blob: d4f06327c810b163e2165455f6c0466c98b42f25 [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
un'ichi Nomura87092692009-03-09 10:40:52 +0100466 ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200467
Li Zefan059ea332009-03-09 10:42:45 +0100468 if (ret < 0) {
469 bio_put(b);
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +0200470 return NULL;
Li Zefan059ea332009-03-09 10:42:45 +0100471 }
Peter Osterlund36763472005-09-06 15:16:42 -0700472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 return b;
475}
476
477/**
478 * bio_get_nr_vecs - return approx number of vecs
479 * @bdev: I/O target
480 *
481 * Return the approximate number of pages we can send to this target.
482 * There's no guarantee that you will be able to fit this number of pages
483 * into a bio, it does not account for dynamic restrictions that vary
484 * on offset.
485 */
486int bio_get_nr_vecs(struct block_device *bdev)
487{
Jens Axboe165125e2007-07-24 09:28:11 +0200488 struct request_queue *q = bdev_get_queue(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 int nr_pages;
490
491 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
492 if (nr_pages > q->max_phys_segments)
493 nr_pages = q->max_phys_segments;
494 if (nr_pages > q->max_hw_segments)
495 nr_pages = q->max_hw_segments;
496
497 return nr_pages;
498}
499
Jens Axboe165125e2007-07-24 09:28:11 +0200500static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
Mike Christiedefd94b2005-12-05 02:37:06 -0600501 *page, unsigned int len, unsigned int offset,
502 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503{
504 int retried_segments = 0;
505 struct bio_vec *bvec;
506
507 /*
508 * cloned bio must not modify vec list
509 */
510 if (unlikely(bio_flagged(bio, BIO_CLONED)))
511 return 0;
512
Jens Axboe80cfd542006-01-06 09:43:28 +0100513 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return 0;
515
Jens Axboe80cfd542006-01-06 09:43:28 +0100516 /*
517 * For filesystems with a blocksize smaller than the pagesize
518 * we will often be called with the same page as last time and
519 * a consecutive offset. Optimize this special case.
520 */
521 if (bio->bi_vcnt > 0) {
522 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
523
524 if (page == prev->bv_page &&
525 offset == prev->bv_offset + prev->bv_len) {
526 prev->bv_len += len;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200527
528 if (q->merge_bvec_fn) {
529 struct bvec_merge_data bvm = {
530 .bi_bdev = bio->bi_bdev,
531 .bi_sector = bio->bi_sector,
532 .bi_size = bio->bi_size,
533 .bi_rw = bio->bi_rw,
534 };
535
536 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
537 prev->bv_len -= len;
538 return 0;
539 }
Jens Axboe80cfd542006-01-06 09:43:28 +0100540 }
541
542 goto done;
543 }
544 }
545
546 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 return 0;
548
549 /*
550 * we might lose a segment or two here, but rather that than
551 * make this too complex.
552 */
553
554 while (bio->bi_phys_segments >= q->max_phys_segments
Mikulas Patocka5df97b92008-08-15 10:20:02 +0200555 || bio->bi_phys_segments >= q->max_hw_segments) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557 if (retried_segments)
558 return 0;
559
560 retried_segments = 1;
561 blk_recount_segments(q, bio);
562 }
563
564 /*
565 * setup the new entry, we might clear it again later if we
566 * cannot add the page
567 */
568 bvec = &bio->bi_io_vec[bio->bi_vcnt];
569 bvec->bv_page = page;
570 bvec->bv_len = len;
571 bvec->bv_offset = offset;
572
573 /*
574 * if queue has other restrictions (eg varying max sector size
575 * depending on offset), it can specify a merge_bvec_fn in the
576 * queue to get further control
577 */
578 if (q->merge_bvec_fn) {
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200579 struct bvec_merge_data bvm = {
580 .bi_bdev = bio->bi_bdev,
581 .bi_sector = bio->bi_sector,
582 .bi_size = bio->bi_size,
583 .bi_rw = bio->bi_rw,
584 };
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 /*
587 * merge_bvec_fn() returns number of bytes it can accept
588 * at this offset
589 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200590 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 bvec->bv_page = NULL;
592 bvec->bv_len = 0;
593 bvec->bv_offset = 0;
594 return 0;
595 }
596 }
597
598 /* If we may be able to merge these biovecs, force a recount */
Mikulas Patockab8b3e162008-08-15 10:15:19 +0200599 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
601
602 bio->bi_vcnt++;
603 bio->bi_phys_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100604 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 bio->bi_size += len;
606 return len;
607}
608
609/**
Mike Christie6e68af62005-11-11 05:30:27 -0600610 * bio_add_pc_page - attempt to add page to bio
Jens Axboefddfdea2006-01-31 15:24:34 +0100611 * @q: the target queue
Mike Christie6e68af62005-11-11 05:30:27 -0600612 * @bio: destination bio
613 * @page: page to add
614 * @len: vec entry length
615 * @offset: vec entry offset
616 *
617 * Attempt to add a page to the bio_vec maplist. This can fail for a
618 * number of reasons, such as the bio being full or target block
619 * device limitations. The target block device must allow bio's
620 * smaller than PAGE_SIZE, so it is always possible to add a single
621 * page to an empty bio. This should only be used by REQ_PC bios.
