blob: a040cde7f6fd567f68a7c28504f9f25d6a274146 [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>
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +010029#include <trace/block.h>
James Bottomley f1970ba2005-06-20 14:06:52 +020030#include <scsi/sg.h> /* for struct sg_iovec */
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Ingo Molnar0bfc2452008-11-26 11:59:56 +010032DEFINE_TRACE(block_split);
33
Jens Axboe392ddc32008-12-23 12:42:54 +010034/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
Denis ChengRq6feef532008-10-09 08:57:05 +020040static mempool_t *bio_split_pool __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * if you change this list, also change bvec_alloc or things will
44 * break badly! cannot be bigger than what you can fit into an
45 * unsigned short
46 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Jens Axboebb799ca2008-12-10 15:35:05 +010048struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
50};
51#undef BV
52
53/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
55 * IO code that does not need private memory pools.
56 */
Martin K. Petersen51d654e2008-06-17 18:59:56 +020057struct bio_set *fs_bio_set;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Jens Axboebb799ca2008-12-10 15:35:05 +010059/*
60 * Our slab pool management
61 */
62struct bio_slab {
63 struct kmem_cache *slab;
64 unsigned int slab_ref;
65 unsigned int slab_size;
66 char name[8];
67};
68static DEFINE_MUTEX(bio_slab_lock);
69static struct bio_slab *bio_slabs;
70static unsigned int bio_slab_nr, bio_slab_max;
71
72static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
73{
74 unsigned int sz = sizeof(struct bio) + extra_size;
75 struct kmem_cache *slab = NULL;
76 struct bio_slab *bslab;
77 unsigned int i, entry = -1;
78
79 mutex_lock(&bio_slab_lock);
80
81 i = 0;
82 while (i < bio_slab_nr) {
83 struct bio_slab *bslab = &bio_slabs[i];
84
85 if (!bslab->slab && entry == -1)
86 entry = i;
87 else if (bslab->slab_size == sz) {
88 slab = bslab->slab;
89 bslab->slab_ref++;
90 break;
91 }
92 i++;
93 }
94
95 if (slab)
96 goto out_unlock;
97
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
99 bio_slab_max <<= 1;
100 bio_slabs = krealloc(bio_slabs,
101 bio_slab_max * sizeof(struct bio_slab),
102 GFP_KERNEL);
103 if (!bio_slabs)
104 goto out_unlock;
105 }
106 if (entry == -1)
107 entry = bio_slab_nr++;
108
109 bslab = &bio_slabs[entry];
110
111 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
112 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
113 if (!slab)
114 goto out_unlock;
115
116 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
117 bslab->slab = slab;
118 bslab->slab_ref = 1;
119 bslab->slab_size = sz;
120out_unlock:
121 mutex_unlock(&bio_slab_lock);
122 return slab;
123}
124
125static void bio_put_slab(struct bio_set *bs)
126{
127 struct bio_slab *bslab = NULL;
128 unsigned int i;
129
130 mutex_lock(&bio_slab_lock);
131
132 for (i = 0; i < bio_slab_nr; i++) {
133 if (bs->bio_slab == bio_slabs[i].slab) {
134 bslab = &bio_slabs[i];
135 break;
136 }
137 }
138
139 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
140 goto out;
141
142 WARN_ON(!bslab->slab_ref);
143
144 if (--bslab->slab_ref)
145 goto out;
146
147 kmem_cache_destroy(bslab->slab);
148 bslab->slab = NULL;
149
150out:
151 mutex_unlock(&bio_slab_lock);
152}
153
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200154unsigned int bvec_nr_vecs(unsigned short idx)
155{
156 return bvec_slabs[idx].nr_vecs;
157}
158
Jens Axboebb799ca2008-12-10 15:35:05 +0100159void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
160{
161 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
162
163 if (idx == BIOVEC_MAX_IDX)
164 mempool_free(bv, bs->bvec_pool);
165 else {
166 struct biovec_slab *bvs = bvec_slabs + idx;
167
168 kmem_cache_free(bvs->slab, bv);
169 }
170}
171
Jens Axboe7ff93452008-12-11 11:53:43 +0100172struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
173 struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
175 struct bio_vec *bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
177 /*
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200178 * If 'bs' is given, lookup the pool and do the mempool alloc.
179 * If not, this is a bio_kmalloc() allocation and just do a
180 * kzalloc() for the exact number of vecs right away.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100182 if (!bs)
Jens Axboed3f76112008-12-23 12:46:21 +0100183 bvl = kmalloc(nr * sizeof(struct bio_vec), gfp_mask);
Jens Axboe7ff93452008-12-11 11:53:43 +0100184
185 /*
186 * see comment near bvec_array define!
187 */
188 switch (nr) {
189 case 1:
190 *idx = 0;
191 break;
192 case 2 ... 4:
193 *idx = 1;
194 break;
195 case 5 ... 16:
196 *idx = 2;
197 break;
198 case 17 ... 64:
199 *idx = 3;
200 break;
201 case 65 ... 128:
202 *idx = 4;
203 break;
204 case 129 ... BIO_MAX_PAGES:
205 *idx = 5;
206 break;
207 default:
208 return NULL;
209 }
210
211 /*
212 * idx now points to the pool we want to allocate from. only the
213 * 1-vec entry pool is mempool backed.
214 */
215 if (*idx == BIOVEC_MAX_IDX) {
216fallback:
217 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
218 } else {
219 struct biovec_slab *bvs = bvec_slabs + *idx;
220 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200222 /*
Jens Axboe7ff93452008-12-11 11:53:43 +0100223 * Make this allocation restricted and don't dump info on
224 * allocation failures, since we'll fallback to the mempool
225 * in case of failure.
