Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Jens Axboe | 0fe2347 | 2006-09-04 15:41:16 +0200 | [diff] [blame] | 2 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
| 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 Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 28 | #include <linux/blktrace_api.h> |
Arnaldo Carvalho de Melo | 5f3ea37 | 2008-10-30 08:34:33 +0100 | [diff] [blame] | 29 | #include <trace/block.h> |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 30 | #include <scsi/sg.h> /* for struct sg_iovec */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | |
Ingo Molnar | 0bfc245 | 2008-11-26 11:59:56 +0100 | [diff] [blame] | 32 | DEFINE_TRACE(block_split); |
| 33 | |
Denis ChengRq | 6feef53 | 2008-10-09 08:57:05 +0200 | [diff] [blame] | 34 | static mempool_t *bio_split_pool __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | /* |
| 37 | * if you change this list, also change bvec_alloc or things will |
| 38 | * break badly! cannot be bigger than what you can fit into an |
| 39 | * unsigned short |
| 40 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 42 | struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
| 44 | }; |
| 45 | #undef BV |
| 46 | |
| 47 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by |
| 49 | * IO code that does not need private memory pools. |
| 50 | */ |
Martin K. Petersen | 51d654e | 2008-06-17 18:59:56 +0200 | [diff] [blame] | 51 | struct bio_set *fs_bio_set; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 53 | /* |
| 54 | * Our slab pool management |
| 55 | */ |
| 56 | struct bio_slab { |
| 57 | struct kmem_cache *slab; |
| 58 | unsigned int slab_ref; |
| 59 | unsigned int slab_size; |
| 60 | char name[8]; |
| 61 | }; |
| 62 | static DEFINE_MUTEX(bio_slab_lock); |
| 63 | static struct bio_slab *bio_slabs; |
| 64 | static unsigned int bio_slab_nr, bio_slab_max; |
| 65 | |
| 66 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) |
| 67 | { |
| 68 | unsigned int sz = sizeof(struct bio) + extra_size; |
| 69 | struct kmem_cache *slab = NULL; |
| 70 | struct bio_slab *bslab; |
| 71 | unsigned int i, entry = -1; |
| 72 | |
| 73 | mutex_lock(&bio_slab_lock); |
| 74 | |
| 75 | i = 0; |
| 76 | while (i < bio_slab_nr) { |
| 77 | struct bio_slab *bslab = &bio_slabs[i]; |
| 78 | |
| 79 | if (!bslab->slab && entry == -1) |
| 80 | entry = i; |
| 81 | else if (bslab->slab_size == sz) { |
| 82 | slab = bslab->slab; |
| 83 | bslab->slab_ref++; |
| 84 | break; |
| 85 | } |
| 86 | i++; |
| 87 | } |
| 88 | |
| 89 | if (slab) |
| 90 | goto out_unlock; |
| 91 | |
| 92 | if (bio_slab_nr == bio_slab_max && entry == -1) { |
| 93 | bio_slab_max <<= 1; |
| 94 | bio_slabs = krealloc(bio_slabs, |
| 95 | bio_slab_max * sizeof(struct bio_slab), |
| 96 | GFP_KERNEL); |
| 97 | if (!bio_slabs) |
| 98 | goto out_unlock; |
| 99 | } |
| 100 | if (entry == -1) |
| 101 | entry = bio_slab_nr++; |
| 102 | |
| 103 | bslab = &bio_slabs[entry]; |
| 104 | |
| 105 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); |
| 106 | slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL); |
| 107 | if (!slab) |
| 108 | goto out_unlock; |
| 109 | |
| 110 | printk("bio: create slab <%s> at %d\n", bslab->name, entry); |
| 111 | bslab->slab = slab; |
| 112 | bslab->slab_ref = 1; |
| 113 | bslab->slab_size = sz; |
| 114 | out_unlock: |
| 115 | mutex_unlock(&bio_slab_lock); |
| 116 | return slab; |
| 117 | } |
| 118 | |
| 119 | static void bio_put_slab(struct bio_set *bs) |
| 120 | { |
| 121 | struct bio_slab *bslab = NULL; |
| 122 | unsigned int i; |
| 123 | |
| 124 | mutex_lock(&bio_slab_lock); |
| 125 | |
| 126 | for (i = 0; i < bio_slab_nr; i++) { |
| 127 | if (bs->bio_slab == bio_slabs[i].slab) { |
| 128 | bslab = &bio_slabs[i]; |
| 129 | break; |
| 130 | } |
| 131 | } |
| 132 | |
| 133 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
| 134 | goto out; |
| 135 | |
| 136 | WARN_ON(!bslab->slab_ref); |
| 137 | |
| 138 | if (--bslab->slab_ref) |
| 139 | goto out; |
| 140 | |
| 141 | kmem_cache_destroy(bslab->slab); |
| 142 | bslab->slab = NULL; |
| 143 | |
| 144 | out: |
| 145 | mutex_unlock(&bio_slab_lock); |
| 146 | } |
| 147 | |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 148 | unsigned int bvec_nr_vecs(unsigned short idx) |
| 149 | { |
| 150 | return bvec_slabs[idx].nr_vecs; |
| 151 | } |
| 152 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 153 | void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx) |
| 154 | { |
| 155 | BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); |
| 156 | |
| 157 | if (idx == BIOVEC_MAX_IDX) |
| 158 | mempool_free(bv, bs->bvec_pool); |
| 159 | else { |
| 160 | struct biovec_slab *bvs = bvec_slabs + idx; |
| 161 | |
| 162 | kmem_cache_free(bvs->slab, bv); |
| 163 | } |
| 164 | } |
| 165 | |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 166 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, |
| 167 | struct bio_set *bs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | { |
| 169 | struct bio_vec *bvl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | |
| 171 | /* |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 172 | * If 'bs' is given, lookup the pool and do the mempool alloc. |
| 173 | * If not, this is a bio_kmalloc() allocation and just do a |
| 174 | * kzalloc() for the exact number of vecs right away. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | */ |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 176 | if (!bs) |
| 177 | bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask); |
| 178 | |
| 179 | /* |
| 180 | * see comment near bvec_array define! |
| 181 | */ |
| 182 | switch (nr) { |
| 183 | case 1: |
| 184 | *idx = 0; |
| 185 | break; |
| 186 | case 2 ... 4: |
| 187 | *idx = 1; |
| 188 | break; |
| 189 | case 5 ... 16: |
| 190 | *idx = 2; |
| 191 | break; |
| 192 | case 17 ... 64: |
| 193 | *idx = 3; |
| 194 | break; |
| 195 | case 65 ... 128: |
| 196 | *idx = 4; |
| 197 | break; |
| 198 | case 129 ... BIO_MAX_PAGES: |
| 199 | *idx = 5; |
| 200 | break; |
| 201 | default: |
| 202 | return NULL; |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * idx now points to the pool we want to allocate from. only the |
| 207 | * 1-vec entry pool is mempool backed. |
| 208 | */ |
| 209 | if (*idx == BIOVEC_MAX_IDX) { |
| 210 | fallback: |
| 211 | bvl = mempool_alloc(bs->bvec_pool, gfp_mask); |
| 212 | } else { |
| 213 | struct biovec_slab *bvs = bvec_slabs + *idx; |
| 214 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 216 | /* |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 217 | * Make this allocation restricted and don't dump info on |
| 218 | * allocation failures, since we'll fallback to the mempool |
| 219 | * in case of failure. |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 220 | */ |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 221 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
| 222 | |
| 223 | /* |
| 224 | * Try a slab allocation. If this fails and __GFP_WAIT |
| 225 | * is set, retry with the 1-entry mempool |
| 226 | */ |
| 227 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
| 228 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { |
| 229 | *idx = BIOVEC_MAX_IDX; |
| 230 | goto fallback; |
| 231 | } |
| 232 | } |
| 233 | |
| 234 | if (bvl) |
| 235 | memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | |
| 237 | return bvl; |
| 238 | } |
| 239 | |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 240 | void bio_free(struct bio *bio, struct bio_set *bs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | { |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 242 | void *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 243 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 244 | if (bio->bi_io_vec) |
| 245 | bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio)); |
Jens Axboe | 992c5dd | 2007-07-18 13:18:08 +0200 | [diff] [blame] | 246 | |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 247 | if (bio_integrity(bio)) |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 248 | bio_integrity_free(bio, bs); |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 249 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 250 | /* |
| 251 | * If we have front padding, adjust the bio pointer before freeing |
| 252 | */ |
| 253 | p = bio; |
| 254 | if (bs->front_pad) |
| 255 | p -= bs->front_pad; |
| 256 | |
| 257 | mempool_free(p, bs->bio_pool); |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 258 | } |
| 259 | |
| 260 | /* |
| 261 | * default destructor for a bio allocated with bio_alloc_bioset() |