622 */
Jens Axboe165125e2007-07-24 09:28:11 +0200623int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
Mike Christie6e68af62005-11-11 05:30:27 -0600624 unsigned int len, unsigned int offset)
625{
Mike Christiedefd94b2005-12-05 02:37:06 -0600626 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
Mike Christie6e68af62005-11-11 05:30:27 -0600627}
628
629/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 * bio_add_page - attempt to add page to bio
631 * @bio: destination bio
632 * @page: page to add
633 * @len: vec entry length
634 * @offset: vec entry offset
635 *
636 * Attempt to add a page to the bio_vec maplist. This can fail for a
637 * number of reasons, such as the bio being full or target block
638 * device limitations. The target block device must allow bio's
639 * smaller than PAGE_SIZE, so it is always possible to add a single
640 * page to an empty bio.
641 */
642int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
643 unsigned int offset)
644{
Mike Christiedefd94b2005-12-05 02:37:06 -0600645 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
646 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649struct bio_map_data {
650 struct bio_vec *iovecs;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200651 struct sg_iovec *sgvecs;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900652 int nr_sgvecs;
653 int is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654};
655
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200656static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900657 struct sg_iovec *iov, int iov_count,
658 int is_our_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
660 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200661 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
662 bmd->nr_sgvecs = iov_count;
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900663 bmd->is_our_pages = is_our_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 bio->bi_private = bmd;
665}
666
667static void bio_free_map_data(struct bio_map_data *bmd)
668{
669 kfree(bmd->iovecs);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200670 kfree(bmd->sgvecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 kfree(bmd);
672}
673
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200674static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
675 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200677 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679 if (!bmd)
680 return NULL;
681
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200682 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200683 if (!bmd->iovecs) {
684 kfree(bmd);
685 return NULL;
686 }
687
FUJITA Tomonori76029ff2008-08-25 20:36:08 +0200688 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200689 if (bmd->sgvecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 return bmd;
691
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200692 kfree(bmd->iovecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 kfree(bmd);
694 return NULL;
695}
696
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200697static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900698 struct sg_iovec *iov, int iov_count, int uncopy,
699 int do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200700{
701 int ret = 0, i;
702 struct bio_vec *bvec;
703 int iov_idx = 0;
704 unsigned int iov_off = 0;
705 int read = bio_data_dir(bio) == READ;
706
707 __bio_for_each_segment(bvec, bio, i, 0) {
708 char *bv_addr = page_address(bvec->bv_page);
FUJITA Tomonoriaefcc282008-08-25 20:36:08 +0200709 unsigned int bv_len = iovecs[i].bv_len;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200710
711 while (bv_len && iov_idx < iov_count) {
712 unsigned int bytes;
713 char *iov_addr;
714
715 bytes = min_t(unsigned int,
716 iov[iov_idx].iov_len - iov_off, bv_len);
717 iov_addr = iov[iov_idx].iov_base + iov_off;
718
719 if (!ret) {
720 if (!read && !uncopy)
721 ret = copy_from_user(bv_addr, iov_addr,
722 bytes);
723 if (read && uncopy)
724 ret = copy_to_user(iov_addr, bv_addr,
725 bytes);
726
727 if (ret)
728 ret = -EFAULT;
729 }
730
731 bv_len -= bytes;
732 bv_addr += bytes;
733 iov_addr += bytes;
734 iov_off += bytes;
735
736 if (iov[iov_idx].iov_len == iov_off) {
737 iov_idx++;
738 iov_off = 0;
739 }
740 }
741
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900742 if (do_free_page)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200743 __free_page(bvec->bv_page);
744 }
745
746 return ret;
747}
748
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749/**
750 * bio_uncopy_user - finish previously mapped bio
751 * @bio: bio being terminated
752 *
753 * Free pages allocated from bio_copy_user() and write back data
754 * to user space in case of a read.
755 */
756int bio_uncopy_user(struct bio *bio)
757{
758 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori81882762008-09-02 16:20:19 +0900759 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
FUJITA Tomonori81882762008-09-02 16:20:19 +0900761 if (!bio_flagged(bio, BIO_NULL_MAPPED))
762 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
763 bmd->nr_sgvecs, 1, bmd->is_our_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 bio_free_map_data(bmd);
765 bio_put(bio);
766 return ret;
767}
768
769/**
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200770 * bio_copy_user_iov - copy user data to bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900772 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200773 * @iov: the iovec.
774 * @iov_count: number of elements in the iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900776 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 *
778 * Prepares and returns a bio for indirect user io, bouncing data
779 * to/from kernel pages as necessary. Must be paired with
780 * call bio_uncopy_user() on io completion.
