Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit bf02c082bf7a464518d45b9c178b8aa83f74dd5d

Authored by Andreas Mohr
Committed by Linus Torvalds
1 parent 49a6cbe1cd
Exists in master and in 7 other branches imx_3.0.35_4.1.0, imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga, smarc-l5.0.0_1.0.0-ga

[PATCH] fs/bio.c: tweaks 

- Calculate a variable in bvec_alloc_bs() only once needed, not earlier
  (bio.o down from 18408 to 18376 Bytes, 32 Bytes saved, probably due to
  data locality improvements).

- Init variable idx to silence a gcc warning which already existed in the
  unmodified original base file (bvec_alloc_bs() handles idx correctly, so
  there's no need for the warning):

	fs/bio.c: In function `bio_alloc_bioset':
	fs/bio.c:169: warning: `idx' may be used uninitialized in this function

Signed-off-by: Andreas Mohr <andi@lisas.de>
Acked-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Showing 1 changed file with 5 additions and 4 deletions Inline Diff

1 /* 1 /*
2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
3 * 3 *
4 * This program is free software; you can redistribute it and/or modify 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 5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation. 6 * published by the Free Software Foundation.
7 * 7 *
8 * This program is distributed in the hope that it will be useful, 8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details. 11 * GNU General Public License for more details.
12 * 12 *
13 * You should have received a copy of the GNU General Public Licens 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 14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- 15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
16 * 16 *
17 */ 17 */
18 #include <linux/mm.h> 18 #include <linux/mm.h>
19 #include <linux/swap.h> 19 #include <linux/swap.h>
20 #include <linux/bio.h> 20 #include <linux/bio.h>
21 #include <linux/blkdev.h> 21 #include <linux/blkdev.h>
22 #include <linux/slab.h> 22 #include <linux/slab.h>
23 #include <linux/init.h> 23 #include <linux/init.h>
24 #include <linux/kernel.h> 24 #include <linux/kernel.h>
25 #include <linux/module.h> 25 #include <linux/module.h>
26 #include <linux/mempool.h> 26 #include <linux/mempool.h>
27 #include <linux/workqueue.h> 27 #include <linux/workqueue.h>
28 #include <linux/blktrace_api.h> 28 #include <linux/blktrace_api.h>
29 #include <scsi/sg.h> /* for struct sg_iovec */ 29 #include <scsi/sg.h> /* for struct sg_iovec */
30 30
31 #define BIO_POOL_SIZE 256 31 #define BIO_POOL_SIZE 256
32 32
33 static kmem_cache_t *bio_slab __read_mostly; 33 static kmem_cache_t *bio_slab __read_mostly;
34 34
35 #define BIOVEC_NR_POOLS 6 35 #define BIOVEC_NR_POOLS 6
36 36
37 /* 37 /*
38 * a small number of entries is fine, not going to be performance critical. 38 * a small number of entries is fine, not going to be performance critical.
39 * basically we just need to survive 39 * basically we just need to survive
40 */ 40 */
41 #define BIO_SPLIT_ENTRIES 8 41 #define BIO_SPLIT_ENTRIES 8
42 mempool_t *bio_split_pool __read_mostly; 42 mempool_t *bio_split_pool __read_mostly;
43 43
44 struct biovec_slab { 44 struct biovec_slab {
45 int nr_vecs; 45 int nr_vecs;
46 char *name; 46 char *name;
47 kmem_cache_t *slab; 47 kmem_cache_t *slab;
48 }; 48 };
49 49
50 /* 50 /*
51 * if you change this list, also change bvec_alloc or things will 51 * if you change this list, also change bvec_alloc or things will
52 * break badly! cannot be bigger than what you can fit into an 52 * break badly! cannot be bigger than what you can fit into an
53 * unsigned short 53 * unsigned short
54 */ 54 */
55 55
56 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } 56 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
57 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { 57 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
58 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), 58 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
59 }; 59 };
60 #undef BV 60 #undef BV
61 61
62 /* 62 /*
63 * bio_set is used to allow other portions of the IO system to 63 * bio_set is used to allow other portions of the IO system to
64 * allocate their own private memory pools for bio and iovec structures. 64 * allocate their own private memory pools for bio and iovec structures.
65 * These memory pools in turn all allocate from the bio_slab 65 * These memory pools in turn all allocate from the bio_slab
66 * and the bvec_slabs[]. 66 * and the bvec_slabs[].
67 */ 67 */
68 struct bio_set { 68 struct bio_set {
69 mempool_t *bio_pool; 69 mempool_t *bio_pool;
70 mempool_t *bvec_pools[BIOVEC_NR_POOLS]; 70 mempool_t *bvec_pools[BIOVEC_NR_POOLS];
71 }; 71 };
72 72
73 /* 73 /*
74 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 74 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
75 * IO code that does not need private memory pools. 75 * IO code that does not need private memory pools.
76 */ 76 */
77 static struct bio_set *fs_bio_set; 77 static struct bio_set *fs_bio_set;
78 78
79 static inline struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) 79 static inline struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
80 { 80 {
81 struct bio_vec *bvl; 81 struct bio_vec *bvl;
82 struct biovec_slab *bp;
83 82
84 /* 83 /*
85 * see comment near bvec_array define! 84 * see comment near bvec_array define!
86 */ 85 */
87 switch (nr) { 86 switch (nr) {
88 case 1 : *idx = 0; break; 87 case 1 : *idx = 0; break;
89 case 2 ... 4: *idx = 1; break; 88 case 2 ... 4: *idx = 1; break;
90 case 5 ... 16: *idx = 2; break; 89 case 5 ... 16: *idx = 2; break;
91 case 17 ... 64: *idx = 3; break; 90 case 17 ... 64: *idx = 3; break;
92 case 65 ... 128: *idx = 4; break; 91 case 65 ... 128: *idx = 4; break;
93 case 129 ... BIO_MAX_PAGES: *idx = 5; break; 92 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
94 default: 93 default:
95 return NULL; 94 return NULL;
96 } 95 }
97 /* 96 /*
98 * idx now points to the pool we want to allocate from 97 * idx now points to the pool we want to allocate from
99 */ 98 */
100 99
101 bp = bvec_slabs + *idx;
102 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); 100 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
103 if (bvl) 101 if (bvl) {
102 struct biovec_slab *bp = bvec_slabs + *idx;
103
104 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec)); 104 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
105 }
105 106
106 return bvl; 107 return bvl;
107 } 108 }
108 109
109 void bio_free(struct bio *bio, struct bio_set *bio_set) 110 void bio_free(struct bio *bio, struct bio_set *bio_set)
110 { 111 {
111 const int pool_idx = BIO_POOL_IDX(bio); 112 const int pool_idx = BIO_POOL_IDX(bio);
112 113
113 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); 114 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
114 115
115 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); 116 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
116 mempool_free(bio, bio_set->bio_pool); 117 mempool_free(bio, bio_set->bio_pool);
117 } 118 }
118 119
119 /* 120 /*
120 * default destructor for a bio allocated with bio_alloc_bioset() 121 * default destructor for a bio allocated with bio_alloc_bioset()
121 */ 122 */
122 static void bio_fs_destructor(struct bio *bio) 123 static void bio_fs_destructor(struct bio *bio)
123 { 124 {
124 bio_free(bio, fs_bio_set); 125 bio_free(bio, fs_bio_set);
125 } 126 }
126 127
127 void bio_init(struct bio *bio) 128 void bio_init(struct bio *bio)
128 { 129 {
129 bio->bi_next = NULL; 130 bio->bi_next = NULL;
130 bio->bi_bdev = NULL; 131 bio->bi_bdev = NULL;
131 bio->bi_flags = 1 << BIO_UPTODATE; 132 bio->bi_flags = 1 << BIO_UPTODATE;
132 bio->bi_rw = 0; 133 bio->bi_rw = 0;
133 bio->bi_vcnt = 0; 134 bio->bi_vcnt = 0;
134 bio->bi_idx = 0; 135 bio->bi_idx = 0;
135 bio->bi_phys_segments = 0; 136 bio->bi_phys_segments = 0;
136 bio->bi_hw_segments = 0; 137 bio->bi_hw_segments = 0;
137 bio->bi_hw_front_size = 0; 138 bio->bi_hw_front_size = 0;
138 bio->bi_hw_back_size = 0; 139 bio->bi_hw_back_size = 0;
139 bio->bi_size = 0; 140 bio->bi_size = 0;
140 bio->bi_max_vecs = 0; 141 bio->bi_max_vecs = 0;
141 bio->bi_end_io = NULL; 142 bio->bi_end_io = NULL;
142 atomic_set(&bio->bi_cnt, 1); 143 atomic_set(&bio->bi_cnt, 1);
143 bio->bi_private = NULL; 144 bio->bi_private = NULL;
144 } 145 }
145 146
146 /** 147 /**
147 * bio_alloc_bioset - allocate a bio for I/O 148 * bio_alloc_bioset - allocate a bio for I/O
148 * @gfp_mask: the GFP_ mask given to the slab allocator 149 * @gfp_mask: the GFP_ mask given to the slab allocator
149 * @nr_iovecs: number of iovecs to pre-allocate 150 * @nr_iovecs: number of iovecs to pre-allocate
150 * @bs: the bio_set to allocate from 151 * @bs: the bio_set to allocate from
151 * 152 *
152 * Description: 153 * Description:
153 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation. 154 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
154 * If %__GFP_WAIT is set then we will block on the internal pool waiting 155 * If %__GFP_WAIT is set then we will block on the internal pool waiting
155 * for a &struct bio to become free. 156 * for a &struct bio to become free.
