| 1 | /* |
| 2 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> |
| 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/uio.h> |
| 23 | #include <linux/iocontext.h> |
| 24 | #include <linux/slab.h> |
| 25 | #include <linux/init.h> |
| 26 | #include <linux/kernel.h> |
| 27 | #include <linux/export.h> |
| 28 | #include <linux/mempool.h> |
| 29 | #include <linux/workqueue.h> |
| 30 | #include <linux/cgroup.h> |
| 31 | |
| 32 | #include <trace/events/block.h> |
| 33 | #include "blk.h" |
| 34 | |
| 35 | /* |
| 36 | * Test patch to inline a certain number of bi_io_vec's inside the bio |
| 37 | * itself, to shrink a bio data allocation from two mempool calls to one |
| 38 | */ |
| 39 | #define BIO_INLINE_VECS 4 |
| 40 | |
| 41 | /* |
| 42 | * if you change this list, also change bvec_alloc or things will |
| 43 | * break badly! cannot be bigger than what you can fit into an |
| 44 | * unsigned short |
| 45 | */ |
| 46 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } |
| 47 | static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { |
| 48 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
| 49 | }; |
| 50 | #undef BV |
| 51 | |
| 52 | /* |
| 53 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by |
| 54 | * IO code that does not need private memory pools. |
| 55 | */ |
| 56 | struct bio_set *fs_bio_set; |
| 57 | EXPORT_SYMBOL(fs_bio_set); |
| 58 | |
| 59 | /* |
| 60 | * Our slab pool management |
| 61 | */ |
| 62 | struct bio_slab { |
| 63 | struct kmem_cache *slab; |
| 64 | unsigned int slab_ref; |
| 65 | unsigned int slab_size; |
| 66 | char name[8]; |
| 67 | }; |
| 68 | static DEFINE_MUTEX(bio_slab_lock); |
| 69 | static struct bio_slab *bio_slabs; |
| 70 | static unsigned int bio_slab_nr, bio_slab_max; |
| 71 | |
| 72 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) |
| 73 | { |
| 74 | unsigned int sz = sizeof(struct bio) + extra_size; |
| 75 | struct kmem_cache *slab = NULL; |
| 76 | struct bio_slab *bslab, *new_bio_slabs; |
| 77 | unsigned int new_bio_slab_max; |
| 78 | unsigned int i, entry = -1; |
| 79 | |
| 80 | mutex_lock(&bio_slab_lock); |
| 81 | |
| 82 | i = 0; |
| 83 | while (i < bio_slab_nr) { |
| 84 | bslab = &bio_slabs[i]; |
| 85 | |
| 86 | if (!bslab->slab && entry == -1) |
| 87 | entry = i; |
| 88 | else if (bslab->slab_size == sz) { |
| 89 | slab = bslab->slab; |
| 90 | bslab->slab_ref++; |
| 91 | break; |
| 92 | } |
| 93 | i++; |
| 94 | } |
| 95 | |
| 96 | if (slab) |
| 97 | goto out_unlock; |
| 98 | |
| 99 | if (bio_slab_nr == bio_slab_max && entry == -1) { |
| 100 | new_bio_slab_max = bio_slab_max << 1; |
| 101 | new_bio_slabs = krealloc(bio_slabs, |
| 102 | new_bio_slab_max * sizeof(struct bio_slab), |
| 103 | GFP_KERNEL); |
| 104 | if (!new_bio_slabs) |
| 105 | goto out_unlock; |
| 106 | bio_slab_max = new_bio_slab_max; |
| 107 | bio_slabs = new_bio_slabs; |
| 108 | } |
| 109 | if (entry == -1) |
| 110 | entry = bio_slab_nr++; |
| 111 | |
| 112 | bslab = &bio_slabs[entry]; |
| 113 | |
| 114 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); |
| 115 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, |
| 116 | SLAB_HWCACHE_ALIGN, NULL); |
| 117 | if (!slab) |
| 118 | goto out_unlock; |
| 119 | |
| 120 | bslab->slab = slab; |
| 121 | bslab->slab_ref = 1; |
| 122 | bslab->slab_size = sz; |
| 123 | out_unlock: |
| 124 | mutex_unlock(&bio_slab_lock); |
| 125 | return slab; |
| 126 | } |
| 127 | |
| 128 | static void bio_put_slab(struct bio_set *bs) |
| 129 | { |
| 130 | struct bio_slab *bslab = NULL; |
| 131 | unsigned int i; |
| 132 | |
| 133 | mutex_lock(&bio_slab_lock); |
| 134 | |
| 135 | for (i = 0; i < bio_slab_nr; i++) { |
| 136 | if (bs->bio_slab == bio_slabs[i].slab) { |
