blob: 124b95c4d58286649d8d0f982edd9074af18ac82 [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))
Jens Axboe7ff93452008-12-11 11:53:43 +0100251 bio_integrity_free(bio, bs);
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{
Jens Axboebb799ca2008-12-10 15:35:05 +0100304 struct bio *bio = NULL;
Jens Axboeb2bf9682009-02-19 08:50:26 +0100305 void *uninitialized_var(p);
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200306
Jens Axboebb799ca2008-12-10 15:35:05 +0100307 if (bs) {
Subhash Peddamallua60e78e2009-02-16 10:27:07 +0100308 p = mempool_alloc(bs->bio_pool, gfp_mask);
Jens Axboebb799ca2008-12-10 15:35:05 +0100309
310 if (p)
311 bio = p + bs->front_pad;
312 } else
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200313 bio = kmalloc(sizeof(*bio), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315 if (likely(bio)) {
316 struct bio_vec *bvl = NULL;
317
318 bio_init(bio);
319 if (likely(nr_iovecs)) {
Jens Axboeeeae1d42008-05-07 13:26:27 +0200320 unsigned long uninitialized_var(idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Jens Axboe392ddc32008-12-23 12:42:54 +0100322 if (nr_iovecs <= BIO_INLINE_VECS) {
323 idx = 0;
324 bvl = bio->bi_inline_vecs;
325 nr_iovecs = BIO_INLINE_VECS;
Jens Axboe392ddc32008-12-23 12:42:54 +0100326 } else {
327 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx,
328 bs);
329 nr_iovecs = bvec_nr_vecs(idx);
330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 if (unlikely(!bvl)) {
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200332 if (bs)
Subhash Peddamallua60e78e2009-02-16 10:27:07 +0100333 mempool_free(p, bs->bio_pool);
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200334 else
335 kfree(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 bio = NULL;
337 goto out;
338 }
339 bio->bi_flags |= idx << BIO_POOL_OFFSET;
Jens Axboe392ddc32008-12-23 12:42:54 +0100340 bio->bi_max_vecs = nr_iovecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 }
342 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 }
344out:
345 return bio;
346}
347
Al Virodd0fc662005-10-07 07:46:04 +0100348struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349{
Peter Osterlund36763472005-09-06 15:16:42 -0700350 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
351
352 if (bio)
353 bio->bi_destructor = bio_fs_destructor;
354
355 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
357
Jens Axboe0a0d96b2008-09-11 13:17:37 +0200358/*
359 * Like bio_alloc(), but doesn't use a mempool backing. This means that
360 * it CAN fail, but while bio_alloc() can only be used for allocations
361 * that have a short (finite) life span, bio_kmalloc() should be used
362 * for more permanent bio allocations (like allocating some bio's for
363 * initalization or setup purposes).
364 */
365struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
366{
367 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL);
368
369 if (bio)
370 bio->bi_destructor = bio_kmalloc_destructor;
371
372 return bio;
373}
374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375void zero_fill_bio(struct bio *bio)
376{
377 unsigned long flags;
378 struct bio_vec *bv;
379 int i;
380
381 bio_for_each_segment(bv, bio, i) {
382 char *data = bvec_kmap_irq(bv, &flags);
383 memset(data, 0, bv->bv_len);
384 flush_dcache_page(bv->bv_page);
385 bvec_kunmap_irq(data, &flags);
386 }
387}
388EXPORT_SYMBOL(zero_fill_bio);
389
390/**
391 * bio_put - release a reference to a bio
392 * @bio: bio to release reference to
393 *
394 * Description:
395 * Put a reference to a &struct bio, either one you have gotten with
396 * bio_alloc or bio_get. The last put of a bio will free it.
397 **/
398void bio_put(struct bio *bio)
399{
400 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
401
402 /*
403 * last put frees it
404 */
405 if (atomic_dec_and_test(&bio->bi_cnt)) {
406 bio->bi_next = NULL;
407 bio->bi_destructor(bio);
408 }
409}
410
Jens Axboe165125e2007-07-24 09:28:11 +0200411inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
413 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
414 blk_recount_segments(q, bio);
415
416 return bio->bi_phys_segments;
417}
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419/**
420 * __bio_clone - clone a bio
421 * @bio: destination bio
422 * @bio_src: bio to clone
423 *
424 * Clone a &bio. Caller will own the returned bio, but not
425 * the actual data it points to. Reference count of returned
426 * bio will be one.
