Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2001 Jens Axboe <axboe@suse.de> |
| 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> |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 28 | #include <scsi/sg.h> /* for struct sg_iovec */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | |
| 30 | #define BIO_POOL_SIZE 256 |
| 31 | |
| 32 | static kmem_cache_t *bio_slab; |
| 33 | |
| 34 | #define BIOVEC_NR_POOLS 6 |
| 35 | |
| 36 | /* |
| 37 | * a small number of entries is fine, not going to be performance critical. |
| 38 | * basically we just need to survive |
| 39 | */ |
| 40 | #define BIO_SPLIT_ENTRIES 8 |
| 41 | mempool_t *bio_split_pool; |
| 42 | |
| 43 | struct biovec_slab { |
| 44 | int nr_vecs; |
| 45 | char *name; |
| 46 | kmem_cache_t *slab; |
| 47 | }; |
| 48 | |
| 49 | /* |
| 50 | * if you change this list, also change bvec_alloc or things will |
| 51 | * break badly! cannot be bigger than what you can fit into an |
| 52 | * unsigned short |
| 53 | */ |
| 54 | |
| 55 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } |
Christoph Lameter | 6c03652 | 2005-07-07 17:56:59 -0700 | [diff] [blame] | 56 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
| 58 | }; |
| 59 | #undef BV |
| 60 | |
| 61 | /* |
| 62 | * bio_set is used to allow other portions of the IO system to |
| 63 | * allocate their own private memory pools for bio and iovec structures. |
| 64 | * These memory pools in turn all allocate from the bio_slab |
| 65 | * and the bvec_slabs[]. |
| 66 | */ |
| 67 | struct bio_set { |
| 68 | mempool_t *bio_pool; |
| 69 | mempool_t *bvec_pools[BIOVEC_NR_POOLS]; |
| 70 | }; |
| 71 | |
| 72 | /* |
| 73 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by |
| 74 | * IO code that does not need private memory pools. |
| 75 | */ |
| 76 | static struct bio_set *fs_bio_set; |
| 77 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 78 | static inline struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | { |
| 80 | struct bio_vec *bvl; |
| 81 | struct biovec_slab *bp; |
| 82 | |
| 83 | /* |
| 84 | * see comment near bvec_array define! |
| 85 | */ |
| 86 | switch (nr) { |
| 87 | case 1 : *idx = 0; break; |
| 88 | case 2 ... 4: *idx = 1; break; |
| 89 | case 5 ... 16: *idx = 2; break; |
| 90 | case 17 ... 64: *idx = 3; break; |
| 91 | case 65 ... 128: *idx = 4; break; |
| 92 | case 129 ... BIO_MAX_PAGES: *idx = 5; break; |
| 93 | default: |
| 94 | return NULL; |
| 95 | } |
| 96 | /* |
| 97 | * idx now points to the pool we want to allocate from |
| 98 | */ |
| 99 | |
| 100 | bp = bvec_slabs + *idx; |
| 101 | bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); |
| 102 | if (bvl) |
| 103 | memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec)); |
| 104 | |
| 105 | return bvl; |
| 106 | } |
| 107 | |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 108 | void bio_free(struct bio *bio, struct bio_set *bio_set) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | { |
| 110 | const int pool_idx = BIO_POOL_IDX(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | |
| 112 | BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); |
| 113 | |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 114 | mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); |
| 115 | mempool_free(bio, bio_set->bio_pool); |
| 116 | } |
| 117 | |
| 118 | /* |
| 119 | * default destructor for a bio allocated with bio_alloc_bioset() |
| 120 | */ |
| 121 | static void bio_fs_destructor(struct bio *bio) |
| 122 | { |
| 123 | bio_free(bio, fs_bio_set); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | } |
| 125 | |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame^] | 126 | void bio_init(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | { |
| 128 | bio->bi_next = NULL; |
Jens Axboe | 0ea60b5 | 2006-01-09 14:45:10 +0100 | [diff] [blame] | 129 | bio->bi_bdev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | bio->bi_flags = 1 << BIO_UPTODATE; |
| 131 | bio->bi_rw = 0; |
| 132 | bio->bi_vcnt = 0; |
| 133 | bio->bi_idx = 0; |
| 134 | bio->bi_phys_segments = 0; |
| 135 | bio->bi_hw_segments = 0; |
| 136 | bio->bi_hw_front_size = 0; |
| 137 | bio->bi_hw_back_size = 0; |
| 138 | bio->bi_size = 0; |
| 139 | bio->bi_max_vecs = 0; |
| 140 | bio->bi_end_io = NULL; |
| 141 | atomic_set(&bio->bi_cnt, 1); |
| 142 | bio->bi_private = NULL; |
| 143 | } |
| 144 | |
| 145 | /** |
| 146 | * bio_alloc_bioset - allocate a bio for I/O |
| 147 | * @gfp_mask: the GFP_ mask given to the slab allocator |
| 148 | * @nr_iovecs: number of iovecs to pre-allocate |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 149 | * @bs: the bio_set to allocate from |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | * |
| 151 | * Description: |
| 152 | * bio_alloc_bioset will first try it's on mempool to satisfy the allocation. |
| 153 | * If %__GFP_WAIT is set then we will block on the internal pool waiting |
| 154 | * for a &struct bio to become free. |
| 155 | * |
| 156 | * allocate bio and iovecs from the memory pools specified by the |
| 157 | * bio_set structure. |
| 158 | **/ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 159 | 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] | 160 | { |
| 161 | struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask); |
| 162 | |
| 163 | if (likely(bio)) { |
| 164 | struct bio_vec *bvl = NULL; |
| 165 | |
| 166 | bio_init(bio); |
| 167 | if (likely(nr_iovecs)) { |
| 168 | unsigned long idx; |
| 169 | |
| 170 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); |
| 171 | if (unlikely(!bvl)) { |
| 172 | mempool_free(bio, bs->bio_pool); |
| 173 | bio = NULL; |
| 174 | goto out; |