| 137 | bslab = &bio_slabs[i]; |
| 138 | break; |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
| 143 | goto out; |
| 144 | |
| 145 | WARN_ON(!bslab->slab_ref); |
| 146 | |
| 147 | if (--bslab->slab_ref) |
| 148 | goto out; |
| 149 | |
| 150 | kmem_cache_destroy(bslab->slab); |
| 151 | bslab->slab = NULL; |
| 152 | |
| 153 | out: |
| 154 | mutex_unlock(&bio_slab_lock); |
| 155 | } |
| 156 | |
| 157 | unsigned int bvec_nr_vecs(unsigned short idx) |
| 158 | { |
| 159 | return bvec_slabs[idx].nr_vecs; |
| 160 | } |
| 161 | |
| 162 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
| 163 | { |
| 164 | if (!idx) |
| 165 | return; |
| 166 | idx--; |
| 167 | |
| 168 | BIO_BUG_ON(idx >= BVEC_POOL_NR); |
| 169 | |
| 170 | if (idx == BVEC_POOL_MAX) { |
| 171 | mempool_free(bv, pool); |
| 172 | } else { |
| 173 | struct biovec_slab *bvs = bvec_slabs + idx; |
| 174 | |
| 175 | kmem_cache_free(bvs->slab, bv); |
| 176 | } |
| 177 | } |
| 178 | |
| 179 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
| 180 | mempool_t *pool) |
| 181 | { |
| 182 | struct bio_vec *bvl; |
| 183 | |
| 184 | /* |
| 185 | * see comment near bvec_array define! |
| 186 | */ |
| 187 | switch (nr) { |
| 188 | case 1: |
| 189 | *idx = 0; |
| 190 | break; |
| 191 | case 2 ... 4: |
| 192 | *idx = 1; |
| 193 | break; |
| 194 | case 5 ... 16: |
| 195 | *idx = 2; |
| 196 | break; |
| 197 | case 17 ... 64: |
| 198 | *idx = 3; |
| 199 | break; |
| 200 | case 65 ... 128: |
| 201 | *idx = 4; |
| 202 | break; |
| 203 | case 129 ... BIO_MAX_PAGES: |
| 204 | *idx = 5; |
| 205 | break; |
| 206 | default: |
| 207 | return NULL; |
| 208 | } |
| 209 | |
| 210 | /* |
| 211 | * idx now points to the pool we want to allocate from. only the |
| 212 | * 1-vec entry pool is mempool backed. |
| 213 | */ |
| 214 | if (*idx == BVEC_POOL_MAX) { |
| 215 | fallback: |
| 216 | bvl = mempool_alloc(pool, gfp_mask); |
| 217 | } else { |
| 218 | struct biovec_slab *bvs = bvec_slabs + *idx; |
| 219 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); |
| 220 | |
| 221 | /* |
| 222 | * Make this allocation restricted and don't dump info on |
| 223 | * allocation failures, since we'll fallback to the mempool |
| 224 | * in case of failure. |
| 225 | */ |
| 226 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
| 227 | |
| 228 | /* |
| 229 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM |
| 230 | * is set, retry with the 1-entry mempool |
| 231 | */ |
| 232 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
| 233 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { |
| 234 | *idx = BVEC_POOL_MAX; |
| 235 | goto fallback; |
| 236 | } |
| 237 | } |
| 238 | |
| 239 | (*idx)++; |
| 240 | return bvl; |
| 241 | } |
| 242 | |
| 243 | void bio_uninit(struct bio *bio) |
| 244 | { |
| 245 | bio_disassociate_task(bio); |
| 246 | } |
| 247 | EXPORT_SYMBOL(bio_uninit); |
| 248 | |
| 249 | static void bio_free(struct bio *bio) |
| 250 | { |
| 251 | struct bio_set *bs = bio->bi_pool; |
| 252 | void *p; |
| 253 | |
| 254 | bio_uninit(bio); |
| 255 | |
| 256 | if (bs) { |
| 257 | bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); |
| 258 | |
| 259 | /* |
| 260 | * If we have front padding, adjust the bio pointer before freeing |
| 261 | */ |
| 262 | p = bio; |
| 263 | p -= bs->front_pad; |
| 264 | |
| 265 | mempool_free(p, bs->bio_pool); |
| 266 | } else { |
| 267 | /* Bio was allocated by bio_kmalloc() */ |
| 268 | kfree(bio); |
| 269 | } |
| 270 | } |
| 271 | |
| 272 | /* |
| 273 | * Users of this function have their own bio allocation. Subsequently, |
| 274 | * they must remember to pair any call to bio_init() with bio_uninit() |