| 262 | */ |
| 263 | static void bio_fs_destructor(struct bio *bio) |
| 264 | { |
| 265 | bio_free(bio, fs_bio_set); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | } |
| 267 | |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 268 | static void bio_kmalloc_destructor(struct bio *bio) |
| 269 | { |
| 270 | kfree(bio->bi_io_vec); |
| 271 | kfree(bio); |
| 272 | } |
| 273 | |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 274 | void bio_init(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | { |
Jens Axboe | 2b94de5 | 2007-07-18 13:14:03 +0200 | [diff] [blame] | 276 | memset(bio, 0, sizeof(*bio)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | bio->bi_flags = 1 << BIO_UPTODATE; |
Jens Axboe | c7c22e4 | 2008-09-13 20:26:01 +0200 | [diff] [blame] | 278 | bio->bi_comp_cpu = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 279 | atomic_set(&bio->bi_cnt, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | } |
| 281 | |
| 282 | /** |
| 283 | * bio_alloc_bioset - allocate a bio for I/O |
| 284 | * @gfp_mask: the GFP_ mask given to the slab allocator |
| 285 | * @nr_iovecs: number of iovecs to pre-allocate |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 286 | * @bs: the bio_set to allocate from. If %NULL, just use kmalloc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | * |
| 288 | * Description: |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 289 | * bio_alloc_bioset will first try its own mempool to satisfy the allocation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | * If %__GFP_WAIT is set then we will block on the internal pool waiting |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 291 | * for a &struct bio to become free. If a %NULL @bs is passed in, we will |
| 292 | * fall back to just using @kmalloc to allocate the required memory. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | * |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 294 | * Note that the caller must set ->bi_destructor on succesful return |
| 295 | * of a bio, to do the appropriate freeing of the bio once the reference |
| 296 | * count drops to zero. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | **/ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 298 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | { |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 300 | struct bio *bio = NULL; |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 301 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 302 | if (bs) { |
| 303 | void *p = mempool_alloc(bs->bio_pool, gfp_mask); |
| 304 | |
| 305 | if (p) |
| 306 | bio = p + bs->front_pad; |
| 307 | } else |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 308 | bio = kmalloc(sizeof(*bio), gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | |
| 310 | if (likely(bio)) { |
| 311 | struct bio_vec *bvl = NULL; |
| 312 | |
| 313 | bio_init(bio); |
| 314 | if (likely(nr_iovecs)) { |
Jens Axboe | eeae1d4 | 2008-05-07 13:26:27 +0200 | [diff] [blame] | 315 | unsigned long uninitialized_var(idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | |
| 317 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); |
| 318 | if (unlikely(!bvl)) { |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 319 | if (bs) |
| 320 | mempool_free(bio, bs->bio_pool); |
| 321 | else |
| 322 | kfree(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | bio = NULL; |
| 324 | goto out; |
| 325 | } |
| 326 | bio->bi_flags |= idx << BIO_POOL_OFFSET; |
Denis ChengRq | 1ac0ae0 | 2008-08-04 11:56:30 +0200 | [diff] [blame] | 327 | bio->bi_max_vecs = bvec_nr_vecs(idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 328 | } |
| 329 | bio->bi_io_vec = bvl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | } |
| 331 | out: |
| 332 | return bio; |
| 333 | } |
| 334 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 335 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | { |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 337 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); |
| 338 | |
| 339 | if (bio) |
| 340 | bio->bi_destructor = bio_fs_destructor; |
| 341 | |
| 342 | return bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | } |
| 344 | |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 345 | /* |
| 346 | * Like bio_alloc(), but doesn't use a mempool backing. This means that |
| 347 | * it CAN fail, but while bio_alloc() can only be used for allocations |
| 348 | * that have a short (finite) life span, bio_kmalloc() should be used |
| 349 | * for more permanent bio allocations (like allocating some bio's for |
| 350 | * initalization or setup purposes). |
| 351 | */ |
| 352 | struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs) |
| 353 | { |
| 354 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL); |
| 355 | |
| 356 | if (bio) |
| 357 | bio->bi_destructor = bio_kmalloc_destructor; |
| 358 | |
| 359 | return bio; |
| 360 | } |
| 361 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | void zero_fill_bio(struct bio *bio) |
| 363 | { |
| 364 | unsigned long flags; |
| 365 | struct bio_vec *bv; |
| 366 | int i; |
| 367 | |
| 368 | bio_for_each_segment(bv, bio, i) { |
| 369 | char *data = bvec_kmap_irq(bv, &flags); |
| 370 | memset(data, 0, bv->bv_len); |
| 371 | flush_dcache_page(bv->bv_page); |
| 372 | bvec_kunmap_irq(data, &flags); |
| 373 | } |
| 374 | } |
| 375 | EXPORT_SYMBOL(zero_fill_bio); |
| 376 | |
| 377 | /** |
| 378 | * bio_put - release a reference to a bio |
| 379 | * @bio: bio to release reference to |
| 380 | * |
| 381 | * Description: |
| 382 | * Put a reference to a &struct bio, either one you have gotten with |
| 383 | * bio_alloc or bio_get. The last put of a bio will free it. |
| 384 | **/ |
| 385 | void bio_put(struct bio *bio) |
| 386 | { |
| 387 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); |
| 388 | |
| 389 | /* |
| 390 | * last put frees it |
| 391 | */ |
| 392 | if (atomic_dec_and_test(&bio->bi_cnt)) { |
| 393 | bio->bi_next = NULL; |
| 394 | bio->bi_destructor(bio); |
| 395 | } |
| 396 | } |
| 397 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 398 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | { |
| 400 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
| 401 | blk_recount_segments(q, bio); |
| 402 | |
| 403 | return bio->bi_phys_segments; |
| 404 | } |
| 405 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | /** |
| 407 | * __bio_clone - clone a bio |
| 408 | * @bio: destination bio |
| 409 | * @bio_src: bio to clone |
| 410 | * |
| 411 | * Clone a &bio. Caller will own the returned bio, but not |
| 412 | * the actual data it points to. Reference count of returned |
| 413 | * bio will be one. |
| 414 | */ |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 415 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | { |
Andrew Morton | e525e15 | 2005-08-07 09:42:12 -0700 | [diff] [blame] | 417 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
| 418 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 419 | |
Jens Axboe | 5d84070 | 2008-01-25 12:44:44 +0100 | [diff] [blame] | 420 | /* |
| 421 | * most users will be overriding ->bi_bdev with a new target, |
| 422 | * so we don't set nor calculate new physical/hw segment counts here |
| 423 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | bio->bi_sector = bio_src->bi_sector; |
| 425 | bio->bi_bdev = bio_src->bi_bdev; |
| 426 | bio->bi_flags |= 1 << BIO_CLONED; |
| 427 | bio->bi_rw = bio_src->bi_rw; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | bio->bi_vcnt = bio_src->bi_vcnt; |
| 429 | bio->bi_size = bio_src->bi_size; |
Andrew Morton | a5453be | 2005-07-28 01:07:18 -0700 | [diff] [blame] | 430 | bio->bi_idx = bio_src->bi_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | } |
| 432 | |
| 433 | /** |
| 434 | * bio_clone - clone a bio |
| 435 | * @bio: bio to clone |
| 436 | * @gfp_mask: allocation priority |
| 437 | * |
| 438 | * Like __bio_clone, only also allocates the returned bio |
| 439 | */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 440 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | { |
| 442 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); |
| 443 | |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 444 | if (!b) |
| 445 | return NULL; |
| 446 | |
| 447 | b->bi_destructor = bio_fs_destructor; |
| 448 | __bio_clone(b, bio); |
| 449 | |
| 450 | if (bio_integrity(bio)) { |
| 451 | int ret; |
| 452 | |
| 453 | ret = bio_integrity_clone(b, bio, fs_bio_set); |
| 454 | |