156 * 157 *
157 * allocate bio and iovecs from the memory pools specified by the 158 * allocate bio and iovecs from the memory pools specified by the
158 * bio_set structure. 159 * bio_set structure.
159 **/ 160 **/
160 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) 161 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
161 { 162 {
162 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask); 163 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
163 164
164 if (likely(bio)) { 165 if (likely(bio)) {
165 struct bio_vec *bvl = NULL; 166 struct bio_vec *bvl = NULL;
166 167
167 bio_init(bio); 168 bio_init(bio);
168 if (likely(nr_iovecs)) { 169 if (likely(nr_iovecs)) {
169 unsigned long idx; 170 unsigned long idx = 0; /* shut up gcc */
170 171
171 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); 172 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
172 if (unlikely(!bvl)) { 173 if (unlikely(!bvl)) {
173 mempool_free(bio, bs->bio_pool); 174 mempool_free(bio, bs->bio_pool);
174 bio = NULL; 175 bio = NULL;
175 goto out; 176 goto out;
176 } 177 }
177 bio->bi_flags |= idx << BIO_POOL_OFFSET; 178 bio->bi_flags |= idx << BIO_POOL_OFFSET;
178 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs; 179 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
179 } 180 }
180 bio->bi_io_vec = bvl; 181 bio->bi_io_vec = bvl;
181 } 182 }
182 out: 183 out:
183 return bio; 184 return bio;
184 } 185 }
185 186
186 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) 187 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
187 { 188 {
188 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); 189 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
189 190
190 if (bio) 191 if (bio)
191 bio->bi_destructor = bio_fs_destructor; 192 bio->bi_destructor = bio_fs_destructor;
192 193
193 return bio; 194 return bio;
194 } 195 }
195 196
196 void zero_fill_bio(struct bio *bio) 197 void zero_fill_bio(struct bio *bio)
197 { 198 {
198 unsigned long flags; 199 unsigned long flags;
199 struct bio_vec *bv; 200 struct bio_vec *bv;
200 int i; 201 int i;
201 202
202 bio_for_each_segment(bv, bio, i) { 203 bio_for_each_segment(bv, bio, i) {
203 char *data = bvec_kmap_irq(bv, &flags); 204 char *data = bvec_kmap_irq(bv, &flags);
204 memset(data, 0, bv->bv_len); 205 memset(data, 0, bv->bv_len);
205 flush_dcache_page(bv->bv_page); 206 flush_dcache_page(bv->bv_page);
206 bvec_kunmap_irq(data, &flags); 207 bvec_kunmap_irq(data, &flags);
207 } 208 }
208 } 209 }
209 EXPORT_SYMBOL(zero_fill_bio); 210 EXPORT_SYMBOL(zero_fill_bio);
210 211
211 /** 212 /**
212 * bio_put - release a reference to a bio 213 * bio_put - release a reference to a bio
213 * @bio: bio to release reference to 214 * @bio: bio to release reference to
214 * 215 *
215 * Description: 216 * Description:
216 * Put a reference to a &struct bio, either one you have gotten with 217 * Put a reference to a &struct bio, either one you have gotten with
217 * bio_alloc or bio_get. The last put of a bio will free it. 218 * bio_alloc or bio_get. The last put of a bio will free it.
218 **/ 219 **/
219 void bio_put(struct bio *bio) 220 void bio_put(struct bio *bio)
220 { 221 {
221 BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); 222 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
222 223
223 /* 224 /*
224 * last put frees it 225 * last put frees it
225 */ 226 */
226 if (atomic_dec_and_test(&bio->bi_cnt)) { 227 if (atomic_dec_and_test(&bio->bi_cnt)) {
227 bio->bi_next = NULL; 228 bio->bi_next = NULL;
228 bio->bi_destructor(bio); 229 bio->bi_destructor(bio);
229 } 230 }
230 } 231 }
231 232
232 inline int bio_phys_segments(request_queue_t *q, struct bio *bio) 233 inline int bio_phys_segments(request_queue_t *q, struct bio *bio)
233 { 234 {
234 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 235 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
235 blk_recount_segments(q, bio); 236 blk_recount_segments(q, bio);
236 237
237 return bio->bi_phys_segments; 238 return bio->bi_phys_segments;
238 } 239 }
239 240
240 inline int bio_hw_segments(request_queue_t *q, struct bio *bio) 241 inline int bio_hw_segments(request_queue_t *q, struct bio *bio)
241 { 242 {
242 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 243 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
243 blk_recount_segments(q, bio); 244 blk_recount_segments(q, bio);
244 245
245 return bio->bi_hw_segments; 246 return bio->bi_hw_segments;
246 } 247 }
247 248
248 /** 249 /**
249 * __bio_clone - clone a bio 250 * __bio_clone - clone a bio
250 * @bio: destination bio 251 * @bio: destination bio
251 * @bio_src: bio to clone 252 * @bio_src: bio to clone
252 * 253 *
253 * Clone a &bio. Caller will own the returned bio, but not 254 * Clone a &bio. Caller will own the returned bio, but not
254 * the actual data it points to. Reference count of returned 255 * the actual data it points to. Reference count of returned
255 * bio will be one. 256 * bio will be one.
256 */ 257 */
257 void __bio_clone(struct bio *bio, struct bio *bio_src) 258 void __bio_clone(struct bio *bio, struct bio *bio_src)
258 { 259 {
259 request_queue_t *q = bdev_get_queue(bio_src->bi_bdev); 260 request_queue_t *q = bdev_get_queue(bio_src->bi_bdev);
260 261
261 memcpy(bio->bi_io_vec, bio_src->bi_io_vec, 262 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
262 bio_src->bi_max_vecs * sizeof(struct bio_vec)); 263 bio_src->bi_max_vecs * sizeof(struct bio_vec));
263 264
264 bio->bi_sector = bio_src->bi_sector; 265 bio->bi_sector = bio_src->bi_sector;
265 bio->bi_bdev = bio_src->bi_bdev; 266 bio->bi_bdev = bio_src->bi_bdev;
266 bio->bi_flags |= 1 << BIO_CLONED; 267 bio->bi_flags |= 1 << BIO_CLONED;
267 bio->bi_rw = bio_src->bi_rw; 268 bio->bi_rw = bio_src->bi_rw;
268 bio->bi_vcnt = bio_src->bi_vcnt; 269 bio->bi_vcnt = bio_src->bi_vcnt;
269 bio->bi_size = bio_src->bi_size; 270 bio->bi_size = bio_src->bi_size;
270 bio->bi_idx = bio_src->bi_idx; 271 bio->bi_idx = bio_src->bi_idx;
271 bio_phys_segments(q, bio); 272 bio_phys_segments(q, bio);
272 bio_hw_segments(q, bio); 273 bio_hw_segments(q, bio);
273 } 274 }
274 275
275 /** 276 /**
276 * bio_clone - clone a bio 277 * bio_clone - clone a bio
277 * @bio: bio to clone 278 * @bio: bio to clone
278 * @gfp_mask: allocation priority 279 * @gfp_mask: allocation priority
279 * 280 *
280 * Like __bio_clone, only also allocates the returned bio 281 * Like __bio_clone, only also allocates the returned bio
281 */ 282 */
282 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) 283 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
283 { 284 {
284 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); 285 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
285 286
286 if (b) { 287 if (b) {
287 b->bi_destructor = bio_fs_destructor; 288 b->bi_destructor = bio_fs_destructor;
288 __bio_clone(b, bio); 289 __bio_clone(b, bio);
289 } 290 }
290 291
291 return b; 292 return b;
292 } 293 }
293 294
294 /** 295 /**
295 * bio_get_nr_vecs - return approx number of vecs 296 * bio_get_nr_vecs - return approx number of vecs
296 * @bdev: I/O target 297 * @bdev: I/O target
297 * 298 *
298 * Return the approximate number of pages we can send to this target. 299 * Return the approximate number of pages we can send to this target.
299 * There's no guarantee that you will be able to fit this number of pages 300 * There's no guarantee that you will be able to fit this number of pages
300 * into a bio, it does not account for dynamic restrictions that vary 301 * into a bio, it does not account for dynamic restrictions that vary
301 * on offset. 302 * on offset.