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200226 */
Jens Axboe7ff93452008-12-11 11:53:43 +0100227 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
228
229 /*
230 * Try a slab allocation. If this fails and __GFP_WAIT
231 * is set, retry with the 1-entry mempool
232 */
233 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
234 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
235 *idx = BIOVEC_MAX_IDX;
236 goto fallback;
237 }
238 }
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 return bvl;
241}
242
Jens Axboe7ff93452008-12-11 11:53:43 +0100243void bio_free(struct bio *bio, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
Jens Axboebb799ca2008-12-10 15:35:05 +0100245 void *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Jens Axboe392ddc32008-12-23 12:42:54 +0100247 if (bio_has_allocated_vec(bio))
Jens Axboebb799ca2008-12-10 15:35:05 +0100248 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
Jens Axboe992c5dd2007-07-18 13:18:08 +0200249
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200250 if (bio_integrity(bio))
Martin K. Petersen6d2a78e2009-03-10 08:27:39 +0100251 bio_integrity_free(bio);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200252
Jens Axboebb799ca2008-12-10 15:35:05 +0100253 /*
254 * If we have front padding, adjust the bio pointer before freeing
255 */
256 p = bio;
257 if (bs->front_pad)
258 p -= bs->front_pad;
259
260 mempool_free(p, bs->bio_pool);
Peter Osterlund36763472005-09-06 15:16:42 -0700261}
262
263/*
264 * default destructor for a bio allocated with bio_alloc_bioset()
265 */
266static void bio_fs_destructor(struct bio *bio)
267{
268 bio_free(bio, fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269}
270
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200271static void bio_kmalloc_destructor(struct bio *bio)
272{
Jens Axboe392ddc32008-12-23 12:42:54 +0100273 if (bio_has_allocated_vec(bio))
274 kfree(bio->bi_io_vec);
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200275 kfree(bio);
276}
277
Arjan van de Ven858119e2006-01-14 13:20:43 -0800278void bio_init(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
Jens Axboe2b94de52007-07-18 13:14:03 +0200280 memset(bio, 0, sizeof(*bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 bio->bi_flags = 1 << BIO_UPTODATE;
Jens Axboec7c22e42008-09-13 20:26:01 +0200282 bio->bi_comp_cpu = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 atomic_set(&bio->bi_cnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
286/**
287 * bio_alloc_bioset - allocate a bio for I/O
288 * @gfp_mask: the GFP_ mask given to the slab allocator
289 * @nr_iovecs: number of iovecs to pre-allocate
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200290 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 *
292 * Description:
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200293 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 * If %__GFP_WAIT is set then we will block on the internal pool waiting
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200295 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
296 * fall back to just using @kmalloc to allocate the required memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 *
Jens Axboebb799ca2008-12-10 15:35:05 +0100298 * Note that the caller must set ->bi_destructor on succesful return
299 * of a bio, to do the appropriate freeing of the bio once the reference
300 * count drops to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 **/
Al Virodd0fc662005-10-07 07:46:04 +0100302struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
Ingo Molnar34053972009-02-21 11:16:36 +0100304 struct bio_vec *bvl = NULL;
Jens Axboebb799ca2008-12-10 15:35:05 +0100305 struct bio *bio = NULL;
Ingo Molnar34053972009-02-21 11:16:36 +0100306 unsigned long idx = 0;
307 void *p = NULL;
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200308
Jens Axboebb799ca2008-12-10 15:35:05 +0100309 if (bs) {
Subhash Peddamallua60e78e2009-02-16 10:27:07 +0100310 p = mempool_alloc(bs->bio_pool, gfp_mask);
Ingo Molnar34053972009-02-21 11:16:36 +0100311 if (!p)
312 goto err;
313 bio = p + bs->front_pad;
314 } else {
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200315 bio = kmalloc(sizeof(*bio), gfp_mask);
Ingo Molnar34053972009-02-21 11:16:36 +0100316 if (!bio)
317 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 }
Ingo Molnar34053972009-02-21 11:16:36 +0100319
320 bio_init(bio);
321
322 if (unlikely(!nr_iovecs))
323 goto out_set;
324
325 if (nr_iovecs <= BIO_INLINE_VECS) {
326 bvl = bio->bi_inline_vecs;
327 nr_iovecs = BIO_INLINE_VECS;
328 } else {
329 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
330 if (unlikely(!bvl))
331 goto err_free;
332
333 nr_iovecs = bvec_nr_vecs(idx);
334 }
335 bio->bi_flags |= idx << BIO_POOL_OFFSET;
336 bio->bi_max_vecs = nr_iovecs;
337out_set:
338 bio->bi_io_vec = bvl;
339
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 return bio;
Ingo Molnar34053972009-02-21 11:16:36 +0100341
342err_free:
343 if (bs)
344 mempool_free(p, bs->bio_pool);
345 else
346 kfree(bio);
347err:
348 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349}
350
Al Virodd0fc662005-10-07 07:46:04 +0100351struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
Peter Osterlund36763472005-09-06 15:16:42 -0700353 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
354
355 if (bio)
356 bio->bi_destructor = bio_fs_destructor;
357
358 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359}
360
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200361/*
362 * Like bio_alloc(), but doesn't use a mempool backing. This means that
363 * it CAN fail, but while bio_alloc() can only be used for allocations
364 * that have a short (finite) life span, bio_kmalloc() should be used
365 * for more permanent bio allocations (like allocating some bio's for
366 * initalization or setup purposes).
