blob: 402cb84a92a1dbd538dc4eead68cd474bea3389b [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) {
81 struct bio_slab *bslab = &bio_slabs[i];
82
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 *
Jens Axboebb799ca2008-12-10 15:35:05 +0100275 * Note that the caller must set ->bi_destructor on succesful return
276 * 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 *
328 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask
329 * contains __GFP_WAIT, the allocation is guaranteed to succeed.
330 *
331 * RETURNS:
332 * Pointer to new bio on success, NULL on failure.
333 */
334struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
335{
336 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
337
338 if (bio)
339 bio->bi_destructor = bio_fs_destructor;
340
341 return bio;
342}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200343EXPORT_SYMBOL(bio_alloc);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200344
345static void bio_kmalloc_destructor(struct bio *bio)
346{
347 if (bio_integrity(bio))
Martin K. Petersen7878cba2009-06-26 15:37:49 +0200348 bio_integrity_free(bio, fs_bio_set);
Tejun Heo451a9eb2009-04-15 19:50:51 +0200349 kfree(bio);
350}
351
Jens Axboe86c824b2009-04-15 09:00:07 +0200352/**
353 * bio_alloc - allocate a bio for I/O
354 * @gfp_mask: the GFP_ mask given to the slab allocator
355 * @nr_iovecs: number of iovecs to pre-allocate
356 *
357 * Description:
358 * bio_alloc will allocate a bio and associated bio_vec array that can hold
359 * at least @nr_iovecs entries. Allocations will be done from the
360 * fs_bio_set. Also see @bio_alloc_bioset.
361 *
362 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
363 * a bio. This is due to the mempool guarantees. To make this work, callers
Nikanth Karthikesan76d93ff2009-04-22 13:38:58 +0530364 * must never allocate more than 1 bio at a time from this pool. Callers
Jens Axboe86c824b2009-04-15 09:00:07 +0200365 * that need to allocate more than 1 bio must always submit the previously
Nikanth Karthikesan76d93ff2009-04-22 13:38:58 +0530366 * allocated bio for IO before attempting to allocate a new one. Failure to
Jens Axboe86c824b2009-04-15 09:00:07 +0200367 * do so can cause livelocks under memory pressure.
368 *
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
410 * bio_alloc or bio_get. The last put of a bio will free it.
411 **/
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;
510 if (nr_pages > queue_max_phys_segments(q))
511 nr_pages = queue_max_phys_segments(q);
512 if (nr_pages > queue_max_hw_segments(q))
513 nr_pages = queue_max_hw_segments(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
515 return nr_pages;
516}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200517EXPORT_SYMBOL(bio_get_nr_vecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Jens Axboe165125e2007-07-24 09:28:11 +0200519static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600520 *page, unsigned int len, unsigned int offset,
521 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522{
523 int retried_segments = 0;
524 struct bio_vec *bvec;
525
526 /*
527 * cloned bio must not modify vec list
528 */
529 if (unlikely(bio_flagged(bio, BIO_CLONED)))
530 return 0;
531
Jens Axboe80cfd542006-01-06 09:43:28 +0100532 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 return 0;
534
Jens Axboe80cfd542006-01-06 09:43:28 +0100535 /*
536 * For filesystems with a blocksize smaller than the pagesize
537 * we will often be called with the same page as last time and
538 * a consecutive offset. Optimize this special case.
539 */
540 if (bio->bi_vcnt > 0) {
541 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
542
543 if (page == prev->bv_page &&
544 offset == prev->bv_offset + prev->bv_len) {
545 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200546
547 if (q->merge_bvec_fn) {
548 struct bvec_merge_data bvm = {
549 .bi_bdev = bio->bi_bdev,
550 .bi_sector = bio->bi_sector,
551 .bi_size = bio->bi_size,
552 .bi_rw = bio->bi_rw,
553 };
554
555 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
556 prev->bv_len -= len;
557 return 0;
558 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100559 }
560
561 goto done;
562 }
563 }
564
565 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 return 0;
567
568 /*
569 * we might lose a segment or two here, but rather that than
570 * make this too complex.