302 */ 303 */
303 int bio_get_nr_vecs(struct block_device *bdev) 304 int bio_get_nr_vecs(struct block_device *bdev)
304 { 305 {
305 request_queue_t *q = bdev_get_queue(bdev); 306 request_queue_t *q = bdev_get_queue(bdev);
306 int nr_pages; 307 int nr_pages;
307 308
308 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; 309 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
309 if (nr_pages > q->max_phys_segments) 310 if (nr_pages > q->max_phys_segments)
310 nr_pages = q->max_phys_segments; 311 nr_pages = q->max_phys_segments;
311 if (nr_pages > q->max_hw_segments) 312 if (nr_pages > q->max_hw_segments)
312 nr_pages = q->max_hw_segments; 313 nr_pages = q->max_hw_segments;
313 314
314 return nr_pages; 315 return nr_pages;
315 } 316 }
316 317
317 static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page 318 static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page
318 *page, unsigned int len, unsigned int offset, 319 *page, unsigned int len, unsigned int offset,
319 unsigned short max_sectors) 320 unsigned short max_sectors)
320 { 321 {
321 int retried_segments = 0; 322 int retried_segments = 0;
322 struct bio_vec *bvec; 323 struct bio_vec *bvec;
323 324
324 /* 325 /*
325 * cloned bio must not modify vec list 326 * cloned bio must not modify vec list
326 */ 327 */
327 if (unlikely(bio_flagged(bio, BIO_CLONED))) 328 if (unlikely(bio_flagged(bio, BIO_CLONED)))
328 return 0; 329 return 0;
329 330
330 if (((bio->bi_size + len) >> 9) > max_sectors) 331 if (((bio->bi_size + len) >> 9) > max_sectors)
331 return 0; 332 return 0;
332 333
333 /* 334 /*
334 * For filesystems with a blocksize smaller than the pagesize 335 * For filesystems with a blocksize smaller than the pagesize
335 * we will often be called with the same page as last time and 336 * we will often be called with the same page as last time and
336 * a consecutive offset. Optimize this special case. 337 * a consecutive offset. Optimize this special case.
337 */ 338 */
338 if (bio->bi_vcnt > 0) { 339 if (bio->bi_vcnt > 0) {
339 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; 340 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
340 341
341 if (page == prev->bv_page && 342 if (page == prev->bv_page &&
342 offset == prev->bv_offset + prev->bv_len) { 343 offset == prev->bv_offset + prev->bv_len) {
343 prev->bv_len += len; 344 prev->bv_len += len;
344 if (q->merge_bvec_fn && 345 if (q->merge_bvec_fn &&
345 q->merge_bvec_fn(q, bio, prev) < len) { 346 q->merge_bvec_fn(q, bio, prev) < len) {
346 prev->bv_len -= len; 347 prev->bv_len -= len;
347 return 0; 348 return 0;
348 } 349 }
349 350
350 goto done; 351 goto done;
351 } 352 }
352 } 353 }
353 354
354 if (bio->bi_vcnt >= bio->bi_max_vecs) 355 if (bio->bi_vcnt >= bio->bi_max_vecs)
355 return 0; 356 return 0;
356 357
357 /* 358 /*
358 * we might lose a segment or two here, but rather that than 359 * we might lose a segment or two here, but rather that than
359 * make this too complex. 360 * make this too complex.
360 */ 361 */
361 362
362 while (bio->bi_phys_segments >= q->max_phys_segments 363 while (bio->bi_phys_segments >= q->max_phys_segments
363 || bio->bi_hw_segments >= q->max_hw_segments 364 || bio->bi_hw_segments >= q->max_hw_segments
364 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) { 365 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
365 366
366 if (retried_segments) 367 if (retried_segments)
367 return 0; 368 return 0;
368 369
369 retried_segments = 1; 370 retried_segments = 1;
370 blk_recount_segments(q, bio); 371 blk_recount_segments(q, bio);
371 } 372 }
372 373
373 /* 374 /*
374 * setup the new entry, we might clear it again later if we 375 * setup the new entry, we might clear it again later if we
375 * cannot add the page 376 * cannot add the page
376 */ 377 */
377 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 378 bvec = &bio->bi_io_vec[bio->bi_vcnt];
378 bvec->bv_page = page; 379 bvec->bv_page = page;
379 bvec->bv_len = len; 380 bvec->bv_len = len;
380 bvec->bv_offset = offset; 381 bvec->bv_offset = offset;
381 382
382 /* 383 /*
383 * if queue has other restrictions (eg varying max sector size 384 * if queue has other restrictions (eg varying max sector size
384 * depending on offset), it can specify a merge_bvec_fn in the 385 * depending on offset), it can specify a merge_bvec_fn in the
385 * queue to get further control 386 * queue to get further control
386 */ 387 */
387 if (q->merge_bvec_fn) { 388 if (q->merge_bvec_fn) {
388 /* 389 /*
389 * merge_bvec_fn() returns number of bytes it can accept 390 * merge_bvec_fn() returns number of bytes it can accept
390 * at this offset 391 * at this offset
391 */ 392 */
392 if (q->merge_bvec_fn(q, bio, bvec) < len) { 393 if (q->merge_bvec_fn(q, bio, bvec) < len) {
393 bvec->bv_page = NULL; 394 bvec->bv_page = NULL;
394 bvec->bv_len = 0; 395 bvec->bv_len = 0;
395 bvec->bv_offset = 0; 396 bvec->bv_offset = 0;
396 return 0; 397 return 0;
397 } 398 }
398 } 399 }
399 400
400 /* If we may be able to merge these biovecs, force a recount */ 401 /* If we may be able to merge these biovecs, force a recount */
401 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) || 402 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
402 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec))) 403 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
403 bio->bi_flags &= ~(1 << BIO_SEG_VALID); 404 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
404 405
405 bio->bi_vcnt++; 406 bio->bi_vcnt++;
406 bio->bi_phys_segments++; 407 bio->bi_phys_segments++;
407 bio->bi_hw_segments++; 408 bio->bi_hw_segments++;
408 done: 409 done:
409 bio->bi_size += len; 410 bio->bi_size += len;
410 return len; 411 return len;
411 } 412 }
412 413
413 /** 414 /**
414 * bio_add_pc_page - attempt to add page to bio 415 * bio_add_pc_page - attempt to add page to bio
415 * @q: the target queue 416 * @q: the target queue
416 * @bio: destination bio 417 * @bio: destination bio
417 * @page: page to add 418 * @page: page to add
418 * @len: vec entry length 419 * @len: vec entry length
419 * @offset: vec entry offset 420 * @offset: vec entry offset
420 * 421 *
421 * Attempt to add a page to the bio_vec maplist. This can fail for a 422 * Attempt to add a page to the bio_vec maplist. This can fail for a
422 * number of reasons, such as the bio being full or target block 423 * number of reasons, such as the bio being full or target block
423 * device limitations. The target block device must allow bio's 424 * device limitations. The target block device must allow bio's
424 * smaller than PAGE_SIZE, so it is always possible to add a single 425 * smaller than PAGE_SIZE, so it is always possible to add a single
425 * page to an empty bio. This should only be used by REQ_PC bios. 426 * page to an empty bio. This should only be used by REQ_PC bios.
426 */ 427 */
427 int bio_add_pc_page(request_queue_t *q, struct bio *bio, struct page *page, 428 int bio_add_pc_page(request_queue_t *q, struct bio *bio, struct page *page,
428 unsigned int len, unsigned int offset) 429 unsigned int len, unsigned int offset)
429 { 430 {
430 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); 431 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
431 } 432 }
432 433
433 /** 434 /**
434 * bio_add_page - attempt to add page to bio 435 * bio_add_page - attempt to add page to bio
435 * @bio: destination bio 436 * @bio: destination bio
436 * @page: page to add 437 * @page: page to add
437 * @len: vec entry length 438 * @len: vec entry length
438 * @offset: vec entry offset 439 * @offset: vec entry offset
439 * 440 *
440 * Attempt to add a page to the bio_vec maplist. This can fail for a 441 * Attempt to add a page to the bio_vec maplist. This can fail for a
441 * number of reasons, such as the bio being full or target block 442 * number of reasons, such as the bio being full or target block
442 * device limitations. The target block device must allow bio's 443 * device limitations. The target block device must allow bio's
443 * smaller than PAGE_SIZE, so it is always possible to add a single 444 * smaller than PAGE_SIZE, so it is always possible to add a single
444 * page to an empty bio. 445 * page to an empty bio.
445 */ 446 */
446 int bio_add_page(struct bio *bio, struct page *page, unsigned int len, 447 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
447 unsigned int offset) 448 unsigned int offset)
448 { 449 {
449 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 450 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
450 return __bio_add_page(q, bio, page, len, offset, q->max_sectors); 451 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
451 } 452 }
452 453
453 struct bio_map_data { 454 struct bio_map_data {
454 struct bio_vec *iovecs; 455 struct bio_vec *iovecs;
455 void __user *userptr; 456 void __user *userptr;
456 }; 457 };
457 458
458 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio) 459 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
459 { 460 {
460 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); 461 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
461 bio->bi_private = bmd; 462 bio->bi_private = bmd;
462 } 463 }
463 464
464 static void bio_free_map_data(struct bio_map_data *bmd) 465 static void bio_free_map_data(struct bio_map_data *bmd)
465 { 466 {
466 kfree(bmd->iovecs); 467 kfree(bmd->iovecs);
467 kfree(bmd); 468 kfree(bmd);
468 } 469 }
469 470
470 static struct bio_map_data *bio_alloc_map_data(int nr_segs) 471 static struct bio_map_data *bio_alloc_map_data(int nr_segs)
471 { 472 {
472 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL); 473 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
473 474
474 if (!bmd) 475 if (!bmd)
475 return NULL; 476 return NULL;
476 477
477 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL); 478 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
478 if (bmd->iovecs) 479 if (bmd->iovecs)
479 return bmd; 480 return bmd;
480 481
481 kfree(bmd); 482 kfree(bmd);
482 return NULL; 483 return NULL;
483 } 484 }
484 485
485 /** 486 /**
486 * bio_uncopy_user - finish previously mapped bio 487 * bio_uncopy_user - finish previously mapped bio
487 * @bio: bio being terminated 488 * @bio: bio being terminated
488 * 489 *
489 * Free pages allocated from bio_copy_user() and write back data 490 * Free pages allocated from bio_copy_user() and write back data
490 * to user space in case of a read. 491 * to user space in case of a read.