781 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900782struct bio *bio_copy_user_iov(struct request_queue *q,
783 struct rq_map_data *map_data,
784 struct sg_iovec *iov, int iov_count,
785 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 struct bio_map_data *bmd;
788 struct bio_vec *bvec;
789 struct page *page;
790 struct bio *bio;
791 int i, ret;
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200792 int nr_pages = 0;
793 unsigned int len = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900794 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200796 for (i = 0; i < iov_count; i++) {
797 unsigned long uaddr;
798 unsigned long end;
799 unsigned long start;
800
801 uaddr = (unsigned long)iov[i].iov_base;
802 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
803 start = uaddr >> PAGE_SHIFT;
804
805 nr_pages += end - start;
806 len += iov[i].iov_len;
807 }
808
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900809 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 if (!bmd)
811 return ERR_PTR(-ENOMEM);
812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900814 bio = bio_alloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (!bio)
816 goto out_bmd;
817
818 bio->bi_rw |= (!write_to_vm << BIO_RW);
819
820 ret = 0;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900821
822 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900823 nr_pages = 1 << map_data->page_order;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900824 i = map_data->offset / PAGE_SIZE;
825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 while (len) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900827 unsigned int bytes = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900829 bytes -= offset;
830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 if (bytes > len)
832 bytes = len;
833
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900834 if (map_data) {
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900835 if (i == map_data->nr_entries * nr_pages) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900836 ret = -ENOMEM;
837 break;
838 }
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900839
840 page = map_data->pages[i / nr_pages];
841 page += (i % nr_pages);
842
843 i++;
844 } else {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900845 page = alloc_page(q->bounce_gfp | gfp_mask);
FUJITA Tomonorie623ddb2008-12-18 14:49:36 +0900846 if (!page) {
847 ret = -ENOMEM;
848 break;
849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900852 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 len -= bytes;
FUJITA Tomonori56c451f2008-12-18 14:49:37 +0900856 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 }
858
859 if (ret)
860 goto cleanup;
861
862 /*
863 * success
864 */
FUJITA Tomonori97ae77a2008-12-18 14:49:38 +0900865 if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900866 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200867 if (ret)
868 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 }
870
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900871 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return bio;
873cleanup:
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900874 if (!map_data)
875 bio_for_each_segment(bvec, bio, i)
876 __free_page(bvec->bv_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878 bio_put(bio);
879out_bmd:
880 bio_free_map_data(bmd);
881 return ERR_PTR(ret);
882}
883
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200884/**
885 * bio_copy_user - copy user data to bio
886 * @q: destination block queue
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900887 * @map_data: pointer to the rq_map_data holding pages (if necessary)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200888 * @uaddr: start of user address
889 * @len: length in bytes
890 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900891 * @gfp_mask: memory allocation flags
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200892 *
893 * Prepares and returns a bio for indirect user io, bouncing data
894 * to/from kernel pages as necessary. Must be paired with
895 * call bio_uncopy_user() on io completion.
896 */
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900897struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
898 unsigned long uaddr, unsigned int len,
899 int write_to_vm, gfp_t gfp_mask)
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200900{
901 struct sg_iovec iov;
902
903 iov.iov_base = (void __user *)uaddr;
904 iov.iov_len = len;
905
FUJITA Tomonori152e2832008-08-28 16:17:06 +0900906 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
FUJITA Tomonoric5dec1c2008-04-11 12:56:49 +0200907}
908
Jens Axboe165125e2007-07-24 09:28:11 +0200909static struct bio *__bio_map_user_iov(struct request_queue *q,
James Bottomley f1970ba2005-06-20 14:06:52 +0200910 struct block_device *bdev,
911 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900912 int write_to_vm, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
James Bottomley f1970ba2005-06-20 14:06:52 +0200914 int i, j;
915 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 struct page **pages;
917 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200918 int cur_page = 0;
919 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
James Bottomley f1970ba2005-06-20 14:06:52 +0200921 for (i = 0; i < iov_count; i++) {
922 unsigned long uaddr = (unsigned long)iov[i].iov_base;
923 unsigned long len = iov[i].iov_len;
924 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
925 unsigned long start = uaddr >> PAGE_SHIFT;
926
927 nr_pages += end - start;
928 /*
Mike Christiead2d7222006-12-01 10:40:20 +0100929 * buffer must be aligned to at least hardsector size for now
James Bottomley f1970ba2005-06-20 14:06:52 +0200930 */
Mike Christiead2d7222006-12-01 10:40:20 +0100931 if (uaddr & queue_dma_alignment(q))
James Bottomley f1970ba2005-06-20 14:06:52 +0200932 return ERR_PTR(-EINVAL);
933 }
934
935 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return ERR_PTR(-EINVAL);
937
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900938 bio = bio_alloc(gfp_mask, nr_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 if (!bio)
940 return ERR_PTR(-ENOMEM);
941
942 ret = -ENOMEM;
FUJITA Tomonoria3bce902008-08-28 16:17:05 +0900943 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (!pages)
945 goto out;
946
James Bottomley f1970ba2005-06-20 14:06:52 +0200947 for (i = 0; i < iov_count; i++) {
948 unsigned long uaddr = (unsigned long)iov[i].iov_base;
949 unsigned long len = iov[i].iov_len;
950 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 unsigned long start = uaddr >> PAGE_SHIFT;
952 const int local_nr_pages = end - start;
953 const int page_limit = cur_page + local_nr_pages;
954
Nick Pigginf5dd33c2008-07-25 19:45:25 -0700955 ret = get_user_pages_fast(uaddr, local_nr_pages,
956 write_to_vm, &pages[cur_page]);
Jens Axboe99172152006-06-16 13:02:29 +0200957 if (ret < local_nr_pages) {
958 ret = -EFAULT;
James Bottomley f1970ba2005-06-20 14:06:52 +0200959 goto out_unmap;
Jens Axboe99172152006-06-16 13:02:29 +0200960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
James Bottomley f1970ba2005-06-20 14:06:52 +0200962 offset = uaddr & ~PAGE_MASK;
963 for (j = cur_page; j < page_limit; j++) {
964 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
James Bottomley f1970ba2005-06-20 14:06:52 +0200966 if (len <= 0)
967 break;
968
969 if (bytes > len)
970 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
James Bottomley f1970ba2005-06-20 14:06:52 +0200972 /*
973 * sorry...
