blob: 5f80848c320c7a66c6425e75aea21533607134fa [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>
Jens Axboe2056a782006-03-23 20:00:26 +010028#include <linux/blktrace_api.h>
James Bottomley f1970ba2005-06-20 14:06:52 +020029#include <scsi/sg.h> /* for struct sg_iovec */
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Li Zefan55782132009-06-09 13:43:05 +080031#include <trace/events/block.h>
Ingo Molnar0bfc2452008-11-26 11:59:56 +010032
Jens Axboe392ddc32008-12-23 12:42:54 +010033/*
34 * Test patch to inline a certain number of bi_io_vec's inside the bio
35 * itself, to shrink a bio data allocation from two mempool calls to one
36 */
37#define BIO_INLINE_VECS 4
38
Denis ChengRq6feef532008-10-09 08:57:05 +020039static mempool_t *bio_split_pool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/*
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
44 * unsigned short
45 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Jens Axboebb799ca2008-12-10 15:35:05 +010047struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
49};
50#undef BV
51
52/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
55 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020056struct bio_set *fs_bio_set;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Jens Axboebb799ca2008-12-10 15:35:05 +010058/*
59 * Our slab pool management
60 */
61struct bio_slab {
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
65 char name[8];
66};
67static DEFINE_MUTEX(bio_slab_lock);
68static struct bio_slab *bio_slabs;
69static unsigned int bio_slab_nr, bio_slab_max;
70
71static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72{
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
75 struct bio_slab *bslab;
76 unsigned int i, entry = -1;
77
78 mutex_lock(&bio_slab_lock);
79
80 i = 0;
81 while (i < bio_slab_nr) {
82 struct bio_slab *bslab = &bio_slabs[i];
83
84 if (!bslab->slab && entry == -1)
85 entry = i;
86 else if (bslab->slab_size == sz) {
87 slab = bslab->slab;
88 bslab->slab_ref++;
89 break;
90 }
91 i++;
92 }
93
94 if (slab)
95 goto out_unlock;
96
97 if (bio_slab_nr == bio_slab_max && entry == -1) {
98 bio_slab_max <<= 1;
99 bio_slabs = krealloc(bio_slabs,
100 bio_slab_max * sizeof(struct bio_slab),
101 GFP_KERNEL);
102 if (!bio_slabs)
103 goto out_unlock;
104 }
105 if (entry == -1)
106 entry = bio_slab_nr++;
107
108 bslab = &bio_slabs[entry];
109
110 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
111 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
112 if (!slab)
113 goto out_unlock;
114
115 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
116 bslab->slab = slab;
117 bslab->slab_ref = 1;
118 bslab->slab_size = sz;
119out_unlock:
120 mutex_unlock(&bio_slab_lock);
121 return slab;
122}
123
124static void bio_put_slab(struct bio_set *bs)
125{
126 struct bio_slab *bslab = NULL;
127 unsigned int i;
128
129 mutex_lock(&bio_slab_lock);
130
131 for (i = 0; i < bio_slab_nr; i++) {
132 if (bs->bio_slab == bio_slabs[i].slab) {
133 bslab = &bio_slabs[i];
134 break;
135 }
136 }
137
138 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
139 goto out;
140
141 WARN_ON(!bslab->slab_ref);
142
143 if (--bslab->slab_ref)
144 goto out;
145
146 kmem_cache_destroy(bslab->slab);
147 bslab->slab = NULL;
148
149out:
150 mutex_unlock(&bio_slab_lock);
151}
152
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200153unsigned int bvec_nr_vecs(unsigned short idx)
154{
155 return bvec_slabs[idx].nr_vecs;
156}
157
Jens Axboebb799ca2008-12-10 15:35:05 +0100158void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
159{
160 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
161
162 if (idx == BIOVEC_MAX_IDX)
163 mempool_free(bv, bs->bvec_pool);
164 else {
165 struct biovec_slab *bvs = bvec_slabs + idx;
166
167 kmem_cache_free(bvs->slab, bv);
168 }
169}
170
Jens Axboe7ff93452008-12-11 11:53:43 +0100171struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
172 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
174 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100177 * see comment near bvec_array define!
178 */
179 switch (nr) {
180 case 1:
181 *idx = 0;
182 break;
183 case 2 ... 4:
184 *idx = 1;
185 break;
186 case 5 ... 16:
187 *idx = 2;
188 break;
189 case 17 ... 64:
190 *idx = 3;
191 break;
192 case 65 ... 128:
193 *idx = 4;
194 break;
195 case 129 ... BIO_MAX_PAGES:
196 *idx = 5;
197 break;
198 default:
199 return NULL;
200 }
201
202 /*
203 * idx now points to the pool we want to allocate from. only the
204 * 1-vec entry pool is mempool backed.