571 */
572
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400573 while (bio->bi_phys_segments >= queue_max_phys_segments(q)
574 || bio->bi_phys_segments >= queue_max_hw_segments(q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
576 if (retried_segments)
577 return 0;
578
579 retried_segments = 1;
580 blk_recount_segments(q, bio);
581 }
582
583 /*
584 * setup the new entry, we might clear it again later if we
585 * cannot add the page
586 */
587 bvec = &bio->bi_io_vec[bio->bi_vcnt];
588 bvec->bv_page = page;
589 bvec->bv_len = len;
590 bvec->bv_offset = offset;
591
592 /*
593 * if queue has other restrictions (eg varying max sector size
594 * depending on offset), it can specify a merge_bvec_fn in the
595 * queue to get further control
596 */
597 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200598 struct bvec_merge_data bvm = {
599 .bi_bdev = bio->bi_bdev,
600 .bi_sector = bio->bi_sector,
601 .bi_size = bio->bi_size,
602 .bi_rw = bio->bi_rw,
603 };
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 /*
606 * merge_bvec_fn() returns number of bytes it can accept
607 * at this offset
608 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200609 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 bvec->bv_page = NULL;
611 bvec->bv_len = 0;
612 bvec->bv_offset = 0;
613 return 0;
614 }
615 }
616
617 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200618 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
620
621 bio->bi_vcnt++;
622 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100623 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 bio->bi_size += len;
625 return len;
626}
627
628/**
Mike Christie6e68af62005-11-11 05:30:27 -0600629 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100630 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600631 * @bio: destination bio
632 * @page: page to add
633 * @len: vec entry length
634 * @offset: vec entry offset
635 *
636 * Attempt to add a page to the bio_vec maplist. This can fail for a
637 * number of reasons, such as the bio being full or target block
638 * device limitations. The target block device must allow bio's
639 * smaller than PAGE_SIZE, so it is always possible to add a single
640 * page to an empty bio. This should only be used by REQ_PC bios.
641 */
Jens Axboe165125e2007-07-24 09:28:11 +0200642int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600643 unsigned int len, unsigned int offset)
644{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400645 return __bio_add_page(q, bio, page, len, offset,
646 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600647}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200648EXPORT_SYMBOL(bio_add_pc_page);
Mike Christie6e68af62005-11-11 05:30:27 -0600649
650/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 * bio_add_page - attempt to add page to bio
652 * @bio: destination bio
653 * @page: page to add
654 * @len: vec entry length
655 * @offset: vec entry offset
656 *
657 * Attempt to add a page to the bio_vec maplist. This can fail for a
658 * number of reasons, such as the bio being full or target block
659 * device limitations. The target block device must allow bio's
660 * smaller than PAGE_SIZE, so it is always possible to add a single
661 * page to an empty bio.
662 */
663int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
664 unsigned int offset)
665{
Mike Christiedefd94b2005-12-05 02:37:06 -0600666 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400667 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200669EXPORT_SYMBOL(bio_add_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
671struct bio_map_data {
672 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200673 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900674 int nr_sgvecs;
675 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676};
677
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200678static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900679 struct sg_iovec *iov, int iov_count,
680 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200683 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
684 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900685 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 bio->bi_private = bmd;
687}
688
689static void bio_free_map_data(struct bio_map_data *bmd)
690{
691 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200692 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 kfree(bmd);
694}
695
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200696static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
697 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200699 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701 if (!bmd)
702 return NULL;
703
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200704 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200705 if (!bmd->iovecs) {
706 kfree(bmd);
707 return NULL;
708 }
709
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200710 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200711 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 return bmd;
713
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200714 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 kfree(bmd);
716 return NULL;
717}
718
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200719static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200720 struct sg_iovec *iov, int iov_count,
721 int to_user, int from_user, int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200722{
723 int ret = 0, i;
724 struct bio_vec *bvec;
725 int iov_idx = 0;
726 unsigned int iov_off = 0;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200727
728 __bio_for_each_segment(bvec, bio, i, 0) {
729 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200730 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200731
732 while (bv_len && iov_idx < iov_count) {
733 unsigned int bytes;
Michal Simek0e0c6212009-06-10 12:57:07 -0700734 char __user *iov_addr;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200735
736 bytes = min_t(unsigned int,
737 iov[iov_idx].iov_len - iov_off, bv_len);
738 iov_addr = iov[iov_idx].iov_base + iov_off;
739
740 if (!ret) {
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200741 if (to_user)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200742 ret = copy_to_user(iov_addr, bv_addr,
743 bytes);
744
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200745 if (from_user)
746 ret = copy_from_user(bv_addr, iov_addr,
747 bytes);
748
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200749 if (ret)
750 ret = -EFAULT;
751 }
752
753 bv_len -= bytes;
754 bv_addr += bytes;
755 iov_addr += bytes;
756 iov_off += bytes;
757
758 if (iov[iov_idx].iov_len == iov_off) {
759 iov_idx++;
760 iov_off = 0;
761 }
762 }
763
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900764 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200765 __free_page(bvec->bv_page);
766 }
767
768 return ret;
769}
770
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771/**
772 * bio_uncopy_user - finish previously mapped bio
773 * @bio: bio being terminated
774 *
775 * Free pages allocated from bio_copy_user() and write back data
776 * to user space in case of a read.