974 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600975 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
976 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200977 break;
978
979 len -= bytes;
980 offset = 0;
981 }
982
983 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200985 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200987 while (j < page_limit)
988 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 }
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 kfree(pages);
992
993 /*
994 * set data direction, and check if mapped pages need bouncing
995 */
996 if (!write_to_vm)
997 bio->bi_rw |= (1 << BIO_RW);
998
James Bottomley f1970ba2005-06-20 14:06:52 +0200999 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1001 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001002
1003 out_unmap:
1004 for (i = 0; i < nr_pages; i++) {
1005 if(!pages[i])
1006 break;
1007 page_cache_release(pages[i]);
1008 }
1009 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 kfree(pages);
1011 bio_put(bio);
1012 return ERR_PTR(ret);
1013}
1014
1015/**
1016 * bio_map_user - map user address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001017 * @q: the struct request_queue for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 * @bdev: destination block device
1019 * @uaddr: start of user address
1020 * @len: length in bytes
1021 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001022 * @gfp_mask: memory allocation flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 *
1024 * Map the user space address into a bio suitable for io to a block
1025 * device. Returns an error pointer in case of error.
1026 */
Jens Axboe165125e2007-07-24 09:28:11 +02001027struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001028 unsigned long uaddr, unsigned int len, int write_to_vm,
1029 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030{
James Bottomley f1970ba2005-06-20 14:06:52 +02001031 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +01001033 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +02001034 iov.iov_len = len;
1035
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001036 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
James Bottomley f1970ba2005-06-20 14:06:52 +02001037}
1038
1039/**
1040 * bio_map_user_iov - map user sg_iovec table into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001041 * @q: the struct request_queue for the bio
James Bottomley f1970ba2005-06-20 14:06:52 +02001042 * @bdev: destination block device
1043 * @iov: the iovec.
1044 * @iov_count: number of elements in the iovec
1045 * @write_to_vm: bool indicating writing to pages or not
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001046 * @gfp_mask: memory allocation flags
James Bottomley f1970ba2005-06-20 14:06:52 +02001047 *
1048 * Map the user space address into a bio suitable for io to a block
1049 * device. Returns an error pointer in case of error.
1050 */
Jens Axboe165125e2007-07-24 09:28:11 +02001051struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
James Bottomley f1970ba2005-06-20 14:06:52 +02001052 struct sg_iovec *iov, int iov_count,
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001053 int write_to_vm, gfp_t gfp_mask)
James Bottomley f1970ba2005-06-20 14:06:52 +02001054{
1055 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +02001056
FUJITA Tomonoria3bce902008-08-28 16:17:05 +09001057 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1058 gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (IS_ERR(bio))
1060 return bio;
1061
1062 /*
1063 * subtle -- if __bio_map_user() ended up bouncing a bio,
1064 * it would normally disappear when its bi_end_io is run.
1065 * however, we need it for the unmap, so grab an extra
1066 * reference to it
1067 */
1068 bio_get(bio);
1069
Mike Christie0e75f902006-12-01 10:40:55 +01001070 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071}
1072
1073static void __bio_unmap_user(struct bio *bio)
1074{
1075 struct bio_vec *bvec;
1076 int i;
1077
1078 /*
1079 * make sure we dirty pages we wrote to
1080 */
1081 __bio_for_each_segment(bvec, bio, i, 0) {
1082 if (bio_data_dir(bio) == READ)
1083 set_page_dirty_lock(bvec->bv_page);
1084
1085 page_cache_release(bvec->bv_page);
1086 }
1087
1088 bio_put(bio);
1089}
1090
1091/**
1092 * bio_unmap_user - unmap a bio
1093 * @bio: the bio being unmapped
1094 *
1095 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1096 * a process context.