491 */ 492 */
492 int bio_uncopy_user(struct bio *bio) 493 int bio_uncopy_user(struct bio *bio)
493 { 494 {
494 struct bio_map_data *bmd = bio->bi_private; 495 struct bio_map_data *bmd = bio->bi_private;
495 const int read = bio_data_dir(bio) == READ; 496 const int read = bio_data_dir(bio) == READ;
496 struct bio_vec *bvec; 497 struct bio_vec *bvec;
497 int i, ret = 0; 498 int i, ret = 0;
498 499
499 __bio_for_each_segment(bvec, bio, i, 0) { 500 __bio_for_each_segment(bvec, bio, i, 0) {
500 char *addr = page_address(bvec->bv_page); 501 char *addr = page_address(bvec->bv_page);
501 unsigned int len = bmd->iovecs[i].bv_len; 502 unsigned int len = bmd->iovecs[i].bv_len;
502 503
503 if (read && !ret && copy_to_user(bmd->userptr, addr, len)) 504 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
504 ret = -EFAULT; 505 ret = -EFAULT;
505 506
506 __free_page(bvec->bv_page); 507 __free_page(bvec->bv_page);
507 bmd->userptr += len; 508 bmd->userptr += len;
508 } 509 }
509 bio_free_map_data(bmd); 510 bio_free_map_data(bmd);
510 bio_put(bio); 511 bio_put(bio);
511 return ret; 512 return ret;
512 } 513 }
513 514
514 /** 515 /**
515 * bio_copy_user - copy user data to bio 516 * bio_copy_user - copy user data to bio
516 * @q: destination block queue 517 * @q: destination block queue
517 * @uaddr: start of user address 518 * @uaddr: start of user address
518 * @len: length in bytes 519 * @len: length in bytes
519 * @write_to_vm: bool indicating writing to pages or not 520 * @write_to_vm: bool indicating writing to pages or not
520 * 521 *
521 * Prepares and returns a bio for indirect user io, bouncing data 522 * Prepares and returns a bio for indirect user io, bouncing data
522 * to/from kernel pages as necessary. Must be paired with 523 * to/from kernel pages as necessary. Must be paired with
523 * call bio_uncopy_user() on io completion. 524 * call bio_uncopy_user() on io completion.
524 */ 525 */
525 struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr, 526 struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
526 unsigned int len, int write_to_vm) 527 unsigned int len, int write_to_vm)
527 { 528 {
528 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 529 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
529 unsigned long start = uaddr >> PAGE_SHIFT; 530 unsigned long start = uaddr >> PAGE_SHIFT;
530 struct bio_map_data *bmd; 531 struct bio_map_data *bmd;
531 struct bio_vec *bvec; 532 struct bio_vec *bvec;
532 struct page *page; 533 struct page *page;
533 struct bio *bio; 534 struct bio *bio;
534 int i, ret; 535 int i, ret;
535 536
536 bmd = bio_alloc_map_data(end - start); 537 bmd = bio_alloc_map_data(end - start);
537 if (!bmd) 538 if (!bmd)
538 return ERR_PTR(-ENOMEM); 539 return ERR_PTR(-ENOMEM);
539 540
540 bmd->userptr = (void __user *) uaddr; 541 bmd->userptr = (void __user *) uaddr;
541 542
542 ret = -ENOMEM; 543 ret = -ENOMEM;
543 bio = bio_alloc(GFP_KERNEL, end - start); 544 bio = bio_alloc(GFP_KERNEL, end - start);
544 if (!bio) 545 if (!bio)
545 goto out_bmd; 546 goto out_bmd;
546 547
547 bio->bi_rw |= (!write_to_vm << BIO_RW); 548 bio->bi_rw |= (!write_to_vm << BIO_RW);
548 549
549 ret = 0; 550 ret = 0;
550 while (len) { 551 while (len) {
551 unsigned int bytes = PAGE_SIZE; 552 unsigned int bytes = PAGE_SIZE;
552 553
553 if (bytes > len) 554 if (bytes > len)
554 bytes = len; 555 bytes = len;
555 556
556 page = alloc_page(q->bounce_gfp | GFP_KERNEL); 557 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
557 if (!page) { 558 if (!page) {
558 ret = -ENOMEM; 559 ret = -ENOMEM;
559 break; 560 break;
560 } 561 }
561 562
562 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) { 563 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
563 ret = -EINVAL; 564 ret = -EINVAL;
564 break; 565 break;
565 } 566 }
566 567
567 len -= bytes; 568 len -= bytes;
568 } 569 }
569 570
570 if (ret) 571 if (ret)
571 goto cleanup; 572 goto cleanup;
572 573
573 /* 574 /*
574 * success 575 * success
575 */ 576 */
576 if (!write_to_vm) { 577 if (!write_to_vm) {
577 char __user *p = (char __user *) uaddr; 578 char __user *p = (char __user *) uaddr;
578 579
579 /* 580 /*
580 * for a write, copy in data to kernel pages 581 * for a write, copy in data to kernel pages
581 */ 582 */
582 ret = -EFAULT; 583 ret = -EFAULT;
583 bio_for_each_segment(bvec, bio, i) { 584 bio_for_each_segment(bvec, bio, i) {
584 char *addr = page_address(bvec->bv_page); 585 char *addr = page_address(bvec->bv_page);
585 586
586 if (copy_from_user(addr, p, bvec->bv_len)) 587 if (copy_from_user(addr, p, bvec->bv_len))
587 goto cleanup; 588 goto cleanup;
588 p += bvec->bv_len; 589 p += bvec->bv_len;
589 } 590 }
590 } 591 }
591 592
592 bio_set_map_data(bmd, bio); 593 bio_set_map_data(bmd, bio);
593 return bio; 594 return bio;
594 cleanup: 595 cleanup:
595 bio_for_each_segment(bvec, bio, i) 596 bio_for_each_segment(bvec, bio, i)
596 __free_page(bvec->bv_page); 597 __free_page(bvec->bv_page);
597 598
598 bio_put(bio); 599 bio_put(bio);
599 out_bmd: 600 out_bmd:
600 bio_free_map_data(bmd); 601 bio_free_map_data(bmd);
601 return ERR_PTR(ret); 602 return ERR_PTR(ret);
602 } 603 }
603 604
604 static struct bio *__bio_map_user_iov(request_queue_t *q, 605 static struct bio *__bio_map_user_iov(request_queue_t *q,
605 struct block_device *bdev, 606 struct block_device *bdev,
606 struct sg_iovec *iov, int iov_count, 607 struct sg_iovec *iov, int iov_count,
607 int write_to_vm) 608 int write_to_vm)
608 { 609 {
609 int i, j; 610 int i, j;
610 int nr_pages = 0; 611 int nr_pages = 0;
611 struct page **pages; 612 struct page **pages;
612 struct bio *bio; 613 struct bio *bio;
613 int cur_page = 0; 614 int cur_page = 0;
614 int ret, offset; 615 int ret, offset;
615 616
616 for (i = 0; i < iov_count; i++) { 617 for (i = 0; i < iov_count; i++) {
617 unsigned long uaddr = (unsigned long)iov[i].iov_base; 618 unsigned long uaddr = (unsigned long)iov[i].iov_base;
618 unsigned long len = iov[i].iov_len; 619 unsigned long len = iov[i].iov_len;
619 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 620 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
620 unsigned long start = uaddr >> PAGE_SHIFT; 621 unsigned long start = uaddr >> PAGE_SHIFT;
621 622
622 nr_pages += end - start; 623 nr_pages += end - start;
623 /* 624 /*
624 * transfer and buffer must be aligned to at least hardsector 625 * transfer and buffer must be aligned to at least hardsector
625 * size for now, in the future we can relax this restriction 626 * size for now, in the future we can relax this restriction
626 */ 627 */
627 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q))) 628 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
628 return ERR_PTR(-EINVAL); 629 return ERR_PTR(-EINVAL);
629 } 630 }
630 631
631 if (!nr_pages) 632 if (!nr_pages)
632 return ERR_PTR(-EINVAL); 633 return ERR_PTR(-EINVAL);
633 634
634 bio = bio_alloc(GFP_KERNEL, nr_pages); 635 bio = bio_alloc(GFP_KERNEL, nr_pages);
635 if (!bio) 636 if (!bio)
636 return ERR_PTR(-ENOMEM); 637 return ERR_PTR(-ENOMEM);
637 638
638 ret = -ENOMEM; 639 ret = -ENOMEM;
639 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); 640 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
640 if (!pages) 641 if (!pages)
641 goto out; 642 goto out;
642 643
643 for (i = 0; i < iov_count; i++) { 644 for (i = 0; i < iov_count; i++) {
644 unsigned long uaddr = (unsigned long)iov[i].iov_base; 645 unsigned long uaddr = (unsigned long)iov[i].iov_base;
645 unsigned long len = iov[i].iov_len; 646 unsigned long len = iov[i].iov_len;
646 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 647 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
647 unsigned long start = uaddr >> PAGE_SHIFT; 648 unsigned long start = uaddr >> PAGE_SHIFT;
648 const int local_nr_pages = end - start; 649 const int local_nr_pages = end - start;
649 const int page_limit = cur_page + local_nr_pages; 650 const int page_limit = cur_page + local_nr_pages;
650 651
651 down_read(&current->mm->mmap_sem); 652 down_read(&current->mm->mmap_sem);
652 ret = get_user_pages(current, current->mm, uaddr, 653 ret = get_user_pages(current, current->mm, uaddr,
653 local_nr_pages, 654 local_nr_pages,
654 write_to_vm, 0, &pages[cur_page], NULL); 655 write_to_vm, 0, &pages[cur_page], NULL);
655 up_read(&current->mm->mmap_sem); 656 up_read(&current->mm->mmap_sem);
656 657
657 if (ret < local_nr_pages) { 658 if (ret < local_nr_pages) {
658 ret = -EFAULT; 659 ret = -EFAULT;
659 goto out_unmap; 660 goto out_unmap;
660 } 661 }
661 662
662 offset = uaddr & ~PAGE_MASK; 663 offset = uaddr & ~PAGE_MASK;
663 for (j = cur_page; j < page_limit; j++) { 664 for (j = cur_page; j < page_limit; j++) {
664 unsigned int bytes = PAGE_SIZE - offset; 665 unsigned int bytes = PAGE_SIZE - offset;
665 666
666 if (len <= 0) 667 if (len <= 0)
667 break; 668 break;
668 669
669 if (bytes > len) 670 if (bytes > len)
670 bytes = len; 671 bytes = len;
671 672
672 /* 673 /*
673 * sorry... 674 * sorry...