| 455 | if (ret < 0) |
| 456 | return NULL; |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 457 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | |
| 459 | return b; |
| 460 | } |
| 461 | |
| 462 | /** |
| 463 | * bio_get_nr_vecs - return approx number of vecs |
| 464 | * @bdev: I/O target |
| 465 | * |
| 466 | * Return the approximate number of pages we can send to this target. |
| 467 | * There's no guarantee that you will be able to fit this number of pages |
| 468 | * into a bio, it does not account for dynamic restrictions that vary |
| 469 | * on offset. |
| 470 | */ |
| 471 | int bio_get_nr_vecs(struct block_device *bdev) |
| 472 | { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 473 | struct request_queue *q = bdev_get_queue(bdev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | int nr_pages; |
| 475 | |
| 476 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 477 | if (nr_pages > q->max_phys_segments) |
| 478 | nr_pages = q->max_phys_segments; |
| 479 | if (nr_pages > q->max_hw_segments) |
| 480 | nr_pages = q->max_hw_segments; |
| 481 | |
| 482 | return nr_pages; |
| 483 | } |
| 484 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 485 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 486 | *page, unsigned int len, unsigned int offset, |
| 487 | unsigned short max_sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | { |
| 489 | int retried_segments = 0; |
| 490 | struct bio_vec *bvec; |
| 491 | |
| 492 | /* |
| 493 | * cloned bio must not modify vec list |
| 494 | */ |
| 495 | if (unlikely(bio_flagged(bio, BIO_CLONED))) |
| 496 | return 0; |
| 497 | |
Jens Axboe | 80cfd54 | 2006-01-06 09:43:28 +0100 | [diff] [blame] | 498 | if (((bio->bi_size + len) >> 9) > max_sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 499 | return 0; |
| 500 | |
Jens Axboe | 80cfd54 | 2006-01-06 09:43:28 +0100 | [diff] [blame] | 501 | /* |
| 502 | * For filesystems with a blocksize smaller than the pagesize |
| 503 | * we will often be called with the same page as last time and |
| 504 | * a consecutive offset. Optimize this special case. |
| 505 | */ |
| 506 | if (bio->bi_vcnt > 0) { |
| 507 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
| 508 | |
| 509 | if (page == prev->bv_page && |
| 510 | offset == prev->bv_offset + prev->bv_len) { |
| 511 | prev->bv_len += len; |
Alasdair G Kergon | cc371e6 | 2008-07-03 09:53:43 +0200 | [diff] [blame] | 512 | |
| 513 | if (q->merge_bvec_fn) { |
| 514 | struct bvec_merge_data bvm = { |
| 515 | .bi_bdev = bio->bi_bdev, |
| 516 | .bi_sector = bio->bi_sector, |
| 517 | .bi_size = bio->bi_size, |
| 518 | .bi_rw = bio->bi_rw, |
| 519 | }; |
| 520 | |
| 521 | if (q->merge_bvec_fn(q, &bvm, prev) < len) { |
| 522 | prev->bv_len -= len; |
| 523 | return 0; |
| 524 | } |
Jens Axboe | 80cfd54 | 2006-01-06 09:43:28 +0100 | [diff] [blame] | 525 | } |
| 526 | |
| 527 | goto done; |
| 528 | } |
| 529 | } |
| 530 | |
| 531 | if (bio->bi_vcnt >= bio->bi_max_vecs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | return 0; |
| 533 | |
| 534 | /* |
| 535 | * we might lose a segment or two here, but rather that than |
| 536 | * make this too complex. |
| 537 | */ |
| 538 | |
| 539 | while (bio->bi_phys_segments >= q->max_phys_segments |
Mikulas Patocka | 5df97b9 | 2008-08-15 10:20:02 +0200 | [diff] [blame] | 540 | || bio->bi_phys_segments >= q->max_hw_segments) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | |
| 542 | if (retried_segments) |
| 543 | return 0; |
| 544 | |
| 545 | retried_segments = 1; |
| 546 | blk_recount_segments(q, bio); |
| 547 | } |
| 548 | |
| 549 | /* |
| 550 | * setup the new entry, we might clear it again later if we |
| 551 | * cannot add the page |
| 552 | */ |
| 553 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
| 554 | bvec->bv_page = page; |
| 555 | bvec->bv_len = len; |
| 556 | bvec->bv_offset = offset; |
| 557 | |
| 558 | /* |
| 559 | * if queue has other restrictions (eg varying max sector size |
| 560 | * depending on offset), it can specify a merge_bvec_fn in the |
| 561 | * queue to get further control |
| 562 | */ |
| 563 | if (q->merge_bvec_fn) { |
Alasdair G Kergon | cc371e6 | 2008-07-03 09:53:43 +0200 | [diff] [blame] | 564 | struct bvec_merge_data bvm = { |
| 565 | .bi_bdev = bio->bi_bdev, |
| 566 | .bi_sector = bio->bi_sector, |
| 567 | .bi_size = bio->bi_size, |
| 568 | .bi_rw = bio->bi_rw, |
| 569 | }; |
| 570 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | /* |
| 572 | * merge_bvec_fn() returns number of bytes it can accept |
| 573 | * at this offset |
| 574 | */ |
Alasdair G Kergon | cc371e6 | 2008-07-03 09:53:43 +0200 | [diff] [blame] | 575 | if (q->merge_bvec_fn(q, &bvm, bvec) < len) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 576 | bvec->bv_page = NULL; |
| 577 | bvec->bv_len = 0; |
| 578 | bvec->bv_offset = 0; |
| 579 | return 0; |
| 580 | } |
| 581 | } |
| 582 | |
| 583 | /* If we may be able to merge these biovecs, force a recount */ |
Mikulas Patocka | b8b3e16 | 2008-08-15 10:15:19 +0200 | [diff] [blame] | 584 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 585 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
| 586 | |
| 587 | bio->bi_vcnt++; |
| 588 | bio->bi_phys_segments++; |
Jens Axboe | 80cfd54 | 2006-01-06 09:43:28 +0100 | [diff] [blame] | 589 | done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 590 | bio->bi_size += len; |
| 591 | return len; |
| 592 | } |
| 593 | |
| 594 | /** |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 595 | * bio_add_pc_page - attempt to add page to bio |
Jens Axboe | fddfdea | 2006-01-31 15:24:34 +0100 | [diff] [blame] | 596 | * @q: the target queue |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 597 | * @bio: destination bio |
| 598 | * @page: page to add |
| 599 | * @len: vec entry length |
| 600 | * @offset: vec entry offset |
| 601 | * |
| 602 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
| 603 | * number of reasons, such as the bio being full or target block |
| 604 | * device limitations. The target block device must allow bio's |
| 605 | * smaller than PAGE_SIZE, so it is always possible to add a single |
| 606 | * page to an empty bio. This should only be used by REQ_PC bios. |
| 607 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 608 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 609 | unsigned int len, unsigned int offset) |
| 610 | { |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 611 | return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 612 | } |
| 613 | |
| 614 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 615 | * bio_add_page - attempt to add page to bio |
| 616 | * @bio: destination bio |
| 617 | * @page: page to add |
| 618 | * @len: vec entry length |
| 619 | * @offset: vec entry offset |
| 620 | * |
| 621 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
| 622 | * number of reasons, such as the bio being full or target block |
| 623 | * device limitations. The target block device must allow bio's |
| 624 | * smaller than PAGE_SIZE, so it is always possible to add a single |
| 625 | * page to an empty bio. |
| 626 | */ |
| 627 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, |
| 628 | unsigned int offset) |
| 629 | { |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 630 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
| 631 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | } |
| 633 | |
| 634 | struct bio_map_data { |
| 635 | struct bio_vec *iovecs; |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 636 | struct sg_iovec *sgvecs; |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 637 | int nr_sgvecs; |
| 638 | int is_our_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | }; |
| 640 | |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 641 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 642 | struct sg_iovec *iov, int iov_count, |
| 643 | int is_our_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | { |
| 645 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 646 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
| 647 | bmd->nr_sgvecs = iov_count; |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 648 | bmd->is_our_pages = is_our_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | bio->bi_private = bmd; |
| 650 | } |
| 651 | |
| 652 | static void bio_free_map_data(struct bio_map_data *bmd) |
| 653 | { |
| 654 | kfree(bmd->iovecs); |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 655 | kfree(bmd->sgvecs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 656 | kfree(bmd); |