1097 *
1098 * bio_unmap_user() may sleep.
1099 */
1100void bio_unmap_user(struct bio *bio)
1101{
1102 __bio_unmap_user(bio);
1103 bio_put(bio);
1104}
1105
NeilBrown6712ecf2007-09-27 12:47:43 +02001106static void bio_map_kern_endio(struct bio *bio, int err)
Jens Axboeb8238252005-06-20 14:05:27 +02001107{
Jens Axboeb8238252005-06-20 14:05:27 +02001108 bio_put(bio);
Jens Axboeb8238252005-06-20 14:05:27 +02001109}
1110
1111
Jens Axboe165125e2007-07-24 09:28:11 +02001112static struct bio *__bio_map_kern(struct request_queue *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -04001113 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001114{
1115 unsigned long kaddr = (unsigned long)data;
1116 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1117 unsigned long start = kaddr >> PAGE_SHIFT;
1118 const int nr_pages = end - start;
1119 int offset, i;
1120 struct bio *bio;
1121
1122 bio = bio_alloc(gfp_mask, nr_pages);
1123 if (!bio)
1124 return ERR_PTR(-ENOMEM);
1125
1126 offset = offset_in_page(kaddr);
1127 for (i = 0; i < nr_pages; i++) {
1128 unsigned int bytes = PAGE_SIZE - offset;
1129
1130 if (len <= 0)
1131 break;
1132
1133 if (bytes > len)
1134 bytes = len;
1135
Mike Christiedefd94b2005-12-05 02:37:06 -06001136 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1137 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +02001138 break;
1139
1140 data += bytes;
1141 len -= bytes;
1142 offset = 0;
1143 }
1144
Jens Axboeb8238252005-06-20 14:05:27 +02001145 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +02001146 return bio;
1147}
1148
1149/**
1150 * bio_map_kern - map kernel address into bio
Jens Axboe165125e2007-07-24 09:28:11 +02001151 * @q: the struct request_queue for the bio
Mike Christie df46b9a2005-06-20 14:04:44 +02001152 * @data: pointer to buffer to map
1153 * @len: length in bytes
1154 * @gfp_mask: allocation flags for bio allocation
1155 *
1156 * Map the kernel address into a bio suitable for io to a block
1157 * device. Returns an error pointer in case of error.
1158 */
Jens Axboe165125e2007-07-24 09:28:11 +02001159struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -04001160 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +02001161{
1162 struct bio *bio;
1163
1164 bio = __bio_map_kern(q, data, len, gfp_mask);
1165 if (IS_ERR(bio))
1166 return bio;
1167
1168 if (bio->bi_size == len)
1169 return bio;
1170
1171 /*
1172 * Don't support partial mappings.
1173 */
1174 bio_put(bio);
1175 return ERR_PTR(-EINVAL);
1176}
1177
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001178static void bio_copy_kern_endio(struct bio *bio, int err)
1179{
1180 struct bio_vec *bvec;
1181 const int read = bio_data_dir(bio) == READ;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001182 struct bio_map_data *bmd = bio->bi_private;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001183 int i;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001184 char *p = bmd->sgvecs[0].iov_base;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001185
1186 __bio_for_each_segment(bvec, bio, i, 0) {
1187 char *addr = page_address(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001188 int len = bmd->iovecs[i].bv_len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001189
1190 if (read && !err)
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001191 memcpy(p, addr, len);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001192
1193 __free_page(bvec->bv_page);
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001194 p += len;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001195 }
1196
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001197 bio_free_map_data(bmd);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001198 bio_put(bio);
1199}
1200
1201/**
1202 * bio_copy_kern - copy kernel address into bio
1203 * @q: the struct request_queue for the bio
1204 * @data: pointer to buffer to copy
1205 * @len: length in bytes
1206 * @gfp_mask: allocation flags for bio and page allocation
Randy Dunlapffee0252008-04-30 09:08:54 +02001207 * @reading: data direction is READ
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001208 *
1209 * copy the kernel address into a bio suitable for io to a block
1210 * device. Returns an error pointer in case of error.
1211 */
1212struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1213 gfp_t gfp_mask, int reading)
1214{
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001215 struct bio *bio;
1216 struct bio_vec *bvec;
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001217 int i;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001218
FUJITA Tomonori4d8ab622008-08-28 15:05:57 +09001219 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1220 if (IS_ERR(bio))
1221 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001222
1223 if (!reading) {
1224 void *p = data;
1225
1226 bio_for_each_segment(bvec, bio, i) {
1227 char *addr = page_address(bvec->bv_page);
1228
1229 memcpy(addr, p, bvec->bv_len);
1230 p += bvec->bv_len;
1231 }
1232 }
1233
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001234 bio->bi_end_io = bio_copy_kern_endio;
FUJITA Tomonori76029ff2008-08-25 20:36:08 +02001235
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001236 return bio;
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001237}
1238
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239/*
1240 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1241 * for performing direct-IO in BIOs.