367 */
368struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
369{
370 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL);
371
372 if (bio)
373 bio->bi_destructor = bio_kmalloc_destructor;
374
375 return bio;
376}
377
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378void zero_fill_bio(struct bio *bio)
379{
380 unsigned long flags;
381 struct bio_vec *bv;
382 int i;
383
384 bio_for_each_segment(bv, bio, i) {
385 char *data = bvec_kmap_irq(bv, &flags);
386 memset(data, 0, bv->bv_len);
387 flush_dcache_page(bv->bv_page);
388 bvec_kunmap_irq(data, &flags);
389 }
390}
391EXPORT_SYMBOL(zero_fill_bio);
392
393/**
394 * bio_put - release a reference to a bio
395 * @bio: bio to release reference to
396 *
397 * Description:
398 * Put a reference to a &struct bio, either one you have gotten with
399 * bio_alloc or bio_get. The last put of a bio will free it.
400 **/
401void bio_put(struct bio *bio)
402{
403 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
404
405 /*
406 * last put frees it
407 */
408 if (atomic_dec_and_test(&bio->bi_cnt)) {
409 bio->bi_next = NULL;
410 bio->bi_destructor(bio);
411 }
412}
413
Jens Axboe165125e2007-07-24 09:28:11 +0200414inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
416 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
417 blk_recount_segments(q, bio);
418
419 return bio->bi_phys_segments;
420}
421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422/**
423 * __bio_clone - clone a bio
424 * @bio: destination bio
425 * @bio_src: bio to clone
426 *
427 * Clone a &bio. Caller will own the returned bio, but not
428 * the actual data it points to. Reference count of returned
429 * bio will be one.
430 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800431void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
Andrew Mortone525e152005-08-07 09:42:12 -0700433 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
434 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Jens Axboe5d840702008-01-25 12:44:44 +0100436 /*
437 * most users will be overriding ->bi_bdev with a new target,
438 * so we don't set nor calculate new physical/hw segment counts here
439 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 bio->bi_sector = bio_src->bi_sector;
441 bio->bi_bdev = bio_src->bi_bdev;
442 bio->bi_flags |= 1 << BIO_CLONED;
443 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 bio->bi_vcnt = bio_src->bi_vcnt;
445 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700446 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447}
448
449/**
450 * bio_clone - clone a bio
451 * @bio: bio to clone
452 * @gfp_mask: allocation priority
453 *
454 * Like __bio_clone, only also allocates the returned bio
455 */
Al Virodd0fc662005-10-07 07:46:04 +0100456struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
459
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200460 if (!b)
461 return NULL;
462
463 b->bi_destructor = bio_fs_destructor;
464 __bio_clone(b, bio);
465
466 if (bio_integrity(bio)) {
467 int ret;
468
Martin K. Petersen6d2a78e2009-03-10 08:27:39 +0100469 ret = bio_integrity_clone(b, bio, gfp_mask);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200470
Li Zefan059ea332009-03-09 10:42:45 +0100471 if (ret < 0) {
472 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200473 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100474 }
Peter Osterlund36763472005-09-06 15:16:42 -0700475 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
477 return b;
478}
479
480/**
481 * bio_get_nr_vecs - return approx number of vecs
482 * @bdev: I/O target
483 *
484 * Return the approximate number of pages we can send to this target.
485 * There's no guarantee that you will be able to fit this number of pages
486 * into a bio, it does not account for dynamic restrictions that vary
487 * on offset.
488 */
489int bio_get_nr_vecs(struct block_device *bdev)
490{
Jens Axboe165125e2007-07-24 09:28:11 +0200491 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 int nr_pages;
493
494 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
495 if (nr_pages > q->max_phys_segments)
496 nr_pages = q->max_phys_segments;
497 if (nr_pages > q->max_hw_segments)
498 nr_pages = q->max_hw_segments;
499
500 return nr_pages;
501}
502
Jens Axboe165125e2007-07-24 09:28:11 +0200503static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600504 *page, unsigned int len, unsigned int offset,
505 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
507 int retried_segments = 0;
508 struct bio_vec *bvec;
509
510 /*
511 * cloned bio must not modify vec list
512 */
513 if (unlikely(bio_flagged(bio, BIO_CLONED)))
514 return 0;
515
Jens Axboe80cfd542006-01-06 09:43:28 +0100516 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 return 0;
518
Jens Axboe80cfd542006-01-06 09:43:28 +0100519 /*
520 * For filesystems with a blocksize smaller than the pagesize
521 * we will often be called with the same page as last time and
522 * a consecutive offset. Optimize this special case.
523 */
524 if (bio->bi_vcnt > 0) {
525 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
526
527 if (page == prev->bv_page &&
528 offset == prev->bv_offset + prev->bv_len) {
529 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200530
531 if (q->merge_bvec_fn) {
532 struct bvec_merge_data bvm = {
533 .bi_bdev = bio->bi_bdev,
534 .bi_sector = bio->bi_sector,
535 .bi_size = bio->bi_size,
536 .bi_rw = bio->bi_rw,
537 };
538
539 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
540 prev->bv_len -= len;
541 return 0;
542 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100543 }
544
545 goto done;
546 }
547 }
548
549 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 return 0;
551
552 /*
553 * we might lose a segment or two here, but rather that than
554 * make this too complex.