| 175 | } |
| 176 | bio->bi_flags |= idx << BIO_POOL_OFFSET; |
| 177 | bio->bi_max_vecs = bvec_slabs[idx].nr_vecs; |
| 178 | } |
| 179 | bio->bi_io_vec = bvl; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | } |
| 181 | out: |
| 182 | return bio; |
| 183 | } |
| 184 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 185 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | { |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 187 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); |
| 188 | |
| 189 | if (bio) |
| 190 | bio->bi_destructor = bio_fs_destructor; |
| 191 | |
| 192 | return bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | } |
| 194 | |
| 195 | void zero_fill_bio(struct bio *bio) |
| 196 | { |
| 197 | unsigned long flags; |
| 198 | struct bio_vec *bv; |
| 199 | int i; |
| 200 | |
| 201 | bio_for_each_segment(bv, bio, i) { |
| 202 | char *data = bvec_kmap_irq(bv, &flags); |
| 203 | memset(data, 0, bv->bv_len); |
| 204 | flush_dcache_page(bv->bv_page); |
| 205 | bvec_kunmap_irq(data, &flags); |
| 206 | } |
| 207 | } |
| 208 | EXPORT_SYMBOL(zero_fill_bio); |
| 209 | |
| 210 | /** |
| 211 | * bio_put - release a reference to a bio |
| 212 | * @bio: bio to release reference to |
| 213 | * |
| 214 | * Description: |
| 215 | * Put a reference to a &struct bio, either one you have gotten with |
| 216 | * bio_alloc or bio_get. The last put of a bio will free it. |
| 217 | **/ |
| 218 | void bio_put(struct bio *bio) |
| 219 | { |
| 220 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); |
| 221 | |
| 222 | /* |
| 223 | * last put frees it |
| 224 | */ |
| 225 | if (atomic_dec_and_test(&bio->bi_cnt)) { |
| 226 | bio->bi_next = NULL; |
| 227 | bio->bi_destructor(bio); |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | inline int bio_phys_segments(request_queue_t *q, struct bio *bio) |
| 232 | { |
| 233 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
| 234 | blk_recount_segments(q, bio); |
| 235 | |
| 236 | return bio->bi_phys_segments; |
| 237 | } |
| 238 | |
| 239 | inline int bio_hw_segments(request_queue_t *q, struct bio *bio) |
| 240 | { |
| 241 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
| 242 | blk_recount_segments(q, bio); |
| 243 | |
| 244 | return bio->bi_hw_segments; |
| 245 | } |
| 246 | |
| 247 | /** |
| 248 | * __bio_clone - clone a bio |
| 249 | * @bio: destination bio |
| 250 | * @bio_src: bio to clone |
| 251 | * |
| 252 | * Clone a &bio. Caller will own the returned bio, but not |
| 253 | * the actual data it points to. Reference count of returned |
| 254 | * bio will be one. |
| 255 | */ |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame^] | 256 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 257 | { |
| 258 | request_queue_t *q = bdev_get_queue(bio_src->bi_bdev); |
| 259 | |
Andrew Morton | e525e15 | 2005-08-07 09:42:12 -0700 | [diff] [blame] | 260 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
| 261 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 262 | |
| 263 | bio->bi_sector = bio_src->bi_sector; |
| 264 | bio->bi_bdev = bio_src->bi_bdev; |
| 265 | bio->bi_flags |= 1 << BIO_CLONED; |
| 266 | bio->bi_rw = bio_src->bi_rw; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | bio->bi_vcnt = bio_src->bi_vcnt; |
| 268 | bio->bi_size = bio_src->bi_size; |
Andrew Morton | a5453be | 2005-07-28 01:07:18 -0700 | [diff] [blame] | 269 | bio->bi_idx = bio_src->bi_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | bio_phys_segments(q, bio); |
| 271 | bio_hw_segments(q, bio); |
| 272 | } |
| 273 | |
| 274 | /** |
| 275 | * bio_clone - clone a bio |
| 276 | * @bio: bio to clone |
| 277 | * @gfp_mask: allocation priority |
| 278 | * |
| 279 | * Like __bio_clone, only also allocates the returned bio |
| 280 | */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 281 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | { |
| 283 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); |
| 284 | |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 285 | if (b) { |
| 286 | b->bi_destructor = bio_fs_destructor; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | __bio_clone(b, bio); |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 288 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | |
| 290 | return b; |
| 291 | } |
| 292 | |
| 293 | /** |
| 294 | * bio_get_nr_vecs - return approx number of vecs |
| 295 | * @bdev: I/O target |
| 296 | * |
| 297 | * Return the approximate number of pages we can send to this target. |
| 298 | * There's no guarantee that you will be able to fit this number of pages |
| 299 | * into a bio, it does not account for dynamic restrictions that vary |
| 300 | * on offset. |
| 301 | */ |
| 302 | int bio_get_nr_vecs(struct block_device *bdev) |
| 303 | { |
| 304 | request_queue_t *q = bdev_get_queue(bdev); |
| 305 | int nr_pages; |
| 306 | |
| 307 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 308 | if (nr_pages > q->max_phys_segments) |
| 309 | nr_pages = q->max_phys_segments; |
| 310 | if (nr_pages > q->max_hw_segments) |
| 311 | nr_pages = q->max_hw_segments; |
| 312 | |
| 313 | return nr_pages; |
| 314 | } |
| 315 | |
| 316 | static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 317 | *page, unsigned int len, unsigned int offset, |
| 318 | unsigned short max_sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | { |
| 320 | int retried_segments = 0; |
| 321 | struct bio_vec *bvec; |
| 322 | |
| 323 | /* |
| 324 | * cloned bio must not modify vec list |
| 325 | */ |
| 326 | if (unlikely(bio_flagged(bio, BIO_CLONED))) |
| 327 | return 0; |
| 328 | |