| 275 | * when IO has completed, or when the bio is released. |
| 276 | */ |
| 277 | void bio_init(struct bio *bio, struct bio_vec *table, |
| 278 | unsigned short max_vecs) |
| 279 | { |
| 280 | memset(bio, 0, sizeof(*bio)); |
| 281 | atomic_set(&bio->__bi_remaining, 1); |
| 282 | atomic_set(&bio->__bi_cnt, 1); |
| 283 | |
| 284 | bio->bi_io_vec = table; |
| 285 | bio->bi_max_vecs = max_vecs; |
| 286 | } |
| 287 | EXPORT_SYMBOL(bio_init); |
| 288 | |
| 289 | /** |
| 290 | * bio_reset - reinitialize a bio |
| 291 | * @bio: bio to reset |
| 292 | * |
| 293 | * Description: |
| 294 | * After calling bio_reset(), @bio will be in the same state as a freshly |
| 295 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are |
| 296 | * preserved are the ones that are initialized by bio_alloc_bioset(). See |
| 297 | * comment in struct bio. |
| 298 | */ |
| 299 | void bio_reset(struct bio *bio) |
| 300 | { |
| 301 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); |
| 302 | |
| 303 | bio_uninit(bio); |
| 304 | |
| 305 | memset(bio, 0, BIO_RESET_BYTES); |
| 306 | bio->bi_flags = flags; |
| 307 | atomic_set(&bio->__bi_remaining, 1); |
| 308 | } |
| 309 | EXPORT_SYMBOL(bio_reset); |
| 310 | |
| 311 | static struct bio *__bio_chain_endio(struct bio *bio) |
| 312 | { |
| 313 | struct bio *parent = bio->bi_private; |
| 314 | |
| 315 | if (!parent->bi_status) |
| 316 | parent->bi_status = bio->bi_status; |
| 317 | bio_put(bio); |
| 318 | return parent; |
| 319 | } |
| 320 | |
| 321 | static void bio_chain_endio(struct bio *bio) |
| 322 | { |
| 323 | bio_endio(__bio_chain_endio(bio)); |
| 324 | } |
| 325 | |
| 326 | /** |
| 327 | * bio_chain - chain bio completions |
| 328 | * @bio: the target bio |
| 329 | * @parent: the @bio's parent bio |
| 330 | * |
| 331 | * The caller won't have a bi_end_io called when @bio completes - instead, |
| 332 | * @parent's bi_end_io won't be called until both @parent and @bio have |
| 333 | * completed; the chained bio will also be freed when it completes. |
| 334 | * |
| 335 | * The caller must not set bi_private or bi_end_io in @bio. |
| 336 | */ |
| 337 | void bio_chain(struct bio *bio, struct bio *parent) |
| 338 | { |
| 339 | BUG_ON(bio->bi_private || bio->bi_end_io); |
| 340 | |
| 341 | bio->bi_private = parent; |
| 342 | bio->bi_end_io = bio_chain_endio; |
| 343 | bio_inc_remaining(parent); |
| 344 | } |
| 345 | EXPORT_SYMBOL(bio_chain); |
| 346 | |
| 347 | static void bio_alloc_rescue(struct work_struct *work) |
| 348 | { |
| 349 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); |
| 350 | struct bio *bio; |
| 351 | |
| 352 | while (1) { |
| 353 | spin_lock(&bs->rescue_lock); |
| 354 | bio = bio_list_pop(&bs->rescue_list); |
| 355 | spin_unlock(&bs->rescue_lock); |
| 356 | |
| 357 | if (!bio) |
| 358 | break; |
| 359 | |
| 360 | generic_make_request(bio); |
| 361 | } |
| 362 | } |
| 363 | |
| 364 | static void punt_bios_to_rescuer(struct bio_set *bs) |
| 365 | { |
| 366 | struct bio_list punt, nopunt; |
| 367 | struct bio *bio; |
| 368 | |
| 369 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
| 370 | return; |
| 371 | /* |
| 372 | * In order to guarantee forward progress we must punt only bios that |
| 373 | * were allocated from this bio_set; otherwise, if there was a bio on |
| 374 | * there for a stacking driver higher up in the stack, processing it |
| 375 | * could require allocating bios from this bio_set, and doing that from |
| 376 | * our own rescuer would be bad. |
| 377 | * |
| 378 | * Since bio lists are singly linked, pop them all instead of trying to |
| 379 | * remove from the middle of the list: |
| 380 | */ |
| 381 | |
| 382 | bio_list_init(&punt); |
| 383 | bio_list_init(&nopunt); |