205 */
206 if (*idx == BIOVEC_MAX_IDX) {
207fallback:
208 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
209 } else {
210 struct biovec_slab *bvs = bvec_slabs + *idx;
211 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200213 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100214 * Make this allocation restricted and don't dump info on
215 * allocation failures, since we'll fallback to the mempool
216 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200217 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100218 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
219
220 /*
221 * Try a slab allocation. If this fails and __GFP_WAIT
222 * is set, retry with the 1-entry mempool
223 */
224 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
225 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
226 *idx = BIOVEC_MAX_IDX;
227 goto fallback;
228 }
229 }
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 return bvl;
232}
233
Jens Axboe7ff93452008-12-11 11:53:43 +0100234void bio_free(struct bio *bio, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Jens Axboebb799ca2008-12-10 15:35:05 +0100236 void *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Jens Axboe392ddc32008-12-23 12:42:54 +0100238 if (bio_has_allocated_vec(bio))
Jens Axboebb799ca2008-12-10 15:35:05 +0100239 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
Jens Axboe992c5dd2007-07-18 13:18:08 +0200240
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200241 if (bio_integrity(bio))
Martin K. Petersen6d2a78e2009-03-10 08:27:39 +0100242 bio_integrity_free(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200243
Jens Axboebb799ca2008-12-10 15:35:05 +0100244 /*
245 * If we have front padding, adjust the bio pointer before freeing
246 */
247 p = bio;
248 if (bs->front_pad)
249 p -= bs->front_pad;
250
251 mempool_free(p, bs->bio_pool);
Peter Osterlund36763472005-09-06 15:16:42 -0700252}
253
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}
261
262/**
263 * bio_alloc_bioset - allocate a bio for I/O
264 * @gfp_mask: the GFP_ mask given to the slab allocator
265 * @nr_iovecs: number of iovecs to pre-allocate
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200266 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 *
268 * Description:
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200269 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 * If %__GFP_WAIT is set then we will block on the internal pool waiting
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200271 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
272 * fall back to just using @kmalloc to allocate the required memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 *
Jens Axboebb799ca2008-12-10 15:35:05 +0100274 * Note that the caller must set ->bi_destructor on succesful return
275 * of a bio, to do the appropriate freeing of the bio once the reference
276 * count drops to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 **/
Al Virodd0fc662005-10-07 07:46:04 +0100278struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200280 unsigned long idx = BIO_POOL_NONE;
Ingo Molnar34053972009-02-21 11:16:36 +0100281 struct bio_vec *bvl = NULL;
Tejun Heo451a9eb2009-04-15 19:50:51 +0200282 struct bio *bio;
283 void *p;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200284
Tejun Heo451a9eb2009-04-15 19:50:51 +0200285 p = mempool_alloc(bs->bio_pool, gfp_mask);
286 if (unlikely(!p))
287 return NULL;
288 bio = p + bs->front_pad;
Ingo Molnar34053972009-02-21 11:16:36 +0100289
290 bio_init(bio);
291
292 if (unlikely(!nr_iovecs))
293 goto out_set;
294
295 if (nr_iovecs <= BIO_INLINE_VECS) {
296 bvl = bio->bi_inline_vecs;
297 nr_iovecs = BIO_INLINE_VECS;
298 } else {
299 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
300 if (unlikely(!bvl))
301 goto err_free;
302
303 nr_iovecs = bvec_nr_vecs(idx);
304 }
Tejun Heo451a9eb2009-04-15 19:50:51 +0200305out_set:
Ingo Molnar34053972009-02-21 11:16:36 +0100306 bio->bi_flags |= idx << BIO_POOL_OFFSET;
307 bio->bi_max_vecs = nr_iovecs;
Ingo Molnar34053972009-02-21 11:16:36 +0100308 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100310
311err_free:
Tejun Heo451a9eb2009-04-15 19:50:51 +0200312 mempool_free(p, bs->bio_pool);
Ingo Molnar34053972009-02-21 11:16:36 +0100313 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314}
315
Tejun Heo451a9eb2009-04-15 19:50:51 +0200316static void bio_fs_destructor(struct bio *bio)
317{
318 bio_free(bio, fs_bio_set);
319}
320
321/**
322 * bio_alloc - allocate a new bio, memory pool backed
323 * @gfp_mask: allocation mask to use
324 * @nr_iovecs: number of iovecs
325 *
326 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask
327 * contains __GFP_WAIT, the allocation is guaranteed to succeed.
328 *
329 * RETURNS:
330 * Pointer to new bio on success, NULL on failure.
331 */
332struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
333{
334 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
335
336 if (bio)
337 bio->bi_destructor = bio_fs_destructor;
338
339 return bio;
340}
341
342static void bio_kmalloc_destructor(struct bio *bio)
343{
344 if (bio_integrity(bio))
345 bio_integrity_free(bio);
346 kfree(bio);
347}
348
Jens Axboe86c824b2009-04-15 09:00:07 +0200349/**
350 * bio_alloc - allocate a bio for I/O
351 * @gfp_mask: the GFP_ mask given to the slab allocator
352 * @nr_iovecs: number of iovecs to pre-allocate
353 *
354 * Description:
355 * bio_alloc will allocate a bio and associated bio_vec array that can hold
356 * at least @nr_iovecs entries. Allocations will be done from the
357 * fs_bio_set. Also see @bio_alloc_bioset.
358 *
359 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
360 * a bio. This is due to the mempool guarantees. To make this work, callers
Nikanth Karthikesan76d93ff2009-04-22 13:38:58 +0530361 * must never allocate more than 1 bio at a time from this pool. Callers
Jens Axboe86c824b2009-04-15 09:00:07 +0200362 * that need to allocate more than 1 bio must always submit the previously
Nikanth Karthikesan76d93ff2009-04-22 13:38:58 +0530363 * allocated bio for IO before attempting to allocate a new one. Failure to
Jens Axboe86c824b2009-04-15 09:00:07 +0200364 * do so can cause livelocks under memory pressure.