674 */ 675 */
675 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < 676 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
676 bytes) 677 bytes)
677 break; 678 break;
678 679
679 len -= bytes; 680 len -= bytes;
680 offset = 0; 681 offset = 0;
681 } 682 }
682 683
683 cur_page = j; 684 cur_page = j;
684 /* 685 /*
685 * release the pages we didn't map into the bio, if any 686 * release the pages we didn't map into the bio, if any
686 */ 687 */
687 while (j < page_limit) 688 while (j < page_limit)
688 page_cache_release(pages[j++]); 689 page_cache_release(pages[j++]);
689 } 690 }
690 691
691 kfree(pages); 692 kfree(pages);
692 693
693 /* 694 /*
694 * set data direction, and check if mapped pages need bouncing 695 * set data direction, and check if mapped pages need bouncing
695 */ 696 */
696 if (!write_to_vm) 697 if (!write_to_vm)
697 bio->bi_rw |= (1 << BIO_RW); 698 bio->bi_rw |= (1 << BIO_RW);
698 699
699 bio->bi_bdev = bdev; 700 bio->bi_bdev = bdev;
700 bio->bi_flags |= (1 << BIO_USER_MAPPED); 701 bio->bi_flags |= (1 << BIO_USER_MAPPED);
701 return bio; 702 return bio;
702 703
703 out_unmap: 704 out_unmap:
704 for (i = 0; i < nr_pages; i++) { 705 for (i = 0; i < nr_pages; i++) {
705 if(!pages[i]) 706 if(!pages[i])
706 break; 707 break;
707 page_cache_release(pages[i]); 708 page_cache_release(pages[i]);
708 } 709 }
709 out: 710 out:
710 kfree(pages); 711 kfree(pages);
711 bio_put(bio); 712 bio_put(bio);
712 return ERR_PTR(ret); 713 return ERR_PTR(ret);
713 } 714 }
714 715
715 /** 716 /**
716 * bio_map_user - map user address into bio 717 * bio_map_user - map user address into bio
717 * @q: the request_queue_t for the bio 718 * @q: the request_queue_t for the bio
718 * @bdev: destination block device 719 * @bdev: destination block device
719 * @uaddr: start of user address 720 * @uaddr: start of user address
720 * @len: length in bytes 721 * @len: length in bytes
721 * @write_to_vm: bool indicating writing to pages or not 722 * @write_to_vm: bool indicating writing to pages or not
722 * 723 *
723 * Map the user space address into a bio suitable for io to a block 724 * Map the user space address into a bio suitable for io to a block
724 * device. Returns an error pointer in case of error. 725 * device. Returns an error pointer in case of error.
725 */ 726 */
726 struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev, 727 struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
727 unsigned long uaddr, unsigned int len, int write_to_vm) 728 unsigned long uaddr, unsigned int len, int write_to_vm)
728 { 729 {
729 struct sg_iovec iov; 730 struct sg_iovec iov;
730 731
731 iov.iov_base = (void __user *)uaddr; 732 iov.iov_base = (void __user *)uaddr;
732 iov.iov_len = len; 733 iov.iov_len = len;
733 734
734 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm); 735 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
735 } 736 }
736 737
737 /** 738 /**
738 * bio_map_user_iov - map user sg_iovec table into bio 739 * bio_map_user_iov - map user sg_iovec table into bio
739 * @q: the request_queue_t for the bio 740 * @q: the request_queue_t for the bio
740 * @bdev: destination block device 741 * @bdev: destination block device
741 * @iov: the iovec. 742 * @iov: the iovec.
742 * @iov_count: number of elements in the iovec 743 * @iov_count: number of elements in the iovec
743 * @write_to_vm: bool indicating writing to pages or not 744 * @write_to_vm: bool indicating writing to pages or not
744 * 745 *
745 * Map the user space address into a bio suitable for io to a block 746 * Map the user space address into a bio suitable for io to a block
746 * device. Returns an error pointer in case of error. 747 * device. Returns an error pointer in case of error.
747 */ 748 */
748 struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev, 749 struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
749 struct sg_iovec *iov, int iov_count, 750 struct sg_iovec *iov, int iov_count,
750 int write_to_vm) 751 int write_to_vm)
751 { 752 {
752 struct bio *bio; 753 struct bio *bio;
753 int len = 0, i; 754 int len = 0, i;
754 755
755 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm); 756 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
756 757
757 if (IS_ERR(bio)) 758 if (IS_ERR(bio))
758 return bio; 759 return bio;
759 760
760 /* 761 /*
761 * subtle -- if __bio_map_user() ended up bouncing a bio, 762 * subtle -- if __bio_map_user() ended up bouncing a bio,
762 * it would normally disappear when its bi_end_io is run. 763 * it would normally disappear when its bi_end_io is run.
763 * however, we need it for the unmap, so grab an extra 764 * however, we need it for the unmap, so grab an extra
764 * reference to it 765 * reference to it
765 */ 766 */
766 bio_get(bio); 767 bio_get(bio);
767 768
768 for (i = 0; i < iov_count; i++) 769 for (i = 0; i < iov_count; i++)
769 len += iov[i].iov_len; 770 len += iov[i].iov_len;
770 771
771 if (bio->bi_size == len) 772 if (bio->bi_size == len)
772 return bio; 773 return bio;
773 774
774 /* 775 /*
775 * don't support partial mappings 776 * don't support partial mappings
776 */ 777 */
777 bio_endio(bio, bio->bi_size, 0); 778 bio_endio(bio, bio->bi_size, 0);
778 bio_unmap_user(bio); 779 bio_unmap_user(bio);
779 return ERR_PTR(-EINVAL); 780 return ERR_PTR(-EINVAL);
780 } 781 }
781 782
782 static void __bio_unmap_user(struct bio *bio) 783 static void __bio_unmap_user(struct bio *bio)
783 { 784 {
784 struct bio_vec *bvec; 785 struct bio_vec *bvec;
785 int i; 786 int i;
786 787
787 /* 788 /*
788 * make sure we dirty pages we wrote to 789 * make sure we dirty pages we wrote to
789 */ 790 */
790 __bio_for_each_segment(bvec, bio, i, 0) { 791 __bio_for_each_segment(bvec, bio, i, 0) {
791 if (bio_data_dir(bio) == READ) 792 if (bio_data_dir(bio) == READ)
792 set_page_dirty_lock(bvec->bv_page); 793 set_page_dirty_lock(bvec->bv_page);
793 794
794 page_cache_release(bvec->bv_page); 795 page_cache_release(bvec->bv_page);
795 } 796 }
796 797
797 bio_put(bio); 798 bio_put(bio);
798 } 799 }
799 800
800 /** 801 /**
801 * bio_unmap_user - unmap a bio 802 * bio_unmap_user - unmap a bio
802 * @bio: the bio being unmapped 803 * @bio: the bio being unmapped
803 * 804 *
804 * Unmap a bio previously mapped by bio_map_user(). Must be called with 805 * Unmap a bio previously mapped by bio_map_user(). Must be called with
805 * a process context. 806 * a process context.
806 * 807 *
807 * bio_unmap_user() may sleep. 808 * bio_unmap_user() may sleep.