427 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800428void __bio_clone(struct bio *bio, struct bio *bio_src)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Andrew Mortone525e152005-08-07 09:42:12 -0700430 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
431 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Jens Axboe5d840702008-01-25 12:44:44 +0100433 /*
434 * most users will be overriding ->bi_bdev with a new target,
435 * so we don't set nor calculate new physical/hw segment counts here
436 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 bio->bi_sector = bio_src->bi_sector;
438 bio->bi_bdev = bio_src->bi_bdev;
439 bio->bi_flags |= 1 << BIO_CLONED;
440 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 bio->bi_vcnt = bio_src->bi_vcnt;
442 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700443 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444}
445
446/**
447 * bio_clone - clone a bio
448 * @bio: bio to clone
449 * @gfp_mask: allocation priority
450 *
451 * Like __bio_clone, only also allocates the returned bio
452 */
Al Virodd0fc662005-10-07 07:46:04 +0100453struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
455 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
456
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200457 if (!b)
458 return NULL;
459
460 b->bi_destructor = bio_fs_destructor;
461 __bio_clone(b, bio);
462
463 if (bio_integrity(bio)) {
464 int ret;
465
466 ret = bio_integrity_clone(b, bio, fs_bio_set);
467
468 if (ret < 0)
469 return NULL;
Peter Osterlund36763472005-09-06 15:16:42 -0700470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 return b;
473}
474
475/**
476 * bio_get_nr_vecs - return approx number of vecs
477 * @bdev: I/O target
478 *
479 * Return the approximate number of pages we can send to this target.
480 * There's no guarantee that you will be able to fit this number of pages
481 * into a bio, it does not account for dynamic restrictions that vary
482 * on offset.
483 */
484int bio_get_nr_vecs(struct block_device *bdev)
485{
Jens Axboe165125e2007-07-24 09:28:11 +0200486 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 int nr_pages;
488
489 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
490 if (nr_pages > q->max_phys_segments)
491 nr_pages = q->max_phys_segments;
492 if (nr_pages > q->max_hw_segments)
493 nr_pages = q->max_hw_segments;
494
495 return nr_pages;
496}
497
Jens Axboe165125e2007-07-24 09:28:11 +0200498static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600499 *page, unsigned int len, unsigned int offset,
500 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
502 int retried_segments = 0;
503 struct bio_vec *bvec;
504
505 /*
506 * cloned bio must not modify vec list
507 */
508 if (unlikely(bio_flagged(bio, BIO_CLONED)))
509 return 0;
510
Jens Axboe80cfd542006-01-06 09:43:28 +0100511 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return 0;
513
Jens Axboe80cfd542006-01-06 09:43:28 +0100514 /*
515 * For filesystems with a blocksize smaller than the pagesize
516 * we will often be called with the same page as last time and
517 * a consecutive offset. Optimize this special case.
518 */
519 if (bio->bi_vcnt > 0) {
520 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
521
522 if (page == prev->bv_page &&
523 offset == prev->bv_offset + prev->bv_len) {
524 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200525
526 if (q->merge_bvec_fn) {
527 struct bvec_merge_data bvm = {
528 .bi_bdev = bio->bi_bdev,
529 .bi_sector = bio->bi_sector,
530 .bi_size = bio->bi_size,
531 .bi_rw = bio->bi_rw,
532 };
533
534 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
535 prev->bv_len -= len;
536 return 0;
537 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100538 }
539
540 goto done;
541 }
542 }
543
544 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 return 0;
546
547 /*
548 * we might lose a segment or two here, but rather that than
549 * make this too complex.
550 */
551
552 while (bio->bi_phys_segments >= q->max_phys_segments
Mikulas Patocka5df97b92008-08-15 10:20:02 +0200553 || bio->bi_phys_segments >= q->max_hw_segments) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 if (retried_segments)
556 return 0;
557
558 retried_segments = 1;
559 blk_recount_segments(q, bio);
560 }
561
562 /*
563 * setup the new entry, we might clear it again later if we
564 * cannot add the page
565 */
566 bvec = &bio->bi_io_vec[bio->bi_vcnt];
567 bvec->bv_page = page;
568 bvec->bv_len = len;
569 bvec->bv_offset = offset;
570
571 /*
572 * if queue has other restrictions (eg varying max sector size
573 * depending on offset), it can specify a merge_bvec_fn in the
574 * queue to get further control
575 */
576 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200577 struct bvec_merge_data bvm = {
578 .bi_bdev = bio->bi_bdev,
579 .bi_sector = bio->bi_sector,
580 .bi_size = bio->bi_size,
581 .bi_rw = bio->bi_rw,
582 };
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 /*
585 * merge_bvec_fn() returns number of bytes it can accept
586 * at this offset
587 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200588 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 bvec->bv_page = NULL;
590 bvec->bv_len = 0;
591 bvec->bv_offset = 0;
592 return 0;
593 }
594 }
595
596 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200597 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
599
600 bio->bi_vcnt++;
601 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100602 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 bio->bi_size += len;
604 return len;
605}
606
607/**
Mike Christie6e68af62005-11-11 05:30:27 -0600608 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100609 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600610 * @bio: destination bio
611 * @page: page to add
612 * @len: vec entry length
613 * @offset: vec entry offset
614 *
615 * Attempt to add a page to the bio_vec maplist. This can fail for a
616 * number of reasons, such as the bio being full or target block
617 * device limitations. The target block device must allow bio's
618 * smaller than PAGE_SIZE, so it is always possible to add a single
619 * page to an empty bio. This should only be used by REQ_PC bios.