1242 *
1243 * The problem is that we cannot run set_page_dirty() from interrupt context
1244 * because the required locks are not interrupt-safe. So what we can do is to
1245 * mark the pages dirty _before_ performing IO. And in interrupt context,
1246 * check that the pages are still dirty. If so, fine. If not, redirty them
1247 * in process context.
1248 *
1249 * We special-case compound pages here: normally this means reads into hugetlb
1250 * pages. The logic in here doesn't really work right for compound pages
1251 * because the VM does not uniformly chase down the head page in all cases.
1252 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1253 * handle them at all. So we skip compound pages here at an early stage.
1254 *
1255 * Note that this code is very hard to test under normal circumstances because
1256 * direct-io pins the pages with get_user_pages(). This makes
1257 * is_page_cache_freeable return false, and the VM will not clean the pages.
1258 * But other code (eg, pdflush) could clean the pages if they are mapped
1259 * pagecache.
1260 *
1261 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1262 * deferred bio dirtying paths.
1263 */
1264
1265/*
1266 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1267 */
1268void bio_set_pages_dirty(struct bio *bio)
1269{
1270 struct bio_vec *bvec = bio->bi_io_vec;
1271 int i;
1272
1273 for (i = 0; i < bio->bi_vcnt; i++) {
1274 struct page *page = bvec[i].bv_page;
1275
1276 if (page && !PageCompound(page))
1277 set_page_dirty_lock(page);
1278 }
1279}
1280
Adrian Bunk86b6c7a2008-02-18 13:48:32 +01001281static void bio_release_pages(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282{
1283 struct bio_vec *bvec = bio->bi_io_vec;
1284 int i;
1285
1286 for (i = 0; i < bio->bi_vcnt; i++) {
1287 struct page *page = bvec[i].bv_page;
1288
1289 if (page)
1290 put_page(page);
1291 }
1292}
1293
1294/*
1295 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1296 * If they are, then fine. If, however, some pages are clean then they must
1297 * have been written out during the direct-IO read. So we take another ref on
1298 * the BIO and the offending pages and re-dirty the pages in process context.
1299 *
1300 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1301 * here on. It will run one page_cache_release() against each page and will
1302 * run one bio_put() against the BIO.
1303 */
1304
David Howells65f27f32006-11-22 14:55:48 +00001305static void bio_dirty_fn(struct work_struct *work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
David Howells65f27f32006-11-22 14:55:48 +00001307static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308static DEFINE_SPINLOCK(bio_dirty_lock);
1309static struct bio *bio_dirty_list;
1310
1311/*
1312 * This runs in process context
1313 */
David Howells65f27f32006-11-22 14:55:48 +00001314static void bio_dirty_fn(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
1316 unsigned long flags;
1317 struct bio *bio;
1318
1319 spin_lock_irqsave(&bio_dirty_lock, flags);
1320 bio = bio_dirty_list;
1321 bio_dirty_list = NULL;
1322 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1323
1324 while (bio) {
1325 struct bio *next = bio->bi_private;
1326
1327 bio_set_pages_dirty(bio);
1328 bio_release_pages(bio);
1329 bio_put(bio);
1330 bio = next;
1331 }
1332}
1333
1334void bio_check_pages_dirty(struct bio *bio)
1335{
1336 struct bio_vec *bvec = bio->bi_io_vec;
1337 int nr_clean_pages = 0;
1338 int i;
1339
1340 for (i = 0; i < bio->bi_vcnt; i++) {
1341 struct page *page = bvec[i].bv_page;
1342
1343 if (PageDirty(page) || PageCompound(page)) {
1344 page_cache_release(page);
1345 bvec[i].bv_page = NULL;
1346 } else {
1347 nr_clean_pages++;
1348 }
1349 }
1350
1351 if (nr_clean_pages) {
1352 unsigned long flags;
1353
1354 spin_lock_irqsave(&bio_dirty_lock, flags);
1355 bio->bi_private = bio_dirty_list;
1356 bio_dirty_list = bio;
1357 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1358 schedule_work(&bio_dirty_work);
1359 } else {
1360 bio_put(bio);
1361 }
1362}
1363
1364/**
1365 * bio_endio - end I/O on a bio
1366 * @bio: bio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 * @error: error, if any
1368 *
1369 * Description:
NeilBrown6712ecf2007-09-27 12:47:43 +02001370 * bio_endio() will end I/O on the whole bio. bio_endio() is the
NeilBrown5bb23a62007-09-27 12:46:13 +02001371 * preferred way to end I/O on a bio, it takes care of clearing
1372 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1373 * established -Exxxx (-EIO, for instance) error values in case
1374 * something went wrong. Noone should call bi_end_io() directly on a