777 */
778int bio_uncopy_user(struct bio *bio)
779{
780 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900781 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
FUJITA Tomonori81882762008-09-02 16:20:19 +0900783 if (!bio_flagged(bio, BIO_NULL_MAPPED))
784 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200785 bmd->nr_sgvecs, bio_data_dir(bio) == READ,
786 0, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 bio_free_map_data(bmd);
788 bio_put(bio);
789 return ret;
790}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200791EXPORT_SYMBOL(bio_uncopy_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
793/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200794 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900796 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200797 * @iov: the iovec.
798 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900800 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 *
802 * Prepares and returns a bio for indirect user io, bouncing data
803 * to/from kernel pages as necessary. Must be paired with
804 * call bio_uncopy_user() on io completion.
805 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900806struct bio *bio_copy_user_iov(struct request_queue *q,
807 struct rq_map_data *map_data,
808 struct sg_iovec *iov, int iov_count,
809 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 struct bio_map_data *bmd;
812 struct bio_vec *bvec;
813 struct page *page;
814 struct bio *bio;
815 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200816 int nr_pages = 0;
817 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900818 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200820 for (i = 0; i < iov_count; i++) {
821 unsigned long uaddr;
822 unsigned long end;
823 unsigned long start;
824
825 uaddr = (unsigned long)iov[i].iov_base;
826 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
827 start = uaddr >> PAGE_SHIFT;
828
829 nr_pages += end - start;
830 len += iov[i].iov_len;
831 }
832
FUJITA Tomonori69838722009-04-28 20:24:29 +0200833 if (offset)
834 nr_pages++;
835
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900836 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (!bmd)
838 return ERR_PTR(-ENOMEM);
839
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900841 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 if (!bio)
843 goto out_bmd;
844
845 bio->bi_rw |= (!write_to_vm << BIO_RW);
846
847 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900848
849 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900850 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900851 i = map_data->offset / PAGE_SIZE;
852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900854 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900856 bytes -= offset;
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 if (bytes > len)
859 bytes = len;
860
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900861 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900862 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900863 ret = -ENOMEM;
864 break;
865 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900866
867 page = map_data->pages[i / nr_pages];
868 page += (i % nr_pages);
869
870 i++;
871 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900872 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900873 if (!page) {
874 ret = -ENOMEM;
875 break;
876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
878
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900879 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
882 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900883 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
885
886 if (ret)
887 goto cleanup;
888
889 /*
890 * success
891 */
FUJITA Tomonoriecb554a2009-07-09 14:46:53 +0200892 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
893 (map_data && map_data->from_user)) {
894 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200895 if (ret)
896 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
898
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900899 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 return bio;
901cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900902 if (!map_data)
903 bio_for_each_segment(bvec, bio, i)
904 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 bio_put(bio);
907out_bmd:
908 bio_free_map_data(bmd);
909 return ERR_PTR(ret);
910}
911
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200912/**
913 * bio_copy_user - copy user data to bio
914 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900915 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200916 * @uaddr: start of user address
917 * @len: length in bytes
918 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900919 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200920 *
921 * Prepares and returns a bio for indirect user io, bouncing data
922 * to/from kernel pages as necessary. Must be paired with
923 * call bio_uncopy_user() on io completion.