| 657 | } |
| 658 | |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 659 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count, |
| 660 | gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 661 | { |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 662 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 663 | |
| 664 | if (!bmd) |
| 665 | return NULL; |
| 666 | |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 667 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 668 | if (!bmd->iovecs) { |
| 669 | kfree(bmd); |
| 670 | return NULL; |
| 671 | } |
| 672 | |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 673 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 674 | if (bmd->sgvecs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | return bmd; |
| 676 | |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 677 | kfree(bmd->iovecs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | kfree(bmd); |
| 679 | return NULL; |
| 680 | } |
| 681 | |
FUJITA Tomonori | aefcc28 | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 682 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 683 | struct sg_iovec *iov, int iov_count, int uncopy, |
| 684 | int do_free_page) |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 685 | { |
| 686 | int ret = 0, i; |
| 687 | struct bio_vec *bvec; |
| 688 | int iov_idx = 0; |
| 689 | unsigned int iov_off = 0; |
| 690 | int read = bio_data_dir(bio) == READ; |
| 691 | |
| 692 | __bio_for_each_segment(bvec, bio, i, 0) { |
| 693 | char *bv_addr = page_address(bvec->bv_page); |
FUJITA Tomonori | aefcc28 | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 694 | unsigned int bv_len = iovecs[i].bv_len; |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 695 | |
| 696 | while (bv_len && iov_idx < iov_count) { |
| 697 | unsigned int bytes; |
| 698 | char *iov_addr; |
| 699 | |
| 700 | bytes = min_t(unsigned int, |
| 701 | iov[iov_idx].iov_len - iov_off, bv_len); |
| 702 | iov_addr = iov[iov_idx].iov_base + iov_off; |
| 703 | |
| 704 | if (!ret) { |
| 705 | if (!read && !uncopy) |
| 706 | ret = copy_from_user(bv_addr, iov_addr, |
| 707 | bytes); |
| 708 | if (read && uncopy) |
| 709 | ret = copy_to_user(iov_addr, bv_addr, |
| 710 | bytes); |
| 711 | |
| 712 | if (ret) |
| 713 | ret = -EFAULT; |
| 714 | } |
| 715 | |
| 716 | bv_len -= bytes; |
| 717 | bv_addr += bytes; |
| 718 | iov_addr += bytes; |
| 719 | iov_off += bytes; |
| 720 | |
| 721 | if (iov[iov_idx].iov_len == iov_off) { |
| 722 | iov_idx++; |
| 723 | iov_off = 0; |
| 724 | } |
| 725 | } |
| 726 | |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 727 | if (do_free_page) |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 728 | __free_page(bvec->bv_page); |
| 729 | } |
| 730 | |
| 731 | return ret; |
| 732 | } |
| 733 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | /** |
| 735 | * bio_uncopy_user - finish previously mapped bio |
| 736 | * @bio: bio being terminated |
| 737 | * |
| 738 | * Free pages allocated from bio_copy_user() and write back data |
| 739 | * to user space in case of a read. |
| 740 | */ |
| 741 | int bio_uncopy_user(struct bio *bio) |
| 742 | { |
| 743 | struct bio_map_data *bmd = bio->bi_private; |
FUJITA Tomonori | 8188276 | 2008-09-02 16:20:19 +0900 | [diff] [blame] | 744 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 745 | |
FUJITA Tomonori | 8188276 | 2008-09-02 16:20:19 +0900 | [diff] [blame] | 746 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) |
| 747 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, |
| 748 | bmd->nr_sgvecs, 1, bmd->is_our_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 749 | bio_free_map_data(bmd); |
| 750 | bio_put(bio); |
| 751 | return ret; |
| 752 | } |
| 753 | |
| 754 | /** |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 755 | * bio_copy_user_iov - copy user data to bio |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 756 | * @q: destination block queue |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 757 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 758 | * @iov: the iovec. |
| 759 | * @iov_count: number of elements in the iovec |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | * @write_to_vm: bool indicating writing to pages or not |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 761 | * @gfp_mask: memory allocation flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | * |
| 763 | * Prepares and returns a bio for indirect user io, bouncing data |
| 764 | * to/from kernel pages as necessary. Must be paired with |
| 765 | * call bio_uncopy_user() on io completion. |
| 766 | */ |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 767 | struct bio *bio_copy_user_iov(struct request_queue *q, |
| 768 | struct rq_map_data *map_data, |
| 769 | struct sg_iovec *iov, int iov_count, |
| 770 | int write_to_vm, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | struct bio_map_data *bmd; |
| 773 | struct bio_vec *bvec; |
| 774 | struct page *page; |
| 775 | struct bio *bio; |
| 776 | int i, ret; |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 777 | int nr_pages = 0; |
| 778 | unsigned int len = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 780 | for (i = 0; i < iov_count; i++) { |
| 781 | unsigned long uaddr; |
| 782 | unsigned long end; |
| 783 | unsigned long start; |
| 784 | |
| 785 | uaddr = (unsigned long)iov[i].iov_base; |
| 786 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 787 | start = uaddr >> PAGE_SHIFT; |
| 788 | |
| 789 | nr_pages += end - start; |
| 790 | len += iov[i].iov_len; |
| 791 | } |
| 792 | |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 793 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 794 | if (!bmd) |
| 795 | return ERR_PTR(-ENOMEM); |
| 796 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | ret = -ENOMEM; |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 798 | bio = bio_alloc(gfp_mask, nr_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | if (!bio) |
| 800 | goto out_bmd; |
| 801 | |
| 802 | bio->bi_rw |= (!write_to_vm << BIO_RW); |
| 803 | |
| 804 | ret = 0; |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 805 | i = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 806 | while (len) { |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 807 | unsigned int bytes; |
| 808 | |
| 809 | if (map_data) |
| 810 | bytes = 1U << (PAGE_SHIFT + map_data->page_order); |
| 811 | else |
| 812 | bytes = PAGE_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | |
| 814 | if (bytes > len) |
| 815 | bytes = len; |
| 816 | |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 817 | if (map_data) { |
| 818 | if (i == map_data->nr_entries) { |
| 819 | ret = -ENOMEM; |
| 820 | break; |
| 821 | } |
| 822 | page = map_data->pages[i++]; |
| 823 | } else |
| 824 | page = alloc_page(q->bounce_gfp | gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | if (!page) { |
| 826 | ret = -ENOMEM; |
| 827 | break; |
| 828 | } |
| 829 | |
Mike Christie | 0e75f90 | 2006-12-01 10:40:55 +0100 | [diff] [blame] | 830 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 831 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 832 | |
| 833 | len -= bytes; |
| 834 | } |
| 835 | |
| 836 | if (ret) |
| 837 | goto cleanup; |
| 838 | |
| 839 | /* |
| 840 | * success |
| 841 | */ |
| 842 | if (!write_to_vm) { |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 843 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0); |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 844 | if (ret) |
| 845 | goto cleanup; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | } |
| 847 | |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 848 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | return bio; |
| 850 | cleanup: |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 851 | if (!map_data) |
| 852 | bio_for_each_segment(bvec, bio, i) |
| 853 | __free_page(bvec->bv_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 854 | |
| 855 | bio_put(bio); |
| 856 | out_bmd: |
| 857 | bio_free_map_data(bmd); |
| 858 | return ERR_PTR(ret); |
| 859 | } |
| 860 | |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 861 | /** |
| 862 | * bio_copy_user - copy user data to bio |
| 863 | * @q: destination block queue |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 864 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 865 | * @uaddr: start of user address |
| 866 | * @len: length in bytes |