365 *
366 **/
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200367struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
368{
Tejun Heo451a9eb2009-04-15 19:50:51 +0200369 struct bio *bio;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200370
Tejun Heo451a9eb2009-04-15 19:50:51 +0200371 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
372 gfp_mask);
373 if (unlikely(!bio))
374 return NULL;
375
376 bio_init(bio);
377 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
378 bio->bi_max_vecs = nr_iovecs;
379 bio->bi_io_vec = bio->bi_inline_vecs;
380 bio->bi_destructor = bio_kmalloc_destructor;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200381
382 return bio;
383}
384
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385void zero_fill_bio(struct bio *bio)
386{
387 unsigned long flags;
388 struct bio_vec *bv;
389 int i;
390
391 bio_for_each_segment(bv, bio, i) {
392 char *data = bvec_kmap_irq(bv, &flags);
393 memset(data, 0, bv->bv_len);
394 flush_dcache_page(bv->bv_page);
395 bvec_kunmap_irq(data, &flags);
396 }
397}
398EXPORT_SYMBOL(zero_fill_bio);
399
400/**
401 * bio_put - release a reference to a bio
402 * @bio: bio to release reference to
403 *
404 * Description:
405 * Put a reference to a &struct bio, either one you have gotten with
406 * bio_alloc or bio_get. The last put of a bio will free it.
407 **/
408void bio_put(struct bio *bio)
409{
410 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
411
412 /*
413 * last put frees it
414 */
415 if (atomic_dec_and_test(&bio->bi_cnt)) {
416 bio->bi_next = NULL;
417 bio->bi_destructor(bio);
418 }
419}
420
Jens Axboe165125e2007-07-24 09:28:11 +0200421inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
423 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
424 blk_recount_segments(q, bio);
425
426 return bio->bi_phys_segments;
427}
428
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429/**
430 * __bio_clone - clone a bio
431 * @bio: destination bio
432 * @bio_src: bio to clone
433 *
434 * Clone a &bio. Caller will own the returned bio, but not
435 * the actual data it points to. Reference count of returned
436 * bio will be one.
437 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800438void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
Andrew Mortone525e152005-08-07 09:42:12 -0700440 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
441 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Jens Axboe5d840702008-01-25 12:44:44 +0100443 /*
444 * most users will be overriding ->bi_bdev with a new target,
445 * so we don't set nor calculate new physical/hw segment counts here
446 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 bio->bi_sector = bio_src->bi_sector;
448 bio->bi_bdev = bio_src->bi_bdev;
449 bio->bi_flags |= 1 << BIO_CLONED;
450 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 bio->bi_vcnt = bio_src->bi_vcnt;
452 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700453 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456/**
457 * bio_clone - clone a bio
458 * @bio: bio to clone
459 * @gfp_mask: allocation priority
460 *
461 * Like __bio_clone, only also allocates the returned bio
462 */
Al Virodd0fc662005-10-07 07:46:04 +0100463struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
465 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
466
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200467 if (!b)
468 return NULL;
469
470 b->bi_destructor = bio_fs_destructor;
471 __bio_clone(b, bio);
472
473 if (bio_integrity(bio)) {
474 int ret;
475
Martin K. Petersen6d2a78e2009-03-10 08:27:39 +0100476 ret = bio_integrity_clone(b, bio, gfp_mask);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200477
Li Zefan059ea332009-03-09 10:42:45 +0100478 if (ret < 0) {
479 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200480 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100481 }
Peter Osterlund36763472005-09-06 15:16:42 -0700482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 return b;
485}
486
487/**
488 * bio_get_nr_vecs - return approx number of vecs
489 * @bdev: I/O target
490 *
491 * Return the approximate number of pages we can send to this target.
492 * There's no guarantee that you will be able to fit this number of pages
493 * into a bio, it does not account for dynamic restrictions that vary
494 * on offset.
495 */
496int bio_get_nr_vecs(struct block_device *bdev)
497{
Jens Axboe165125e2007-07-24 09:28:11 +0200498 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 int nr_pages;
500
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400501 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
502 if (nr_pages > queue_max_phys_segments(q))
503 nr_pages = queue_max_phys_segments(q);
504 if (nr_pages > queue_max_hw_segments(q))
505 nr_pages = queue_max_hw_segments(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
507 return nr_pages;
508}
509
Jens Axboe165125e2007-07-24 09:28:11 +0200510static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600511 *page, unsigned int len, unsigned int offset,
512 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
514 int retried_segments = 0;
515 struct bio_vec *bvec;
516
517 /*
518 * cloned bio must not modify vec list
519 */
520 if (unlikely(bio_flagged(bio, BIO_CLONED)))
521 return 0;
522
Jens Axboe80cfd542006-01-06 09:43:28 +0100523 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 return 0;
525
Jens Axboe80cfd542006-01-06 09:43:28 +0100526 /*
527 * For filesystems with a blocksize smaller than the pagesize
528 * we will often be called with the same page as last time and
529 * a consecutive offset. Optimize this special case.
530 */
531 if (bio->bi_vcnt > 0) {
532 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
533
534 if (page == prev->bv_page &&
535 offset == prev->bv_offset + prev->bv_len) {
536 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200537
538 if (q->merge_bvec_fn) {
539 struct bvec_merge_data bvm = {
540 .bi_bdev = bio->bi_bdev,
541 .bi_sector = bio->bi_sector,
542 .bi_size = bio->bi_size,
543 .bi_rw = bio->bi_rw,
544 };
545
546 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
547 prev->bv_len -= len;
548 return 0;
549 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100550 }
551
552 goto done;
553 }
554 }
555
556 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 return 0;
558
559 /*
560 * we might lose a segment or two here, but rather that than
561 * make this too complex.