620 */
Jens Axboe165125e2007-07-24 09:28:11 +0200621int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600622 unsigned int len, unsigned int offset)
623{
Mike Christiedefd94b2005-12-05 02:37:06 -0600624 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
Mike Christie6e68af62005-11-11 05:30:27 -0600625}
626
627/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 * bio_add_page - attempt to add page to bio
629 * @bio: destination bio
630 * @page: page to add
631 * @len: vec entry length
632 * @offset: vec entry offset
633 *
634 * Attempt to add a page to the bio_vec maplist. This can fail for a
635 * number of reasons, such as the bio being full or target block
636 * device limitations. The target block device must allow bio's
637 * smaller than PAGE_SIZE, so it is always possible to add a single
638 * page to an empty bio.
639 */
640int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
641 unsigned int offset)
642{
Mike Christiedefd94b2005-12-05 02:37:06 -0600643 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
644 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647struct bio_map_data {
648 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200649 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900650 int nr_sgvecs;
651 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652};
653
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200654static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900655 struct sg_iovec *iov, int iov_count,
656 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
658 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200659 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
660 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900661 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 bio->bi_private = bmd;
663}
664
665static void bio_free_map_data(struct bio_map_data *bmd)
666{
667 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200668 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 kfree(bmd);
670}
671
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200672static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
673 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200675 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 if (!bmd)
678 return NULL;
679
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200680 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200681 if (!bmd->iovecs) {
682 kfree(bmd);
683 return NULL;
684 }
685
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200686 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200687 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 return bmd;
689
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200690 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 kfree(bmd);
692 return NULL;
693}
694
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200695static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900696 struct sg_iovec *iov, int iov_count, int uncopy,
697 int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200698{
699 int ret = 0, i;
700 struct bio_vec *bvec;
701 int iov_idx = 0;
702 unsigned int iov_off = 0;
703 int read = bio_data_dir(bio) == READ;
704
705 __bio_for_each_segment(bvec, bio, i, 0) {
706 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200707 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200708
709 while (bv_len && iov_idx < iov_count) {
710 unsigned int bytes;
711 char *iov_addr;
712
713 bytes = min_t(unsigned int,
714 iov[iov_idx].iov_len - iov_off, bv_len);
715 iov_addr = iov[iov_idx].iov_base + iov_off;
716
717 if (!ret) {
718 if (!read && !uncopy)
719 ret = copy_from_user(bv_addr, iov_addr,
720 bytes);
721 if (read && uncopy)
722 ret = copy_to_user(iov_addr, bv_addr,
723 bytes);
724
725 if (ret)
726 ret = -EFAULT;
727 }
728
729 bv_len -= bytes;
730 bv_addr += bytes;
731 iov_addr += bytes;
732 iov_off += bytes;
733
734 if (iov[iov_idx].iov_len == iov_off) {
735 iov_idx++;
736 iov_off = 0;
737 }
738 }
739
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900740 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200741 __free_page(bvec->bv_page);
742 }
743
744 return ret;
745}
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747/**
748 * bio_uncopy_user - finish previously mapped bio
749 * @bio: bio being terminated
750 *
751 * Free pages allocated from bio_copy_user() and write back data
752 * to user space in case of a read.
753 */
754int bio_uncopy_user(struct bio *bio)
755{
756 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900757 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
FUJITA Tomonori81882762008-09-02 16:20:19 +0900759 if (!bio_flagged(bio, BIO_NULL_MAPPED))
760 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
761 bmd->nr_sgvecs, 1, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 bio_free_map_data(bmd);
763 bio_put(bio);
764 return ret;
765}
766
767/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200768 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900770 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200771 * @iov: the iovec.
772 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900774 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
776 * Prepares and returns a bio for indirect user io, bouncing data
777 * to/from kernel pages as necessary. Must be paired with
778 * call bio_uncopy_user() on io completion.
