#include <linux/workqueue.h>
Tejun Heo852c7882012-03-05 13:15:27 -080029#include <linux/cgroup.h>
James Bottomley f1970ba2005-06-20 14:06:52 +020030#include <scsi/sg.h> /* for struct sg_iovec */
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Li Zefan55782132009-06-09 13:43:05 +080032#include <trace/events/block.h>
Ingo Molnar0bfc2452008-11-26 11:59:56 +010033
Jens Axboe392ddc32008-12-23 12:42:54 +010034/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
Denis ChengRq6feef532008-10-09 08:57:05 +020040static mempool_t *bio_split_pool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * if you change this list, also change bvec_alloc or things will
44 * break badly! cannot be bigger than what you can fit into an
45 * unsigned short
46 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Martin K. Petersendf677142011-03-08 08:28:01 +010048static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
50};
51#undef BV
52
53/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
55 * IO code that does not need private memory pools.
56 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020057struct bio_set *fs_bio_set;
Kent Overstreet3f86a822012-09-06 15:35:01 -070058EXPORT_SYMBOL(fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Jens Axboebb799ca2008-12-10 15:35:05 +010060/*
61 * Our slab pool management
62 */
63struct bio_slab {
64 struct kmem_cache *slab;
65 unsigned int slab_ref;
66 unsigned int slab_size;
67 char name[8];
68};
69static DEFINE_MUTEX(bio_slab_lock);
70static struct bio_slab *bio_slabs;
71static unsigned int bio_slab_nr, bio_slab_max;
72
73static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
74{
75 unsigned int sz = sizeof(struct bio) + extra_size;
76 struct kmem_cache *slab = NULL;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +020077 struct bio_slab *bslab, *new_bio_slabs;
Anna Leuschner386bc352012-10-22 21:53:36 +020078 unsigned int new_bio_slab_max;
Jens Axboebb799ca2008-12-10 15:35:05 +010079 unsigned int i, entry = -1;
80
81 mutex_lock(&bio_slab_lock);
82
83 i = 0;
84 while (i < bio_slab_nr) {
Thiago Farinaf06f1352010-01-19 14:07:09 +010085 bslab = &bio_slabs[i];
Jens Axboebb799ca2008-12-10 15:35:05 +010086
87 if (!bslab->slab && entry == -1)
88 entry = i;
89 else if (bslab->slab_size == sz) {
90 slab = bslab->slab;
91 bslab->slab_ref++;
92 break;
93 }
94 i++;
95 }
96
97 if (slab)
98 goto out_unlock;
99
100 if (bio_slab_nr == bio_slab_max && entry == -1) {
Anna Leuschner386bc352012-10-22 21:53:36 +0200101 new_bio_slab_max = bio_slab_max << 1;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200102 new_bio_slabs = krealloc(bio_slabs,
Anna Leuschner386bc352012-10-22 21:53:36 +0200103 new_bio_slab_max * sizeof(struct bio_slab),
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200104 GFP_KERNEL);
105 if (!new_bio_slabs)
Jens Axboebb799ca2008-12-10 15:35:05 +0100106 goto out_unlock;
Anna Leuschner386bc352012-10-22 21:53:36 +0200107 bio_slab_max = new_bio_slab_max;
Alexey Khoroshilov389d7b22012-08-09 15:19:25 +0200108 bio_slabs = new_bio_slabs;
Jens Axboebb799ca2008-12-10 15:35:05 +0100109 }
110 if (entry == -1)
111 entry = bio_slab_nr++;
112
113 bslab = &bio_slabs[entry];
114
115 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
116 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
117 if (!slab)
118 goto out_unlock;
119
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -0700120 printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry);
Jens Axboebb799ca2008-12-10 15:35:05 +0100121 bslab->slab = slab;
122 bslab->slab_ref = 1;
123 bslab->slab_size = sz;
124out_unlock:
125 mutex_unlock(&bio_slab_lock);
126 return slab;
127}
128
129static void bio_put_slab(struct bio_set *bs)
130{
131 struct bio_slab *bslab = NULL;
132 unsigned int i;
133
134 mutex_lock(&bio_slab_lock);
135
136 for (i = 0; i < bio_slab_nr; i++) {
137 if (bs->bio_slab == bio_slabs[i].slab) {
138 bslab = &bio_slabs[i];
139 break;
140 }
141 }
142
143 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
144 goto out;
145
146 WARN_ON(!bslab->slab_ref);
147
148 if (--bslab->slab_ref)
149 goto out;
150
151 kmem_cache_destroy(bslab->slab);
152 bslab->slab = NULL;
153
154out:
155 mutex_unlock(&bio_slab_lock);
156}
157
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200158unsigned int bvec_nr_vecs(unsigned short idx)
159{
160 return bvec_slabs[idx].nr_vecs;
161}
162
Jens Axboebb799ca2008-12-10 15:35:05 +0100163void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
164{
165 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
166
167 if (idx == BIOVEC_MAX_IDX)
168 mempool_free(bv, bs->bvec_pool);
169 else {
170 struct biovec_slab *bvs = bvec_slabs + idx;
171
172 kmem_cache_free(bvs->slab, bv);
173 }
174}
175
Jens Axboe7ff93452008-12-11 11:53:43 +0100176struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
177 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
179 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
181 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100182 * see comment near bvec_array define!