562 */
563
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400564 while (bio->bi_phys_segments >= queue_max_phys_segments(q)
565 || bio->bi_phys_segments >= queue_max_hw_segments(q)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
567 if (retried_segments)
568 return 0;
569
570 retried_segments = 1;
571 blk_recount_segments(q, bio);
572 }
573
574 /*
575 * setup the new entry, we might clear it again later if we
576 * cannot add the page
577 */
578 bvec = &bio->bi_io_vec[bio->bi_vcnt];
579 bvec->bv_page = page;
580 bvec->bv_len = len;
581 bvec->bv_offset = offset;
582
583 /*
584 * if queue has other restrictions (eg varying max sector size
585 * depending on offset), it can specify a merge_bvec_fn in the
586 * queue to get further control
587 */
588 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200589 struct bvec_merge_data bvm = {
590 .bi_bdev = bio->bi_bdev,
591 .bi_sector = bio->bi_sector,
592 .bi_size = bio->bi_size,
593 .bi_rw = bio->bi_rw,
594 };
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 /*
597 * merge_bvec_fn() returns number of bytes it can accept
598 * at this offset
599 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200600 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 bvec->bv_page = NULL;
602 bvec->bv_len = 0;
603 bvec->bv_offset = 0;
604 return 0;
605 }
606 }
607
608 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200609 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
611
612 bio->bi_vcnt++;
613 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100614 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 bio->bi_size += len;
616 return len;
617}
618
619/**
Mike Christie6e68af62005-11-11 05:30:27 -0600620 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100621 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600622 * @bio: destination bio
623 * @page: page to add
624 * @len: vec entry length
625 * @offset: vec entry offset
626 *
627 * Attempt to add a page to the bio_vec maplist. This can fail for a
628 * number of reasons, such as the bio being full or target block
629 * device limitations. The target block device must allow bio's
630 * smaller than PAGE_SIZE, so it is always possible to add a single
631 * page to an empty bio. This should only be used by REQ_PC bios.
632 */
Jens Axboe165125e2007-07-24 09:28:11 +0200633int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600634 unsigned int len, unsigned int offset)
635{
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400636 return __bio_add_page(q, bio, page, len, offset,
637 queue_max_hw_sectors(q));
Mike Christie6e68af62005-11-11 05:30:27 -0600638}
639
640/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 * bio_add_page - attempt to add page to bio
642 * @bio: destination bio
643 * @page: page to add
644 * @len: vec entry length
645 * @offset: vec entry offset
646 *
647 * Attempt to add a page to the bio_vec maplist. This can fail for a
648 * number of reasons, such as the bio being full or target block
649 * device limitations. The target block device must allow bio's
650 * smaller than PAGE_SIZE, so it is always possible to add a single
651 * page to an empty bio.
652 */
653int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
654 unsigned int offset)
655{
Mike Christiedefd94b2005-12-05 02:37:06 -0600656 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400657 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
660struct bio_map_data {
661 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200662 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900663 int nr_sgvecs;
664 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665};
666
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200667static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900668 struct sg_iovec *iov, int iov_count,
669 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
671 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200672 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
673 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900674 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 bio->bi_private = bmd;
676}
677
678static void bio_free_map_data(struct bio_map_data *bmd)
679{
680 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200681 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 kfree(bmd);
683}
684
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200685static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
686 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200688 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690 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 Tomonori152e2832008-08-28 16:17:06 +0900709 struct sg_iovec *iov, int iov_count, int uncopy,
710 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;
716 int read = bio_data_dir(bio) == READ;
717
718 __bio_for_each_segment(bvec, bio, i, 0) {
719 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200720 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200721
722 while (bv_len && iov_idx < iov_count) {
723 unsigned int bytes;
Michal Simek0e0c6212009-06-10 12:57:07 -0700724 char __user *iov_addr;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200725
726 bytes = min_t(unsigned int,
727 iov[iov_idx].iov_len - iov_off, bv_len);
728 iov_addr = iov[iov_idx].iov_base + iov_off;
729
730 if (!ret) {
731 if (!read && !uncopy)
732 ret = copy_from_user(bv_addr, iov_addr,
733 bytes);
734 if (read && uncopy)
735 ret = copy_to_user(iov_addr, bv_addr,
736 bytes);
737
738 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,
774 bmd->nr_sgvecs, 1, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 bio_free_map_data(bmd);
776 bio_put(bio);
777 return ret;
778}
779
780/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200781 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900783 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200784 * @iov: the iovec.
785 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900787 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 *
789 * Prepares and returns a bio for indirect user io, bouncing data
790 * to/from kernel pages as necessary. Must be paired with
791 * call bio_uncopy_user() on io completion.
792 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900793struct bio *bio_copy_user_iov(struct request_queue *q,
794 struct rq_map_data *map_data,
795 struct sg_iovec *iov, int iov_count,
796 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 struct bio_map_data *bmd;
799 struct bio_vec *bvec;
800 struct page *page;
801 struct bio *bio;
802 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200803 int nr_pages = 0;
804 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900805 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200807 for (i = 0; i < iov_count; i++) {
808 unsigned long uaddr;
809 unsigned long end;
810 unsigned long start;
811
812 uaddr = (unsigned long)iov[i].iov_base;
813 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
814 start = uaddr >> PAGE_SHIFT;
815
816 nr_pages += end - start;
817 len += iov[i].iov_len;
818 }
819
FUJITA Tomonori69838722009-04-28 20:24:29 +0200820 if (offset)
821 nr_pages++;
822
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900823 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 if (!bmd)
825 return ERR_PTR(-ENOMEM);
826
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 ret = -ENOMEM;
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900828 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 if (!bio)
830 goto out_bmd;
831
832 bio->bi_rw |= (!write_to_vm << BIO_RW);
833
834 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900835
836 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900837 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900838 i = map_data->offset / PAGE_SIZE;
839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900841 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900843 bytes -= offset;
844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 if (bytes > len)
846 bytes = len;
847
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900848 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900849 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900850 ret = -ENOMEM;
851 break;
852 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900853
854 page = map_data->pages[i / nr_pages];
855 page += (i % nr_pages);
856
857 i++;
858 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900859 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900860 if (!page) {
861 ret = -ENOMEM;
862 break;
863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 }
865
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900866 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900870 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 }
872
873 if (ret)
874 goto cleanup;
875
876 /*
877 * success
878 */
FUJITA Tomonori97ae77a2008-12-18 14:49:38 +0900879 if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900880 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200881 if (ret)
882 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 }
884
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900885 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 return bio;
887cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900888 if (!map_data)
889 bio_for_each_segment(bvec, bio, i)
890 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 bio_put(bio);
893out_bmd:
894 bio_free_map_data(bmd);
895 return ERR_PTR(ret);
896}
897
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200898/**
899 * bio_copy_user - copy user data to bio
900 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900901 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200902 * @uaddr: start of user address
903 * @len: length in bytes
904 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900905 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200906 *
907 * Prepares and returns a bio for indirect user io, bouncing data
908 * to/from kernel pages as necessary. Must be paired with
909 * call bio_uncopy_user() on io completion.