1375 * bio unless they own it and thus know that it has an end_io
1376 * function.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 **/
NeilBrown6712ecf2007-09-27 12:47:43 +02001378void bio_endio(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
1380 if (error)
1381 clear_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9cc54d42007-09-27 12:46:12 +02001382 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1383 error = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
NeilBrown5bb23a62007-09-27 12:46:13 +02001385 if (bio->bi_end_io)
NeilBrown6712ecf2007-09-27 12:47:43 +02001386 bio->bi_end_io(bio, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
1389void bio_pair_release(struct bio_pair *bp)
1390{
1391 if (atomic_dec_and_test(&bp->cnt)) {
1392 struct bio *master = bp->bio1.bi_private;
1393
NeilBrown6712ecf2007-09-27 12:47:43 +02001394 bio_endio(master, bp->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 mempool_free(bp, bp->bio2.bi_private);
1396 }
1397}
1398
NeilBrown6712ecf2007-09-27 12:47:43 +02001399static void bio_pair_end_1(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1402
1403 if (err)
1404 bp->error = err;
1405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407}
1408
NeilBrown6712ecf2007-09-27 12:47:43 +02001409static void bio_pair_end_2(struct bio *bi, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410{
1411 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1412
1413 if (err)
1414 bp->error = err;
1415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 bio_pair_release(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417}
1418
1419/*
1420 * split a bio - only worry about a bio with a single page
1421 * in it's iovec
1422 */
Denis ChengRq6feef532008-10-09 08:57:05 +02001423struct bio_pair *bio_split(struct bio *bi, int first_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
Denis ChengRq6feef532008-10-09 08:57:05 +02001425 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
1427 if (!bp)
1428 return bp;
1429
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +01001430 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
Jens Axboe2056a782006-03-23 20:00:26 +01001431 bi->bi_sector + first_sectors);
1432
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 BUG_ON(bi->bi_vcnt != 1);
1434 BUG_ON(bi->bi_idx != 0);
1435 atomic_set(&bp->cnt, 3);
1436 bp->error = 0;
1437 bp->bio1 = *bi;
1438 bp->bio2 = *bi;
1439 bp->bio2.bi_sector += first_sectors;
1440 bp->bio2.bi_size -= first_sectors << 9;
1441 bp->bio1.bi_size = first_sectors << 9;
1442
1443 bp->bv1 = bi->bi_io_vec[0];
1444 bp->bv2 = bi->bi_io_vec[0];
1445 bp->bv2.bv_offset += first_sectors << 9;
1446 bp->bv2.bv_len -= first_sectors << 9;
1447 bp->bv1.bv_len = first_sectors << 9;
1448
1449 bp->bio1.bi_io_vec = &bp->bv1;
1450 bp->bio2.bi_io_vec = &bp->bv2;
1451
NeilBrowna2eb0c12006-05-22 22:35:27 -07001452 bp->bio1.bi_max_vecs = 1;
1453 bp->bio2.bi_max_vecs = 1;
1454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 bp->bio1.bi_end_io = bio_pair_end_1;
1456 bp->bio2.bi_end_io = bio_pair_end_2;
1457
1458 bp->bio1.bi_private = bi;
Denis ChengRq6feef532008-10-09 08:57:05 +02001459 bp->bio2.bi_private = bio_split_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001461 if (bio_integrity(bi))
1462 bio_integrity_split(bi, bp, first_sectors);
1463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return bp;
1465}
1466
Martin K. Petersenad3316b2008-10-01 22:42:53 -04001467/**
1468 * bio_sector_offset - Find hardware sector offset in bio
1469 * @bio: bio to inspect
1470 * @index: bio_vec index
1471 * @offset: offset in bv_page
1472 *
1473 * Return the number of hardware sectors between beginning of bio
1474 * and an end point indicated by a bio_vec index and an offset
1475 * within that vector's page.
1476 */
1477sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1478 unsigned int offset)
1479{
1480 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1481 struct bio_vec *bv;
1482 sector_t sectors;
1483 int i;
1484
1485 sectors = 0;
1486
1487 if (index >= bio->bi_idx)
1488 index = bio->bi_vcnt - 1;
1489
1490 __bio_for_each_segment(bv, bio, i, 0) {
1491 if (i == index) {
1492 if (offset > bv->bv_offset)
1493 sectors += (offset - bv->bv_offset) / sector_sz;
1494 break;
1495 }
1496
1497 sectors += bv->bv_len / sector_sz;
1498 }
1499
1500 return sectors;
1501}
1502EXPORT_SYMBOL(bio_sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
1504/*
1505 * create memory pools for biovec's in a bio_set.
1506 * use the global biovec slabs created for general use.