183 */
184 switch (nr) {
185 case 1:
186 *idx = 0;
187 break;
188 case 2 ... 4:
189 *idx = 1;
190 break;
191 case 5 ... 16:
192 *idx = 2;
193 break;
194 case 17 ... 64:
195 *idx = 3;
196 break;
197 case 65 ... 128:
198 *idx = 4;
199 break;
200 case 129 ... BIO_MAX_PAGES:
201 *idx = 5;
202 break;
203 default:
204 return NULL;
205 }
206
207 /*
208 * idx now points to the pool we want to allocate from. only the
209 * 1-vec entry pool is mempool backed.
210 */
211 if (*idx == BIOVEC_MAX_IDX) {
212fallback:
213 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
214 } else {
215 struct biovec_slab *bvs = bvec_slabs + *idx;
216 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200218 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100219 * Make this allocation restricted and don't dump info on
220 * allocation failures, since we'll fallback to the mempool
221 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200222 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100223 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
224
225 /*
226 * Try a slab allocation. If this fails and __GFP_WAIT
227 * is set, retry with the 1-entry mempool
228 */
229 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
230 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
231 *idx = BIOVEC_MAX_IDX;
232 goto fallback;
233 }
234 }
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 return bvl;
237}
238
Kent Overstreet4254bba2012-09-06 15:35:00 -0700239static void __bio_free(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Kent Overstreet4254bba2012-09-06 15:35:00 -0700241 bio_disassociate_task(bio);
Jens Axboe992c5dd2007-07-18 13:18:08 +0200242
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200243 if (bio_integrity(bio))
Kent Overstreet1e2a410f2012-09-06 15:34:56 -0700244 bio_integrity_free(bio);
Kent Overstreet4254bba2012-09-06 15:35:00 -0700245}
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200246
Kent Overstreet4254bba2012-09-06 15:35:00 -0700247static void bio_free(struct bio *bio)
248{
249 struct bio_set *bs = bio->bi_pool;
250 void *p;
251
252 __bio_free(bio);
253
254 if (bs) {
255 if (bio_has_allocated_vec(bio))
256 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
257
258 /*
259 * If we have front padding, adjust the bio pointer before freeing
260 */
261 p = bio;
Jens Axboebb799ca2008-12-10 15:35:05 +0100262 p -= bs->front_pad;
263
Kent Overstreet4254bba2012-09-06 15:35:00 -0700264 mempool_free(p, bs->bio_pool);
265 } else {
266 /* Bio was allocated by bio_kmalloc() */
267 kfree(bio);
268 }
Peter Osterlund36763472005-09-06 15:16:42 -0700269}
270
Arjan van de Ven858119e2006-01-14 13:20:43 -0800271void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Jens Axboe2b94de52007-07-18 13:14:03 +0200273 memset(bio, 0, sizeof(*bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 bio->bi_flags = 1 << BIO_UPTODATE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 atomic_set(&bio->bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200277EXPORT_SYMBOL(bio_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279/**
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700280 * bio_reset - reinitialize a bio
281 * @bio: bio to reset
282 *
283 * Description:
284 * After calling bio_reset(), @bio will be in the same state as a freshly
285 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
286 * preserved are the ones that are initialized by bio_alloc_bioset(). See
287 * comment in struct bio.
288 */
289void bio_reset(struct bio *bio)
290{
291 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
292
Kent Overstreet4254bba2012-09-06 15:35:00 -0700293 __bio_free(bio);
Kent Overstreetf44b48c2012-09-06 15:34:58 -0700294
295 memset(bio, 0, BIO_RESET_BYTES);
296 bio->bi_flags = flags|(1 << BIO_UPTODATE);
297}
298EXPORT_SYMBOL(bio_reset);
299
300/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 * bio_alloc_bioset - allocate a bio for I/O
302 * @gfp_mask: the GFP_ mask given to the slab allocator
303 * @nr_iovecs: number of iovecs to pre-allocate
Jaak Ristiojadb18efa2010-01-15 12:05:07 +0200304 * @bs: the bio_set to allocate from.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 *
306 * Description:
Kent Overstreet3f86a822012-09-06 15:35:01 -0700307 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
308 * backed by the @bs's mempool.