910 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900911struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
912 unsigned long uaddr, unsigned int len,
913 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200914{
915 struct sg_iovec iov;
916
917 iov.iov_base = (void __user *)uaddr;
918 iov.iov_len = len;
919
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900920 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200921}
922
Jens Axboe165125e2007-07-24 09:28:11 +0200923static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200924 struct block_device *bdev,
925 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900926 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927{
James Bottomley f1970ba2005-06-20 14:06:52 +0200928 int i, j;
929 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 struct page **pages;
931 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200932 int cur_page = 0;
933 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
James Bottomley f1970ba2005-06-20 14:06:52 +0200935 for (i = 0; i < iov_count; i++) {
936 unsigned long uaddr = (unsigned long)iov[i].iov_base;
937 unsigned long len = iov[i].iov_len;
938 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
939 unsigned long start = uaddr >> PAGE_SHIFT;
940
941 nr_pages += end - start;
942 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100943 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200944 */
Mike Christiead2d7222006-12-01 10:40:20 +0100945 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200946 return ERR_PTR(-EINVAL);
947 }
948
949 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 return ERR_PTR(-EINVAL);
951
Tejun Heoa9e9dc22009-04-15 22:10:27 +0900952 bio = bio_kmalloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (!bio)
954 return ERR_PTR(-ENOMEM);
955
956 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900957 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 if (!pages)
959 goto out;
960
James Bottomley f1970ba2005-06-20 14:06:52 +0200961 for (i = 0; i < iov_count; i++) {
962 unsigned long uaddr = (unsigned long)iov[i].iov_base;
963 unsigned long len = iov[i].iov_len;
964 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
965 unsigned long start = uaddr >> PAGE_SHIFT;
966 const int local_nr_pages = end - start;
967 const int page_limit = cur_page + local_nr_pages;
968
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700969 ret = get_user_pages_fast(uaddr, local_nr_pages,
970 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200971 if (ret < local_nr_pages) {
972 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200973 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
James Bottomley f1970ba2005-06-20 14:06:52 +0200976 offset = uaddr & ~PAGE_MASK;
977 for (j = cur_page; j < page_limit; j++) {
978 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
James Bottomley f1970ba2005-06-20 14:06:52 +0200980 if (len <= 0)
981 break;
982
983 if (bytes > len)
984 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
James Bottomley f1970ba2005-06-20 14:06:52 +0200986 /*
987 * sorry...
988 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600989 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
990 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200991 break;
992
993 len -= bytes;
994 offset = 0;
995 }
996
997 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200999 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 */
James Bottomley f1970ba2005-06-20 14:06:52 +02001001 while (j < page_limit)
1002 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 kfree(pages);
1006
1007 /*
1008 * set data direction, and check if mapped pages need bouncing
1009 */
1010 if (!write_to_vm)
1011 bio->bi_rw |= (1 << BIO_RW);
1012
James Bottomley f1970ba2005-06-20 14:06:52 +02001013 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1015 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001016
1017 out_unmap:
1018 for (i = 0; i < nr_pages; i++) {
1019 if(!pages[i])
1020 break;
1021 page_cache_release(pages[i]);
1022 }
1023 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 kfree(pages);
1025 bio_put(bio);
1026 return ERR_PTR(ret);
1027}
1028
1029/**
1030 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001031 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 * @bdev: destination block device
1033 * @uaddr: start of user address
1034 * @len: length in bytes
1035 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001036 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 *
1038 * Map the user space address into a bio suitable for io to a block
1039 * device. Returns an error pointer in case of error.
1040 */
Jens Axboe165125e2007-07-24 09:28:11 +02001041struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001042 unsigned long uaddr, unsigned int len, int write_to_vm,
1043 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
James Bottomley f1970ba2005-06-20 14:06:52 +02001045 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001047 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001048 iov.iov_len = len;
1049
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001050 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001051}
1052
1053/**
1054 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001055 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001056 * @bdev: destination block device
1057 * @iov: the iovec.
1058 * @iov_count: number of elements in the iovec
1059 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001060 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001061 *
1062 * Map the user space address into a bio suitable for io to a block
1063 * device. Returns an error pointer in case of error.
1064 */
Jens Axboe165125e2007-07-24 09:28:11 +02001065struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001066 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001067 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001068{
1069 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001070
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001071 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1072 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (IS_ERR(bio))
1074 return bio;
1075
1076 /*
1077 * subtle -- if __bio_map_user() ended up bouncing a bio,
1078 * it would normally disappear when its bi_end_io is run.