779 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900780struct bio *bio_copy_user_iov(struct request_queue *q,
781 struct rq_map_data *map_data,
782 struct sg_iovec *iov, int iov_count,
783 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 struct bio_map_data *bmd;
786 struct bio_vec *bvec;
787 struct page *page;
788 struct bio *bio;
789 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200790 int nr_pages = 0;
791 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900792 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200794 for (i = 0; i < iov_count; i++) {
795 unsigned long uaddr;
796 unsigned long end;
797 unsigned long start;
798
799 uaddr = (unsigned long)iov[i].iov_base;
800 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
801 start = uaddr >> PAGE_SHIFT;
802
803 nr_pages += end - start;
804 len += iov[i].iov_len;
805 }
806
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900807 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (!bmd)
809 return ERR_PTR(-ENOMEM);
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900812 bio = bio_alloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 if (!bio)
814 goto out_bmd;
815
816 bio->bi_rw |= (!write_to_vm << BIO_RW);
817
818 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900819
820 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900821 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900822 i = map_data->offset / PAGE_SIZE;
823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900825 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900827 bytes -= offset;
828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 if (bytes > len)
830 bytes = len;
831
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900832 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900833 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900834 ret = -ENOMEM;
835 break;
836 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900837
838 page = map_data->pages[i / nr_pages];
839 page += (i % nr_pages);
840
841 i++;
842 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900843 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900844 if (!page) {
845 ret = -ENOMEM;
846 break;
847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 }
849
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900850 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
853 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900854 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 }
856
857 if (ret)
858 goto cleanup;
859
860 /*
861 * success
862 */
FUJITA Tomonori97ae77a2008-12-18 14:49:38 +0900863 if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900864 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200865 if (ret)
866 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 }
868
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900869 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return bio;
871cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900872 if (!map_data)
873 bio_for_each_segment(bvec, bio, i)
874 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
876 bio_put(bio);
877out_bmd:
878 bio_free_map_data(bmd);
879 return ERR_PTR(ret);
880}
881
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200882/**
883 * bio_copy_user - copy user data to bio
884 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900885 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200886 * @uaddr: start of user address
887 * @len: length in bytes
888 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900889 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200890 *
891 * Prepares and returns a bio for indirect user io, bouncing data
892 * to/from kernel pages as necessary. Must be paired with
893 * call bio_uncopy_user() on io completion.
894 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900895struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
896 unsigned long uaddr, unsigned int len,
897 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200898{
899 struct sg_iovec iov;
900
901 iov.iov_base = (void __user *)uaddr;
902 iov.iov_len = len;
903
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900904 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200905}
906
Jens Axboe165125e2007-07-24 09:28:11 +0200907static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200908 struct block_device *bdev,
909 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900910 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
James Bottomley f1970ba2005-06-20 14:06:52 +0200912 int i, j;
913 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 struct page **pages;
915 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200916 int cur_page = 0;
917 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
James Bottomley f1970ba2005-06-20 14:06:52 +0200919 for (i = 0; i < iov_count; i++) {
920 unsigned long uaddr = (unsigned long)iov[i].iov_base;
921 unsigned long len = iov[i].iov_len;
922 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
923 unsigned long start = uaddr >> PAGE_SHIFT;
924
925 nr_pages += end - start;
926 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100927 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200928 */
Mike Christiead2d7222006-12-01 10:40:20 +0100929 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200930 return ERR_PTR(-EINVAL);
931 }
932
933 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 return ERR_PTR(-EINVAL);
935
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900936 bio = bio_alloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 if (!bio)
938 return ERR_PTR(-ENOMEM);
939
940 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900941 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (!pages)
943 goto out;
944
James Bottomley f1970ba2005-06-20 14:06:52 +0200945 for (i = 0; i < iov_count; i++) {
946 unsigned long uaddr = (unsigned long)iov[i].iov_base;
947 unsigned long len = iov[i].iov_len;
948 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
949 unsigned long start = uaddr >> PAGE_SHIFT;
950 const int local_nr_pages = end - start;
951 const int page_limit = cur_page + local_nr_pages;
952
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700953 ret = get_user_pages_fast(uaddr, local_nr_pages,
954 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200955 if (ret < local_nr_pages) {
956 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200957 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
James Bottomley f1970ba2005-06-20 14:06:52 +0200960 offset = uaddr & ~PAGE_MASK;
961 for (j = cur_page; j < page_limit; j++) {
962 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
James Bottomley f1970ba2005-06-20 14:06:52 +0200964 if (len <= 0)
965 break;
966
967 if (bytes > len)
968 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
James Bottomley f1970ba2005-06-20 14:06:52 +0200970 /*
971 * sorry...