309 *
310 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
311 * able to allocate a bio. This is due to the mempool guarantees. To make this
312 * work, callers must never allocate more than 1 bio at a time from this pool.
313 * Callers that need to allocate more than 1 bio must always submit the
314 * previously allocated bio for IO before attempting to allocate a new one.
315 * Failure to do so can cause deadlocks under memory pressure.
316 *
317 * RETURNS:
318 * Pointer to new bio on success, NULL on failure.
319 */
Al Virodd0fc662005-10-07 07:46:04 +0100320struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
Kent Overstreet3f86a822012-09-06 15:35:01 -0700322 unsigned front_pad;
323 unsigned inline_vecs;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200324 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100325 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200326 struct bio *bio;
327 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200328
Kent Overstreet3f86a822012-09-06 15:35:01 -0700329 if (!bs) {
330 if (nr_iovecs > UIO_MAXIOV)
331 return NULL;
332
333 p = kmalloc(sizeof(struct bio) +
334 nr_iovecs * sizeof(struct bio_vec),
335 gfp_mask);
336 front_pad = 0;
337 inline_vecs = nr_iovecs;
338 } else {
339 p = mempool_alloc(bs->bio_pool, gfp_mask);
340 front_pad = bs->front_pad;
341 inline_vecs = BIO_INLINE_VECS;
342 }
343
Tejun Heo451a9eb2009-04-15 19:50:51 +0200344 if (unlikely(!p))
345 return NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100346
Kent Overstreet3f86a822012-09-06 15:35:01 -0700347 bio = p + front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100348 bio_init(bio);
349
Kent Overstreet3f86a822012-09-06 15:35:01 -0700350 if (nr_iovecs > inline_vecs) {
Ingo Molnar34053972009-02-21 11:16:36 +0100351 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
352 if (unlikely(!bvl))
353 goto err_free;
Kent Overstreet3f86a822012-09-06 15:35:01 -0700354 } else if (nr_iovecs) {
355 bvl = bio->bi_inline_vecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100356 }
Kent Overstreet3f86a822012-09-06 15:35:01 -0700357
358 bio->bi_pool = bs;
Ingo Molnar34053972009-02-21 11:16:36 +0100359 bio->bi_flags |= idx << BIO_POOL_OFFSET;
360 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100361 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100363
364err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200365 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100366 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200368EXPORT_SYMBOL(bio_alloc_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370void zero_fill_bio(struct bio *bio)
371{
372 unsigned long flags;
373 struct bio_vec *bv;
374 int i;
375
376 bio_for_each_segment(bv, bio, i) {
377 char *data = bvec_kmap_irq(bv, &flags);
378 memset(data, 0, bv->bv_len);
379 flush_dcache_page(bv->bv_page);
380 bvec_kunmap_irq(data, &flags);
381 }
382}
383EXPORT_SYMBOL(zero_fill_bio);
384
385/**
386 * bio_put - release a reference to a bio
387 * @bio: bio to release reference to
388 *
389 * Description:
390 * Put a reference to a &struct bio, either one you have gotten with
Alberto Bertogliad0bf112009-11-02 11:39:22 +0100391 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 **/
393void bio_put(struct bio *bio)
394{
395 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
396
397 /*
398 * last put frees it
399 */
Kent Overstreet4254bba2012-09-06 15:35:00 -0700400 if (atomic_dec_and_test(&bio->bi_cnt))
401 bio_free(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200403EXPORT_SYMBOL(bio_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
Jens Axboe165125e2007-07-24 09:28:11 +0200405inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
407 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
408 blk_recount_segments(q, bio);
409
410 return bio->bi_phys_segments;
411}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200412EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414/**
415 * __bio_clone - clone a bio
416 * @bio: destination bio
417 * @bio_src: bio to clone
418 *
419 * Clone a &bio. Caller will own the returned bio, but not
420 * the actual data it points to. Reference count of returned
421 * bio will be one.