| 384 | |
| 385 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
| 386 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
| 387 | current->bio_list[0] = nopunt; |
| 388 | |
| 389 | bio_list_init(&nopunt); |
| 390 | while ((bio = bio_list_pop(¤t->bio_list[1]))) |
| 391 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
| 392 | current->bio_list[1] = nopunt; |
| 393 | |
| 394 | spin_lock(&bs->rescue_lock); |
| 395 | bio_list_merge(&bs->rescue_list, &punt); |
| 396 | spin_unlock(&bs->rescue_lock); |
| 397 | |
| 398 | queue_work(bs->rescue_workqueue, &bs->rescue_work); |
| 399 | } |
| 400 | |
| 401 | /** |
| 402 | * bio_alloc_bioset - allocate a bio for I/O |
| 403 | * @gfp_mask: the GFP_* mask given to the slab allocator |
| 404 | * @nr_iovecs: number of iovecs to pre-allocate |
| 405 | * @bs: the bio_set to allocate from. |
| 406 | * |
| 407 | * Description: |
| 408 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
| 409 | * backed by the @bs's mempool. |
| 410 | * |
| 411 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will |
| 412 | * always be able to allocate a bio. This is due to the mempool guarantees. |
| 413 | * To make this work, callers must never allocate more than 1 bio at a time |
| 414 | * from this pool. Callers that need to allocate more than 1 bio must always |
| 415 | * submit the previously allocated bio for IO before attempting to allocate |
| 416 | * a new one. Failure to do so can cause deadlocks under memory pressure. |
| 417 | * |
| 418 | * Note that when running under generic_make_request() (i.e. any block |
| 419 | * driver), bios are not submitted until after you return - see the code in |
| 420 | * generic_make_request() that converts recursion into iteration, to prevent |
| 421 | * stack overflows. |
| 422 | * |
| 423 | * This would normally mean allocating multiple bios under |
| 424 | * generic_make_request() would be susceptible to deadlocks, but we have |
| 425 | * deadlock avoidance code that resubmits any blocked bios from a rescuer |
| 426 | * thread. |
| 427 | * |
| 428 | * However, we do not guarantee forward progress for allocations from other |
| 429 | * mempools. Doing multiple allocations from the same mempool under |
| 430 | * generic_make_request() should be avoided - instead, use bio_set's front_pad |
| 431 | * for per bio allocations. |
| 432 | * |
| 433 | * RETURNS: |
| 434 | * Pointer to new bio on success, NULL on failure. |
| 435 | */ |
| 436 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, |
| 437 | struct bio_set *bs) |
| 438 | { |
| 439 | gfp_t saved_gfp = gfp_mask; |
| 440 | unsigned front_pad; |
| 441 | unsigned inline_vecs; |
| 442 | struct bio_vec *bvl = NULL; |
| 443 | struct bio *bio; |
| 444 | void *p; |
| 445 | |
| 446 | if (!bs) { |
| 447 | if (nr_iovecs > UIO_MAXIOV) |
| 448 | return NULL; |
| 449 | |
| 450 | p = kmalloc(sizeof(struct bio) + |
| 451 | nr_iovecs * sizeof(struct bio_vec), |
| 452 | gfp_mask); |
| 453 | front_pad = 0; |
| 454 | inline_vecs = nr_iovecs; |
| 455 | } else { |
| 456 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
| 457 | if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0)) |
| 458 | return NULL; |
| 459 | /* |
| 460 | * generic_make_request() converts recursion to iteration; this |
| 461 | * means if we're running beneath it, any bios we allocate and |
| 462 | * submit will not be submitted (and thus freed) until after we |
| 463 | * return. |
| 464 | * |
| 465 | * This exposes us to a potential deadlock if we allocate |
| 466 | * multiple bios from the same bio_set() while running |
| 467 | * underneath generic_make_request(). If we were to allocate |