Jens Axboe | 80cfd54 | 2006-01-06 09:43:28 +0100 | [diff] [blame] | 329 | if (((bio->bi_size + len) >> 9) > max_sectors) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | return 0; |
| 331 | |
Jens Axboe | 80cfd54 | 2006-01-06 09:43:28 +0100 | [diff] [blame] | 332 | /* |
| 333 | * For filesystems with a blocksize smaller than the pagesize |
| 334 | * we will often be called with the same page as last time and |
| 335 | * a consecutive offset. Optimize this special case. |
| 336 | */ |
| 337 | if (bio->bi_vcnt > 0) { |
| 338 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
| 339 | |
| 340 | if (page == prev->bv_page && |
| 341 | offset == prev->bv_offset + prev->bv_len) { |
| 342 | prev->bv_len += len; |
| 343 | if (q->merge_bvec_fn && |
| 344 | q->merge_bvec_fn(q, bio, prev) < len) { |
| 345 | prev->bv_len -= len; |
| 346 | return 0; |
| 347 | } |
| 348 | |
| 349 | goto done; |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | if (bio->bi_vcnt >= bio->bi_max_vecs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | return 0; |
| 355 | |
| 356 | /* |
| 357 | * we might lose a segment or two here, but rather that than |
| 358 | * make this too complex. |
| 359 | */ |
| 360 | |
| 361 | while (bio->bi_phys_segments >= q->max_phys_segments |
| 362 | || bio->bi_hw_segments >= q->max_hw_segments |
| 363 | || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) { |
| 364 | |
| 365 | if (retried_segments) |
| 366 | return 0; |
| 367 | |
| 368 | retried_segments = 1; |
| 369 | blk_recount_segments(q, bio); |
| 370 | } |
| 371 | |
| 372 | /* |
| 373 | * setup the new entry, we might clear it again later if we |
| 374 | * cannot add the page |
| 375 | */ |
| 376 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
| 377 | bvec->bv_page = page; |
| 378 | bvec->bv_len = len; |
| 379 | bvec->bv_offset = offset; |
| 380 | |
| 381 | /* |
| 382 | * if queue has other restrictions (eg varying max sector size |
| 383 | * depending on offset), it can specify a merge_bvec_fn in the |
| 384 | * queue to get further control |
| 385 | */ |
| 386 | if (q->merge_bvec_fn) { |
| 387 | /* |
| 388 | * merge_bvec_fn() returns number of bytes it can accept |
| 389 | * at this offset |
| 390 | */ |
| 391 | if (q->merge_bvec_fn(q, bio, bvec) < len) { |
| 392 | bvec->bv_page = NULL; |
| 393 | bvec->bv_len = 0; |
| 394 | bvec->bv_offset = 0; |
| 395 | return 0; |
| 396 | } |
| 397 | } |
| 398 | |
| 399 | /* If we may be able to merge these biovecs, force a recount */ |
| 400 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) || |
| 401 | BIOVEC_VIRT_MERGEABLE(bvec-1, bvec))) |
| 402 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
| 403 | |
| 404 | bio->bi_vcnt++; |
| 405 | bio->bi_phys_segments++; |
| 406 | bio->bi_hw_segments++; |
Jens Axboe | 80cfd54 | 2006-01-06 09:43:28 +0100 | [diff] [blame] | 407 | done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | bio->bi_size += len; |
| 409 | return len; |
| 410 | } |
| 411 | |
| 412 | /** |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 413 | * bio_add_pc_page - attempt to add page to bio |
| 414 | * @bio: destination bio |
| 415 | * @page: page to add |
| 416 | * @len: vec entry length |
| 417 | * @offset: vec entry offset |
| 418 | * |
| 419 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
| 420 | * number of reasons, such as the bio being full or target block |
| 421 | * device limitations. The target block device must allow bio's |
| 422 | * smaller than PAGE_SIZE, so it is always possible to add a single |
| 423 | * page to an empty bio. This should only be used by REQ_PC bios. |
| 424 | */ |
| 425 | int bio_add_pc_page(request_queue_t *q, struct bio *bio, struct page *page, |
| 426 | unsigned int len, unsigned int offset) |
| 427 | { |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 428 | 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] | 429 | } |
| 430 | |
| 431 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | * bio_add_page - attempt to add page to bio |
| 433 | * @bio: destination bio |
| 434 | * @page: page to add |
| 435 | * @len: vec entry length |
| 436 | * @offset: vec entry offset |
| 437 | * |
| 438 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
| 439 | * number of reasons, such as the bio being full or target block |
| 440 | * device limitations. The target block device must allow bio's |
| 441 | * smaller than PAGE_SIZE, so it is always possible to add a single |
| 442 | * page to an empty bio. |
| 443 | */ |
| 444 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, |
| 445 | unsigned int offset) |
| 446 | { |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 447 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
| 448 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 449 | } |
| 450 | |
| 451 | struct bio_map_data { |
| 452 | struct bio_vec *iovecs; |
| 453 | void __user *userptr; |
| 454 | }; |
| 455 | |
| 456 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio) |
| 457 | { |
| 458 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); |
| 459 | bio->bi_private = bmd; |
| 460 | } |
| 461 | |
| 462 | static void bio_free_map_data(struct bio_map_data *bmd) |
| 463 | { |
| 464 | kfree(bmd->iovecs); |
| 465 | kfree(bmd); |
| 466 | } |
| 467 | |
| 468 | static struct bio_map_data *bio_alloc_map_data(int nr_segs) |
| 469 | { |
| 470 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL); |
| 471 | |
| 472 | if (!bmd) |
| 473 | return NULL; |
| 474 | |
| 475 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL); |
| 476 | if (bmd->iovecs) |
| 477 | return bmd; |
| 478 | |
| 479 | kfree(bmd); |
| 480 | return NULL; |
| 481 | } |
| 482 | |
| 483 | /** |