555 */
556
557 while (bio->bi_phys_segments >= q->max_phys_segments
Mikulas Patocka5df97b92008-08-15 10:20:02 +0200558 || bio->bi_phys_segments >= q->max_hw_segments) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560 if (retried_segments)
561 return 0;
562
563 retried_segments = 1;
564 blk_recount_segments(q, bio);
565 }
566
567 /*
568 * setup the new entry, we might clear it again later if we
569 * cannot add the page
570 */
571 bvec = &bio->bi_io_vec[bio->bi_vcnt];
572 bvec->bv_page = page;
573 bvec->bv_len = len;
574 bvec->bv_offset = offset;
575
576 /*
577 * if queue has other restrictions (eg varying max sector size
578 * depending on offset), it can specify a merge_bvec_fn in the
579 * queue to get further control
580 */
581 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200582 struct bvec_merge_data bvm = {
583 .bi_bdev = bio->bi_bdev,
584 .bi_sector = bio->bi_sector,
585 .bi_size = bio->bi_size,
586 .bi_rw = bio->bi_rw,
587 };
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 /*
590 * merge_bvec_fn() returns number of bytes it can accept
591 * at this offset
592 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200593 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 bvec->bv_page = NULL;
595 bvec->bv_len = 0;
596 bvec->bv_offset = 0;
597 return 0;
598 }
599 }
600
601 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200602 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
604
605 bio->bi_vcnt++;
606 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100607 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 bio->bi_size += len;
609 return len;
610}
611
612/**
Mike Christie6e68af62005-11-11 05:30:27 -0600613 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100614 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600615 * @bio: destination bio
616 * @page: page to add
617 * @len: vec entry length
618 * @offset: vec entry offset
619 *
620 * Attempt to add a page to the bio_vec maplist. This can fail for a
621 * number of reasons, such as the bio being full or target block
622 * device limitations. The target block device must allow bio's
623 * smaller than PAGE_SIZE, so it is always possible to add a single
624 * page to an empty bio. This should only be used by REQ_PC bios.
625 */
Jens Axboe165125e2007-07-24 09:28:11 +0200626int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600627 unsigned int len, unsigned int offset)
628{
Mike Christiedefd94b2005-12-05 02:37:06 -0600629 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
Mike Christie6e68af62005-11-11 05:30:27 -0600630}
631
632/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 * bio_add_page - attempt to add page to bio
634 * @bio: destination bio
635 * @page: page to add
636 * @len: vec entry length
637 * @offset: vec entry offset
638 *
639 * Attempt to add a page to the bio_vec maplist. This can fail for a
640 * number of reasons, such as the bio being full or target block
641 * device limitations. The target block device must allow bio's
642 * smaller than PAGE_SIZE, so it is always possible to add a single
643 * page to an empty bio.
644 */
645int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
646 unsigned int offset)
647{
Mike Christiedefd94b2005-12-05 02:37:06 -0600648 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
649 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
651
652struct bio_map_data {
653 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200654 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900655 int nr_sgvecs;
656 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657};
658
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200659static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900660 struct sg_iovec *iov, int iov_count,
661 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
663 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200664 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
665 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900666 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 bio->bi_private = bmd;
668}
669
670static void bio_free_map_data(struct bio_map_data *bmd)
671{
672 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200673 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 kfree(bmd);
675}
676
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200677static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
678 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200680 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 if (!bmd)
683 return NULL;
684
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200685 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200686 if (!bmd->iovecs) {
687 kfree(bmd);
688 return NULL;
689 }
690
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200691 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200692 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return bmd;
694
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200695 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 kfree(bmd);
697 return NULL;
698}
699
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200700static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900701 struct sg_iovec *iov, int iov_count, int uncopy,
702 int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200703{
704 int ret = 0, i;
705 struct bio_vec *bvec;
706 int iov_idx = 0;
707 unsigned int iov_off = 0;
708 int read = bio_data_dir(bio) == READ;
709
710 __bio_for_each_segment(bvec, bio, i, 0) {
711 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200712 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200713
714 while (bv_len && iov_idx < iov_count) {
715 unsigned int bytes;
716 char *iov_addr;
717
718 bytes = min_t(unsigned int,
719 iov[iov_idx].iov_len - iov_off, bv_len);
720 iov_addr = iov[iov_idx].iov_base + iov_off;
721
722 if (!ret) {
723 if (!read && !uncopy)
724 ret = copy_from_user(bv_addr, iov_addr,
725 bytes);
726 if (read && uncopy)
727 ret = copy_to_user(iov_addr, bv_addr,
728 bytes);
729
730 if (ret)
731 ret = -EFAULT;
732 }
733
734 bv_len -= bytes;
735 bv_addr += bytes;
736 iov_addr += bytes;
737 iov_off += bytes;
738
739 if (iov[iov_idx].iov_len == iov_off) {
740 iov_idx++;
741 iov_off = 0;
742 }
743 }
744
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900745 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200746 __free_page(bvec->bv_page);
747 }
748
749 return ret;
750}
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752/**
753 * bio_uncopy_user - finish previously mapped bio
754 * @bio: bio being terminated
755 *
756 * Free pages allocated from bio_copy_user() and write back data
757 * to user space in case of a read.
758 */
759int bio_uncopy_user(struct bio *bio)
760{
761 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900762 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
FUJITA Tomonori81882762008-09-02 16:20:19 +0900764 if (!bio_flagged(bio, BIO_NULL_MAPPED))
765 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
766 bmd->nr_sgvecs, 1, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 bio_free_map_data(bmd);
768 bio_put(bio);
769 return ret;
770}
771
772/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200773 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900775 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200776 * @iov: the iovec.
777 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900779 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 *
781 * Prepares and returns a bio for indirect user io, bouncing data
782 * to/from kernel pages as necessary. Must be paired with
783 * call bio_uncopy_user() on io completion.