924 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900925struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
926 unsigned long uaddr, unsigned int len,
927 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200928{
929 struct sg_iovec iov;
930
931 iov.iov_base = (void __user *)uaddr;
932 iov.iov_len = len;
933
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900934 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200935}
H Hartley Sweetena112a712009-09-26 16:19:21 +0200936EXPORT_SYMBOL(bio_copy_user);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200937
Jens Axboe165125e2007-07-24 09:28:11 +0200938static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200939 struct block_device *bdev,
940 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900941 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
James Bottomley f1970ba2005-06-20 14:06:52 +0200943 int i, j;
944 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 struct page **pages;
946 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200947 int cur_page = 0;
948 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
James Bottomley f1970ba2005-06-20 14:06:52 +0200950 for (i = 0; i < iov_count; i++) {
951 unsigned long uaddr = (unsigned long)iov[i].iov_base;
952 unsigned long len = iov[i].iov_len;
953 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
954 unsigned long start = uaddr >> PAGE_SHIFT;
955
956 nr_pages += end - start;
957 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100958 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200959 */
Mike Christiead2d7222006-12-01 10:40:20 +0100960 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200961 return ERR_PTR(-EINVAL);
962 }
963
964 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 return ERR_PTR(-EINVAL);
966
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900967 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 if (!bio)
969 return ERR_PTR(-ENOMEM);
970
971 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900972 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 if (!pages)
974 goto out;
975
James Bottomley f1970ba2005-06-20 14:06:52 +0200976 for (i = 0; i < iov_count; i++) {
977 unsigned long uaddr = (unsigned long)iov[i].iov_base;
978 unsigned long len = iov[i].iov_len;
979 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
980 unsigned long start = uaddr >> PAGE_SHIFT;
981 const int local_nr_pages = end - start;
982 const int page_limit = cur_page + local_nr_pages;
983
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700984 ret = get_user_pages_fast(uaddr, local_nr_pages,
985 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200986 if (ret < local_nr_pages) {
987 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200988 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
James Bottomley f1970ba2005-06-20 14:06:52 +0200991 offset = uaddr & ~PAGE_MASK;
992 for (j = cur_page; j < page_limit; j++) {
993 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
James Bottomley f1970ba2005-06-20 14:06:52 +0200995 if (len <= 0)
996 break;
997
998 if (bytes > len)
999 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
James Bottomley f1970ba2005-06-20 14:06:52 +02001001 /*
1002 * sorry...
1003 */
Mike Christiedefd94b2005-12-05 02:37:06 -06001004 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1005 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +02001006 break;
1007
1008 len -= bytes;
1009 offset = 0;
1010 }
1011
1012 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 /*
James Bottomley f1970ba2005-06-20 14:06:52 +02001014 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001016 while (j < page_limit)
1017 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 kfree(pages);
1021
1022 /*
1023 * set data direction, and check if mapped pages need bouncing
1024 */
1025 if (!write_to_vm)
1026 bio->bi_rw |= (1 << BIO_RW);
1027
James Bottomley f1970ba2005-06-20 14:06:52 +02001028 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1030 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001031
1032 out_unmap:
1033 for (i = 0; i < nr_pages; i++) {
1034 if(!pages[i])
1035 break;
1036 page_cache_release(pages[i]);
1037 }
1038 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 kfree(pages);
1040 bio_put(bio);
1041 return ERR_PTR(ret);
1042}
1043
1044/**
1045 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001046 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 * @bdev: destination block device
1048 * @uaddr: start of user address
1049 * @len: length in bytes
1050 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001051 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 *
1053 * Map the user space address into a bio suitable for io to a block
1054 * device. Returns an error pointer in case of error.
1055 */
Jens Axboe165125e2007-07-24 09:28:11 +02001056struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001057 unsigned long uaddr, unsigned int len, int write_to_vm,
1058 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
James Bottomley f1970ba2005-06-20 14:06:52 +02001060 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001062 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001063 iov.iov_len = len;
1064
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001065 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001066}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001067EXPORT_SYMBOL(bio_map_user);
James Bottomley f1970ba2005-06-20 14:06:52 +02001068
1069/**
1070 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001071 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001072 * @bdev: destination block device
1073 * @iov: the iovec.
1074 * @iov_count: number of elements in the iovec
1075 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001076 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001077 *
1078 * Map the user space address into a bio suitable for io to a block
1079 * device. Returns an error pointer in case of error.