972 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600973 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
974 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200975 break;
976
977 len -= bytes;
978 offset = 0;
979 }
980
981 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200983 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200985 while (j < page_limit)
986 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 }
988
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 kfree(pages);
990
991 /*
992 * set data direction, and check if mapped pages need bouncing
993 */
994 if (!write_to_vm)
995 bio->bi_rw |= (1 << BIO_RW);
996
James Bottomley f1970ba2005-06-20 14:06:52 +0200997 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 bio->bi_flags |= (1 << BIO_USER_MAPPED);
999 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001000
1001 out_unmap:
1002 for (i = 0; i < nr_pages; i++) {
1003 if(!pages[i])
1004 break;
1005 page_cache_release(pages[i]);
1006 }
1007 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 kfree(pages);
1009 bio_put(bio);
1010 return ERR_PTR(ret);
1011}
1012
1013/**
1014 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001015 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 * @bdev: destination block device
1017 * @uaddr: start of user address
1018 * @len: length in bytes
1019 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001020 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 *
1022 * Map the user space address into a bio suitable for io to a block
1023 * device. Returns an error pointer in case of error.
1024 */
Jens Axboe165125e2007-07-24 09:28:11 +02001025struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001026 unsigned long uaddr, unsigned int len, int write_to_vm,
1027 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
James Bottomley f1970ba2005-06-20 14:06:52 +02001029 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001031 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001032 iov.iov_len = len;
1033
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001034 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001035}
1036
1037/**
1038 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001039 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001040 * @bdev: destination block device
1041 * @iov: the iovec.
1042 * @iov_count: number of elements in the iovec
1043 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001044 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001045 *
1046 * Map the user space address into a bio suitable for io to a block
1047 * device. Returns an error pointer in case of error.
1048 */
Jens Axboe165125e2007-07-24 09:28:11 +02001049struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001050 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001051 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001052{
1053 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001054
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001055 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1056 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (IS_ERR(bio))
1058 return bio;
1059
1060 /*
1061 * subtle -- if __bio_map_user() ended up bouncing a bio,
1062 * it would normally disappear when its bi_end_io is run.
1063 * however, we need it for the unmap, so grab an extra
1064 * reference to it
1065 */
1066 bio_get(bio);
1067
Mike Christie0e75f902006-12-01 10:40:55 +01001068 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069}
1070
1071static void __bio_unmap_user(struct bio *bio)
1072{
1073 struct bio_vec *bvec;
1074 int i;
1075
1076 /*
1077 * make sure we dirty pages we wrote to
1078 */
1079 __bio_for_each_segment(bvec, bio, i, 0) {
1080 if (bio_data_dir(bio) == READ)
1081 set_page_dirty_lock(bvec->bv_page);
1082
1083 page_cache_release(bvec->bv_page);
1084 }
1085
1086 bio_put(bio);
1087}
1088
1089/**
1090 * bio_unmap_user - unmap a bio
1091 * @bio: the bio being unmapped
1092 *
1093 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1094 * a process context.
1095 *
1096 * bio_unmap_user() may sleep.
1097 */
1098void bio_unmap_user(struct bio *bio)
1099{
1100 __bio_unmap_user(bio);
1101 bio_put(bio);
1102}
1103
NeilBrown6712ecf2007-09-27 12:47:43 +02001104static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001105{
Jens Axboeb8238252005-06-20 14:05:27 +02001106 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001107}
1108
1109
Jens Axboe165125e2007-07-24 09:28:11 +02001110static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001111 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001112{
1113 unsigned long kaddr = (unsigned long)data;
1114 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1115 unsigned long start = kaddr >> PAGE_SHIFT;
1116 const int nr_pages = end - start;
1117 int offset, i;
1118 struct bio *bio;
1119
1120 bio = bio_alloc(gfp_mask, nr_pages);
1121 if (!bio)
1122 return ERR_PTR(-ENOMEM);
1123
1124 offset = offset_in_page(kaddr);
1125 for (i = 0; i < nr_pages; i++) {
1126 unsigned int bytes = PAGE_SIZE - offset;
1127
1128 if (len <= 0)
1129 break;
1130
1131 if (bytes > len)
1132 bytes = len;
1133
Mike Christiedefd94b2005-12-05 02:37:06 -06001134 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1135 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001136 break;
1137
1138 data += bytes;
1139 len -= bytes;
1140 offset = 0;
1141 }
1142
Jens Axboeb8238252005-06-20 14:05:27 +02001143 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001144 return bio;
1145}
1146
1147/**
1148 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001149 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001150 * @data: pointer to buffer to map
1151 * @len: length in bytes
1152 * @gfp_mask: allocation flags for bio allocation
1153 *
1154 * Map the kernel address into a bio suitable for io to a block
1155 * device. Returns an error pointer in case of error.
1156 */
Jens Axboe165125e2007-07-24 09:28:11 +02001157struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001158 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001159{
1160 struct bio *bio;
1161
1162 bio = __bio_map_kern(q, data, len, gfp_mask);
1163 if (IS_ERR(bio))
1164 return bio;
1165
1166 if (bio->bi_size == len)
1167 return bio;
1168
1169 /*
1170 * Don't support partial mappings.