784 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900785struct bio *bio_copy_user_iov(struct request_queue *q,
786 struct rq_map_data *map_data,
787 struct sg_iovec *iov, int iov_count,
788 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 struct bio_map_data *bmd;
791 struct bio_vec *bvec;
792 struct page *page;
793 struct bio *bio;
794 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200795 int nr_pages = 0;
796 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900797 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200799 for (i = 0; i < iov_count; i++) {
800 unsigned long uaddr;
801 unsigned long end;
802 unsigned long start;
803
804 uaddr = (unsigned long)iov[i].iov_base;
805 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
806 start = uaddr >> PAGE_SHIFT;
807
808 nr_pages += end - start;
809 len += iov[i].iov_len;
810 }
811
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900812 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 if (!bmd)
814 return ERR_PTR(-ENOMEM);
815
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900817 bio = bio_alloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (!bio)
819 goto out_bmd;
820
821 bio->bi_rw |= (!write_to_vm << BIO_RW);
822
823 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900824
825 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900826 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900827 i = map_data->offset / PAGE_SIZE;
828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900830 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900832 bytes -= offset;
833
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 if (bytes > len)
835 bytes = len;
836
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900837 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900838 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900839 ret = -ENOMEM;
840 break;
841 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900842
843 page = map_data->pages[i / nr_pages];
844 page += (i % nr_pages);
845
846 i++;
847 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900848 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900849 if (!page) {
850 ret = -ENOMEM;
851 break;
852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 }
854
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900855 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900859 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
861
862 if (ret)
863 goto cleanup;
864
865 /*
866 * success
867 */
FUJITA Tomonori97ae77a2008-12-18 14:49:38 +0900868 if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900869 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200870 if (ret)
871 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 }
873
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900874 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 return bio;
876cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900877 if (!map_data)
878 bio_for_each_segment(bvec, bio, i)
879 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 bio_put(bio);
882out_bmd:
883 bio_free_map_data(bmd);
884 return ERR_PTR(ret);
885}
886
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200887/**
888 * bio_copy_user - copy user data to bio
889 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900890 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200891 * @uaddr: start of user address
892 * @len: length in bytes
893 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900894 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200895 *
896 * Prepares and returns a bio for indirect user io, bouncing data
897 * to/from kernel pages as necessary. Must be paired with
898 * call bio_uncopy_user() on io completion.
899 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900900struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
901 unsigned long uaddr, unsigned int len,
902 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200903{
904 struct sg_iovec iov;
905
906 iov.iov_base = (void __user *)uaddr;
907 iov.iov_len = len;
908
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900909 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200910}
911
Jens Axboe165125e2007-07-24 09:28:11 +0200912static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200913 struct block_device *bdev,
914 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900915 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916{
James Bottomley f1970ba2005-06-20 14:06:52 +0200917 int i, j;
918 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 struct page **pages;
920 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200921 int cur_page = 0;
922 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
James Bottomley f1970ba2005-06-20 14:06:52 +0200924 for (i = 0; i < iov_count; i++) {
925 unsigned long uaddr = (unsigned long)iov[i].iov_base;
926 unsigned long len = iov[i].iov_len;
927 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
928 unsigned long start = uaddr >> PAGE_SHIFT;
929
930 nr_pages += end - start;
931 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100932 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200933 */
Mike Christiead2d7222006-12-01 10:40:20 +0100934 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200935 return ERR_PTR(-EINVAL);
936 }
937
938 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 return ERR_PTR(-EINVAL);
940
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900941 bio = bio_alloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (!bio)
943 return ERR_PTR(-ENOMEM);
944
945 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900946 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (!pages)
948 goto out;
949
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 const int local_nr_pages = end - start;
956 const int page_limit = cur_page + local_nr_pages;
957
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700958 ret = get_user_pages_fast(uaddr, local_nr_pages,
959 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200960 if (ret < local_nr_pages) {
961 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200962 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
James Bottomley f1970ba2005-06-20 14:06:52 +0200965 offset = uaddr & ~PAGE_MASK;
966 for (j = cur_page; j < page_limit; j++) {
967 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
James Bottomley f1970ba2005-06-20 14:06:52 +0200969 if (len <= 0)
970 break;
971
972 if (bytes > len)
973 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
James Bottomley f1970ba2005-06-20 14:06:52 +0200975 /*
976 * sorry...
977 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600978 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
979 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200980 break;
981
982 len -= bytes;
983 offset = 0;
984 }
985
986 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200988 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200990 while (j < page_limit)
991 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 }
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 kfree(pages);
995
996 /*
997 * set data direction, and check if mapped pages need bouncing
998 */
999 if (!write_to_vm)
1000 bio->bi_rw |= (1 << BIO_RW);
1001
James Bottomley f1970ba2005-06-20 14:06:52 +02001002 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1004 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001005
1006 out_unmap:
1007 for (i = 0; i < nr_pages; i++) {
1008 if(!pages[i])
1009 break;
1010 page_cache_release(pages[i]);
1011 }
1012 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 kfree(pages);
1014 bio_put(bio);
1015 return ERR_PTR(ret);
1016}
1017
1018/**
1019 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001020 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 * @bdev: destination block device
1022 * @uaddr: start of user address
1023 * @len: length in bytes
1024 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001025 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 *
1027 * Map the user space address into a bio suitable for io to a block
1028 * device. Returns an error pointer in case of error.
1029 */
Jens Axboe165125e2007-07-24 09:28:11 +02001030struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001031 unsigned long uaddr, unsigned int len, int write_to_vm,
1032 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
James Bottomley f1970ba2005-06-20 14:06:52 +02001034 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001036 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001037 iov.iov_len = len;
1038
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001039 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001040}
1041
1042/**
1043 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001044 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001045 * @bdev: destination block device
1046 * @iov: the iovec.
1047 * @iov_count: number of elements in the iovec
1048 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001049 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001050 *
1051 * Map the user space address into a bio suitable for io to a block
1052 * device. Returns an error pointer in case of error.
1053 */
Jens Axboe165125e2007-07-24 09:28:11 +02001054struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001055 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001056 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001057{
1058 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001059
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001060 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1061 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 if (IS_ERR(bio))
1063 return bio;
1064
1065 /*
1066 * subtle -- if __bio_map_user() ended up bouncing a bio,
1067 * it would normally disappear when its bi_end_io is run.