1079 * however, we need it for the unmap, so grab an extra
1080 * reference to it
1081 */
1082 bio_get(bio);
1083
Mike Christie0e75f902006-12-01 10:40:55 +01001084 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085}
1086
1087static void __bio_unmap_user(struct bio *bio)
1088{
1089 struct bio_vec *bvec;
1090 int i;
1091
1092 /*
1093 * make sure we dirty pages we wrote to
1094 */
1095 __bio_for_each_segment(bvec, bio, i, 0) {
1096 if (bio_data_dir(bio) == READ)
1097 set_page_dirty_lock(bvec->bv_page);
1098
1099 page_cache_release(bvec->bv_page);
1100 }
1101
1102 bio_put(bio);
1103}
1104
1105/**
1106 * bio_unmap_user - unmap a bio
1107 * @bio: the bio being unmapped
1108 *
1109 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1110 * a process context.
1111 *
1112 * bio_unmap_user() may sleep.
1113 */
1114void bio_unmap_user(struct bio *bio)
1115{
1116 __bio_unmap_user(bio);
1117 bio_put(bio);
1118}
1119
NeilBrown6712ecf2007-09-27 12:47:43 +02001120static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001121{
Jens Axboeb8238252005-06-20 14:05:27 +02001122 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001123}
1124
1125
Jens Axboe165125e2007-07-24 09:28:11 +02001126static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001127 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001128{
1129 unsigned long kaddr = (unsigned long)data;
1130 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1131 unsigned long start = kaddr >> PAGE_SHIFT;
1132 const int nr_pages = end - start;
1133 int offset, i;
1134 struct bio *bio;
1135
Tejun Heoa9e9dc22009-04-15 22:10:27 +09001136 bio = bio_kmalloc(gfp_mask, nr_pages);
Mike Christie df46b9a2005-06-20 14:04:44 +02001137 if (!bio)
1138 return ERR_PTR(-ENOMEM);
1139
1140 offset = offset_in_page(kaddr);
1141 for (i = 0; i < nr_pages; i++) {
1142 unsigned int bytes = PAGE_SIZE - offset;
1143
1144 if (len <= 0)
1145 break;
1146
1147 if (bytes > len)
1148 bytes = len;
1149
Mike Christiedefd94b2005-12-05 02:37:06 -06001150 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1151 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001152 break;
1153
1154 data += bytes;
1155 len -= bytes;
1156 offset = 0;
1157 }
1158
Jens Axboeb8238252005-06-20 14:05:27 +02001159 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001160 return bio;
1161}
1162
1163/**
1164 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001165 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001166 * @data: pointer to buffer to map
1167 * @len: length in bytes
1168 * @gfp_mask: allocation flags for bio allocation
1169 *
1170 * Map the kernel address into a bio suitable for io to a block
1171 * device. Returns an error pointer in case of error.
1172 */
Jens Axboe165125e2007-07-24 09:28:11 +02001173struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001174 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001175{
1176 struct bio *bio;
1177
1178 bio = __bio_map_kern(q, data, len, gfp_mask);
1179 if (IS_ERR(bio))
1180 return bio;
1181
1182 if (bio->bi_size == len)
1183 return bio;
1184
1185 /*
1186 * Don't support partial mappings.
1187 */
1188 bio_put(bio);
1189 return ERR_PTR(-EINVAL);
1190}
1191
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001192static void bio_copy_kern_endio(struct bio *bio, int err)
1193{
1194 struct bio_vec *bvec;
1195 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001196 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001197 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001198 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001199
1200 __bio_for_each_segment(bvec, bio, i, 0) {
1201 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001202 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001203
Tejun Heo4fc981e2009-05-19 18:33:06 +09001204 if (read)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001205 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001206
1207 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001208 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001209 }
1210
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001211 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001212 bio_put(bio);
1213}
1214
1215/**
1216 * bio_copy_kern - copy kernel address into bio
1217 * @q: the struct request_queue for the bio
1218 * @data: pointer to buffer to copy
1219 * @len: length in bytes
1220 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001221 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001222 *
1223 * copy the kernel address into a bio suitable for io to a block
1224 * device. Returns an error pointer in case of error.
1225 */
1226struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1227 gfp_t gfp_mask, int reading)
1228{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001229 struct bio *bio;
1230 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001231 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001232
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001233 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1234 if (IS_ERR(bio))
1235 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001236
1237 if (!reading) {
1238 void *p = data;
1239
1240 bio_for_each_segment(bvec, bio, i) {
1241 char *addr = page_address(bvec->bv_page);
1242
1243 memcpy(addr, p, bvec->bv_len);
1244 p += bvec->bv_len;
1245 }
1246 }
1247
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001248 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001249
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001250 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001251}
1252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253/*
1254 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1255 * for performing direct-IO in BIOs.
1256 *
1257 * The problem is that we cannot run set_page_dirty() from interrupt context
1258 * because the required locks are not interrupt-safe. So what we can do is to
1259 * mark the pages dirty _before_ performing IO. And in interrupt context,
1260 * check that the pages are still dirty. If so, fine. If not, redirty them
1261 * in process context.
1262 *
1263 * We special-case compound pages here: normally this means reads into hugetlb
1264 * pages. The logic in here doesn't really work right for compound pages
1265 * because the VM does not uniformly chase down the head page in all cases.
1266 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1267 * handle them at all. So we skip compound pages here at an early stage.
1268 *
1269 * Note that this code is very hard to test under normal circumstances because
1270 * direct-io pins the pages with get_user_pages(). This makes
1271 * is_page_cache_freeable return false, and the VM will not clean the pages.
1272 * But other code (eg, pdflush) could clean the pages if they are mapped
1273 * pagecache.
1274 *
1275 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1276 * deferred bio dirtying paths.