| 484 | * bio_uncopy_user - finish previously mapped bio |
| 485 | * @bio: bio being terminated |
| 486 | * |
| 487 | * Free pages allocated from bio_copy_user() and write back data |
| 488 | * to user space in case of a read. |
| 489 | */ |
| 490 | int bio_uncopy_user(struct bio *bio) |
| 491 | { |
| 492 | struct bio_map_data *bmd = bio->bi_private; |
| 493 | const int read = bio_data_dir(bio) == READ; |
| 494 | struct bio_vec *bvec; |
| 495 | int i, ret = 0; |
| 496 | |
| 497 | __bio_for_each_segment(bvec, bio, i, 0) { |
| 498 | char *addr = page_address(bvec->bv_page); |
| 499 | unsigned int len = bmd->iovecs[i].bv_len; |
| 500 | |
| 501 | if (read && !ret && copy_to_user(bmd->userptr, addr, len)) |
| 502 | ret = -EFAULT; |
| 503 | |
| 504 | __free_page(bvec->bv_page); |
| 505 | bmd->userptr += len; |
| 506 | } |
| 507 | bio_free_map_data(bmd); |
| 508 | bio_put(bio); |
| 509 | return ret; |
| 510 | } |
| 511 | |
| 512 | /** |
| 513 | * bio_copy_user - copy user data to bio |
| 514 | * @q: destination block queue |
| 515 | * @uaddr: start of user address |
| 516 | * @len: length in bytes |
| 517 | * @write_to_vm: bool indicating writing to pages or not |
| 518 | * |
| 519 | * Prepares and returns a bio for indirect user io, bouncing data |
| 520 | * to/from kernel pages as necessary. Must be paired with |
| 521 | * call bio_uncopy_user() on io completion. |
| 522 | */ |
| 523 | struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr, |
| 524 | unsigned int len, int write_to_vm) |
| 525 | { |
| 526 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 527 | unsigned long start = uaddr >> PAGE_SHIFT; |
| 528 | struct bio_map_data *bmd; |
| 529 | struct bio_vec *bvec; |
| 530 | struct page *page; |
| 531 | struct bio *bio; |
| 532 | int i, ret; |
| 533 | |
| 534 | bmd = bio_alloc_map_data(end - start); |
| 535 | if (!bmd) |
| 536 | return ERR_PTR(-ENOMEM); |
| 537 | |
| 538 | bmd->userptr = (void __user *) uaddr; |
| 539 | |
| 540 | ret = -ENOMEM; |
| 541 | bio = bio_alloc(GFP_KERNEL, end - start); |
| 542 | if (!bio) |
| 543 | goto out_bmd; |
| 544 | |
| 545 | bio->bi_rw |= (!write_to_vm << BIO_RW); |
| 546 | |
| 547 | ret = 0; |
| 548 | while (len) { |
| 549 | unsigned int bytes = PAGE_SIZE; |
| 550 | |
| 551 | if (bytes > len) |
| 552 | bytes = len; |
| 553 | |
| 554 | page = alloc_page(q->bounce_gfp | GFP_KERNEL); |
| 555 | if (!page) { |
| 556 | ret = -ENOMEM; |
| 557 | break; |
| 558 | } |
| 559 | |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 560 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 561 | ret = -EINVAL; |
| 562 | break; |
| 563 | } |
| 564 | |
| 565 | len -= bytes; |
| 566 | } |
| 567 | |
| 568 | if (ret) |
| 569 | goto cleanup; |
| 570 | |
| 571 | /* |
| 572 | * success |
| 573 | */ |
| 574 | if (!write_to_vm) { |
| 575 | char __user *p = (char __user *) uaddr; |
| 576 | |
| 577 | /* |
| 578 | * for a write, copy in data to kernel pages |
| 579 | */ |
| 580 | ret = -EFAULT; |
| 581 | bio_for_each_segment(bvec, bio, i) { |
| 582 | char *addr = page_address(bvec->bv_page); |
| 583 | |
| 584 | if (copy_from_user(addr, p, bvec->bv_len)) |
| 585 | goto cleanup; |
| 586 | p += bvec->bv_len; |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | bio_set_map_data(bmd, bio); |
| 591 | return bio; |
| 592 | cleanup: |
| 593 | bio_for_each_segment(bvec, bio, i) |
| 594 | __free_page(bvec->bv_page); |
| 595 | |
| 596 | bio_put(bio); |
| 597 | out_bmd: |
| 598 | bio_free_map_data(bmd); |
| 599 | return ERR_PTR(ret); |
| 600 | } |
| 601 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 602 | static struct bio *__bio_map_user_iov(request_queue_t *q, |
| 603 | struct block_device *bdev, |
| 604 | struct sg_iovec *iov, int iov_count, |
| 605 | int write_to_vm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 606 | { |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 607 | int i, j; |
| 608 | int nr_pages = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 609 | struct page **pages; |
| 610 | struct bio *bio; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 611 | int cur_page = 0; |
| 612 | int ret, offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 614 | for (i = 0; i < iov_count; i++) { |
| 615 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
| 616 | unsigned long len = iov[i].iov_len; |
| 617 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 618 | unsigned long start = uaddr >> PAGE_SHIFT; |
| 619 | |
| 620 | nr_pages += end - start; |
| 621 | /* |
| 622 | * transfer and buffer must be aligned to at least hardsector |
| 623 | * size for now, in the future we can relax this restriction |
| 624 | */ |
| 625 | if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q))) |
| 626 | return ERR_PTR(-EINVAL); |
| 627 | } |
| 628 | |
| 629 | if (!nr_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | return ERR_PTR(-EINVAL); |
| 631 | |
| 632 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
| 633 | if (!bio) |
| 634 | return ERR_PTR(-ENOMEM); |
| 635 | |
| 636 | ret = -ENOMEM; |
| 637 | pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL); |
| 638 | if (!pages) |
| 639 | goto out; |
| 640 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 641 | memset(pages, 0, nr_pages * sizeof(struct page *)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 643 | for (i = 0; i < iov_count; i++) { |
| 644 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
| 645 | unsigned long len = iov[i].iov_len; |
| 646 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 647 | unsigned long start = uaddr >> PAGE_SHIFT; |
| 648 | const int local_nr_pages = end - start; |
| 649 | const int page_limit = cur_page + local_nr_pages; |