1068 * however, we need it for the unmap, so grab an extra
1069 * reference to it
1070 */
1071 bio_get(bio);
1072
Mike Christie0e75f902006-12-01 10:40:55 +01001073 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
1075
1076static void __bio_unmap_user(struct bio *bio)
1077{
1078 struct bio_vec *bvec;
1079 int i;
1080
1081 /*
1082 * make sure we dirty pages we wrote to
1083 */
1084 __bio_for_each_segment(bvec, bio, i, 0) {
1085 if (bio_data_dir(bio) == READ)
1086 set_page_dirty_lock(bvec->bv_page);
1087
1088 page_cache_release(bvec->bv_page);
1089 }
1090
1091 bio_put(bio);
1092}
1093
1094/**
1095 * bio_unmap_user - unmap a bio
1096 * @bio: the bio being unmapped
1097 *
1098 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1099 * a process context.
1100 *
1101 * bio_unmap_user() may sleep.
1102 */
1103void bio_unmap_user(struct bio *bio)
1104{
1105 __bio_unmap_user(bio);
1106 bio_put(bio);
1107}
1108
NeilBrown6712ecf2007-09-27 12:47:43 +02001109static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001110{
Jens Axboeb8238252005-06-20 14:05:27 +02001111 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001112}
1113
1114
Jens Axboe165125e2007-07-24 09:28:11 +02001115static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001116 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001117{
1118 unsigned long kaddr = (unsigned long)data;
1119 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1120 unsigned long start = kaddr >> PAGE_SHIFT;
1121 const int nr_pages = end - start;
1122 int offset, i;
1123 struct bio *bio;
1124
1125 bio = bio_alloc(gfp_mask, nr_pages);
1126 if (!bio)
1127 return ERR_PTR(-ENOMEM);
1128
1129 offset = offset_in_page(kaddr);
1130 for (i = 0; i < nr_pages; i++) {
1131 unsigned int bytes = PAGE_SIZE - offset;
1132
1133 if (len <= 0)
1134 break;
1135
1136 if (bytes > len)
1137 bytes = len;
1138
Mike Christiedefd94b2005-12-05 02:37:06 -06001139 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1140 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001141 break;
1142
1143 data += bytes;
1144 len -= bytes;
1145 offset = 0;
1146 }
1147
Jens Axboeb8238252005-06-20 14:05:27 +02001148 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001149 return bio;
1150}
1151
1152/**
1153 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001154 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001155 * @data: pointer to buffer to map
1156 * @len: length in bytes
1157 * @gfp_mask: allocation flags for bio allocation
1158 *
1159 * Map the kernel address into a bio suitable for io to a block
1160 * device. Returns an error pointer in case of error.
1161 */
Jens Axboe165125e2007-07-24 09:28:11 +02001162struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001163 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001164{
1165 struct bio *bio;
1166
1167 bio = __bio_map_kern(q, data, len, gfp_mask);
1168 if (IS_ERR(bio))
1169 return bio;
1170
1171 if (bio->bi_size == len)
1172 return bio;
1173
1174 /*
1175 * Don't support partial mappings.
1176 */
1177 bio_put(bio);
1178 return ERR_PTR(-EINVAL);
1179}
1180
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001181static void bio_copy_kern_endio(struct bio *bio, int err)
1182{
1183 struct bio_vec *bvec;
1184 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001185 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001186 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001187 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001188
1189 __bio_for_each_segment(bvec, bio, i, 0) {
1190 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001191 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001192
1193 if (read && !err)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001194 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001195
1196 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001197 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001198 }
1199
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001200 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001201 bio_put(bio);
1202}
1203
1204/**
1205 * bio_copy_kern - copy kernel address into bio
1206 * @q: the struct request_queue for the bio
1207 * @data: pointer to buffer to copy
1208 * @len: length in bytes
1209 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001210 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001211 *
1212 * copy the kernel address into a bio suitable for io to a block
1213 * device. Returns an error pointer in case of error.
1214 */
1215struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1216 gfp_t gfp_mask, int reading)
1217{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001218 struct bio *bio;
1219 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001220 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001221
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001222 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1223 if (IS_ERR(bio))
1224 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001225
1226 if (!reading) {
1227 void *p = data;
1228
1229 bio_for_each_segment(bvec, bio, i) {
1230 char *addr = page_address(bvec->bv_page);
1231
1232 memcpy(addr, p, bvec->bv_len);
1233 p += bvec->bv_len;
1234 }
1235 }
1236
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001237 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001238
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001239 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001240}
1241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242/*
1243 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1244 * for performing direct-IO in BIOs.
1245 *
1246 * The problem is that we cannot run set_page_dirty() from interrupt context
1247 * because the required locks are not interrupt-safe. So what we can do is to
1248 * mark the pages dirty _before_ performing IO. And in interrupt context,
1249 * check that the pages are still dirty. If so, fine. If not, redirty them
1250 * in process context.
1251 *
1252 * We special-case compound pages here: normally this means reads into hugetlb
1253 * pages. The logic in here doesn't really work right for compound pages
1254 * because the VM does not uniformly chase down the head page in all cases.
1255 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1256 * handle them at all. So we skip compound pages here at an early stage.
1257 *
1258 * Note that this code is very hard to test under normal circumstances because
1259 * direct-io pins the pages with get_user_pages(). This makes
1260 * is_page_cache_freeable return false, and the VM will not clean the pages.
1261 * But other code (eg, pdflush) could clean the pages if they are mapped
1262 * pagecache.
1263 *
1264 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1265 * deferred bio dirtying paths.