1277 */
1278
1279/*
1280 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1281 */
1282void bio_set_pages_dirty(struct bio *bio)
1283{
1284 struct bio_vec *bvec = bio->bi_io_vec;
1285 int i;
1286
1287 for (i = 0; i < bio->bi_vcnt; i++) {
1288 struct page *page = bvec[i].bv_page;
1289
1290 if (page && !PageCompound(page))
1291 set_page_dirty_lock(page);
1292 }
1293}
1294
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001295static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296{
1297 struct bio_vec *bvec = bio->bi_io_vec;
1298 int i;
1299
1300 for (i = 0; i < bio->bi_vcnt; i++) {
1301 struct page *page = bvec[i].bv_page;
1302
1303 if (page)
1304 put_page(page);
1305 }
1306}
1307
1308/*
1309 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1310 * If they are, then fine. If, however, some pages are clean then they must
1311 * have been written out during the direct-IO read. So we take another ref on
1312 * the BIO and the offending pages and re-dirty the pages in process context.
1313 *
1314 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1315 * here on. It will run one page_cache_release() against each page and will
1316 * run one bio_put() against the BIO.
1317 */
1318
David Howells65f27f32006-11-22 14:55:48 +00001319static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
David Howells65f27f32006-11-22 14:55:48 +00001321static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322static DEFINE_SPINLOCK(bio_dirty_lock);
1323static struct bio *bio_dirty_list;
1324
1325/*
1326 * This runs in process context
1327 */
David Howells65f27f32006-11-22 14:55:48 +00001328static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329{
1330 unsigned long flags;
1331 struct bio *bio;
1332
1333 spin_lock_irqsave(&bio_dirty_lock, flags);
1334 bio = bio_dirty_list;
1335 bio_dirty_list = NULL;
1336 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1337
1338 while (bio) {
1339 struct bio *next = bio->bi_private;
1340
1341 bio_set_pages_dirty(bio);
1342 bio_release_pages(bio);
1343 bio_put(bio);
1344 bio = next;
1345 }
1346}
1347
1348void bio_check_pages_dirty(struct bio *bio)
1349{
1350 struct bio_vec *bvec = bio->bi_io_vec;
1351 int nr_clean_pages = 0;
1352 int i;
1353
1354 for (i = 0; i < bio->bi_vcnt; i++) {
1355 struct page *page = bvec[i].bv_page;
1356
1357 if (PageDirty(page) || PageCompound(page)) {
1358 page_cache_release(page);
1359 bvec[i].bv_page = NULL;
1360 } else {
1361 nr_clean_pages++;
1362 }
1363 }
1364
1365 if (nr_clean_pages) {
1366 unsigned long flags;
1367
1368 spin_lock_irqsave(&bio_dirty_lock, flags);
1369 bio->bi_private = bio_dirty_list;
1370 bio_dirty_list = bio;
1371 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1372 schedule_work(&bio_dirty_work);
1373 } else {
1374 bio_put(bio);
1375 }
1376}
1377
1378/**
1379 * bio_endio - end I/O on a bio
1380 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 * @error: error, if any
1382 *
1383 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001384 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001385 * preferred way to end I/O on a bio, it takes care of clearing
1386 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1387 * established -Exxxx (-EIO, for instance) error values in case
1388 * something went wrong. Noone should call bi_end_io() directly on a
1389 * bio unless they own it and thus know that it has an end_io
1390 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001392void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
1394 if (error)
1395 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001396 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1397 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
NeilBrown5bb23a62007-09-27 12:46:13 +02001399 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001400 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
1402
1403void bio_pair_release(struct bio_pair *bp)
1404{
1405 if (atomic_dec_and_test(&bp->cnt)) {
1406 struct bio *master = bp->bio1.bi_private;
1407
NeilBrown6712ecf2007-09-27 12:47:43 +02001408 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 mempool_free(bp, bp->bio2.bi_private);
1410 }
1411}
1412
NeilBrown6712ecf2007-09-27 12:47:43 +02001413static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414{
1415 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1416
1417 if (err)
1418 bp->error = err;
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421}
1422
NeilBrown6712ecf2007-09-27 12:47:43 +02001423static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
1425 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1426
1427 if (err)
1428 bp->error = err;
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431}
1432
1433/*
Alberto Bertoglic7eee1b2009-01-25 23:36:14 -02001434 * split a bio - only worry about a bio with a single page in its iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001436struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Denis ChengRq6feef532008-10-09 08:57:05 +02001438 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440 if (!bp)
1441 return bp;
1442
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001443 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001444 bi->bi_sector + first_sectors);
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 BUG_ON(bi->bi_vcnt != 1);
1447 BUG_ON(bi->bi_idx != 0);
1448 atomic_set(&bp->cnt, 3);
1449 bp->error = 0;
1450 bp->bio1 = *bi;
1451 bp->bio2 = *bi;
1452 bp->bio2.bi_sector += first_sectors;
1453 bp->bio2.bi_size -= first_sectors << 9;
1454 bp->bio1.bi_size = first_sectors << 9;
1455
1456 bp->bv1 = bi->bi_io_vec[0];
1457 bp->bv2 = bi->bi_io_vec[0];
1458 bp->bv2.bv_offset += first_sectors << 9;
1459 bp->bv2.bv_len -= first_sectors << 9;
1460 bp->bv1.bv_len = first_sectors << 9;
1461
1462 bp->bio1.bi_io_vec = &bp->bv1;
1463 bp->bio2.bi_io_vec = &bp->bv2;
1464
NeilBrowna2eb0c12006-05-22 22:35:27 -07001465 bp->bio1.bi_max_vecs = 1;
1466 bp->bio2.bi_max_vecs = 1;
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 bp->bio1.bi_end_io = bio_pair_end_1;
1469 bp->bio2.bi_end_io = bio_pair_end_2;
1470
1471 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001472 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001474 if (bio_integrity(bi))
1475 bio_integrity_split(bi, bp, first_sectors);
1476
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return bp;
1478}
1479
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001480/**
1481 * bio_sector_offset - Find hardware sector offset in bio
1482 * @bio: bio to inspect
1483 * @index: bio_vec index
1484 * @offset: offset in bv_page
1485 *
1486 * Return the number of hardware sectors between beginning of bio
1487 * and an end point indicated by a bio_vec index and an offset
1488 * within that vector's page.