1080 */
Jens Axboe165125e2007-07-24 09:28:11 +02001081struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001082 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001083 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001084{
1085 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001086
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001087 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1088 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (IS_ERR(bio))
1090 return bio;
1091
1092 /*
1093 * subtle -- if __bio_map_user() ended up bouncing a bio,
1094 * it would normally disappear when its bi_end_io is run.
1095 * however, we need it for the unmap, so grab an extra
1096 * reference to it
1097 */
1098 bio_get(bio);
1099
Mike Christie0e75f902006-12-01 10:40:55 +01001100 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101}
1102
1103static void __bio_unmap_user(struct bio *bio)
1104{
1105 struct bio_vec *bvec;
1106 int i;
1107
1108 /*
1109 * make sure we dirty pages we wrote to
1110 */
1111 __bio_for_each_segment(bvec, bio, i, 0) {
1112 if (bio_data_dir(bio) == READ)
1113 set_page_dirty_lock(bvec->bv_page);
1114
1115 page_cache_release(bvec->bv_page);
1116 }
1117
1118 bio_put(bio);
1119}
1120
1121/**
1122 * bio_unmap_user - unmap a bio
1123 * @bio: the bio being unmapped
1124 *
1125 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1126 * a process context.
1127 *
1128 * bio_unmap_user() may sleep.
1129 */
1130void bio_unmap_user(struct bio *bio)
1131{
1132 __bio_unmap_user(bio);
1133 bio_put(bio);
1134}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001135EXPORT_SYMBOL(bio_unmap_user);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
NeilBrown6712ecf2007-09-27 12:47:43 +02001137static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001138{
Jens Axboeb8238252005-06-20 14:05:27 +02001139 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001140}
1141
Jens Axboe165125e2007-07-24 09:28:11 +02001142static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001143 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001144{
1145 unsigned long kaddr = (unsigned long)data;
1146 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1147 unsigned long start = kaddr >> PAGE_SHIFT;
1148 const int nr_pages = end - start;
1149 int offset, i;
1150 struct bio *bio;
1151
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001152 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001153 if (!bio)
1154 return ERR_PTR(-ENOMEM);
1155
1156 offset = offset_in_page(kaddr);
1157 for (i = 0; i < nr_pages; i++) {
1158 unsigned int bytes = PAGE_SIZE - offset;
1159
1160 if (len <= 0)
1161 break;
1162
1163 if (bytes > len)
1164 bytes = len;
1165
Mike Christiedefd94b2005-12-05 02:37:06 -06001166 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1167 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001168 break;
1169
1170 data += bytes;
1171 len -= bytes;
1172 offset = 0;
1173 }
1174
Jens Axboeb8238252005-06-20 14:05:27 +02001175 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001176 return bio;
1177}
1178
1179/**
1180 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001181 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001182 * @data: pointer to buffer to map
1183 * @len: length in bytes
1184 * @gfp_mask: allocation flags for bio allocation
1185 *
1186 * Map the kernel address into a bio suitable for io to a block
1187 * device. Returns an error pointer in case of error.
1188 */
Jens Axboe165125e2007-07-24 09:28:11 +02001189struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001190 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001191{
1192 struct bio *bio;
1193
1194 bio = __bio_map_kern(q, data, len, gfp_mask);
1195 if (IS_ERR(bio))
1196 return bio;
1197
1198 if (bio->bi_size == len)
1199 return bio;
1200
1201 /*
1202 * Don't support partial mappings.
1203 */
1204 bio_put(bio);
1205 return ERR_PTR(-EINVAL);
1206}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001207EXPORT_SYMBOL(bio_map_kern);
Mike Christie df46b9a2005-06-20 14:04:44 +02001208
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001209static void bio_copy_kern_endio(struct bio *bio, int err)
1210{
1211 struct bio_vec *bvec;
1212 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001213 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001214 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001215 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001216
1217 __bio_for_each_segment(bvec, bio, i, 0) {
1218 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001219 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001220
Tejun Heo4fc981e2009-05-19 18:33:06 +09001221 if (read)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001222 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001223
1224 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001225 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001226 }
1227
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001228 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001229 bio_put(bio);
1230}
1231
1232/**
1233 * bio_copy_kern - copy kernel address into bio
1234 * @q: the struct request_queue for the bio
1235 * @data: pointer to buffer to copy
1236 * @len: length in bytes
1237 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001238 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001239 *
1240 * copy the kernel address into a bio suitable for io to a block
1241 * device. Returns an error pointer in case of error.