1266 */
1267
1268/*
1269 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1270 */
1271void bio_set_pages_dirty(struct bio *bio)
1272{
1273 struct bio_vec *bvec = bio->bi_io_vec;
1274 int i;
1275
1276 for (i = 0; i < bio->bi_vcnt; i++) {
1277 struct page *page = bvec[i].bv_page;
1278
1279 if (page && !PageCompound(page))
1280 set_page_dirty_lock(page);
1281 }
1282}
1283
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001284static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285{
1286 struct bio_vec *bvec = bio->bi_io_vec;
1287 int i;
1288
1289 for (i = 0; i < bio->bi_vcnt; i++) {
1290 struct page *page = bvec[i].bv_page;
1291
1292 if (page)
1293 put_page(page);
1294 }
1295}
1296
1297/*
1298 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1299 * If they are, then fine. If, however, some pages are clean then they must
1300 * have been written out during the direct-IO read. So we take another ref on
1301 * the BIO and the offending pages and re-dirty the pages in process context.
1302 *
1303 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1304 * here on. It will run one page_cache_release() against each page and will
1305 * run one bio_put() against the BIO.
1306 */
1307
David Howells65f27f32006-11-22 14:55:48 +00001308static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
David Howells65f27f32006-11-22 14:55:48 +00001310static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311static DEFINE_SPINLOCK(bio_dirty_lock);
1312static struct bio *bio_dirty_list;
1313
1314/*
1315 * This runs in process context
1316 */
David Howells65f27f32006-11-22 14:55:48 +00001317static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318{
1319 unsigned long flags;
1320 struct bio *bio;
1321
1322 spin_lock_irqsave(&bio_dirty_lock, flags);
1323 bio = bio_dirty_list;
1324 bio_dirty_list = NULL;
1325 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1326
1327 while (bio) {
1328 struct bio *next = bio->bi_private;
1329
1330 bio_set_pages_dirty(bio);
1331 bio_release_pages(bio);
1332 bio_put(bio);
1333 bio = next;
1334 }
1335}
1336
1337void bio_check_pages_dirty(struct bio *bio)
1338{
1339 struct bio_vec *bvec = bio->bi_io_vec;
1340 int nr_clean_pages = 0;
1341 int i;
1342
1343 for (i = 0; i < bio->bi_vcnt; i++) {
1344 struct page *page = bvec[i].bv_page;
1345
1346 if (PageDirty(page) || PageCompound(page)) {
1347 page_cache_release(page);
1348 bvec[i].bv_page = NULL;
1349 } else {
1350 nr_clean_pages++;
1351 }
1352 }
1353
1354 if (nr_clean_pages) {
1355 unsigned long flags;
1356
1357 spin_lock_irqsave(&bio_dirty_lock, flags);
1358 bio->bi_private = bio_dirty_list;
1359 bio_dirty_list = bio;
1360 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1361 schedule_work(&bio_dirty_work);
1362 } else {
1363 bio_put(bio);
1364 }
1365}
1366
1367/**
1368 * bio_endio - end I/O on a bio
1369 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 * @error: error, if any
1371 *
1372 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001373 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001374 * preferred way to end I/O on a bio, it takes care of clearing
1375 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1376 * established -Exxxx (-EIO, for instance) error values in case
1377 * something went wrong. Noone should call bi_end_io() directly on a
1378 * bio unless they own it and thus know that it has an end_io
1379 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001381void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
1383 if (error)
1384 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001385 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1386 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
NeilBrown5bb23a62007-09-27 12:46:13 +02001388 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001389 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390}
1391
1392void bio_pair_release(struct bio_pair *bp)
1393{
1394 if (atomic_dec_and_test(&bp->cnt)) {
1395 struct bio *master = bp->bio1.bi_private;
1396
NeilBrown6712ecf2007-09-27 12:47:43 +02001397 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 mempool_free(bp, bp->bio2.bi_private);
1399 }
1400}
1401
NeilBrown6712ecf2007-09-27 12:47:43 +02001402static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403{
1404 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1405
1406 if (err)
1407 bp->error = err;
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410}
1411
NeilBrown6712ecf2007-09-27 12:47:43 +02001412static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
1414 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1415
1416 if (err)
1417 bp->error = err;
1418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420}
1421
1422/*
1423 * split a bio - only worry about a bio with a single page
1424 * in it's iovec
1425 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001426struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427{
Denis ChengRq6feef532008-10-09 08:57:05 +02001428 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 if (!bp)
1431 return bp;
1432
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001433 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001434 bi->bi_sector + first_sectors);
1435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 BUG_ON(bi->bi_vcnt != 1);
1437 BUG_ON(bi->bi_idx != 0);
1438 atomic_set(&bp->cnt, 3);
1439 bp->error = 0;
1440 bp->bio1 = *bi;
1441 bp->bio2 = *bi;
1442 bp->bio2.bi_sector += first_sectors;
1443 bp->bio2.bi_size -= first_sectors << 9;
1444 bp->bio1.bi_size = first_sectors << 9;
1445
1446 bp->bv1 = bi->bi_io_vec[0];
1447 bp->bv2 = bi->bi_io_vec[0];
1448 bp->bv2.bv_offset += first_sectors << 9;
1449 bp->bv2.bv_len -= first_sectors << 9;
1450 bp->bv1.bv_len = first_sectors << 9;
1451
1452 bp->bio1.bi_io_vec = &bp->bv1;
1453 bp->bio2.bi_io_vec = &bp->bv2;
1454
NeilBrowna2eb0c12006-05-22 22:35:27 -07001455 bp->bio1.bi_max_vecs = 1;
1456 bp->bio2.bi_max_vecs = 1;
1457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 bp->bio1.bi_end_io = bio_pair_end_1;
1459 bp->bio2.bi_end_io = bio_pair_end_2;
1460
1461 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001462 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001464 if (bio_integrity(bi))
1465 bio_integrity_split(bi, bp, first_sectors);
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 return bp;
1468}
1469
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001470/**
1471 * bio_sector_offset - Find hardware sector offset in bio
1472 * @bio: bio to inspect
1473 * @index: bio_vec index
1474 * @offset: offset in bv_page
1475 *
1476 * Return the number of hardware sectors between beginning of bio
1477 * and an end point indicated by a bio_vec index and an offset
1478 * within that vector's page.