1171 */
1172 bio_put(bio);
1173 return ERR_PTR(-EINVAL);
1174}
1175
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001176static void bio_copy_kern_endio(struct bio *bio, int err)
1177{
1178 struct bio_vec *bvec;
1179 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001180 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001181 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001182 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001183
1184 __bio_for_each_segment(bvec, bio, i, 0) {
1185 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001186 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001187
1188 if (read && !err)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001189 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001190
1191 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001192 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001193 }
1194
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001195 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001196 bio_put(bio);
1197}
1198
1199/**
1200 * bio_copy_kern - copy kernel address into bio
1201 * @q: the struct request_queue for the bio
1202 * @data: pointer to buffer to copy
1203 * @len: length in bytes
1204 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001205 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001206 *
1207 * copy the kernel address into a bio suitable for io to a block
1208 * device. Returns an error pointer in case of error.
1209 */
1210struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1211 gfp_t gfp_mask, int reading)
1212{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001213 struct bio *bio;
1214 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001215 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001216
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001217 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1218 if (IS_ERR(bio))
1219 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001220
1221 if (!reading) {
1222 void *p = data;
1223
1224 bio_for_each_segment(bvec, bio, i) {
1225 char *addr = page_address(bvec->bv_page);
1226
1227 memcpy(addr, p, bvec->bv_len);
1228 p += bvec->bv_len;
1229 }
1230 }
1231
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001232 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001233
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001234 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001235}
1236
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237/*
1238 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1239 * for performing direct-IO in BIOs.
1240 *
1241 * The problem is that we cannot run set_page_dirty() from interrupt context
1242 * because the required locks are not interrupt-safe. So what we can do is to
1243 * mark the pages dirty _before_ performing IO. And in interrupt context,
1244 * check that the pages are still dirty. If so, fine. If not, redirty them
1245 * in process context.
1246 *
1247 * We special-case compound pages here: normally this means reads into hugetlb
1248 * pages. The logic in here doesn't really work right for compound pages
1249 * because the VM does not uniformly chase down the head page in all cases.
1250 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1251 * handle them at all. So we skip compound pages here at an early stage.
1252 *
1253 * Note that this code is very hard to test under normal circumstances because
1254 * direct-io pins the pages with get_user_pages(). This makes
1255 * is_page_cache_freeable return false, and the VM will not clean the pages.
1256 * But other code (eg, pdflush) could clean the pages if they are mapped
1257 * pagecache.
1258 *
1259 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1260 * deferred bio dirtying paths.
1261 */
1262
1263/*
1264 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1265 */
1266void bio_set_pages_dirty(struct bio *bio)
1267{
1268 struct bio_vec *bvec = bio->bi_io_vec;
1269 int i;
1270
1271 for (i = 0; i < bio->bi_vcnt; i++) {
1272 struct page *page = bvec[i].bv_page;
1273
1274 if (page && !PageCompound(page))
1275 set_page_dirty_lock(page);
1276 }
1277}
1278
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001279static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280{
1281 struct bio_vec *bvec = bio->bi_io_vec;
1282 int i;
1283
1284 for (i = 0; i < bio->bi_vcnt; i++) {
1285 struct page *page = bvec[i].bv_page;
1286
1287 if (page)
1288 put_page(page);
1289 }
1290}
1291
1292/*
1293 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1294 * If they are, then fine. If, however, some pages are clean then they must
1295 * have been written out during the direct-IO read. So we take another ref on
1296 * the BIO and the offending pages and re-dirty the pages in process context.
1297 *
1298 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1299 * here on. It will run one page_cache_release() against each page and will
1300 * run one bio_put() against the BIO.