| 650 | |
| 651 | down_read(¤t->mm->mmap_sem); |
| 652 | ret = get_user_pages(current, current->mm, uaddr, |
| 653 | local_nr_pages, |
| 654 | write_to_vm, 0, &pages[cur_page], NULL); |
| 655 | up_read(¤t->mm->mmap_sem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 656 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 657 | if (ret < local_nr_pages) |
| 658 | goto out_unmap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 659 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 661 | offset = uaddr & ~PAGE_MASK; |
| 662 | for (j = cur_page; j < page_limit; j++) { |
| 663 | unsigned int bytes = PAGE_SIZE - offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 665 | if (len <= 0) |
| 666 | break; |
| 667 | |
| 668 | if (bytes > len) |
| 669 | bytes = len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 671 | /* |
| 672 | * sorry... |
| 673 | */ |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 674 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
| 675 | bytes) |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 676 | break; |
| 677 | |
| 678 | len -= bytes; |
| 679 | offset = 0; |
| 680 | } |
| 681 | |
| 682 | cur_page = j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | /* |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 684 | * release the pages we didn't map into the bio, if any |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | */ |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 686 | while (j < page_limit) |
| 687 | page_cache_release(pages[j++]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | } |
| 689 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | kfree(pages); |
| 691 | |
| 692 | /* |
| 693 | * set data direction, and check if mapped pages need bouncing |
| 694 | */ |
| 695 | if (!write_to_vm) |
| 696 | bio->bi_rw |= (1 << BIO_RW); |
| 697 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 698 | bio->bi_bdev = bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 699 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
| 700 | return bio; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 701 | |
| 702 | out_unmap: |
| 703 | for (i = 0; i < nr_pages; i++) { |
| 704 | if(!pages[i]) |
| 705 | break; |
| 706 | page_cache_release(pages[i]); |
| 707 | } |
| 708 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 709 | kfree(pages); |
| 710 | bio_put(bio); |
| 711 | return ERR_PTR(ret); |
| 712 | } |
| 713 | |
| 714 | /** |
| 715 | * bio_map_user - map user address into bio |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 716 | * @q: the request_queue_t for the bio |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | * @bdev: destination block device |
| 718 | * @uaddr: start of user address |
| 719 | * @len: length in bytes |
| 720 | * @write_to_vm: bool indicating writing to pages or not |
| 721 | * |
| 722 | * Map the user space address into a bio suitable for io to a block |
| 723 | * device. Returns an error pointer in case of error. |
| 724 | */ |
| 725 | struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev, |
| 726 | unsigned long uaddr, unsigned int len, int write_to_vm) |
| 727 | { |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 728 | struct sg_iovec iov; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 729 | |
viro@ZenIV.linux.org.uk | 3f70353 | 2005-09-09 16:53:56 +0100 | [diff] [blame] | 730 | iov.iov_base = (void __user *)uaddr; |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 731 | iov.iov_len = len; |
| 732 | |
| 733 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm); |
| 734 | } |
| 735 | |
| 736 | /** |
| 737 | * bio_map_user_iov - map user sg_iovec table into bio |
| 738 | * @q: the request_queue_t for the bio |
| 739 | * @bdev: destination block device |
| 740 | * @iov: the iovec. |
| 741 | * @iov_count: number of elements in the iovec |
| 742 | * @write_to_vm: bool indicating writing to pages or not |
| 743 | * |
| 744 | * Map the user space address into a bio suitable for io to a block |
| 745 | * device. Returns an error pointer in case of error. |
| 746 | */ |
| 747 | struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev, |
| 748 | struct sg_iovec *iov, int iov_count, |
| 749 | int write_to_vm) |
| 750 | { |
| 751 | struct bio *bio; |
| 752 | int len = 0, i; |
| 753 | |
| 754 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 755 | |
| 756 | if (IS_ERR(bio)) |
| 757 | return bio; |
| 758 | |
| 759 | /* |
| 760 | * subtle -- if __bio_map_user() ended up bouncing a bio, |
| 761 | * it would normally disappear when its bi_end_io is run. |
| 762 | * however, we need it for the unmap, so grab an extra |
| 763 | * reference to it |
| 764 | */ |
| 765 | bio_get(bio); |
| 766 | |
James Bottomley | f1970ba | 2005-06-20 14:06:52 +0200 | [diff] [blame] | 767 | for (i = 0; i < iov_count; i++) |
| 768 | len += iov[i].iov_len; |
| 769 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | if (bio->bi_size == len) |
| 771 | return bio; |
| 772 | |
| 773 | /* |
| 774 | * don't support partial mappings |
| 775 | */ |
| 776 | bio_endio(bio, bio->bi_size, 0); |
| 777 | bio_unmap_user(bio); |
| 778 | return ERR_PTR(-EINVAL); |
| 779 | } |
| 780 | |
| 781 | static void __bio_unmap_user(struct bio *bio) |
| 782 | { |
| 783 | struct bio_vec *bvec; |
| 784 | int i; |
| 785 | |
| 786 | /* |
| 787 | * make sure we dirty pages we wrote to |
| 788 | */ |
| 789 | __bio_for_each_segment(bvec, bio, i, 0) { |
| 790 | if (bio_data_dir(bio) == READ) |
| 791 | set_page_dirty_lock(bvec->bv_page); |
| 792 | |
| 793 | page_cache_release(bvec->bv_page); |
| 794 | } |
| 795 | |
| 796 | bio_put(bio); |