| 867 | * @write_to_vm: bool indicating writing to pages or not |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 868 | * @gfp_mask: memory allocation flags |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 869 | * |
| 870 | * Prepares and returns a bio for indirect user io, bouncing data |
| 871 | * to/from kernel pages as necessary. Must be paired with |
| 872 | * call bio_uncopy_user() on io completion. |
| 873 | */ |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 874 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, |
| 875 | unsigned long uaddr, unsigned int len, |
| 876 | int write_to_vm, gfp_t gfp_mask) |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 877 | { |
| 878 | struct sg_iovec iov; |
| 879 | |
| 880 | iov.iov_base = (void __user *)uaddr; |
| 881 | iov.iov_len = len; |
| 882 | |
FUJITA Tomonori | 152e283 | 2008-08-28 16:17:06 +0900 | [diff] [blame] | 883 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); |
FUJITA Tomonori | c5dec1c | 2008-04-11 12:56:49 +0200 | [diff] [blame] | 884 | } |
| 885 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 886 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 887 | struct block_device *bdev, |
| 888 | struct sg_iovec *iov, int iov_count, |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 889 | int write_to_vm, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | { |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 891 | int i, j; |
| 892 | int nr_pages = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | struct page **pages; |
| 894 | struct bio *bio; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 895 | int cur_page = 0; |
| 896 | int ret, offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 897 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 898 | for (i = 0; i < iov_count; i++) { |
| 899 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
| 900 | unsigned long len = iov[i].iov_len; |
| 901 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 902 | unsigned long start = uaddr >> PAGE_SHIFT; |
| 903 | |
| 904 | nr_pages += end - start; |
| 905 | /* |
Mike Christie | ad2d722 | 2006-12-01 10:40:20 +0100 | [diff] [blame] | 906 | * buffer must be aligned to at least hardsector size for now |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 907 | */ |
Mike Christie | ad2d722 | 2006-12-01 10:40:20 +0100 | [diff] [blame] | 908 | if (uaddr & queue_dma_alignment(q)) |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 909 | return ERR_PTR(-EINVAL); |
| 910 | } |
| 911 | |
| 912 | if (!nr_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 913 | return ERR_PTR(-EINVAL); |
| 914 | |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 915 | bio = bio_alloc(gfp_mask, nr_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | if (!bio) |
| 917 | return ERR_PTR(-ENOMEM); |
| 918 | |
| 919 | ret = -ENOMEM; |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 920 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | if (!pages) |
| 922 | goto out; |
| 923 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 924 | for (i = 0; i < iov_count; i++) { |
| 925 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
| 926 | unsigned long len = iov[i].iov_len; |
| 927 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 928 | unsigned long start = uaddr >> PAGE_SHIFT; |
| 929 | const int local_nr_pages = end - start; |
| 930 | const int page_limit = cur_page + local_nr_pages; |
| 931 | |
Nick Piggin | f5dd33c | 2008-07-25 19:45:25 -0700 | [diff] [blame] | 932 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
| 933 | write_to_vm, &pages[cur_page]); |
Jens Axboe | 9917215 | 2006-06-16 13:02:29 +0200 | [diff] [blame] | 934 | if (ret < local_nr_pages) { |
| 935 | ret = -EFAULT; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 936 | goto out_unmap; |
Jens Axboe | 9917215 | 2006-06-16 13:02:29 +0200 | [diff] [blame] | 937 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 939 | offset = uaddr & ~PAGE_MASK; |
| 940 | for (j = cur_page; j < page_limit; j++) { |
| 941 | unsigned int bytes = PAGE_SIZE - offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 943 | if (len <= 0) |
| 944 | break; |
| 945 | |
| 946 | if (bytes > len) |
| 947 | bytes = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 949 | /* |
| 950 | * sorry... |
| 951 | */ |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 952 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
| 953 | bytes) |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 954 | break; |
| 955 | |
| 956 | len -= bytes; |
| 957 | offset = 0; |
| 958 | } |
| 959 | |
| 960 | cur_page = j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | /* |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 962 | * release the pages we didn't map into the bio, if any |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 963 | */ |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 964 | while (j < page_limit) |
| 965 | page_cache_release(pages[j++]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 966 | } |
| 967 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 968 | kfree(pages); |
| 969 | |
| 970 | /* |
| 971 | * set data direction, and check if mapped pages need bouncing |
| 972 | */ |
| 973 | if (!write_to_vm) |
| 974 | bio->bi_rw |= (1 << BIO_RW); |
| 975 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 976 | bio->bi_bdev = bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 977 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
| 978 | return bio; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 979 | |
| 980 | out_unmap: |
| 981 | for (i = 0; i < nr_pages; i++) { |
| 982 | if(!pages[i]) |
| 983 | break; |
| 984 | page_cache_release(pages[i]); |
| 985 | } |
| 986 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | kfree(pages); |
| 988 | bio_put(bio); |
| 989 | return ERR_PTR(ret); |
| 990 | } |
| 991 | |
| 992 | /** |
| 993 | * bio_map_user - map user address into bio |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 994 | * @q: the struct request_queue for the bio |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 995 | * @bdev: destination block device |
| 996 | * @uaddr: start of user address |
| 997 | * @len: length in bytes |
| 998 | * @write_to_vm: bool indicating writing to pages or not |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 999 | * @gfp_mask: memory allocation flags |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | * |
| 1001 | * Map the user space address into a bio suitable for io to a block |
| 1002 | * device. Returns an error pointer in case of error. |
| 1003 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1004 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 1005 | unsigned long uaddr, unsigned int len, int write_to_vm, |
| 1006 | gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | { |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1008 | struct sg_iovec iov; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | |
viro@ZenIV.linux.org.uk | 3f70353 | 2005-09-09 16:53:56 +0100 | [diff] [blame] | 1010 | iov.iov_base = (void __user *)uaddr; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1011 | iov.iov_len = len; |
| 1012 | |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 1013 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1014 | } |
| 1015 | |
| 1016 | /** |
| 1017 | * bio_map_user_iov - map user sg_iovec table into bio |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1018 | * @q: the struct request_queue for the bio |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1019 | * @bdev: destination block device |
| 1020 | * @iov: the iovec. |
| 1021 | * @iov_count: number of elements in the iovec |
| 1022 | * @write_to_vm: bool indicating writing to pages or not |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 1023 | * @gfp_mask: memory allocation flags |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1024 | * |
| 1025 | * Map the user space address into a bio suitable for io to a block |
| 1026 | * device. Returns an error pointer in case of error. |
| 1027 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1028 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1029 | struct sg_iovec *iov, int iov_count, |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 1030 | int write_to_vm, gfp_t gfp_mask) |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1031 | { |
| 1032 | struct bio *bio; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 1033 | |
FUJITA Tomonori | a3bce90 | 2008-08-28 16:17:05 +0900 | [diff] [blame] | 1034 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, |