1301 */
1302
David Howells65f27f32006-11-22 14:55:48 +00001303static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
David Howells65f27f32006-11-22 14:55:48 +00001305static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306static DEFINE_SPINLOCK(bio_dirty_lock);
1307static struct bio *bio_dirty_list;
1308
1309/*
1310 * This runs in process context
1311 */
David Howells65f27f32006-11-22 14:55:48 +00001312static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
1314 unsigned long flags;
1315 struct bio *bio;
1316
1317 spin_lock_irqsave(&bio_dirty_lock, flags);
1318 bio = bio_dirty_list;
1319 bio_dirty_list = NULL;
1320 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1321
1322 while (bio) {
1323 struct bio *next = bio->bi_private;
1324
1325 bio_set_pages_dirty(bio);
1326 bio_release_pages(bio);
1327 bio_put(bio);
1328 bio = next;
1329 }
1330}
1331
1332void bio_check_pages_dirty(struct bio *bio)
1333{
1334 struct bio_vec *bvec = bio->bi_io_vec;
1335 int nr_clean_pages = 0;
1336 int i;
1337
1338 for (i = 0; i < bio->bi_vcnt; i++) {
1339 struct page *page = bvec[i].bv_page;
1340
1341 if (PageDirty(page) || PageCompound(page)) {
1342 page_cache_release(page);
1343 bvec[i].bv_page = NULL;
1344 } else {
1345 nr_clean_pages++;
1346 }
1347 }
1348
1349 if (nr_clean_pages) {
1350 unsigned long flags;
1351
1352 spin_lock_irqsave(&bio_dirty_lock, flags);
1353 bio->bi_private = bio_dirty_list;
1354 bio_dirty_list = bio;
1355 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1356 schedule_work(&bio_dirty_work);
1357 } else {
1358 bio_put(bio);
1359 }
1360}
1361
1362/**
1363 * bio_endio - end I/O on a bio
1364 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 * @error: error, if any
1366 *
1367 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001368 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001369 * preferred way to end I/O on a bio, it takes care of clearing
1370 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1371 * established -Exxxx (-EIO, for instance) error values in case
1372 * something went wrong. Noone should call bi_end_io() directly on a
1373 * bio unless they own it and thus know that it has an end_io
1374 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001376void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
1378 if (error)
1379 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001380 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1381 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
NeilBrown5bb23a62007-09-27 12:46:13 +02001383 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001384 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385}
1386
1387void bio_pair_release(struct bio_pair *bp)
1388{
1389 if (atomic_dec_and_test(&bp->cnt)) {
1390 struct bio *master = bp->bio1.bi_private;
1391
NeilBrown6712ecf2007-09-27 12:47:43 +02001392 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 mempool_free(bp, bp->bio2.bi_private);
1394 }
1395}
1396
NeilBrown6712ecf2007-09-27 12:47:43 +02001397static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398{
1399 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1400
1401 if (err)
1402 bp->error = err;
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405}
1406
NeilBrown6712ecf2007-09-27 12:47:43 +02001407static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
1409 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1410
1411 if (err)
1412 bp->error = err;
1413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
1416
1417/*
1418 * split a bio - only worry about a bio with a single page
1419 * in it's iovec
1420 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001421struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422{
Denis ChengRq6feef532008-10-09 08:57:05 +02001423 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
1425 if (!bp)
1426 return bp;
1427
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001428 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001429 bi->bi_sector + first_sectors);
1430
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 BUG_ON(bi->bi_vcnt != 1);
1432 BUG_ON(bi->bi_idx != 0);
1433 atomic_set(&bp->cnt, 3);
1434 bp->error = 0;
1435 bp->bio1 = *bi;
1436 bp->bio2 = *bi;
1437 bp->bio2.bi_sector += first_sectors;
1438 bp->bio2.bi_size -= first_sectors << 9;
1439 bp->bio1.bi_size = first_sectors << 9;
1440
1441 bp->bv1 = bi->bi_io_vec[0];
1442 bp->bv2 = bi->bi_io_vec[0];
1443 bp->bv2.bv_offset += first_sectors << 9;
1444 bp->bv2.bv_len -= first_sectors << 9;
1445 bp->bv1.bv_len = first_sectors << 9;
1446
1447 bp->bio1.bi_io_vec = &bp->bv1;
1448 bp->bio2.bi_io_vec = &bp->bv2;
1449
NeilBrowna2eb0c12006-05-22 22:35:27 -07001450 bp->bio1.bi_max_vecs = 1;
1451 bp->bio2.bi_max_vecs = 1;
1452
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 bp->bio1.bi_end_io = bio_pair_end_1;
1454 bp->bio2.bi_end_io = bio_pair_end_2;
1455
1456 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001457 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001459 if (bio_integrity(bi))
1460 bio_integrity_split(bi, bp, first_sectors);
1461
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 return bp;
1463}
1464
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001465/**
1466 * bio_sector_offset - Find hardware sector offset in bio
1467 * @bio: bio to inspect
1468 * @index: bio_vec index
1469 * @offset: offset in bv_page
1470 *
1471 * Return the number of hardware sectors between beginning of bio
1472 * and an end point indicated by a bio_vec index and an offset
1473 * within that vector's page.