1242 */
1243struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1244 gfp_t gfp_mask, int reading)
1245{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001246 struct bio *bio;
1247 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001248 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001249
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001250 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1251 if (IS_ERR(bio))
1252 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001253
1254 if (!reading) {
1255 void *p = data;
1256
1257 bio_for_each_segment(bvec, bio, i) {
1258 char *addr = page_address(bvec->bv_page);
1259
1260 memcpy(addr, p, bvec->bv_len);
1261 p += bvec->bv_len;
1262 }
1263 }
1264
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001265 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001266
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001267 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001268}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001269EXPORT_SYMBOL(bio_copy_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271/*
1272 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1273 * for performing direct-IO in BIOs.
1274 *
1275 * The problem is that we cannot run set_page_dirty() from interrupt context
1276 * because the required locks are not interrupt-safe. So what we can do is to
1277 * mark the pages dirty _before_ performing IO. And in interrupt context,
1278 * check that the pages are still dirty. If so, fine. If not, redirty them
1279 * in process context.
1280 *
1281 * We special-case compound pages here: normally this means reads into hugetlb
1282 * pages. The logic in here doesn't really work right for compound pages
1283 * because the VM does not uniformly chase down the head page in all cases.
1284 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1285 * handle them at all. So we skip compound pages here at an early stage.
1286 *
1287 * Note that this code is very hard to test under normal circumstances because
1288 * direct-io pins the pages with get_user_pages(). This makes
1289 * is_page_cache_freeable return false, and the VM will not clean the pages.
1290 * But other code (eg, pdflush) could clean the pages if they are mapped
1291 * pagecache.
1292 *
1293 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1294 * deferred bio dirtying paths.
1295 */
1296
1297/*
1298 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1299 */
1300void bio_set_pages_dirty(struct bio *bio)
1301{
1302 struct bio_vec *bvec = bio->bi_io_vec;
1303 int i;
1304
1305 for (i = 0; i < bio->bi_vcnt; i++) {
1306 struct page *page = bvec[i].bv_page;
1307
1308 if (page && !PageCompound(page))
1309 set_page_dirty_lock(page);
1310 }
1311}
1312
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001313static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314{
1315 struct bio_vec *bvec = bio->bi_io_vec;
1316 int i;
1317
1318 for (i = 0; i < bio->bi_vcnt; i++) {
1319 struct page *page = bvec[i].bv_page;
1320
1321 if (page)
1322 put_page(page);
1323 }
1324}
1325
1326/*
1327 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1328 * If they are, then fine. If, however, some pages are clean then they must
1329 * have been written out during the direct-IO read. So we take another ref on
1330 * the BIO and the offending pages and re-dirty the pages in process context.
1331 *
1332 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1333 * here on. It will run one page_cache_release() against each page and will
1334 * run one bio_put() against the BIO.