808 */ 809 */
809 void bio_unmap_user(struct bio *bio) 810 void bio_unmap_user(struct bio *bio)
810 { 811 {
811 __bio_unmap_user(bio); 812 __bio_unmap_user(bio);
812 bio_put(bio); 813 bio_put(bio);
813 } 814 }
814 815
815 static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err) 816 static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err)
816 { 817 {
817 if (bio->bi_size) 818 if (bio->bi_size)
818 return 1; 819 return 1;
819 820
820 bio_put(bio); 821 bio_put(bio);
821 return 0; 822 return 0;
822 } 823 }
823 824
824 825
825 static struct bio *__bio_map_kern(request_queue_t *q, void *data, 826 static struct bio *__bio_map_kern(request_queue_t *q, void *data,
826 unsigned int len, gfp_t gfp_mask) 827 unsigned int len, gfp_t gfp_mask)
827 { 828 {
828 unsigned long kaddr = (unsigned long)data; 829 unsigned long kaddr = (unsigned long)data;
829 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 830 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
830 unsigned long start = kaddr >> PAGE_SHIFT; 831 unsigned long start = kaddr >> PAGE_SHIFT;
831 const int nr_pages = end - start; 832 const int nr_pages = end - start;
832 int offset, i; 833 int offset, i;
833 struct bio *bio; 834 struct bio *bio;
834 835
835 bio = bio_alloc(gfp_mask, nr_pages); 836 bio = bio_alloc(gfp_mask, nr_pages);
836 if (!bio) 837 if (!bio)
837 return ERR_PTR(-ENOMEM); 838 return ERR_PTR(-ENOMEM);
838 839
839 offset = offset_in_page(kaddr); 840 offset = offset_in_page(kaddr);
840 for (i = 0; i < nr_pages; i++) { 841 for (i = 0; i < nr_pages; i++) {
841 unsigned int bytes = PAGE_SIZE - offset; 842 unsigned int bytes = PAGE_SIZE - offset;
842 843
843 if (len <= 0) 844 if (len <= 0)
844 break; 845 break;
845 846
846 if (bytes > len) 847 if (bytes > len)
847 bytes = len; 848 bytes = len;
848 849
849 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 850 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
850 offset) < bytes) 851 offset) < bytes)
851 break; 852 break;
852 853
853 data += bytes; 854 data += bytes;
854 len -= bytes; 855 len -= bytes;
855 offset = 0; 856 offset = 0;
856 } 857 }
857 858
858 bio->bi_end_io = bio_map_kern_endio; 859 bio->bi_end_io = bio_map_kern_endio;
859 return bio; 860 return bio;
860 } 861 }
861 862
862 /** 863 /**
863 * bio_map_kern - map kernel address into bio 864 * bio_map_kern - map kernel address into bio
864 * @q: the request_queue_t for the bio 865 * @q: the request_queue_t for the bio
865 * @data: pointer to buffer to map 866 * @data: pointer to buffer to map
866 * @len: length in bytes 867 * @len: length in bytes
867 * @gfp_mask: allocation flags for bio allocation 868 * @gfp_mask: allocation flags for bio allocation
868 * 869 *
869 * Map the kernel address into a bio suitable for io to a block 870 * Map the kernel address into a bio suitable for io to a block
870 * device. Returns an error pointer in case of error. 871 * device. Returns an error pointer in case of error.
871 */ 872 */
872 struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len, 873 struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len,
873 gfp_t gfp_mask) 874 gfp_t gfp_mask)
874 { 875 {
875 struct bio *bio; 876 struct bio *bio;
876 877
877 bio = __bio_map_kern(q, data, len, gfp_mask); 878 bio = __bio_map_kern(q, data, len, gfp_mask);
878 if (IS_ERR(bio)) 879 if (IS_ERR(bio))
879 return bio; 880 return bio;
880 881
881 if (bio->bi_size == len) 882 if (bio->bi_size == len)
882 return bio; 883 return bio;
883 884
884 /* 885 /*
885 * Don't support partial mappings. 886 * Don't support partial mappings.
886 */ 887 */
887 bio_put(bio); 888 bio_put(bio);
888 return ERR_PTR(-EINVAL); 889 return ERR_PTR(-EINVAL);
889 } 890 }
890 891
891 /* 892 /*
892 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 893 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
893 * for performing direct-IO in BIOs. 894 * for performing direct-IO in BIOs.
894 * 895 *
895 * The problem is that we cannot run set_page_dirty() from interrupt context 896 * The problem is that we cannot run set_page_dirty() from interrupt context
896 * because the required locks are not interrupt-safe. So what we can do is to 897 * because the required locks are not interrupt-safe. So what we can do is to
897 * mark the pages dirty _before_ performing IO. And in interrupt context, 898 * mark the pages dirty _before_ performing IO. And in interrupt context,
898 * check that the pages are still dirty. If so, fine. If not, redirty them 899 * check that the pages are still dirty. If so, fine. If not, redirty them
899 * in process context. 900 * in process context.
900 * 901 *
901 * We special-case compound pages here: normally this means reads into hugetlb 902 * We special-case compound pages here: normally this means reads into hugetlb
902 * pages. The logic in here doesn't really work right for compound pages 903 * pages. The logic in here doesn't really work right for compound pages
903 * because the VM does not uniformly chase down the head page in all cases. 904 * because the VM does not uniformly chase down the head page in all cases.
904 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 905 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
905 * handle them at all. So we skip compound pages here at an early stage. 906 * handle them at all. So we skip compound pages here at an early stage.
906 * 907 *
907 * Note that this code is very hard to test under normal circumstances because 908 * Note that this code is very hard to test under normal circumstances because
908 * direct-io pins the pages with get_user_pages(). This makes 909 * direct-io pins the pages with get_user_pages(). This makes
909 * is_page_cache_freeable return false, and the VM will not clean the pages. 910 * is_page_cache_freeable return false, and the VM will not clean the pages.
910 * But other code (eg, pdflush) could clean the pages if they are mapped 911 * But other code (eg, pdflush) could clean the pages if they are mapped
911 * pagecache. 912 * pagecache.
912 * 913 *
913 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 914 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
914 * deferred bio dirtying paths. 915 * deferred bio dirtying paths.
915 */ 916 */
916 917
917 /* 918 /*
918 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 919 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
919 */ 920 */
920 void bio_set_pages_dirty(struct bio *bio) 921 void bio_set_pages_dirty(struct bio *bio)
921 { 922 {
922 struct bio_vec *bvec = bio->bi_io_vec; 923 struct bio_vec *bvec = bio->bi_io_vec;
923 int i; 924 int i;
924 925
925 for (i = 0; i < bio->bi_vcnt; i++) { 926 for (i = 0; i < bio->bi_vcnt; i++) {
926 struct page *page = bvec[i].bv_page; 927 struct page *page = bvec[i].bv_page;
927 928
928 if (page && !PageCompound(page)) 929 if (page && !PageCompound(page))
929 set_page_dirty_lock(page); 930 set_page_dirty_lock(page);
930 } 931 }
931 } 932 }
932 933
933 static void bio_release_pages(struct bio *bio) 934 static void bio_release_pages(struct bio *bio)
934 { 935 {
935 struct bio_vec *bvec = bio->bi_io_vec; 936 struct bio_vec *bvec = bio->bi_io_vec;
936 int i; 937 int i;
937 938
938 for (i = 0; i < bio->bi_vcnt; i++) { 939 for (i = 0; i < bio->bi_vcnt; i++) {
939 struct page *page = bvec[i].bv_page; 940 struct page *page = bvec[i].bv_page;
940 941
941 if (page) 942 if (page)
942 put_page(page); 943 put_page(page);
943 } 944 }
944 } 945 }
945 946
946 /* 947 /*
947 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 948 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
948 * If they are, then fine. If, however, some pages are clean then they must 949 * If they are, then fine. If, however, some pages are clean then they must
949 * have been written out during the direct-IO read. So we take another ref on 950 * have been written out during the direct-IO read. So we take another ref on
950 * the BIO and the offending pages and re-dirty the pages in process context. 951 * the BIO and the offending pages and re-dirty the pages in process context.
951 * 952 *
952 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 953 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
953 * here on. It will run one page_cache_release() against each page and will 954 * here on. It will run one page_cache_release() against each page and will
954 * run one bio_put() against the BIO. 955 * run one bio_put() against the BIO.