1479 */
1480sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1481 unsigned int offset)
1482{
1483 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1484 struct bio_vec *bv;
1485 sector_t sectors;
1486 int i;
1487
1488 sectors = 0;
1489
1490 if (index >= bio->bi_idx)
1491 index = bio->bi_vcnt - 1;
1492
1493 __bio_for_each_segment(bv, bio, i, 0) {
1494 if (i == index) {
1495 if (offset > bv->bv_offset)
1496 sectors += (offset - bv->bv_offset) / sector_sz;
1497 break;
1498 }
1499
1500 sectors += bv->bv_len / sector_sz;
1501 }
1502
1503 return sectors;
1504}
1505EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
1507/*
1508 * create memory pools for biovec's in a bio_set.
1509 * use the global biovec slabs created for general use.
1510 */
Jens Axboe59725112007-04-02 10:06:42 +02001511static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
Jens Axboe7ff93452008-12-11 11:53:43 +01001513 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Jens Axboe7ff93452008-12-11 11:53:43 +01001515 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1516 if (!bs->bvec_pool)
1517 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 return 0;
1520}
1521
1522static void biovec_free_pools(struct bio_set *bs)
1523{
Jens Axboe7ff93452008-12-11 11:53:43 +01001524 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525}
1526
1527void bioset_free(struct bio_set *bs)
1528{
1529 if (bs->bio_pool)
1530 mempool_destroy(bs->bio_pool);
1531
1532 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001533 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
1535 kfree(bs);
1536}
1537
Jens Axboebb799ca2008-12-10 15:35:05 +01001538/**
1539 * bioset_create - Create a bio_set
1540 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1541 * @front_pad: Number of bytes to allocate in front of the returned bio
1542 *
1543 * Description:
1544 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1545 * to ask for a number of bytes to be allocated in front of the bio.
1546 * Front pad allocation is useful for embedding the bio inside
1547 * another structure, to avoid allocating extra data to go with the bio.
1548 * Note that the bio must be embedded at the END of that structure always,
1549 * or things will break badly.
1550 */
1551struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552{
Jens Axboe392ddc32008-12-23 12:42:54 +01001553 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001554 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
Jens Axboe1b434492008-10-22 20:32:58 +02001556 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 if (!bs)
1558 return NULL;
1559
Jens Axboebb799ca2008-12-10 15:35:05 +01001560 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001561
Jens Axboe392ddc32008-12-23 12:42:54 +01001562 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001563 if (!bs->bio_slab) {
1564 kfree(bs);
1565 return NULL;
1566 }
1567
1568 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 if (!bs->bio_pool)
1570 goto bad;
1571
Jens Axboebb799ca2008-12-10 15:35:05 +01001572 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 return bs;
1574
1575bad:
1576 bioset_free(bs);
1577 return NULL;
1578}
1579
1580static void __init biovec_init_slabs(void)
1581{
1582 int i;
1583
1584 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1585 int size;
1586 struct biovec_slab *bvs = bvec_slabs + i;
1587
Jens Axboea7fcd372008-12-05 16:10:29 +01001588#ifndef CONFIG_BLK_DEV_INTEGRITY
1589 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1590 bvs->slab = NULL;
1591 continue;
1592 }
1593#endif
1594
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 size = bvs->nr_vecs * sizeof(struct bio_vec);
1596 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001597 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599}
1600
1601static int __init init_bio(void)
1602{
Jens Axboebb799ca2008-12-10 15:35:05 +01001603 bio_slab_max = 2;
1604 bio_slab_nr = 0;
1605 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1606 if (!bio_slabs)
1607 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
1609 biovec_init_slabs();
1610
Jens Axboebb799ca2008-12-10 15:35:05 +01001611 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (!fs_bio_set)
1613 panic("bio: can't allocate bios\n");
1614
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001615 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1616 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 if (!bio_split_pool)
1618 panic("bio: can't create split pool\n");
1619
1620 return 0;
1621}
1622
1623subsys_initcall(init_bio);
1624
1625EXPORT_SYMBOL(bio_alloc);
Jens Axboe0a0d96b2008-09-11 13:17:37 +02001626EXPORT_SYMBOL(bio_kmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001628EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629EXPORT_SYMBOL(bio_endio);
1630EXPORT_SYMBOL(bio_init);
1631EXPORT_SYMBOL(__bio_clone);
1632EXPORT_SYMBOL(bio_clone);
1633EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001635EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636EXPORT_SYMBOL(bio_get_nr_vecs);
Jens Axboe40044ce2008-03-17 21:14:40 +01001637EXPORT_SYMBOL(bio_map_user);
1638EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001639EXPORT_SYMBOL(bio_map_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001640EXPORT_SYMBOL(bio_copy_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641EXPORT_SYMBOL(bio_pair_release);
1642EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643EXPORT_SYMBOL(bio_copy_user);
1644EXPORT_SYMBOL(bio_uncopy_user);
1645EXPORT_SYMBOL(bioset_create);
1646EXPORT_SYMBOL(bioset_free);
1647EXPORT_SYMBOL(bio_alloc_bioset);