1335 */
1336
David Howells65f27f32006-11-22 14:55:48 +00001337static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
David Howells65f27f32006-11-22 14:55:48 +00001339static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340static DEFINE_SPINLOCK(bio_dirty_lock);
1341static struct bio *bio_dirty_list;
1342
1343/*
1344 * This runs in process context
1345 */
David Howells65f27f32006-11-22 14:55:48 +00001346static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 unsigned long flags;
1349 struct bio *bio;
1350
1351 spin_lock_irqsave(&bio_dirty_lock, flags);
1352 bio = bio_dirty_list;
1353 bio_dirty_list = NULL;
1354 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1355
1356 while (bio) {
1357 struct bio *next = bio->bi_private;
1358
1359 bio_set_pages_dirty(bio);
1360 bio_release_pages(bio);
1361 bio_put(bio);
1362 bio = next;
1363 }
1364}
1365
1366void bio_check_pages_dirty(struct bio *bio)
1367{
1368 struct bio_vec *bvec = bio->bi_io_vec;
1369 int nr_clean_pages = 0;
1370 int i;
1371
1372 for (i = 0; i < bio->bi_vcnt; i++) {
1373 struct page *page = bvec[i].bv_page;
1374
1375 if (PageDirty(page) || PageCompound(page)) {
1376 page_cache_release(page);
1377 bvec[i].bv_page = NULL;
1378 } else {
1379 nr_clean_pages++;
1380 }
1381 }
1382
1383 if (nr_clean_pages) {
1384 unsigned long flags;
1385
1386 spin_lock_irqsave(&bio_dirty_lock, flags);
1387 bio->bi_private = bio_dirty_list;
1388 bio_dirty_list = bio;
1389 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1390 schedule_work(&bio_dirty_work);
1391 } else {
1392 bio_put(bio);
1393 }
1394}
1395
1396/**
1397 * bio_endio - end I/O on a bio
1398 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 * @error: error, if any
1400 *
1401 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001402 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001403 * preferred way to end I/O on a bio, it takes care of clearing
1404 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1405 * established -Exxxx (-EIO, for instance) error values in case
1406 * something went wrong. Noone should call bi_end_io() directly on a
1407 * bio unless they own it and thus know that it has an end_io
1408 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001410void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
1412 if (error)
1413 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001414 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1415 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
NeilBrown5bb23a62007-09-27 12:46:13 +02001417 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001418 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001420EXPORT_SYMBOL(bio_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
1422void bio_pair_release(struct bio_pair *bp)
1423{
1424 if (atomic_dec_and_test(&bp->cnt)) {
1425 struct bio *master = bp->bio1.bi_private;
1426
NeilBrown6712ecf2007-09-27 12:47:43 +02001427 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 mempool_free(bp, bp->bio2.bi_private);
1429 }
1430}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001431EXPORT_SYMBOL(bio_pair_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
NeilBrown6712ecf2007-09-27 12:47:43 +02001433static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434{
1435 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1436
1437 if (err)
1438 bp->error = err;
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441}
1442
NeilBrown6712ecf2007-09-27 12:47:43 +02001443static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
1445 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1446
1447 if (err)
1448 bp->error = err;
1449
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451}
1452
1453/*
Alberto Bertoglic7eee1b2009-01-25 23:36:14 -02001454 * split a bio - only worry about a bio with a single page in its iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001456struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
Denis ChengRq6feef532008-10-09 08:57:05 +02001458 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
1460 if (!bp)
1461 return bp;
1462
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001463 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001464 bi->bi_sector + first_sectors);
1465
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 BUG_ON(bi->bi_vcnt != 1);
1467 BUG_ON(bi->bi_idx != 0);
1468 atomic_set(&bp->cnt, 3);
1469 bp->error = 0;
1470 bp->bio1 = *bi;
1471 bp->bio2 = *bi;
1472 bp->bio2.bi_sector += first_sectors;
1473 bp->bio2.bi_size -= first_sectors << 9;
1474 bp->bio1.bi_size = first_sectors << 9;
1475
1476 bp->bv1 = bi->bi_io_vec[0];
1477 bp->bv2 = bi->bi_io_vec[0];
1478 bp->bv2.bv_offset += first_sectors << 9;
1479 bp->bv2.bv_len -= first_sectors << 9;
1480 bp->bv1.bv_len = first_sectors << 9;
1481
1482 bp->bio1.bi_io_vec = &bp->bv1;
1483 bp->bio2.bi_io_vec = &bp->bv2;
1484
NeilBrowna2eb0c12006-05-22 22:35:27 -07001485 bp->bio1.bi_max_vecs = 1;
1486 bp->bio2.bi_max_vecs = 1;
1487
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 bp->bio1.bi_end_io = bio_pair_end_1;
1489 bp->bio2.bi_end_io = bio_pair_end_2;
1490
1491 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001492 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001494 if (bio_integrity(bi))
1495 bio_integrity_split(bi, bp, first_sectors);
1496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 return bp;
1498}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001499EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001501/**
1502 * bio_sector_offset - Find hardware sector offset in bio
1503 * @bio: bio to inspect
1504 * @index: bio_vec index
1505 * @offset: offset in bv_page
1506 *
1507 * Return the number of hardware sectors between beginning of bio
1508 * and an end point indicated by a bio_vec index and an offset
1509 * within that vector's page.