| 1035 | gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | if (IS_ERR(bio)) |
| 1037 | return bio; |
| 1038 | |
| 1039 | /* |
| 1040 | * subtle -- if __bio_map_user() ended up bouncing a bio, |
| 1041 | * it would normally disappear when its bi_end_io is run. |
| 1042 | * however, we need it for the unmap, so grab an extra |
| 1043 | * reference to it |
| 1044 | */ |
| 1045 | bio_get(bio); |
| 1046 | |
Mike Christie | 0e75f90 | 2006-12-01 10:40:55 +0100 | [diff] [blame] | 1047 | return bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1048 | } |
| 1049 | |
| 1050 | static void __bio_unmap_user(struct bio *bio) |
| 1051 | { |
| 1052 | struct bio_vec *bvec; |
| 1053 | int i; |
| 1054 | |
| 1055 | /* |
| 1056 | * make sure we dirty pages we wrote to |
| 1057 | */ |
| 1058 | __bio_for_each_segment(bvec, bio, i, 0) { |
| 1059 | if (bio_data_dir(bio) == READ) |
| 1060 | set_page_dirty_lock(bvec->bv_page); |
| 1061 | |
| 1062 | page_cache_release(bvec->bv_page); |
| 1063 | } |
| 1064 | |
| 1065 | bio_put(bio); |
| 1066 | } |
| 1067 | |
| 1068 | /** |
| 1069 | * bio_unmap_user - unmap a bio |
| 1070 | * @bio: the bio being unmapped |
| 1071 | * |
| 1072 | * Unmap a bio previously mapped by bio_map_user(). Must be called with |
| 1073 | * a process context. |
| 1074 | * |
| 1075 | * bio_unmap_user() may sleep. |
| 1076 | */ |
| 1077 | void bio_unmap_user(struct bio *bio) |
| 1078 | { |
| 1079 | __bio_unmap_user(bio); |
| 1080 | bio_put(bio); |
| 1081 | } |
| 1082 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1083 | static void bio_map_kern_endio(struct bio *bio, int err) |
Jens Axboe | b823825 | 2005-06-20 14:05:27 +0200 | [diff] [blame] | 1084 | { |
Jens Axboe | b823825 | 2005-06-20 14:05:27 +0200 | [diff] [blame] | 1085 | bio_put(bio); |
Jens Axboe | b823825 | 2005-06-20 14:05:27 +0200 | [diff] [blame] | 1086 | } |
| 1087 | |
| 1088 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1089 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
Al Viro | 27496a8 | 2005-10-21 03:20:48 -0400 | [diff] [blame] | 1090 | unsigned int len, gfp_t gfp_mask) |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 1091 | { |
| 1092 | unsigned long kaddr = (unsigned long)data; |
| 1093 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1094 | unsigned long start = kaddr >> PAGE_SHIFT; |
| 1095 | const int nr_pages = end - start; |
| 1096 | int offset, i; |
| 1097 | struct bio *bio; |
| 1098 | |
| 1099 | bio = bio_alloc(gfp_mask, nr_pages); |
| 1100 | if (!bio) |
| 1101 | return ERR_PTR(-ENOMEM); |
| 1102 | |
| 1103 | offset = offset_in_page(kaddr); |
| 1104 | for (i = 0; i < nr_pages; i++) { |
| 1105 | unsigned int bytes = PAGE_SIZE - offset; |
| 1106 | |
| 1107 | if (len <= 0) |
| 1108 | break; |
| 1109 | |
| 1110 | if (bytes > len) |
| 1111 | bytes = len; |
| 1112 | |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 1113 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
| 1114 | offset) < bytes) |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 1115 | break; |
| 1116 | |
| 1117 | data += bytes; |
| 1118 | len -= bytes; |
| 1119 | offset = 0; |
| 1120 | } |
| 1121 | |
Jens Axboe | b823825 | 2005-06-20 14:05:27 +0200 | [diff] [blame] | 1122 | bio->bi_end_io = bio_map_kern_endio; |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 1123 | return bio; |
| 1124 | } |
| 1125 | |
| 1126 | /** |
| 1127 | * bio_map_kern - map kernel address into bio |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1128 | * @q: the struct request_queue for the bio |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 1129 | * @data: pointer to buffer to map |
| 1130 | * @len: length in bytes |
| 1131 | * @gfp_mask: allocation flags for bio allocation |
| 1132 | * |
| 1133 | * Map the kernel address into a bio suitable for io to a block |
| 1134 | * device. Returns an error pointer in case of error. |
| 1135 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1136 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
Al Viro | 27496a8 | 2005-10-21 03:20:48 -0400 | [diff] [blame] | 1137 | gfp_t gfp_mask) |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 1138 | { |
| 1139 | struct bio *bio; |
| 1140 | |
| 1141 | bio = __bio_map_kern(q, data, len, gfp_mask); |
| 1142 | if (IS_ERR(bio)) |
| 1143 | return bio; |
| 1144 | |
| 1145 | if (bio->bi_size == len) |
| 1146 | return bio; |
| 1147 | |
| 1148 | /* |
| 1149 | * Don't support partial mappings. |
| 1150 | */ |
| 1151 | bio_put(bio); |
| 1152 | return ERR_PTR(-EINVAL); |
| 1153 | } |
| 1154 | |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1155 | static void bio_copy_kern_endio(struct bio *bio, int err) |
| 1156 | { |
| 1157 | struct bio_vec *bvec; |
| 1158 | const int read = bio_data_dir(bio) == READ; |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 1159 | struct bio_map_data *bmd = bio->bi_private; |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1160 | int i; |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 1161 | char *p = bmd->sgvecs[0].iov_base; |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1162 | |
| 1163 | __bio_for_each_segment(bvec, bio, i, 0) { |
| 1164 | char *addr = page_address(bvec->bv_page); |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 1165 | int len = bmd->iovecs[i].bv_len; |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1166 | |
| 1167 | if (read && !err) |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 1168 | memcpy(p, addr, len); |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1169 | |
| 1170 | __free_page(bvec->bv_page); |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 1171 | p += len; |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1172 | } |
| 1173 | |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 1174 | bio_free_map_data(bmd); |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1175 | bio_put(bio); |
| 1176 | } |
| 1177 | |
| 1178 | /** |
| 1179 | * bio_copy_kern - copy kernel address into bio |
| 1180 | * @q: the struct request_queue for the bio |
| 1181 | * @data: pointer to buffer to copy |
| 1182 | * @len: length in bytes |
| 1183 | * @gfp_mask: allocation flags for bio and page allocation |
Randy Dunlap | ffee025 | 2008-04-30 09:08:54 +0200 | [diff] [blame] | 1184 | * @reading: data direction is READ |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1185 | * |
| 1186 | * copy the kernel address into a bio suitable for io to a block |
| 1187 | * device. Returns an error pointer in case of error. |
| 1188 | */ |
| 1189 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, |
| 1190 | gfp_t gfp_mask, int reading) |
| 1191 | { |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1192 | struct bio *bio; |
| 1193 | struct bio_vec *bvec; |
FUJITA Tomonori | 4d8ab62 | 2008-08-28 15:05:57 +0900 | [diff] [blame] | 1194 | int i; |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1195 | |
FUJITA Tomonori | 4d8ab62 | 2008-08-28 15:05:57 +0900 | [diff] [blame] | 1196 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); |
| 1197 | if (IS_ERR(bio)) |
| 1198 | return bio; |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1199 | |
| 1200 | if (!reading) { |
| 1201 | void *p = data; |
| 1202 | |
| 1203 | bio_for_each_segment(bvec, bio, i) { |
| 1204 | char *addr = page_address(bvec->bv_page); |
| 1205 | |
| 1206 | memcpy(addr, p, bvec->bv_len); |
| 1207 | p += bvec->bv_len; |
| 1208 | } |
| 1209 | } |
| 1210 | |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1211 | bio->bi_end_io = bio_copy_kern_endio; |
FUJITA Tomonori | 76029ff | 2008-08-25 20:36:08 +0200 | [diff] [blame] | 1212 | |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1213 | return bio; |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1214 | } |
| 1215 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | /* |
| 1217 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions |
| 1218 | * for performing direct-IO in BIOs. |
| 1219 | * |
| 1220 | * The problem is that we cannot run set_page_dirty() from interrupt context |
| 1221 | * because the required locks are not interrupt-safe. So what we can do is to |
| 1222 | * mark the pages dirty _before_ performing IO. And in interrupt context, |
| 1223 | * check that the pages are still dirty. If so, fine. If not, redirty them |
| 1224 | * in process context. |
| 1225 | * |
| 1226 | * We special-case compound pages here: normally this means reads into hugetlb |
| 1227 | * pages. The logic in here doesn't really work right for compound pages |
| 1228 | * because the VM does not uniformly chase down the head page in all cases. |