| 797 | } |
| 798 | |
| 799 | /** |
| 800 | * bio_unmap_user - unmap a bio |
| 801 | * @bio: the bio being unmapped |
| 802 | * |
| 803 | * Unmap a bio previously mapped by bio_map_user(). Must be called with |
| 804 | * a process context. |
| 805 | * |
| 806 | * bio_unmap_user() may sleep. |
| 807 | */ |
| 808 | void bio_unmap_user(struct bio *bio) |
| 809 | { |
| 810 | __bio_unmap_user(bio); |
| 811 | bio_put(bio); |
| 812 | } |
| 813 | |
Jens Axboe | b823825 | 2005-06-20 14:05:27 +0200 | [diff] [blame] | 814 | static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err) |
| 815 | { |
| 816 | if (bio->bi_size) |
| 817 | return 1; |
| 818 | |
| 819 | bio_put(bio); |
| 820 | return 0; |
| 821 | } |
| 822 | |
| 823 | |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 824 | static struct bio *__bio_map_kern(request_queue_t *q, void *data, |
Al Viro | 27496a8 | 2005-10-21 03:20:48 -0400 | [diff] [blame] | 825 | unsigned int len, gfp_t gfp_mask) |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 826 | { |
| 827 | unsigned long kaddr = (unsigned long)data; |
| 828 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 829 | unsigned long start = kaddr >> PAGE_SHIFT; |
| 830 | const int nr_pages = end - start; |
| 831 | int offset, i; |
| 832 | struct bio *bio; |
| 833 | |
| 834 | bio = bio_alloc(gfp_mask, nr_pages); |
| 835 | if (!bio) |
| 836 | return ERR_PTR(-ENOMEM); |
| 837 | |
| 838 | offset = offset_in_page(kaddr); |
| 839 | for (i = 0; i < nr_pages; i++) { |
| 840 | unsigned int bytes = PAGE_SIZE - offset; |
| 841 | |
| 842 | if (len <= 0) |
| 843 | break; |
| 844 | |
| 845 | if (bytes > len) |
| 846 | bytes = len; |
| 847 | |
Mike Christie | defd94b | 2005-12-05 02:37:06 -0600 | [diff] [blame] | 848 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
| 849 | offset) < bytes) |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 850 | break; |
| 851 | |
| 852 | data += bytes; |
| 853 | len -= bytes; |
| 854 | offset = 0; |
| 855 | } |
| 856 | |
Jens Axboe | b823825 | 2005-06-20 14:05:27 +0200 | [diff] [blame] | 857 | bio->bi_end_io = bio_map_kern_endio; |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 858 | return bio; |
| 859 | } |
| 860 | |
| 861 | /** |
| 862 | * bio_map_kern - map kernel address into bio |
| 863 | * @q: the request_queue_t for the bio |
| 864 | * @data: pointer to buffer to map |
| 865 | * @len: length in bytes |
| 866 | * @gfp_mask: allocation flags for bio allocation |
| 867 | * |
| 868 | * Map the kernel address into a bio suitable for io to a block |
| 869 | * device. Returns an error pointer in case of error. |
| 870 | */ |
| 871 | struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len, |
Al Viro | 27496a8 | 2005-10-21 03:20:48 -0400 | [diff] [blame] | 872 | gfp_t gfp_mask) |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 873 | { |
| 874 | struct bio *bio; |
| 875 | |
| 876 | bio = __bio_map_kern(q, data, len, gfp_mask); |
| 877 | if (IS_ERR(bio)) |
| 878 | return bio; |
| 879 | |
| 880 | if (bio->bi_size == len) |
| 881 | return bio; |
| 882 | |
| 883 | /* |
| 884 | * Don't support partial mappings. |
| 885 | */ |
| 886 | bio_put(bio); |
| 887 | return ERR_PTR(-EINVAL); |
| 888 | } |
| 889 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | /* |
| 891 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions |
| 892 | * for performing direct-IO in BIOs. |
| 893 | * |
| 894 | * The problem is that we cannot run set_page_dirty() from interrupt context |
| 895 | * because the required locks are not interrupt-safe. So what we can do is to |
| 896 | * mark the pages dirty _before_ performing IO. And in interrupt context, |
| 897 | * check that the pages are still dirty. If so, fine. If not, redirty them |
| 898 | * in process context. |
| 899 | * |
| 900 | * We special-case compound pages here: normally this means reads into hugetlb |
| 901 | * pages. The logic in here doesn't really work right for compound pages |
| 902 | * because the VM does not uniformly chase down the head page in all cases. |
| 903 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't |
| 904 | * handle them at all. So we skip compound pages here at an early stage. |
| 905 | * |
| 906 | * Note that this code is very hard to test under normal circumstances because |
| 907 | * direct-io pins the pages with get_user_pages(). This makes |
| 908 | * is_page_cache_freeable return false, and the VM will not clean the pages. |
| 909 | * But other code (eg, pdflush) could clean the pages if they are mapped |
| 910 | * pagecache. |
| 911 | * |
| 912 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the |
| 913 | * deferred bio dirtying paths. |
| 914 | */ |
| 915 | |
| 916 | /* |
| 917 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. |
| 918 | */ |
| 919 | void bio_set_pages_dirty(struct bio *bio) |
| 920 | { |
| 921 | struct bio_vec *bvec = bio->bi_io_vec; |
| 922 | int i; |
| 923 | |
| 924 | for (i = 0; i < bio->bi_vcnt; i++) { |
| 925 | struct page *page = bvec[i].bv_page; |
| 926 | |
| 927 | if (page && !PageCompound(page)) |
| 928 | set_page_dirty_lock(page); |
| 929 | } |
| 930 | } |
| 931 | |
| 932 | static void bio_release_pages(struct bio *bio) |
| 933 | { |
| 934 | struct bio_vec *bvec = bio->bi_io_vec; |
| 935 | int i; |
| 936 | |
| 937 | for (i = 0; i < bio->bi_vcnt; i++) { |
| 938 | struct page *page = bvec[i].bv_page; |
| 939 | |
| 940 | if (page) |
| 941 | put_page(page); |
| 942 | } |
| 943 | } |
| 944 | |
| 945 | /* |
| 946 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. |
| 947 | * If they are, then fine. If, however, some pages are clean then they must |