422 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800423void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
Andrew Mortone525e152005-08-07 09:42:12 -0700425 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
426 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
Jens Axboe5d840702008-01-25 12:44:44 +0100428 /*
429 * most users will be overriding ->bi_bdev with a new target,
430 * so we don't set nor calculate new physical/hw segment counts here
431 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 bio->bi_sector = bio_src->bi_sector;
433 bio->bi_bdev = bio_src->bi_bdev;
434 bio->bi_flags |= 1 << BIO_CLONED;
435 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 bio->bi_vcnt = bio_src->bi_vcnt;
437 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700438 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200440EXPORT_SYMBOL(__bio_clone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
442/**
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700443 * bio_clone_bioset - clone a bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * @bio: bio to clone
445 * @gfp_mask: allocation priority
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700446 * @bs: bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 *
448 * Like __bio_clone, only also allocates the returned bio
449 */
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700450struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask,
451 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700453 struct bio *b;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700455 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200456 if (!b)
457 return NULL;
458
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200459 __bio_clone(b, bio);
460
461 if (bio_integrity(bio)) {
462 int ret;
463
Kent Overstreet1e2a410f2012-09-06 15:34:56 -0700464 ret = bio_integrity_clone(b, bio, gfp_mask);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200465
Li Zefan059ea332009-03-09 10:42:45 +0100466 if (ret < 0) {
467 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200468 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100469 }
Peter Osterlund36763472005-09-06 15:16:42 -0700470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 return b;
473}
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700474EXPORT_SYMBOL(bio_clone_bioset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476/**
477 * bio_get_nr_vecs - return approx number of vecs
478 * @bdev: I/O target
479 *
480 * Return the approximate number of pages we can send to this target.
481 * There's no guarantee that you will be able to fit this number of pages
482 * into a bio, it does not account for dynamic restrictions that vary
483 * on offset.
484 */
485int bio_get_nr_vecs(struct block_device *bdev)
486{
Jens Axboe165125e2007-07-24 09:28:11 +0200487 struct request_queue *q = bdev_get_queue(bdev);
Bernd Schubertf908ee92012-05-11 16:36:44 +0200488 int nr_pages;
489
490 nr_pages = min_t(unsigned,
Kent Overstreet5abebfd2012-02-08 22:07:18 +0100491 queue_max_segments(q),
492 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
Bernd Schubertf908ee92012-05-11 16:36:44 +0200493
494 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200497EXPORT_SYMBOL(bio_get_nr_vecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Jens Axboe165125e2007-07-24 09:28:11 +0200499static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600500 *page, unsigned int len, unsigned int offset,
501 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 int retried_segments = 0;
504 struct bio_vec *bvec;
505
506 /*
507 * cloned bio must not modify vec list
508 */
509 if (unlikely(bio_flagged(bio, BIO_CLONED)))
510 return 0;
511
Jens Axboe80cfd542006-01-06 09:43:28 +0100512 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 return 0;
514
Jens Axboe80cfd542006-01-06 09:43:28 +0100515 /*
516 * For filesystems with a blocksize smaller than the pagesize
517 * we will often be called with the same page as last time and
518 * a consecutive offset. Optimize this special case.
519 */
520 if (bio->bi_vcnt > 0) {
521 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
522
523 if (page == prev->bv_page &&
524 offset == prev->bv_offset + prev->bv_len) {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300525 unsigned int prev_bv_len = prev->bv_len;
Jens Axboe80cfd542006-01-06 09:43:28 +0100526 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200527
528 if (q->merge_bvec_fn) {
529 struct bvec_merge_data bvm = {
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300530 /* prev_bvec is already charged in
531 bi_size, discharge it in order to
532 simulate merging updated prev_bvec
533 as new bvec. */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200534 .bi_bdev = bio->bi_bdev,
535 .bi_sector = bio->bi_sector,
Dmitry Monakhov1d616582010-01-27 22:44:36 +0300536 .bi_size = bio->bi_size - prev_bv_len,
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200537 .bi_rw = bio->bi_rw,
538 };
539
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300540 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200541 prev->bv_len -= len;
542 return 0;
543 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100544 }
545
546 goto done;
547 }
548 }
549
550 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 return 0;
552
553 /*
554 * we might lose a segment or two here, but rather that than
555 * make this too complex.
556 */
557
Martin K. Petersen8a783622010-02-26 00:20:39 -0500558 while (bio->bi_phys_segments >= queue_max_segments(q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560 if (retried_segments)
561 return 0;
562
563 retried_segments = 1;
564 blk_recount_segments(q, bio);
565 }
566
567 /*
568 * setup the new entry, we might clear it again later if we
569 * cannot add the page
570 */
571 bvec = &bio->bi_io_vec[bio->bi_vcnt];
572 bvec->bv_page = page;
573 bvec->bv_len = len;
574 bvec->bv_offset = offset;
575
576 /*
577 * if queue has other restrictions (eg varying max sector size
578 * depending on offset), it can specify a merge_bvec_fn in the
579 * queue to get further control
580 */
581 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200582 struct bvec_merge_data bvm = {
583 .bi_bdev = bio->bi_bdev,
584 .bi_sector = bio->bi_sector,
585 .bi_size = bio->bi_size,
586 .bi_rw = bio->bi_rw,
587 };
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 /*
590 * merge_bvec_fn() returns number of bytes it can accept
591 * at this offset
592 */
Dmitry Monakhov8bf8c372010-03-03 06:28:06 +0300593 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 bvec->bv_page = NULL;
595 bvec->bv_len = 0;
596 bvec->bv_offset = 0;
597 return 0;
598 }
599 }
600
601 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200602 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
604
605 bio->bi_vcnt++;
606 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100607 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 bio->bi_size += len;
609 return len;
610}
611
612/**
Mike Christie6e68af62005-11-11 05:30:27 -0600613 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100614 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600615 * @bio: destination bio
616 * @page: page to add
617 * @len: vec entry length
618 * @offset: vec entry offset
619 *
620 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200621 * number of reasons, such as the bio being full or target block device
622 * limitations. The target block device must allow bio's up to PAGE_SIZE,
623 * so it is always possible to add a single page to an empty bio.