| 1229 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't |
| 1230 | * handle them at all. So we skip compound pages here at an early stage. |
| 1231 | * |
| 1232 | * Note that this code is very hard to test under normal circumstances because |
| 1233 | * direct-io pins the pages with get_user_pages(). This makes |
| 1234 | * is_page_cache_freeable return false, and the VM will not clean the pages. |
| 1235 | * But other code (eg, pdflush) could clean the pages if they are mapped |
| 1236 | * pagecache. |
| 1237 | * |
| 1238 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the |
| 1239 | * deferred bio dirtying paths. |
| 1240 | */ |
| 1241 | |
| 1242 | /* |
| 1243 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. |
| 1244 | */ |
| 1245 | void bio_set_pages_dirty(struct bio *bio) |
| 1246 | { |
| 1247 | struct bio_vec *bvec = bio->bi_io_vec; |
| 1248 | int i; |
| 1249 | |
| 1250 | for (i = 0; i < bio->bi_vcnt; i++) { |
| 1251 | struct page *page = bvec[i].bv_page; |
| 1252 | |
| 1253 | if (page && !PageCompound(page)) |
| 1254 | set_page_dirty_lock(page); |
| 1255 | } |
| 1256 | } |
| 1257 | |
Adrian Bunk | 86b6c7a | 2008-02-18 13:48:32 +0100 | [diff] [blame] | 1258 | static void bio_release_pages(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1259 | { |
| 1260 | struct bio_vec *bvec = bio->bi_io_vec; |
| 1261 | int i; |
| 1262 | |
| 1263 | for (i = 0; i < bio->bi_vcnt; i++) { |
| 1264 | struct page *page = bvec[i].bv_page; |
| 1265 | |
| 1266 | if (page) |
| 1267 | put_page(page); |
| 1268 | } |
| 1269 | } |
| 1270 | |
| 1271 | /* |
| 1272 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. |
| 1273 | * If they are, then fine. If, however, some pages are clean then they must |
| 1274 | * have been written out during the direct-IO read. So we take another ref on |
| 1275 | * the BIO and the offending pages and re-dirty the pages in process context. |
| 1276 | * |
| 1277 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from |
| 1278 | * here on. It will run one page_cache_release() against each page and will |
| 1279 | * run one bio_put() against the BIO. |
| 1280 | */ |
| 1281 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 1282 | static void bio_dirty_fn(struct work_struct *work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1283 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 1284 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1285 | static DEFINE_SPINLOCK(bio_dirty_lock); |
| 1286 | static struct bio *bio_dirty_list; |
| 1287 | |
| 1288 | /* |
| 1289 | * This runs in process context |
| 1290 | */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 1291 | static void bio_dirty_fn(struct work_struct *work) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | { |
| 1293 | unsigned long flags; |
| 1294 | struct bio *bio; |
| 1295 | |
| 1296 | spin_lock_irqsave(&bio_dirty_lock, flags); |
| 1297 | bio = bio_dirty_list; |
| 1298 | bio_dirty_list = NULL; |
| 1299 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
| 1300 | |
| 1301 | while (bio) { |
| 1302 | struct bio *next = bio->bi_private; |
| 1303 | |
| 1304 | bio_set_pages_dirty(bio); |
| 1305 | bio_release_pages(bio); |
| 1306 | bio_put(bio); |
| 1307 | bio = next; |
| 1308 | } |
| 1309 | } |
| 1310 | |
| 1311 | void bio_check_pages_dirty(struct bio *bio) |
| 1312 | { |
| 1313 | struct bio_vec *bvec = bio->bi_io_vec; |
| 1314 | int nr_clean_pages = 0; |
| 1315 | int i; |
| 1316 | |
| 1317 | for (i = 0; i < bio->bi_vcnt; i++) { |
| 1318 | struct page *page = bvec[i].bv_page; |
| 1319 | |
| 1320 | if (PageDirty(page) || PageCompound(page)) { |
| 1321 | page_cache_release(page); |
| 1322 | bvec[i].bv_page = NULL; |
| 1323 | } else { |
| 1324 | nr_clean_pages++; |
| 1325 | } |
| 1326 | } |
| 1327 | |
| 1328 | if (nr_clean_pages) { |
| 1329 | unsigned long flags; |
| 1330 | |
| 1331 | spin_lock_irqsave(&bio_dirty_lock, flags); |
| 1332 | bio->bi_private = bio_dirty_list; |
| 1333 | bio_dirty_list = bio; |
| 1334 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
| 1335 | schedule_work(&bio_dirty_work); |
| 1336 | } else { |
| 1337 | bio_put(bio); |
| 1338 | } |
| 1339 | } |
| 1340 | |
| 1341 | /** |
| 1342 | * bio_endio - end I/O on a bio |
| 1343 | * @bio: bio |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | * @error: error, if any |
| 1345 | * |
| 1346 | * Description: |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1347 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 1348 | * preferred way to end I/O on a bio, it takes care of clearing |
| 1349 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the |
| 1350 | * established -Exxxx (-EIO, for instance) error values in case |
| 1351 | * something went wrong. Noone should call bi_end_io() directly on a |
| 1352 | * bio unless they own it and thus know that it has an end_io |
| 1353 | * function. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | **/ |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1355 | void bio_endio(struct bio *bio, int error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | { |
| 1357 | if (error) |
| 1358 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
NeilBrown | 9cc54d4 | 2007-09-27 12:46:12 +0200 | [diff] [blame] | 1359 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
| 1360 | error = -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | |
NeilBrown | 5bb23a6 | 2007-09-27 12:46:13 +0200 | [diff] [blame] | 1362 | if (bio->bi_end_io) |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1363 | bio->bi_end_io(bio, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1364 | } |
| 1365 | |
| 1366 | void bio_pair_release(struct bio_pair *bp) |
| 1367 | { |
| 1368 | if (atomic_dec_and_test(&bp->cnt)) { |
| 1369 | struct bio *master = bp->bio1.bi_private; |
| 1370 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1371 | bio_endio(master, bp->error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | mempool_free(bp, bp->bio2.bi_private); |
| 1373 | } |
| 1374 | } |
| 1375 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1376 | static void bio_pair_end_1(struct bio *bi, int err) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1377 | { |
| 1378 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); |
| 1379 | |
| 1380 | if (err) |
| 1381 | bp->error = err; |
| 1382 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1383 | bio_pair_release(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1384 | } |
| 1385 | |
NeilBrown | 6712ecf | 2007-09-27 12:47:43 +0200 | [diff] [blame] | 1386 | static void bio_pair_end_2(struct bio *bi, int err) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1387 | { |
| 1388 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); |
| 1389 | |
| 1390 | if (err) |
| 1391 | bp->error = err; |
| 1392 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1393 | bio_pair_release(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | } |
| 1395 | |
| 1396 | /* |
| 1397 | * split a bio - only worry about a bio with a single page |
| 1398 | * in it's iovec |
| 1399 | */ |
Denis ChengRq | 6feef53 | 2008-10-09 08:57:05 +0200 | [diff] [blame] | 1400 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1401 | { |
Denis ChengRq | 6feef53 | 2008-10-09 08:57:05 +0200 | [diff] [blame] | 1402 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | |
| 1404 | if (!bp) |
| 1405 | return bp; |
| 1406 | |
Arnaldo Carvalho de Melo | 5f3ea37 | 2008-10-30 08:34:33 +0100 | [diff] [blame] | 1407 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, |
Jens Axboe | 2056a78 | 2006-03-23 20:00:26 +0100 | [diff] [blame] | 1408 | bi->bi_sector + first_sectors); |
| 1409 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1410 | BUG_ON(bi->bi_vcnt != 1); |
| 1411 | BUG_ON(bi->bi_idx != 0); |
| 1412 | atomic_set(&bp->cnt, 3); |
| 1413 | bp->error = 0; |
| 1414 | bp->bio1 = *bi; |
| 1415 | bp->bio2 = *bi; |
| 1416 | bp->bio2.bi_sector += first_sectors; |
| 1417 | bp->bio2.bi_size -= first_sectors << 9; |
| 1418 | bp->bio1.bi_size = first_sectors << 9; |
| 1419 | |
| 1420 | bp->bv1 = bi->bi_io_vec[0]; |
| 1421 | bp->bv2 = bi->bi_io_vec[0]; |
| 1422 | bp->bv2.bv_offset += first_sectors << 9; |
| 1423 | bp->bv2.bv_len -= first_sectors << 9; |
| 1424 | bp->bv1.bv_len = first_sectors << 9; |
| 1425 | |
| 1426 | bp->bio1.bi_io_vec = &bp->bv1; |
| 1427 | bp->bio2.bi_io_vec = &bp->bv2; |
| 1428 | |
NeilBrown | a2eb0c1 | 2006-05-22 22:35:27 -0700 | [diff] [blame] | 1429 | bp->bio1.bi_max_vecs = 1; |
| 1430 | bp->bio2.bi_max_vecs = 1; |