| 948 | * have been written out during the direct-IO read. So we take another ref on |
| 949 | * the BIO and the offending pages and re-dirty the pages in process context. |
| 950 | * |
| 951 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from |
| 952 | * here on. It will run one page_cache_release() against each page and will |
| 953 | * run one bio_put() against the BIO. |
| 954 | */ |
| 955 | |
| 956 | static void bio_dirty_fn(void *data); |
| 957 | |
| 958 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL); |
| 959 | static DEFINE_SPINLOCK(bio_dirty_lock); |
| 960 | static struct bio *bio_dirty_list; |
| 961 | |
| 962 | /* |
| 963 | * This runs in process context |
| 964 | */ |
| 965 | static void bio_dirty_fn(void *data) |
| 966 | { |
| 967 | unsigned long flags; |
| 968 | struct bio *bio; |
| 969 | |
| 970 | spin_lock_irqsave(&bio_dirty_lock, flags); |
| 971 | bio = bio_dirty_list; |
| 972 | bio_dirty_list = NULL; |
| 973 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
| 974 | |
| 975 | while (bio) { |
| 976 | struct bio *next = bio->bi_private; |
| 977 | |
| 978 | bio_set_pages_dirty(bio); |
| 979 | bio_release_pages(bio); |
| 980 | bio_put(bio); |
| 981 | bio = next; |
| 982 | } |
| 983 | } |
| 984 | |
| 985 | void bio_check_pages_dirty(struct bio *bio) |
| 986 | { |
| 987 | struct bio_vec *bvec = bio->bi_io_vec; |
| 988 | int nr_clean_pages = 0; |
| 989 | int i; |
| 990 | |
| 991 | for (i = 0; i < bio->bi_vcnt; i++) { |
| 992 | struct page *page = bvec[i].bv_page; |
| 993 | |
| 994 | if (PageDirty(page) || PageCompound(page)) { |
| 995 | page_cache_release(page); |
| 996 | bvec[i].bv_page = NULL; |
| 997 | } else { |
| 998 | nr_clean_pages++; |
| 999 | } |
| 1000 | } |
| 1001 | |
| 1002 | if (nr_clean_pages) { |
| 1003 | unsigned long flags; |
| 1004 | |
| 1005 | spin_lock_irqsave(&bio_dirty_lock, flags); |
| 1006 | bio->bi_private = bio_dirty_list; |
| 1007 | bio_dirty_list = bio; |
| 1008 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
| 1009 | schedule_work(&bio_dirty_work); |
| 1010 | } else { |
| 1011 | bio_put(bio); |
| 1012 | } |
| 1013 | } |
| 1014 | |
| 1015 | /** |
| 1016 | * bio_endio - end I/O on a bio |
| 1017 | * @bio: bio |
| 1018 | * @bytes_done: number of bytes completed |
| 1019 | * @error: error, if any |
| 1020 | * |
| 1021 | * Description: |
| 1022 | * bio_endio() will end I/O on @bytes_done number of bytes. This may be |
| 1023 | * just a partial part of the bio, or it may be the whole bio. bio_endio() |
| 1024 | * is the preferred way to end I/O on a bio, it takes care of decrementing |
| 1025 | * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and |
| 1026 | * and one of the established -Exxxx (-EIO, for instance) error values in |
| 1027 | * case something went wrong. Noone should call bi_end_io() directly on |
| 1028 | * a bio unless they own it and thus know that it has an end_io function. |
| 1029 | **/ |
| 1030 | void bio_endio(struct bio *bio, unsigned int bytes_done, int error) |
| 1031 | { |
| 1032 | if (error) |
| 1033 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
| 1034 | |
| 1035 | if (unlikely(bytes_done > bio->bi_size)) { |
| 1036 | printk("%s: want %u bytes done, only %u left\n", __FUNCTION__, |
| 1037 | bytes_done, bio->bi_size); |
| 1038 | bytes_done = bio->bi_size; |
| 1039 | } |
| 1040 | |
| 1041 | bio->bi_size -= bytes_done; |
| 1042 | bio->bi_sector += (bytes_done >> 9); |
| 1043 | |
| 1044 | if (bio->bi_end_io) |
| 1045 | bio->bi_end_io(bio, bytes_done, error); |
| 1046 | } |
| 1047 | |
| 1048 | void bio_pair_release(struct bio_pair *bp) |
| 1049 | { |
| 1050 | if (atomic_dec_and_test(&bp->cnt)) { |
| 1051 | struct bio *master = bp->bio1.bi_private; |
| 1052 | |
| 1053 | bio_endio(master, master->bi_size, bp->error); |
| 1054 | mempool_free(bp, bp->bio2.bi_private); |
| 1055 | } |
| 1056 | } |
| 1057 | |
| 1058 | static int bio_pair_end_1(struct bio * bi, unsigned int done, int err) |
| 1059 | { |
| 1060 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); |
| 1061 | |
| 1062 | if (err) |
| 1063 | bp->error = err; |
| 1064 | |
| 1065 | if (bi->bi_size) |
| 1066 | return 1; |
| 1067 | |
| 1068 | bio_pair_release(bp); |
| 1069 | return 0; |
| 1070 | } |
| 1071 | |
| 1072 | static int bio_pair_end_2(struct bio * bi, unsigned int done, int err) |
| 1073 | { |
| 1074 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); |
| 1075 | |
| 1076 | if (err) |
| 1077 | bp->error = err; |
| 1078 | |
| 1079 | if (bi->bi_size) |
| 1080 | return 1; |
| 1081 | |
| 1082 | bio_pair_release(bp); |
| 1083 | return 0; |
| 1084 | } |
| 1085 | |
| 1086 | /* |
| 1087 | * split a bio - only worry about a bio with a single page |
| 1088 | * in it's iovec |
| 1089 | */ |
| 1090 | struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) |
| 1091 | { |
| 1092 | struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); |
| 1093 | |
| 1094 | if (!bp) |
| 1095 | return bp; |
| 1096 | |
| 1097 | BUG_ON(bi->bi_vcnt != 1); |
| 1098 | BUG_ON(bi->bi_idx != 0); |
| 1099 | atomic_set(&bp->cnt, 3); |
| 1100 | bp->error = 0; |
| 1101 | bp->bio1 = *bi; |
| 1102 | bp->bio2 = *bi; |
| 1103 | bp->bio2.bi_sector += first_sectors; |
| 1104 | bp->bio2.bi_size -= first_sectors << 9; |
| 1105 | bp->bio1.bi_size = first_sectors << 9; |
| 1106 | |
| 1107 | bp->bv1 = bi->bi_io_vec[0]; |
| 1108 | bp->bv2 = bi->bi_io_vec[0]; |
| 1109 | bp->bv2.bv_offset += first_sectors << 9; |
| 1110 | bp->bv2.bv_len -= first_sectors << 9; |
| 1111 | bp->bv1.bv_len = first_sectors << 9; |
| 1112 | |
| 1113 | bp->bio1.bi_io_vec = &bp->bv1; |
| 1114 | bp->bio2.bi_io_vec = &bp->bv2; |
| 1115 | |
| 1116 | bp->bio1.bi_end_io = bio_pair_end_1; |