624 *
625 * This should only be used by REQ_PC bios.
Mike Christie6e68af62005-11-11 05:30:27 -0600626 */
Jens Axboe165125e2007-07-24 09:28:11 +0200627int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600628 unsigned int len, unsigned int offset)
629{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400630 return __bio_add_page(q, bio, page, len, offset,
631 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600632}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200633EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600634
635/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 * bio_add_page - attempt to add page to bio
637 * @bio: destination bio
638 * @page: page to add
639 * @len: vec entry length
640 * @offset: vec entry offset
641 *
642 * Attempt to add a page to the bio_vec maplist. This can fail for a
Andreas Gruenbacherc6428082011-05-27 14:52:09 +0200643 * number of reasons, such as the bio being full or target block device
644 * limitations. The target block device must allow bio's up to PAGE_SIZE,
645 * so it is always possible to add a single page to an empty bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 */
647int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
648 unsigned int offset)
649{
Mike Christiedefd94b2005-12-05 02:37:06 -0600650 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400651 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200653EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
655struct bio_map_data {
656 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200657 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900658 int nr_sgvecs;
659 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660};
661
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200662static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900663 struct sg_iovec *iov, int iov_count,
664 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
666 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200667 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
668 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900669 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 bio->bi_private = bmd;
671}
672
673static void bio_free_map_data(struct bio_map_data *bmd)
674{
675 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200676 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 kfree(bmd);
678}
679
Dan Carpenter121f0992011-11-16 09:21:50 +0100680static struct bio_map_data *bio_alloc_map_data(int nr_segs,
681 unsigned int iov_count,
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200682 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
Jens Axboef3f63c12010-10-29 11:46:56 -0600684 struct bio_map_data *bmd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Jens Axboef3f63c12010-10-29 11:46:56 -0600686 if (iov_count > UIO_MAXIOV)
687 return NULL;
688
689 bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (!bmd)
691 return NULL;
692
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200693 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200694 if (!bmd->iovecs) {
695 kfree(bmd);
696 return NULL;
697 }
698
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200699 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200700 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return bmd;
702
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200703 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 kfree(bmd);
705 return NULL;
706}
707
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200708static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200709 struct sg_iovec *iov, int iov_count,
710 int to_user, int from_user, int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200711{
712 int ret = 0, i;
713 struct bio_vec *bvec;
714 int iov_idx = 0;
715 unsigned int iov_off = 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200716
717 __bio_for_each_segment(bvec, bio, i, 0) {
718 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200719 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200720
721 while (bv_len && iov_idx < iov_count) {
722 unsigned int bytes;
Michal Simek0e0c6212009-06-10 12:57:07 -0700723 char __user *iov_addr;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200724
725 bytes = min_t(unsigned int,
726 iov[iov_idx].iov_len - iov_off, bv_len);
727 iov_addr = iov[iov_idx].iov_base + iov_off;
728
729 if (!ret) {
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200730 if (to_user)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200731 ret = copy_to_user(iov_addr, bv_addr,
732 bytes);
733
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200734 if (from_user)
735 ret = copy_from_user(bv_addr, iov_addr,
736 bytes);
737
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200738 if (ret)
739 ret = -EFAULT;
740 }
741
742 bv_len -= bytes;
743 bv_addr += bytes;
744 iov_addr += bytes;
745 iov_off += bytes;
746
747 if (iov[iov_idx].iov_len == iov_off) {
748 iov_idx++;
749 iov_off = 0;
750 }
751 }
752
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900753 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200754 __free_page(bvec->bv_page);
755 }
756
757 return ret;
758}
759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760/**
761 * bio_uncopy_user - finish previously mapped bio
762 * @bio: bio being terminated
763 *
764 * Free pages allocated from bio_copy_user() and write back data
765 * to user space in case of a read.
766 */
767int bio_uncopy_user(struct bio *bio)
768{
769 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900770 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
FUJITA Tomonori81882762008-09-02 16:20:19 +0900772 if (!bio_flagged(bio, BIO_NULL_MAPPED))
773 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200774 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
775 0, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 bio_free_map_data(bmd);
777 bio_put(bio);
778 return ret;
779}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200780EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
782/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200783 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900785 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200786 * @iov: the iovec.