| 1431 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1432 | bp->bio1.bi_end_io = bio_pair_end_1; |
| 1433 | bp->bio2.bi_end_io = bio_pair_end_2; |
| 1434 | |
| 1435 | bp->bio1.bi_private = bi; |
Denis ChengRq | 6feef53 | 2008-10-09 08:57:05 +0200 | [diff] [blame] | 1436 | bp->bio2.bi_private = bio_split_pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 1438 | if (bio_integrity(bi)) |
| 1439 | bio_integrity_split(bi, bp, first_sectors); |
| 1440 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1441 | return bp; |
| 1442 | } |
| 1443 | |
Martin K. Petersen | ad3316b | 2008-10-01 22:42:53 -0400 | [diff] [blame] | 1444 | /** |
| 1445 | * bio_sector_offset - Find hardware sector offset in bio |
| 1446 | * @bio: bio to inspect |
| 1447 | * @index: bio_vec index |
| 1448 | * @offset: offset in bv_page |
| 1449 | * |
| 1450 | * Return the number of hardware sectors between beginning of bio |
| 1451 | * and an end point indicated by a bio_vec index and an offset |
| 1452 | * within that vector's page. |
| 1453 | */ |
| 1454 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, |
| 1455 | unsigned int offset) |
| 1456 | { |
| 1457 | unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue); |
| 1458 | struct bio_vec *bv; |
| 1459 | sector_t sectors; |
| 1460 | int i; |
| 1461 | |
| 1462 | sectors = 0; |
| 1463 | |
| 1464 | if (index >= bio->bi_idx) |
| 1465 | index = bio->bi_vcnt - 1; |
| 1466 | |
| 1467 | __bio_for_each_segment(bv, bio, i, 0) { |
| 1468 | if (i == index) { |
| 1469 | if (offset > bv->bv_offset) |
| 1470 | sectors += (offset - bv->bv_offset) / sector_sz; |
| 1471 | break; |
| 1472 | } |
| 1473 | |
| 1474 | sectors += bv->bv_len / sector_sz; |
| 1475 | } |
| 1476 | |
| 1477 | return sectors; |
| 1478 | } |
| 1479 | EXPORT_SYMBOL(bio_sector_offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1480 | |
| 1481 | /* |
| 1482 | * create memory pools for biovec's in a bio_set. |
| 1483 | * use the global biovec slabs created for general use. |
| 1484 | */ |
Jens Axboe | 5972511 | 2007-04-02 10:06:42 +0200 | [diff] [blame] | 1485 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | { |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 1487 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1488 | |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 1489 | bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab); |
| 1490 | if (!bs->bvec_pool) |
| 1491 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1493 | return 0; |
| 1494 | } |
| 1495 | |
| 1496 | static void biovec_free_pools(struct bio_set *bs) |
| 1497 | { |
Jens Axboe | 7ff9345 | 2008-12-11 11:53:43 +0100 | [diff] [blame] | 1498 | mempool_destroy(bs->bvec_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1499 | } |
| 1500 | |
| 1501 | void bioset_free(struct bio_set *bs) |
| 1502 | { |
| 1503 | if (bs->bio_pool) |
| 1504 | mempool_destroy(bs->bio_pool); |
| 1505 | |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 1506 | bioset_integrity_free(bs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1507 | biovec_free_pools(bs); |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1508 | bio_put_slab(bs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1509 | |
| 1510 | kfree(bs); |
| 1511 | } |
| 1512 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1513 | /** |
| 1514 | * bioset_create - Create a bio_set |
| 1515 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
| 1516 | * @front_pad: Number of bytes to allocate in front of the returned bio |
| 1517 | * |
| 1518 | * Description: |
| 1519 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller |
| 1520 | * to ask for a number of bytes to be allocated in front of the bio. |
| 1521 | * Front pad allocation is useful for embedding the bio inside |
| 1522 | * another structure, to avoid allocating extra data to go with the bio. |
| 1523 | * Note that the bio must be embedded at the END of that structure always, |
| 1524 | * or things will break badly. |
| 1525 | */ |
| 1526 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | { |
Jens Axboe | 1b43449 | 2008-10-22 20:32:58 +0200 | [diff] [blame] | 1528 | struct bio_set *bs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1529 | |
Jens Axboe | 1b43449 | 2008-10-22 20:32:58 +0200 | [diff] [blame] | 1530 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | if (!bs) |
| 1532 | return NULL; |
| 1533 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1534 | bs->front_pad = front_pad; |
Jens Axboe | 1b43449 | 2008-10-22 20:32:58 +0200 | [diff] [blame] | 1535 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1536 | bs->bio_slab = bio_find_or_create_slab(front_pad); |
| 1537 | if (!bs->bio_slab) { |
| 1538 | kfree(bs); |
| 1539 | return NULL; |
| 1540 | } |
| 1541 | |
| 1542 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1543 | if (!bs->bio_pool) |
| 1544 | goto bad; |
| 1545 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1546 | if (bioset_integrity_create(bs, pool_size)) |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 1547 | goto bad; |
| 1548 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1549 | if (!biovec_create_pools(bs, pool_size)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1550 | return bs; |
| 1551 | |
| 1552 | bad: |
| 1553 | bioset_free(bs); |
| 1554 | return NULL; |
| 1555 | } |
| 1556 | |
| 1557 | static void __init biovec_init_slabs(void) |
| 1558 | { |
| 1559 | int i; |
| 1560 | |
| 1561 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
| 1562 | int size; |
| 1563 | struct biovec_slab *bvs = bvec_slabs + i; |
| 1564 | |
| 1565 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
| 1566 | bvs->slab = kmem_cache_create(bvs->name, size, 0, |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 1567 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1568 | } |
| 1569 | } |
| 1570 | |
| 1571 | static int __init init_bio(void) |
| 1572 | { |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1573 | bio_slab_max = 2; |
| 1574 | bio_slab_nr = 0; |
| 1575 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); |
| 1576 | if (!bio_slabs) |
| 1577 | panic("bio: can't allocate bios\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1578 | |
Martin K. Petersen | 7ba1ba1 | 2008-06-30 20:04:41 +0200 | [diff] [blame] | 1579 | bio_integrity_init_slab(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1580 | biovec_init_slabs(); |
| 1581 | |
Jens Axboe | bb799ca | 2008-12-10 15:35:05 +0100 | [diff] [blame^] | 1582 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1583 | if (!fs_bio_set) |
| 1584 | panic("bio: can't allocate bios\n"); |
| 1585 | |
Matthew Dobson | 0eaae62a | 2006-03-26 01:37:47 -0800 | [diff] [blame] | 1586 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
| 1587 | sizeof(struct bio_pair)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1588 | if (!bio_split_pool) |
| 1589 | panic("bio: can't create split pool\n"); |
| 1590 | |
| 1591 | return 0; |
| 1592 | } |
| 1593 | |
| 1594 | subsys_initcall(init_bio); |
| 1595 | |
| 1596 | EXPORT_SYMBOL(bio_alloc); |
Jens Axboe | 0a0d96b | 2008-09-11 13:17:37 +0200 | [diff] [blame] | 1597 | EXPORT_SYMBOL(bio_kmalloc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1598 | EXPORT_SYMBOL(bio_put); |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 1599 | EXPORT_SYMBOL(bio_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1600 | EXPORT_SYMBOL(bio_endio); |
| 1601 | EXPORT_SYMBOL(bio_init); |
| 1602 | EXPORT_SYMBOL(__bio_clone); |
| 1603 | EXPORT_SYMBOL(bio_clone); |
| 1604 | EXPORT_SYMBOL(bio_phys_segments); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1605 | EXPORT_SYMBOL(bio_add_page); |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 1606 | EXPORT_SYMBOL(bio_add_pc_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1607 | EXPORT_SYMBOL(bio_get_nr_vecs); |
Jens Axboe | 40044ce | 2008-03-17 21:14:40 +0100 | [diff] [blame] | 1608 | EXPORT_SYMBOL(bio_map_user); |
| 1609 | EXPORT_SYMBOL(bio_unmap_user); |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 1610 | EXPORT_SYMBOL(bio_map_kern); |
FUJITA Tomonori | 68154e9 | 2008-04-25 12:47:50 +0200 | [diff] [blame] | 1611 | EXPORT_SYMBOL(bio_copy_kern); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1612 | EXPORT_SYMBOL(bio_pair_release); |
| 1613 | EXPORT_SYMBOL(bio_split); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | EXPORT_SYMBOL(bio_copy_user); |
| 1615 | EXPORT_SYMBOL(bio_uncopy_user); |
| 1616 | EXPORT_SYMBOL(bioset_create); |
| 1617 | EXPORT_SYMBOL(bioset_free); |
| 1618 | EXPORT_SYMBOL(bio_alloc_bioset); |