1510 */
1511sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1512 unsigned int offset)
1513{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001514 unsigned int sector_sz;
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001515 struct bio_vec *bv;
1516 sector_t sectors;
1517 int i;
1518
Martin K. Petersene1defc42009-05-22 17:17:49 -04001519 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001520 sectors = 0;
1521
1522 if (index >= bio->bi_idx)
1523 index = bio->bi_vcnt - 1;
1524
1525 __bio_for_each_segment(bv, bio, i, 0) {
1526 if (i == index) {
1527 if (offset > bv->bv_offset)
1528 sectors += (offset - bv->bv_offset) / sector_sz;
1529 break;
1530 }
1531
1532 sectors += bv->bv_len / sector_sz;
1533 }
1534
1535 return sectors;
1536}
1537EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
1539/*
1540 * create memory pools for biovec's in a bio_set.
1541 * use the global biovec slabs created for general use.
1542 */
Jens Axboe59725112007-04-02 10:06:42 +02001543static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544{
Jens Axboe7ff93452008-12-11 11:53:43 +01001545 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Jens Axboe7ff93452008-12-11 11:53:43 +01001547 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1548 if (!bs->bvec_pool)
1549 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 return 0;
1552}
1553
1554static void biovec_free_pools(struct bio_set *bs)
1555{
Jens Axboe7ff93452008-12-11 11:53:43 +01001556 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557}
1558
1559void bioset_free(struct bio_set *bs)
1560{
1561 if (bs->bio_pool)
1562 mempool_destroy(bs->bio_pool);
1563
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001564 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001566 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568 kfree(bs);
1569}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001570EXPORT_SYMBOL(bioset_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Jens Axboebb799ca2008-12-10 15:35:05 +01001572/**
1573 * bioset_create - Create a bio_set
1574 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1575 * @front_pad: Number of bytes to allocate in front of the returned bio
1576 *
1577 * Description:
1578 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1579 * to ask for a number of bytes to be allocated in front of the bio.
1580 * Front pad allocation is useful for embedding the bio inside
1581 * another structure, to avoid allocating extra data to go with the bio.
1582 * Note that the bio must be embedded at the END of that structure always,
1583 * or things will break badly.
1584 */
1585struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
Jens Axboe392ddc32008-12-23 12:42:54 +01001587 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001588 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
Jens Axboe1b434492008-10-22 20:32:58 +02001590 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 if (!bs)
1592 return NULL;
1593
Jens Axboebb799ca2008-12-10 15:35:05 +01001594 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001595
Jens Axboe392ddc32008-12-23 12:42:54 +01001596 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001597 if (!bs->bio_slab) {
1598 kfree(bs);
1599 return NULL;
1600 }
1601
1602 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 if (!bs->bio_pool)
1604 goto bad;
1605
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001606 if (bioset_integrity_create(bs, pool_size))
1607 goto bad;
1608
Jens Axboebb799ca2008-12-10 15:35:05 +01001609 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 return bs;
1611
1612bad:
1613 bioset_free(bs);
1614 return NULL;
1615}
H Hartley Sweetena112a712009-09-26 16:19:21 +02001616EXPORT_SYMBOL(bioset_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
1618static void __init biovec_init_slabs(void)
1619{
1620 int i;
1621
1622 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1623 int size;
1624 struct biovec_slab *bvs = bvec_slabs + i;
1625
Jens Axboea7fcd372008-12-05 16:10:29 +01001626#ifndef CONFIG_BLK_DEV_INTEGRITY
1627 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1628 bvs->slab = NULL;
1629 continue;
1630 }
1631#endif
1632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 size = bvs->nr_vecs * sizeof(struct bio_vec);
1634 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001635 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 }
1637}
1638
1639static int __init init_bio(void)
1640{
Jens Axboebb799ca2008-12-10 15:35:05 +01001641 bio_slab_max = 2;
1642 bio_slab_nr = 0;
1643 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1644 if (!bio_slabs)
1645 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
Martin K. Petersen7878cba2009-06-26 15:37:49 +02001647 bio_integrity_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 biovec_init_slabs();
1649
Jens Axboebb799ca2008-12-10 15:35:05 +01001650 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (!fs_bio_set)
1652 panic("bio: can't allocate bios\n");
1653
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001654 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1655 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 if (!bio_split_pool)
1657 panic("bio: can't create split pool\n");
1658
1659 return 0;
1660}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661subsys_initcall(init_bio);