787 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900789 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 *
791 * Prepares and returns a bio for indirect user io, bouncing data
792 * to/from kernel pages as necessary. Must be paired with
793 * call bio_uncopy_user() on io completion.
794 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900795struct bio *bio_copy_user_iov(struct request_queue *q,
796 struct rq_map_data *map_data,
797 struct sg_iovec *iov, int iov_count,
798 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 struct bio_map_data *bmd;
801 struct bio_vec *bvec;
802 struct page *page;
803 struct bio *bio;
804 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200805 int nr_pages = 0;
806 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900807 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200809 for (i = 0; i < iov_count; i++) {
810 unsigned long uaddr;
811 unsigned long end;
812 unsigned long start;
813
814 uaddr = (unsigned long)iov[i].iov_base;
815 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
816 start = uaddr >> PAGE_SHIFT;
817
Jens Axboecb4644c2010-11-10 14:36:25 +0100818 /*
819 * Overflow, abort
820 */
821 if (end < start)
822 return ERR_PTR(-EINVAL);
823
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200824 nr_pages += end - start;
825 len += iov[i].iov_len;
826 }
827
FUJITA Tomonori69838722009-04-28 20:24:29 +0200828 if (offset)
829 nr_pages++;
830
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900831 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 if (!bmd)
833 return ERR_PTR(-ENOMEM);
834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900836 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (!bio)
838 goto out_bmd;
839
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200840 if (!write_to_vm)
841 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900844
845 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900846 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900847 i = map_data->offset / PAGE_SIZE;
848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900850 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900852 bytes -= offset;
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 if (bytes > len)
855 bytes = len;
856
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900857 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900858 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900859 ret = -ENOMEM;
860 break;
861 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900862
863 page = map_data->pages[i / nr_pages];
864 page += (i % nr_pages);
865
866 i++;
867 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900868 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900869 if (!page) {
870 ret = -ENOMEM;
871 break;
872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
874
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900875 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900879 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 }
881
882 if (ret)
883 goto cleanup;
884
885 /*
886 * success
887 */
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200888 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
889 (map_data && map_data->from_user)) {
890 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200891 if (ret)
892 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 }
894
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900895 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 return bio;
897cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900898 if (!map_data)
899 bio_for_each_segment(bvec, bio, i)
900 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 bio_put(bio);
903out_bmd:
904 bio_free_map_data(bmd);
905 return ERR_PTR(ret);
906}
907
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200908/**
909 * bio_copy_user - copy user data to bio
910 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900911 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200912 * @uaddr: start of user address
913 * @len: length in bytes
914 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900915 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200916 *
917 * Prepares and returns a bio for indirect user io, bouncing data
918 * to/from kernel pages as necessary. Must be paired with
919 * call bio_uncopy_user() on io completion.
920 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900921struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
922 unsigned long uaddr, unsigned int len,
923 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200924{
925 struct sg_iovec iov;
926
927 iov.iov_base = (void __user *)uaddr;
928 iov.iov_len = len;
929
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900930 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200931}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200932EXPORT_SYMBOL(bio_copy_user);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200933
Jens Axboe165125e2007-07-24 09:28:11 +0200934static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200935 struct block_device *bdev,
936 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900937 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
James Bottomley f1970ba2005-06-20 14:06:52 +0200939 int i, j;
940 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 struct page **pages;
942 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200943 int cur_page = 0;
944 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
James Bottomley f1970ba2005-06-20 14:06:52 +0200946 for (i = 0; i < iov_count; i++) {
947 unsigned long uaddr = (unsigned long)iov[i].iov_base;
948 unsigned long len = iov[i].iov_len;
949 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
950 unsigned long start = uaddr >> PAGE_SHIFT;
951
Jens Axboecb4644c2010-11-10 14:36:25 +0100952 /*
953 * Overflow, abort
954 */
955 if (end < start)
956 return ERR_PTR(-EINVAL);
957
James Bottomley f1970ba2005-06-20 14:06:52 +0200958 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;
Jens Axboecb4644c2010-11-10 14:36:25 +0100985
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)
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001028 bio->bi_rw |= REQ_WRITE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
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.