1474 */
1475sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1476 unsigned int offset)
1477{
1478 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1479 struct bio_vec *bv;
1480 sector_t sectors;
1481 int i;
1482
1483 sectors = 0;
1484
1485 if (index >= bio->bi_idx)
1486 index = bio->bi_vcnt - 1;
1487
1488 __bio_for_each_segment(bv, bio, i, 0) {
1489 if (i == index) {
1490 if (offset > bv->bv_offset)
1491 sectors += (offset - bv->bv_offset) / sector_sz;
1492 break;
1493 }
1494
1495 sectors += bv->bv_len / sector_sz;
1496 }
1497
1498 return sectors;
1499}
1500EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
1502/*
1503 * create memory pools for biovec's in a bio_set.
1504 * use the global biovec slabs created for general use.
1505 */
Jens Axboe59725112007-04-02 10:06:42 +02001506static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507{
Jens Axboe7ff93452008-12-11 11:53:43 +01001508 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
Jens Axboe7ff93452008-12-11 11:53:43 +01001510 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1511 if (!bs->bvec_pool)
1512 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 return 0;
1515}
1516
1517static void biovec_free_pools(struct bio_set *bs)
1518{
Jens Axboe7ff93452008-12-11 11:53:43 +01001519 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520}
1521
1522void bioset_free(struct bio_set *bs)
1523{
1524 if (bs->bio_pool)
1525 mempool_destroy(bs->bio_pool);
1526
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001527 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001529 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531 kfree(bs);
1532}
1533
Jens Axboebb799ca2008-12-10 15:35:05 +01001534/**
1535 * bioset_create - Create a bio_set
1536 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1537 * @front_pad: Number of bytes to allocate in front of the returned bio
1538 *
1539 * Description:
1540 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1541 * to ask for a number of bytes to be allocated in front of the bio.
1542 * Front pad allocation is useful for embedding the bio inside
1543 * another structure, to avoid allocating extra data to go with the bio.
1544 * Note that the bio must be embedded at the END of that structure always,
1545 * or things will break badly.
1546 */
1547struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548{
Jens Axboe392ddc32008-12-23 12:42:54 +01001549 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001550 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551
Jens Axboe1b434492008-10-22 20:32:58 +02001552 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 if (!bs)
1554 return NULL;
1555
Jens Axboebb799ca2008-12-10 15:35:05 +01001556 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001557
Jens Axboe392ddc32008-12-23 12:42:54 +01001558 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001559 if (!bs->bio_slab) {
1560 kfree(bs);
1561 return NULL;
1562 }
1563
1564 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 if (!bs->bio_pool)
1566 goto bad;
1567
Jens Axboebb799ca2008-12-10 15:35:05 +01001568 if (bioset_integrity_create(bs, pool_size))
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001569 goto bad;
1570
Jens Axboebb799ca2008-12-10 15:35:05 +01001571 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 return bs;
1573
1574bad:
1575 bioset_free(bs);
1576 return NULL;
1577}
1578
1579static void __init biovec_init_slabs(void)
1580{
1581 int i;
1582
1583 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1584 int size;
1585 struct biovec_slab *bvs = bvec_slabs + i;
1586
1587 size = bvs->nr_vecs * sizeof(struct bio_vec);
1588 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001589 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
1591}
1592
1593static int __init init_bio(void)
1594{
Jens Axboebb799ca2008-12-10 15:35:05 +01001595 bio_slab_max = 2;
1596 bio_slab_nr = 0;
1597 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1598 if (!bio_slabs)
1599 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001601 bio_integrity_init_slab();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 biovec_init_slabs();
1603
Jens Axboebb799ca2008-12-10 15:35:05 +01001604 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 if (!fs_bio_set)
1606 panic("bio: can't allocate bios\n");
1607
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001608 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1609 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 if (!bio_split_pool)
1611 panic("bio: can't create split pool\n");
1612
1613 return 0;
1614}
1615
1616subsys_initcall(init_bio);
1617
1618EXPORT_SYMBOL(bio_alloc);
Jens Axboe0a0d96b2008-09-11 13:17:37 +02001619EXPORT_SYMBOL(bio_kmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001621EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622EXPORT_SYMBOL(bio_endio);
1623EXPORT_SYMBOL(bio_init);
1624EXPORT_SYMBOL(__bio_clone);
1625EXPORT_SYMBOL(bio_clone);
1626EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001628EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629EXPORT_SYMBOL(bio_get_nr_vecs);
Jens Axboe40044ce2008-03-17 21:14:40 +01001630EXPORT_SYMBOL(bio_map_user);
1631EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001632EXPORT_SYMBOL(bio_map_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001633EXPORT_SYMBOL(bio_copy_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634EXPORT_SYMBOL(bio_pair_release);
1635EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636EXPORT_SYMBOL(bio_copy_user);
1637EXPORT_SYMBOL(bio_uncopy_user);
1638EXPORT_SYMBOL(bioset_create);
1639EXPORT_SYMBOL(bioset_free);
1640EXPORT_SYMBOL(bio_alloc_bioset);