1507 */
Jens Axboe59725112007-04-02 10:06:42 +02001508static int biovec_create_pools(struct bio_set *bs, int pool_entries)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
Jens Axboe7ff93452008-12-11 11:53:43 +01001510 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
Jens Axboe7ff93452008-12-11 11:53:43 +01001512 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1513 if (!bs->bvec_pool)
1514 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return 0;
1517}
1518
1519static void biovec_free_pools(struct bio_set *bs)
1520{
Jens Axboe7ff93452008-12-11 11:53:43 +01001521 mempool_destroy(bs->bvec_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522}
1523
1524void bioset_free(struct bio_set *bs)
1525{
1526 if (bs->bio_pool)
1527 mempool_destroy(bs->bio_pool);
1528
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001529 bioset_integrity_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 biovec_free_pools(bs);
Jens Axboebb799ca2008-12-10 15:35:05 +01001531 bio_put_slab(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
1533 kfree(bs);
1534}
1535
Jens Axboebb799ca2008-12-10 15:35:05 +01001536/**
1537 * bioset_create - Create a bio_set
1538 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1539 * @front_pad: Number of bytes to allocate in front of the returned bio
1540 *
1541 * Description:
1542 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1543 * to ask for a number of bytes to be allocated in front of the bio.
1544 * Front pad allocation is useful for embedding the bio inside
1545 * another structure, to avoid allocating extra data to go with the bio.
1546 * Note that the bio must be embedded at the END of that structure always,
1547 * or things will break badly.
1548 */
1549struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
Jens Axboe392ddc32008-12-23 12:42:54 +01001551 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
Jens Axboe1b434492008-10-22 20:32:58 +02001552 struct bio_set *bs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Jens Axboe1b434492008-10-22 20:32:58 +02001554 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (!bs)
1556 return NULL;
1557
Jens Axboebb799ca2008-12-10 15:35:05 +01001558 bs->front_pad = front_pad;
Jens Axboe1b434492008-10-22 20:32:58 +02001559
Jens Axboe392ddc32008-12-23 12:42:54 +01001560 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
Jens Axboebb799ca2008-12-10 15:35:05 +01001561 if (!bs->bio_slab) {
1562 kfree(bs);
1563 return NULL;
1564 }
1565
1566 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 if (!bs->bio_pool)
1568 goto bad;
1569
Jens Axboebb799ca2008-12-10 15:35:05 +01001570 if (bioset_integrity_create(bs, pool_size))
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001571 goto bad;
1572
Jens Axboebb799ca2008-12-10 15:35:05 +01001573 if (!biovec_create_pools(bs, pool_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 return bs;
1575
1576bad:
1577 bioset_free(bs);
1578 return NULL;
1579}
1580
1581static void __init biovec_init_slabs(void)
1582{
1583 int i;
1584
1585 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1586 int size;
1587 struct biovec_slab *bvs = bvec_slabs + i;
1588
1589 size = bvs->nr_vecs * sizeof(struct bio_vec);
1590 bvs->slab = kmem_cache_create(bvs->name, size, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09001591 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 }
1593}
1594
1595static int __init init_bio(void)
1596{
Jens Axboebb799ca2008-12-10 15:35:05 +01001597 bio_slab_max = 2;
1598 bio_slab_nr = 0;
1599 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1600 if (!bio_slabs)
1601 panic("bio: can't allocate bios\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602
Martin K. Petersen7ba1ba12008-06-30 20:04:41 +02001603 bio_integrity_init_slab();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 biovec_init_slabs();
1605
Jens Axboebb799ca2008-12-10 15:35:05 +01001606 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 if (!fs_bio_set)
1608 panic("bio: can't allocate bios\n");
1609
Matthew Dobson0eaae62a2006-03-26 01:37:47 -08001610 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1611 sizeof(struct bio_pair));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (!bio_split_pool)
1613 panic("bio: can't create split pool\n");
1614
1615 return 0;
1616}
1617
1618subsys_initcall(init_bio);
1619
1620EXPORT_SYMBOL(bio_alloc);
Jens Axboe0a0d96b2008-09-11 13:17:37 +02001621EXPORT_SYMBOL(bio_kmalloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001623EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624EXPORT_SYMBOL(bio_endio);
1625EXPORT_SYMBOL(bio_init);
1626EXPORT_SYMBOL(__bio_clone);
1627EXPORT_SYMBOL(bio_clone);
1628EXPORT_SYMBOL(bio_phys_segments);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001630EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631EXPORT_SYMBOL(bio_get_nr_vecs);
Jens Axboe40044ce2008-03-17 21:14:40 +01001632EXPORT_SYMBOL(bio_map_user);
1633EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001634EXPORT_SYMBOL(bio_map_kern);
FUJITA Tomonori68154e92008-04-25 12:47:50 +02001635EXPORT_SYMBOL(bio_copy_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636EXPORT_SYMBOL(bio_pair_release);
1637EXPORT_SYMBOL(bio_split);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638EXPORT_SYMBOL(bio_copy_user);
1639EXPORT_SYMBOL(bio_uncopy_user);
1640EXPORT_SYMBOL(bioset_create);
1641EXPORT_SYMBOL(bioset_free);
1642EXPORT_SYMBOL(bio_alloc_bioset);