955 */ 956 */
956 957
957 static void bio_dirty_fn(void *data); 958 static void bio_dirty_fn(void *data);
958 959
959 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL); 960 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
960 static DEFINE_SPINLOCK(bio_dirty_lock); 961 static DEFINE_SPINLOCK(bio_dirty_lock);
961 static struct bio *bio_dirty_list; 962 static struct bio *bio_dirty_list;
962 963
963 /* 964 /*
964 * This runs in process context 965 * This runs in process context
965 */ 966 */
966 static void bio_dirty_fn(void *data) 967 static void bio_dirty_fn(void *data)
967 { 968 {
968 unsigned long flags; 969 unsigned long flags;
969 struct bio *bio; 970 struct bio *bio;
970 971
971 spin_lock_irqsave(&bio_dirty_lock, flags); 972 spin_lock_irqsave(&bio_dirty_lock, flags);
972 bio = bio_dirty_list; 973 bio = bio_dirty_list;
973 bio_dirty_list = NULL; 974 bio_dirty_list = NULL;
974 spin_unlock_irqrestore(&bio_dirty_lock, flags); 975 spin_unlock_irqrestore(&bio_dirty_lock, flags);
975 976
976 while (bio) { 977 while (bio) {
977 struct bio *next = bio->bi_private; 978 struct bio *next = bio->bi_private;
978 979
979 bio_set_pages_dirty(bio); 980 bio_set_pages_dirty(bio);
980 bio_release_pages(bio); 981 bio_release_pages(bio);
981 bio_put(bio); 982 bio_put(bio);
982 bio = next; 983 bio = next;
983 } 984 }
984 } 985 }
985 986
986 void bio_check_pages_dirty(struct bio *bio) 987 void bio_check_pages_dirty(struct bio *bio)
987 { 988 {
988 struct bio_vec *bvec = bio->bi_io_vec; 989 struct bio_vec *bvec = bio->bi_io_vec;
989 int nr_clean_pages = 0; 990 int nr_clean_pages = 0;
990 int i; 991 int i;
991 992
992 for (i = 0; i < bio->bi_vcnt; i++) { 993 for (i = 0; i < bio->bi_vcnt; i++) {
993 struct page *page = bvec[i].bv_page; 994 struct page *page = bvec[i].bv_page;
994 995
995 if (PageDirty(page) || PageCompound(page)) { 996 if (PageDirty(page) || PageCompound(page)) {
996 page_cache_release(page); 997 page_cache_release(page);
997 bvec[i].bv_page = NULL; 998 bvec[i].bv_page = NULL;
998 } else { 999 } else {
999 nr_clean_pages++; 1000 nr_clean_pages++;
1000 } 1001 }
1001 } 1002 }
1002 1003
1003 if (nr_clean_pages) { 1004 if (nr_clean_pages) {
1004 unsigned long flags; 1005 unsigned long flags;
1005 1006
1006 spin_lock_irqsave(&bio_dirty_lock, flags); 1007 spin_lock_irqsave(&bio_dirty_lock, flags);
1007 bio->bi_private = bio_dirty_list; 1008 bio->bi_private = bio_dirty_list;
1008 bio_dirty_list = bio; 1009 bio_dirty_list = bio;
1009 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1010 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1010 schedule_work(&bio_dirty_work); 1011 schedule_work(&bio_dirty_work);
1011 } else { 1012 } else {
1012 bio_put(bio); 1013 bio_put(bio);
1013 } 1014 }
1014 } 1015 }
1015 1016
1016 /** 1017 /**
1017 * bio_endio - end I/O on a bio 1018 * bio_endio - end I/O on a bio
1018 * @bio: bio 1019 * @bio: bio
1019 * @bytes_done: number of bytes completed 1020 * @bytes_done: number of bytes completed
1020 * @error: error, if any 1021 * @error: error, if any
1021 * 1022 *
1022 * Description: 1023 * Description:
1023 * bio_endio() will end I/O on @bytes_done number of bytes. This may be 1024 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
1024 * just a partial part of the bio, or it may be the whole bio. bio_endio() 1025 * just a partial part of the bio, or it may be the whole bio. bio_endio()
1025 * is the preferred way to end I/O on a bio, it takes care of decrementing 1026 * is the preferred way to end I/O on a bio, it takes care of decrementing
1026 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and 1027 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
1027 * and one of the established -Exxxx (-EIO, for instance) error values in 1028 * and one of the established -Exxxx (-EIO, for instance) error values in
1028 * case something went wrong. Noone should call bi_end_io() directly on 1029 * case something went wrong. Noone should call bi_end_io() directly on
1029 * a bio unless they own it and thus know that it has an end_io function. 1030 * a bio unless they own it and thus know that it has an end_io function.
1030 **/ 1031 **/
1031 void bio_endio(struct bio *bio, unsigned int bytes_done, int error) 1032 void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
1032 { 1033 {
1033 if (error) 1034 if (error)
1034 clear_bit(BIO_UPTODATE, &bio->bi_flags); 1035 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1035 1036
1036 if (unlikely(bytes_done > bio->bi_size)) { 1037 if (unlikely(bytes_done > bio->bi_size)) {
1037 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__, 1038 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
1038 bytes_done, bio->bi_size); 1039 bytes_done, bio->bi_size);
1039 bytes_done = bio->bi_size; 1040 bytes_done = bio->bi_size;
1040 } 1041 }
1041 1042
1042 bio->bi_size -= bytes_done; 1043 bio->bi_size -= bytes_done;
1043 bio->bi_sector += (bytes_done >> 9); 1044 bio->bi_sector += (bytes_done >> 9);
1044 1045
1045 if (bio->bi_end_io) 1046 if (bio->bi_end_io)
1046 bio->bi_end_io(bio, bytes_done, error); 1047 bio->bi_end_io(bio, bytes_done, error);
1047 } 1048 }
1048 1049
1049 void bio_pair_release(struct bio_pair *bp) 1050 void bio_pair_release(struct bio_pair *bp)
1050 { 1051 {
1051 if (atomic_dec_and_test(&bp->cnt)) { 1052 if (atomic_dec_and_test(&bp->cnt)) {
1052 struct bio *master = bp->bio1.bi_private; 1053 struct bio *master = bp->bio1.bi_private;
1053 1054
1054 bio_endio(master, master->bi_size, bp->error); 1055 bio_endio(master, master->bi_size, bp->error);
1055 mempool_free(bp, bp->bio2.bi_private); 1056 mempool_free(bp, bp->bio2.bi_private);
1056 } 1057 }
1057 } 1058 }
1058 1059
1059 static int bio_pair_end_1(struct bio * bi, unsigned int done, int err) 1060 static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
1060 { 1061 {
1061 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); 1062 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1062 1063
1063 if (err) 1064 if (err)
1064 bp->error = err; 1065 bp->error = err;
1065 1066
1066 if (bi->bi_size) 1067 if (bi->bi_size)
1067 return 1; 1068 return 1;
1068 1069
1069 bio_pair_release(bp); 1070 bio_pair_release(bp);
1070 return 0; 1071 return 0;
1071 } 1072 }
1072 1073
1073 static int bio_pair_end_2(struct bio * bi, unsigned int done, int err) 1074 static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
1074 { 1075 {
1075 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); 1076 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1076 1077
1077 if (err) 1078 if (err)
1078 bp->error = err; 1079 bp->error = err;
1079 1080
1080 if (bi->bi_size) 1081 if (bi->bi_size)
1081 return 1; 1082 return 1;
1082 1083
1083 bio_pair_release(bp); 1084 bio_pair_release(bp);
1084 return 0; 1085 return 0;
1085 } 1086 }
1086 1087
1087 /* 1088 /*
1088 * split a bio - only worry about a bio with a single page 1089 * split a bio - only worry about a bio with a single page
1089 * in it's iovec 1090 * in it's iovec
1090 */ 1091 */
1091 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) 1092 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1092 { 1093 {
1093 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); 1094 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1094 1095
1095 if (!bp) 1096 if (!bp)
1096 return bp; 1097 return bp;
1097 1098
1098 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi, 1099 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1099 bi->bi_sector + first_sectors); 1100 bi->bi_sector + first_sectors);
1100 1101
1101 BUG_ON(bi->bi_vcnt != 1); 1102 BUG_ON(bi->bi_vcnt != 1);
1102 BUG_ON(bi->bi_idx != 0); 1103 BUG_ON(bi->bi_idx != 0);
1103 atomic_set(&bp->cnt, 3); 1104 atomic_set(&bp->cnt, 3);
1104 bp->error = 0; 1105 bp->error = 0;
1105 bp->bio1 = *bi; 1106 bp->bio1 = *bi;
1106 bp->bio2 = *bi; 1107 bp->bio2 = *bi;
1107 bp->bio2.bi_sector += first_sectors; 1108 bp->bio2.bi_sector += first_sectors;
1108 bp->bio2.bi_size -= first_sectors << 9; 1109 bp->bio2.bi_size -= first_sectors << 9;
1109 bp->bio1.bi_size = first_sectors << 9; 1110 bp->bio1.bi_size = first_sectors << 9;
1110 1111
1111 bp->bv1 = bi->bi_io_vec[0]; 1112 bp->bv1 = bi->bi_io_vec[0];
1112 bp->bv2 = bi->bi_io_vec[0]; 1113 bp->bv2 = bi->bi_io_vec[0];
1113 bp->bv2.bv_offset += first_sectors << 9; 1114 bp->bv2.bv_offset += first_sectors << 9;
1114 bp->bv2.bv_len -= first_sectors << 9; 1115 bp->bv2.bv_len -= first_sectors << 9;
1115 bp->bv1.bv_len = first_sectors << 9; 1116 bp->bv1.bv_len = first_sectors << 9;
1116 1117
1117 bp->bio1.bi_io_vec = &bp->bv1; 1118 bp->bio1.bi_io_vec = &bp->bv1;
1118 bp->bio2.bi_io_vec = &bp->bv2; 1119 bp->bio2.bi_io_vec = &bp->bv2;
1119 1120
1120 bp->bio1.bi_max_vecs = 1; 1121 bp->bio1.bi_max_vecs = 1;
1121 bp->bio2.bi_max_vecs = 1; 1122 bp->bio2.bi_max_vecs = 1;
1122 1123
1123 bp->bio1.bi_end_io = bio_pair_end_1; 1124 bp->bio1.bi_end_io = bio_pair_end_1;
1124 bp->bio2.bi_end_io = bio_pair_end_2; 1125 bp->bio2.bi_end_io = bio_pair_end_2;
1125 1126
1126 bp->bio1.bi_private = bi; 1127 bp->bio1.bi_private = bi;
1127 bp->bio2.bi_private = pool; 1128 bp->bio2.bi_private = pool;
1128 1129
1129 return bp; 1130 return bp;
1130 } 1131 }
1131 1132
1132 1133
1133 /* 1134 /*
1134 * create memory pools for biovec's in a bio_set. 1135 * create memory pools for biovec's in a bio_set.