Artem Bityutskiy0d5c3eb2012-07-25 18:12:08 +03001292 * But other code (eg, flusher threads) could clean the pages if they are mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 * 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
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001420 * something went wrong. No one should call bi_end_io() directly on a
NeilBrown5bb23a62007-09-27 12:46:13 +02001421 * 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
Tejun Heo3a366e62013-01-11 13:06:33 -08001431 trace_block_bio_complete(bio, error);
1432
NeilBrown5bb23a62007-09-27 12:46:13 +02001433 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001434 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001436EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
1438void bio_pair_release(struct bio_pair *bp)
1439{
1440 if (atomic_dec_and_test(&bp->cnt)) {
1441 struct bio *master = bp->bio1.bi_private;
1442
NeilBrown6712ecf2007-09-27 12:47:43 +02001443 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 mempool_free(bp, bp->bio2.bi_private);
1445 }
1446}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001447EXPORT_SYMBOL(bio_pair_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
NeilBrown6712ecf2007-09-27 12:47:43 +02001449static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
1451 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1452
1453 if (err)
1454 bp->error = err;
1455
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
1458
NeilBrown6712ecf2007-09-27 12:47:43 +02001459static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460{
1461 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1462
1463 if (err)
1464 bp->error = err;
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468
1469/*
Alberto Bertoglic7eee1b2009-01-25 23:36:14 -02001470 * split a bio - only worry about a bio with a single page in its iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001472struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
Denis ChengRq6feef532008-10-09 08:57:05 +02001474 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476 if (!bp)
1477 return bp;
1478
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001479 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001480 bi->bi_sector + first_sectors);
1481
Shaohua Li02f39392012-09-28 10:38:48 +02001482 BUG_ON(bi->bi_vcnt != 1 && bi->bi_vcnt != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 BUG_ON(bi->bi_idx != 0);
1484 atomic_set(&bp->cnt, 3);
1485 bp->error = 0;
1486 bp->bio1 = *bi;
1487 bp->bio2 = *bi;
1488 bp->bio2.bi_sector += first_sectors;
1489 bp->bio2.bi_size -= first_sectors << 9;
1490 bp->bio1.bi_size = first_sectors << 9;
1491
Shaohua Li02f39392012-09-28 10:38:48 +02001492 if (bi->bi_vcnt != 0) {
1493 bp->bv1 = bi->bi_io_vec[0];
1494 bp->bv2 = bi->bi_io_vec[0];
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001495
Shaohua Li02f39392012-09-28 10:38:48 +02001496 if (bio_is_rw(bi)) {
1497 bp->bv2.bv_offset += first_sectors << 9;
1498 bp->bv2.bv_len -= first_sectors << 9;
1499 bp->bv1.bv_len = first_sectors << 9;
1500 }
1501
1502 bp->bio1.bi_io_vec = &bp->bv1;
1503 bp->bio2.bi_io_vec = &bp->bv2;
1504
1505 bp->bio1.bi_max_vecs = 1;
1506 bp->bio2.bi_max_vecs = 1;
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001507 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 bp->bio1.bi_end_io = bio_pair_end_1;
1510 bp->bio2.bi_end_io = bio_pair_end_2;
1511
1512 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001513 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001515 if (bio_integrity(bi))
1516 bio_integrity_split(bi, bp, first_sectors);
1517
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 return bp;
1519}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001520EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001522/**
1523 * bio_sector_offset - Find hardware sector offset in bio
1524 * @bio: bio to inspect
1525 * @index: bio_vec index
1526 * @offset: offset in bv_page
1527 *
1528 * Return the number of hardware sectors between beginning of bio
1529 * and an end point indicated by a bio_vec index and an offset
1530 * within that vector's page.
1531 */
1532sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1533 unsigned int offset)
1534{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001535 unsigned int sector_sz;
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001536 struct bio_vec *bv;
1537 sector_t sectors;
1538 int i;
1539
Martin K. Petersene1defc42009-05-22 17:17:49 -04001540 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001541 sectors = 0;
1542
1543 if (index >= bio->bi_idx)
1544 index = bio->bi_vcnt - 1;
1545
1546 __bio_for_each_segment(bv, bio, i, 0) {
1547 if (i == index) {
1548 if (offset > bv->bv_offset)
1549 sectors += (offset - bv->bv_offset) / sector_sz;
1550 break;
1551 }
1552
1553 sectors += bv->bv_len / sector_sz;
1554 }
1555
1556 return sectors;
1557}
1558EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
1560/*
1561 * create memory pools for biovec's in a bio_set.
1562 * use the global biovec slabs created for general use.
1563 */
Jens Axboe59725112007-04-02 10:06:42 +02001564static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565{
Jens Axboe7ff93452008-12-11 11:53:43 +01001566 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Jens Axboe7ff93452008-12-11 11:53:43 +01001568 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1569 if (!bs->bvec_pool)
1570 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 return 0;
1573}
1574
1575static void biovec_free_pools(struct bio_set *bs)
1576{
Jens Axboe7ff93452008-12-11 11:53:43 +01001577 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578}
1579
1580void bioset_free(struct bio_set *bs)
1581{
1582 if (bs->bio_pool)
1583 mempool_destroy(bs->bio_pool);
1584
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001585 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001587 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589 kfree(bs);
1590}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001591EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
Jens Axboebb799ca2008-12-10 15:35:05 +01001593/**
1594 * bioset_create - Create a bio_set
1595 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1596 * @front_pad: Number of bytes to allocate in front of the returned bio
1597 *
1598 * Description:
1599 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1600 * to ask for a number of bytes to be allocated in front of the bio.