1489 */
1490sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1491 unsigned int offset)
1492{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001493 unsigned int sector_sz;
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001494 struct bio_vec *bv;
1495 sector_t sectors;
1496 int i;
1497
Martin K. Petersene1defc42009-05-22 17:17:49 -04001498 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001499 sectors = 0;
1500
1501 if (index >= bio->bi_idx)
1502 index = bio->bi_vcnt - 1;
1503
1504 __bio_for_each_segment(bv, bio, i, 0) {
1505 if (i == index) {
1506 if (offset > bv->bv_offset)
1507 sectors += (offset - bv->bv_offset) / sector_sz;
1508 break;
1509 }
1510
1511 sectors += bv->bv_len / sector_sz;
1512 }
1513
1514 return sectors;
1515}
1516EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
1518/*
1519 * create memory pools for biovec's in a bio_set.
1520 * use the global biovec slabs created for general use.
1521 */
Jens Axboe59725112007-04-02 10:06:42 +02001522static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523{
Jens Axboe7ff93452008-12-11 11:53:43 +01001524 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Jens Axboe7ff93452008-12-11 11:53:43 +01001526 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1527 if (!bs->bvec_pool)
1528 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 return 0;
1531}
1532
1533static void biovec_free_pools(struct bio_set *bs)
1534{
Jens Axboe7ff93452008-12-11 11:53:43 +01001535 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536}
1537
1538void bioset_free(struct bio_set *bs)
1539{
1540 if (bs->bio_pool)
1541 mempool_destroy(bs->bio_pool);
1542
1543 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001544 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
1546 kfree(bs);
1547}
1548
Jens Axboebb799ca2008-12-10 15:35:05 +01001549/**
1550 * bioset_create - Create a bio_set
1551 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1552 * @front_pad: Number of bytes to allocate in front of the returned bio
1553 *
1554 * Description:
1555 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1556 * to ask for a number of bytes to be allocated in front of the bio.
1557 * Front pad allocation is useful for embedding the bio inside
1558 * another structure, to avoid allocating extra data to go with the bio.
1559 * Note that the bio must be embedded at the END of that structure always,
1560 * or things will break badly.
1561 */
1562struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
Jens Axboe392ddc32008-12-23 12:42:54 +01001564 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001565 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Jens Axboe1b434492008-10-22 20:32:58 +02001567 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 if (!bs)
1569 return NULL;
1570
Jens Axboebb799ca2008-12-10 15:35:05 +01001571 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001572
Jens Axboe392ddc32008-12-23 12:42:54 +01001573 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001574 if (!bs->bio_slab) {
1575 kfree(bs);
1576 return NULL;
1577 }
1578
1579 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 if (!bs->bio_pool)
1581 goto bad;
1582
Jens Axboebb799ca2008-12-10 15:35:05 +01001583 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 return bs;
1585
1586bad:
1587 bioset_free(bs);
1588 return NULL;
1589}
1590
1591static void __init biovec_init_slabs(void)
1592{
1593 int i;
1594
1595 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1596 int size;
1597 struct biovec_slab *bvs = bvec_slabs + i;
1598
Jens Axboea7fcd372008-12-05 16:10:29 +01001599#ifndef CONFIG_BLK_DEV_INTEGRITY
1600 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1601 bvs->slab = NULL;
1602 continue;
1603 }
1604#endif
1605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 size = bvs->nr_vecs * sizeof(struct bio_vec);
1607 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001608 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 }
1610}
1611
1612static int __init init_bio(void)
1613{
Jens Axboebb799ca2008-12-10 15:35:05 +01001614 bio_slab_max = 2;
1615 bio_slab_nr = 0;
1616 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1617 if (!bio_slabs)
1618 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
1620 biovec_init_slabs();
1621
Jens Axboebb799ca2008-12-10 15:35:05 +01001622 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 if (!fs_bio_set)
1624 panic("bio: can't allocate bios\n");
1625
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001626 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1627 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 if (!bio_split_pool)
1629 panic("bio: can't create split pool\n");
1630
1631 return 0;
1632}
1633
1634subsys_initcall(init_bio);
1635
1636EXPORT_SYMBOL(bio_alloc);
Jens Axboe0a0d96b2008-09-11 13:17:37 +02001637EXPORT_SYMBOL(bio_kmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001639EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640EXPORT_SYMBOL(bio_endio);
1641EXPORT_SYMBOL(bio_init);
1642EXPORT_SYMBOL(__bio_clone);
1643EXPORT_SYMBOL(bio_clone);
1644EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001646EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647EXPORT_SYMBOL(bio_get_nr_vecs);
Jens Axboe40044ce2008-03-17 21:14:40 +01001648EXPORT_SYMBOL(bio_map_user);
1649EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001650EXPORT_SYMBOL(bio_map_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001651EXPORT_SYMBOL(bio_copy_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652EXPORT_SYMBOL(bio_pair_release);
1653EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654EXPORT_SYMBOL(bio_copy_user);
1655EXPORT_SYMBOL(bio_uncopy_user);
1656EXPORT_SYMBOL(bioset_create);
1657EXPORT_SYMBOL(bioset_free);
1658EXPORT_SYMBOL(bio_alloc_bioset);