1135 * use the global biovec slabs created for general use. 1136 * use the global biovec slabs created for general use.
1136 */ 1137 */
1137 static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale) 1138 static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
1138 { 1139 {
1139 int i; 1140 int i;
1140 1141
1141 for (i = 0; i < BIOVEC_NR_POOLS; i++) { 1142 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1142 struct biovec_slab *bp = bvec_slabs + i; 1143 struct biovec_slab *bp = bvec_slabs + i;
1143 mempool_t **bvp = bs->bvec_pools + i; 1144 mempool_t **bvp = bs->bvec_pools + i;
1144 1145
1145 if (pool_entries > 1 && i >= scale) 1146 if (pool_entries > 1 && i >= scale)
1146 pool_entries >>= 1; 1147 pool_entries >>= 1;
1147 1148
1148 *bvp = mempool_create_slab_pool(pool_entries, bp->slab); 1149 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
1149 if (!*bvp) 1150 if (!*bvp)
1150 return -ENOMEM; 1151 return -ENOMEM;
1151 } 1152 }
1152 return 0; 1153 return 0;
1153 } 1154 }
1154 1155
1155 static void biovec_free_pools(struct bio_set *bs) 1156 static void biovec_free_pools(struct bio_set *bs)
1156 { 1157 {
1157 int i; 1158 int i;
1158 1159
1159 for (i = 0; i < BIOVEC_NR_POOLS; i++) { 1160 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1160 mempool_t *bvp = bs->bvec_pools[i]; 1161 mempool_t *bvp = bs->bvec_pools[i];
1161 1162
1162 if (bvp) 1163 if (bvp)
1163 mempool_destroy(bvp); 1164 mempool_destroy(bvp);
1164 } 1165 }
1165 1166
1166 } 1167 }
1167 1168
1168 void bioset_free(struct bio_set *bs) 1169 void bioset_free(struct bio_set *bs)
1169 { 1170 {
1170 if (bs->bio_pool) 1171 if (bs->bio_pool)
1171 mempool_destroy(bs->bio_pool); 1172 mempool_destroy(bs->bio_pool);
1172 1173
1173 biovec_free_pools(bs); 1174 biovec_free_pools(bs);
1174 1175
1175 kfree(bs); 1176 kfree(bs);
1176 } 1177 }
1177 1178
1178 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale) 1179 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
1179 { 1180 {
1180 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); 1181 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1181 1182
1182 if (!bs) 1183 if (!bs)
1183 return NULL; 1184 return NULL;
1184 1185
1185 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); 1186 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
1186 if (!bs->bio_pool) 1187 if (!bs->bio_pool)
1187 goto bad; 1188 goto bad;
1188 1189
1189 if (!biovec_create_pools(bs, bvec_pool_size, scale)) 1190 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1190 return bs; 1191 return bs;
1191 1192
1192 bad: 1193 bad:
1193 bioset_free(bs); 1194 bioset_free(bs);
1194 return NULL; 1195 return NULL;
1195 } 1196 }
1196 1197
1197 static void __init biovec_init_slabs(void) 1198 static void __init biovec_init_slabs(void)
1198 { 1199 {
1199 int i; 1200 int i;
1200 1201
1201 for (i = 0; i < BIOVEC_NR_POOLS; i++) { 1202 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1202 int size; 1203 int size;
1203 struct biovec_slab *bvs = bvec_slabs + i; 1204 struct biovec_slab *bvs = bvec_slabs + i;
1204 1205
1205 size = bvs->nr_vecs * sizeof(struct bio_vec); 1206 size = bvs->nr_vecs * sizeof(struct bio_vec);
1206 bvs->slab = kmem_cache_create(bvs->name, size, 0, 1207 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1207 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 1208 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1208 } 1209 }
1209 } 1210 }
1210 1211
1211 static int __init init_bio(void) 1212 static int __init init_bio(void)
1212 { 1213 {
1213 int megabytes, bvec_pool_entries; 1214 int megabytes, bvec_pool_entries;
1214 int scale = BIOVEC_NR_POOLS; 1215 int scale = BIOVEC_NR_POOLS;
1215 1216
1216 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0, 1217 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1217 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); 1218 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1218 1219
1219 biovec_init_slabs(); 1220 biovec_init_slabs();
1220 1221
1221 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT); 1222 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1222 1223
1223 /* 1224 /*
1224 * find out where to start scaling 1225 * find out where to start scaling
1225 */ 1226 */
1226 if (megabytes <= 16) 1227 if (megabytes <= 16)
1227 scale = 0; 1228 scale = 0;
1228 else if (megabytes <= 32) 1229 else if (megabytes <= 32)
1229 scale = 1; 1230 scale = 1;
1230 else if (megabytes <= 64) 1231 else if (megabytes <= 64)
1231 scale = 2; 1232 scale = 2;
1232 else if (megabytes <= 96) 1233 else if (megabytes <= 96)
1233 scale = 3; 1234 scale = 3;
1234 else if (megabytes <= 128) 1235 else if (megabytes <= 128)
1235 scale = 4; 1236 scale = 4;
1236 1237
1237 /* 1238 /*
1238 * Limit number of entries reserved -- mempools are only used when 1239 * Limit number of entries reserved -- mempools are only used when
1239 * the system is completely unable to allocate memory, so we only 1240 * the system is completely unable to allocate memory, so we only
1240 * need enough to make progress. 1241 * need enough to make progress.
1241 */ 1242 */
1242 bvec_pool_entries = 1 + scale; 1243 bvec_pool_entries = 1 + scale;
1243 1244
1244 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale); 1245 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1245 if (!fs_bio_set) 1246 if (!fs_bio_set)
1246 panic("bio: can't allocate bios\n"); 1247 panic("bio: can't allocate bios\n");
1247 1248
1248 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, 1249 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1249 sizeof(struct bio_pair)); 1250 sizeof(struct bio_pair));
1250 if (!bio_split_pool) 1251 if (!bio_split_pool)
1251 panic("bio: can't create split pool\n"); 1252 panic("bio: can't create split pool\n");
1252 1253
1253 return 0; 1254 return 0;
1254 } 1255 }
1255 1256
1256 subsys_initcall(init_bio); 1257 subsys_initcall(init_bio);
1257 1258
1258 EXPORT_SYMBOL(bio_alloc); 1259 EXPORT_SYMBOL(bio_alloc);
1259 EXPORT_SYMBOL(bio_put); 1260 EXPORT_SYMBOL(bio_put);
1260 EXPORT_SYMBOL(bio_free); 1261 EXPORT_SYMBOL(bio_free);
1261 EXPORT_SYMBOL(bio_endio); 1262 EXPORT_SYMBOL(bio_endio);
1262 EXPORT_SYMBOL(bio_init); 1263 EXPORT_SYMBOL(bio_init);
1263 EXPORT_SYMBOL(__bio_clone); 1264 EXPORT_SYMBOL(__bio_clone);
1264 EXPORT_SYMBOL(bio_clone); 1265 EXPORT_SYMBOL(bio_clone);
1265 EXPORT_SYMBOL(bio_phys_segments); 1266 EXPORT_SYMBOL(bio_phys_segments);
1266 EXPORT_SYMBOL(bio_hw_segments); 1267 EXPORT_SYMBOL(bio_hw_segments);
1267 EXPORT_SYMBOL(bio_add_page); 1268 EXPORT_SYMBOL(bio_add_page);
1268 EXPORT_SYMBOL(bio_add_pc_page); 1269 EXPORT_SYMBOL(bio_add_pc_page);
1269 EXPORT_SYMBOL(bio_get_nr_vecs); 1270 EXPORT_SYMBOL(bio_get_nr_vecs);
1270 EXPORT_SYMBOL(bio_map_user); 1271 EXPORT_SYMBOL(bio_map_user);
1271 EXPORT_SYMBOL(bio_unmap_user); 1272 EXPORT_SYMBOL(bio_unmap_user);
1272 EXPORT_SYMBOL(bio_map_kern); 1273 EXPORT_SYMBOL(bio_map_kern);
1273 EXPORT_SYMBOL(bio_pair_release); 1274 EXPORT_SYMBOL(bio_pair_release);
1274 EXPORT_SYMBOL(bio_split); 1275 EXPORT_SYMBOL(bio_split);
1275 EXPORT_SYMBOL(bio_split_pool); 1276 EXPORT_SYMBOL(bio_split_pool);
1276 EXPORT_SYMBOL(bio_copy_user); 1277 EXPORT_SYMBOL(bio_copy_user);
1277 EXPORT_SYMBOL(bio_uncopy_user); 1278 EXPORT_SYMBOL(bio_uncopy_user);
1278 EXPORT_SYMBOL(bioset_create); 1279 EXPORT_SYMBOL(bioset_create);
1279 EXPORT_SYMBOL(bioset_free); 1280 EXPORT_SYMBOL(bioset_free);