| 1117 | bp->bio2.bi_end_io = bio_pair_end_2; |
| 1118 | |
| 1119 | bp->bio1.bi_private = bi; |
| 1120 | bp->bio2.bi_private = pool; |
| 1121 | |
| 1122 | return bp; |
| 1123 | } |
| 1124 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 1125 | static void *bio_pair_alloc(gfp_t gfp_flags, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | { |
| 1127 | return kmalloc(sizeof(struct bio_pair), gfp_flags); |
| 1128 | } |
| 1129 | |
| 1130 | static void bio_pair_free(void *bp, void *data) |
| 1131 | { |
| 1132 | kfree(bp); |
| 1133 | } |
| 1134 | |
| 1135 | |
| 1136 | /* |
| 1137 | * create memory pools for biovec's in a bio_set. |
| 1138 | * use the global biovec slabs created for general use. |
| 1139 | */ |
| 1140 | static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale) |
| 1141 | { |
| 1142 | int i; |
| 1143 | |
| 1144 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
| 1145 | struct biovec_slab *bp = bvec_slabs + i; |
| 1146 | mempool_t **bvp = bs->bvec_pools + i; |
| 1147 | |
| 1148 | if (i >= scale) |
| 1149 | pool_entries >>= 1; |
| 1150 | |
| 1151 | *bvp = mempool_create(pool_entries, mempool_alloc_slab, |
| 1152 | mempool_free_slab, bp->slab); |
| 1153 | if (!*bvp) |
| 1154 | return -ENOMEM; |
| 1155 | } |
| 1156 | return 0; |
| 1157 | } |
| 1158 | |
| 1159 | static void biovec_free_pools(struct bio_set *bs) |
| 1160 | { |
| 1161 | int i; |
| 1162 | |
| 1163 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
| 1164 | mempool_t *bvp = bs->bvec_pools[i]; |
| 1165 | |
| 1166 | if (bvp) |
| 1167 | mempool_destroy(bvp); |
| 1168 | } |
| 1169 | |
| 1170 | } |
| 1171 | |
| 1172 | void bioset_free(struct bio_set *bs) |
| 1173 | { |
| 1174 | if (bs->bio_pool) |
| 1175 | mempool_destroy(bs->bio_pool); |
| 1176 | |
| 1177 | biovec_free_pools(bs); |
| 1178 | |
| 1179 | kfree(bs); |
| 1180 | } |
| 1181 | |
| 1182 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale) |
| 1183 | { |
| 1184 | struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL); |
| 1185 | |
| 1186 | if (!bs) |
| 1187 | return NULL; |
| 1188 | |
| 1189 | memset(bs, 0, sizeof(*bs)); |
| 1190 | bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab, |
| 1191 | mempool_free_slab, bio_slab); |
| 1192 | |
| 1193 | if (!bs->bio_pool) |
| 1194 | goto bad; |
| 1195 | |
| 1196 | if (!biovec_create_pools(bs, bvec_pool_size, scale)) |
| 1197 | return bs; |
| 1198 | |
| 1199 | bad: |
| 1200 | bioset_free(bs); |
| 1201 | return NULL; |
| 1202 | } |
| 1203 | |
| 1204 | static void __init biovec_init_slabs(void) |
| 1205 | { |
| 1206 | int i; |
| 1207 | |
| 1208 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
| 1209 | int size; |
| 1210 | struct biovec_slab *bvs = bvec_slabs + i; |
| 1211 | |
| 1212 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
| 1213 | bvs->slab = kmem_cache_create(bvs->name, size, 0, |
| 1214 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
| 1215 | } |
| 1216 | } |
| 1217 | |
| 1218 | static int __init init_bio(void) |
| 1219 | { |
| 1220 | int megabytes, bvec_pool_entries; |
| 1221 | int scale = BIOVEC_NR_POOLS; |
| 1222 | |
| 1223 | bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0, |
| 1224 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
| 1225 | |
| 1226 | biovec_init_slabs(); |
| 1227 | |
| 1228 | megabytes = nr_free_pages() >> (20 - PAGE_SHIFT); |
| 1229 | |
| 1230 | /* |
| 1231 | * find out where to start scaling |
| 1232 | */ |
| 1233 | if (megabytes <= 16) |
| 1234 | scale = 0; |
| 1235 | else if (megabytes <= 32) |
| 1236 | scale = 1; |
| 1237 | else if (megabytes <= 64) |
| 1238 | scale = 2; |
| 1239 | else if (megabytes <= 96) |
| 1240 | scale = 3; |
| 1241 | else if (megabytes <= 128) |
| 1242 | scale = 4; |
| 1243 | |
| 1244 | /* |
| 1245 | * scale number of entries |
| 1246 | */ |
| 1247 | bvec_pool_entries = megabytes * 2; |
| 1248 | if (bvec_pool_entries > 256) |
| 1249 | bvec_pool_entries = 256; |
| 1250 | |
| 1251 | fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale); |
| 1252 | if (!fs_bio_set) |
| 1253 | panic("bio: can't allocate bios\n"); |
| 1254 | |
| 1255 | bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES, |
| 1256 | bio_pair_alloc, bio_pair_free, NULL); |
| 1257 | if (!bio_split_pool) |
| 1258 | panic("bio: can't create split pool\n"); |
| 1259 | |
| 1260 | return 0; |
| 1261 | } |
| 1262 | |
| 1263 | subsys_initcall(init_bio); |
| 1264 | |
| 1265 | EXPORT_SYMBOL(bio_alloc); |
| 1266 | EXPORT_SYMBOL(bio_put); |
Peter Osterlund | 3676347 | 2005-09-06 15:16:42 -0700 | [diff] [blame] | 1267 | EXPORT_SYMBOL(bio_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1268 | EXPORT_SYMBOL(bio_endio); |
| 1269 | EXPORT_SYMBOL(bio_init); |
| 1270 | EXPORT_SYMBOL(__bio_clone); |
| 1271 | EXPORT_SYMBOL(bio_clone); |
| 1272 | EXPORT_SYMBOL(bio_phys_segments); |
| 1273 | EXPORT_SYMBOL(bio_hw_segments); |
| 1274 | EXPORT_SYMBOL(bio_add_page); |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 1275 | EXPORT_SYMBOL(bio_add_pc_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | EXPORT_SYMBOL(bio_get_nr_vecs); |
| 1277 | EXPORT_SYMBOL(bio_map_user); |
| 1278 | EXPORT_SYMBOL(bio_unmap_user); |
Mike Christie | df46b9a | 2005-06-20 14:04:44 +0200 | [diff] [blame] | 1279 | EXPORT_SYMBOL(bio_map_kern); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | EXPORT_SYMBOL(bio_pair_release); |
| 1281 | EXPORT_SYMBOL(bio_split); |
| 1282 | EXPORT_SYMBOL(bio_split_pool); |
| 1283 | EXPORT_SYMBOL(bio_copy_user); |
| 1284 | EXPORT_SYMBOL(bio_uncopy_user); |
| 1285 | EXPORT_SYMBOL(bioset_create); |
| 1286 | EXPORT_SYMBOL(bioset_free); |
| 1287 | EXPORT_SYMBOL(bio_alloc_bioset); |