1601 * Front pad allocation is useful for embedding the bio inside
1602 * another structure, to avoid allocating extra data to go with the bio.
1603 * Note that the bio must be embedded at the END of that structure always,
1604 * or things will break badly.
1605 */
1606struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607{
Jens Axboe392ddc32008-12-23 12:42:54 +01001608 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001609 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Jens Axboe1b434492008-10-22 20:32:58 +02001611 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (!bs)
1613 return NULL;
1614
Jens Axboebb799ca2008-12-10 15:35:05 +01001615 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001616
Jens Axboe392ddc32008-12-23 12:42:54 +01001617 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001618 if (!bs->bio_slab) {
1619 kfree(bs);
1620 return NULL;
1621 }
1622
1623 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 if (!bs->bio_pool)
1625 goto bad;
1626
Jens Axboebb799ca2008-12-10 15:35:05 +01001627 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 return bs;
1629
1630bad:
1631 bioset_free(bs);
1632 return NULL;
1633}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001634EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Tejun Heo852c7882012-03-05 13:15:27 -08001636#ifdef CONFIG_BLK_CGROUP
1637/**
1638 * bio_associate_current - associate a bio with %current
1639 * @bio: target bio
1640 *
1641 * Associate @bio with %current if it hasn't been associated yet. Block
1642 * layer will treat @bio as if it were issued by %current no matter which
1643 * task actually issues it.
1644 *
1645 * This function takes an extra reference of @task's io_context and blkcg
1646 * which will be put when @bio is released. The caller must own @bio,
1647 * ensure %current->io_context exists, and is responsible for synchronizing
1648 * calls to this function.
1649 */
1650int bio_associate_current(struct bio *bio)
1651{
1652 struct io_context *ioc;
1653 struct cgroup_subsys_state *css;
1654
1655 if (bio->bi_ioc)
1656 return -EBUSY;
1657
1658 ioc = current->io_context;
1659 if (!ioc)
1660 return -ENOENT;
1661
1662 /* acquire active ref on @ioc and associate */
1663 get_io_context_active(ioc);
1664 bio->bi_ioc = ioc;
1665
1666 /* associate blkcg if exists */
1667 rcu_read_lock();
1668 css = task_subsys_state(current, blkio_subsys_id);
1669 if (css && css_tryget(css))
1670 bio->bi_css = css;
1671 rcu_read_unlock();
1672
1673 return 0;
1674}
1675
1676/**
1677 * bio_disassociate_task - undo bio_associate_current()
1678 * @bio: target bio
1679 */
1680void bio_disassociate_task(struct bio *bio)
1681{
1682 if (bio->bi_ioc) {
1683 put_io_context(bio->bi_ioc);
1684 bio->bi_ioc = NULL;
1685 }
1686 if (bio->bi_css) {
1687 css_put(bio->bi_css);
1688 bio->bi_css = NULL;
1689 }
1690}
1691
1692#endif /* CONFIG_BLK_CGROUP */
1693
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694static void __init biovec_init_slabs(void)
1695{
1696 int i;
1697
1698 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1699 int size;
1700 struct biovec_slab *bvs = bvec_slabs + i;
1701
Jens Axboea7fcd372008-12-05 16:10:29 +01001702 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1703 bvs->slab = NULL;
1704 continue;
1705 }
Jens Axboea7fcd372008-12-05 16:10:29 +01001706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 size = bvs->nr_vecs * sizeof(struct bio_vec);
1708 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001709 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 }
1711}
1712
1713static int __init init_bio(void)
1714{
Jens Axboebb799ca2008-12-10 15:35:05 +01001715 bio_slab_max = 2;
1716 bio_slab_nr = 0;
1717 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1718 if (!bio_slabs)
1719 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001721 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 biovec_init_slabs();
1723
Jens Axboebb799ca2008-12-10 15:35:05 +01001724 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 if (!fs_bio_set)
1726 panic("bio: can't allocate bios\n");
1727
Martin K. Petersena91a2782011-03-17 11:11:05 +01001728 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
1729 panic("bio: can't create integrity pool\n");
1730
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001731 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1732 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 if (!bio_split_pool)
1734 panic("bio: can't create split pool\n");
1735
1736 return 0;
1737}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738subsys_initcall(init_bio);