| 468 | * multiple bios (say a stacking block driver that was splitting |
| 469 | * bios), we would deadlock if we exhausted the mempool's |
| 470 | * reserve. |
| 471 | * |
| 472 | * We solve this, and guarantee forward progress, with a rescuer |
| 473 | * workqueue per bio_set. If we go to allocate and there are |
| 474 | * bios on current->bio_list, we first try the allocation |
| 475 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those |
| 476 | * bios we would be blocking to the rescuer workqueue before |
| 477 | * we retry with the original gfp_flags. |
| 478 | */ |
| 479 | |
| 480 | if (current->bio_list && |
| 481 | (!bio_list_empty(¤t->bio_list[0]) || |
| 482 | !bio_list_empty(¤t->bio_list[1])) && |
| 483 | bs->rescue_workqueue) |
| 484 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; |
| 485 | |
| 486 | p = mempool_alloc(bs->bio_pool, gfp_mask); |
| 487 | if (!p && gfp_mask != saved_gfp) { |
| 488 | punt_bios_to_rescuer(bs); |
| 489 | gfp_mask = saved_gfp; |
| 490 | p = mempool_alloc(bs->bio_pool, gfp_mask); |
| 491 | } |
| 492 | |
| 493 | front_pad = bs->front_pad; |
| 494 | inline_vecs = BIO_INLINE_VECS; |
| 495 | } |
| 496 | |
| 497 | if (unlikely(!p)) |
| 498 | return NULL; |
| 499 | |
| 500 | bio = p + front_pad; |
| 501 | bio_init(bio, NULL, 0); |
| 502 | |
| 503 | if (nr_iovecs > inline_vecs) { |
| 504 | unsigned long idx = 0; |
| 505 | |
| 506 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); |
| 507 | if (!bvl && gfp_mask != saved_gfp) { |
| 508 | punt_bios_to_rescuer(bs); |
| 509 | gfp_mask = saved_gfp; |
| 510 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); |
| 511 | } |
| 512 | |
| 513 | if (unlikely(!bvl)) |
| 514 | goto err_free; |
| 515 | |
| 516 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; |
| 517 | } else if (nr_iovecs) { |
| 518 | bvl = bio->bi_inline_vecs; |
| 519 | } |
| 520 | |
| 521 | bio->bi_pool = bs; |
| 522 | bio->bi_max_vecs = nr_iovecs; |
| 523 | bio->bi_io_vec = bvl; |
| 524 | return bio; |
| 525 | |
| 526 | err_free: |
| 527 | mempool_free(p, bs->bio_pool); |
| 528 | return NULL; |
| 529 | } |
| 530 | EXPORT_SYMBOL(bio_alloc_bioset); |
| 531 | |
| 532 | void zero_fill_bio(struct bio *bio) |
| 533 | { |
| 534 | unsigned long flags; |
| 535 | struct bio_vec bv; |
| 536 | struct bvec_iter iter; |
| 537 | |
| 538 | bio_for_each_segment(bv, bio, iter) { |
| 539 | char *data = bvec_kmap_irq(&bv, &flags); |
| 540 | memset(data, 0, bv.bv_len); |
| 541 | flush_dcache_page(bv.bv_page); |
| 542 | bvec_kunmap_irq(data, &flags); |
| 543 | } |
| 544 | } |
| 545 | EXPORT_SYMBOL(zero_fill_bio); |
| 546 | |
| 547 | /** |
| 548 | * bio_put - release a reference to a bio |
| 549 | * @bio: bio to release reference to |
| 550 | * |
| 551 | * Description: |
| 552 | * Put a reference to a &struct bio, either one you have gotten with |
| 553 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
| 554 | **/ |
| 555 | void bio_put(struct bio *bio) |
| 556 | { |
| 557 | if (!bio_flagged(bio, BIO_REFFED)) |
| 558 | bio_free(bio); |
| 559 | else { |
| 560 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); |
| 561 | |
| 562 | /* |
| 563 | * last put frees it |
| 564 | */ |
| 565 | if (atomic_dec_and_test(&bio->__bi_cnt)) |
| 566 | bio_free(bio); |
| 567 | } |
| 568 | } |
| 569 | EXPORT_SYMBOL(bio_put); |
| 570 | |
| 571 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
| 572 | { |
| 573 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
| 574 | blk_recount_segments(q, bio); |
| 575 | |
| 576 | return bio->bi_phys_segments; |
| 577 | } |
| 578 | EXPORT_SYMBOL(bio_phys_segments); |
| 579 | |
| 580 | /** |
| 581 | * __bio_clone_fast - clone a bio that shares the original bio's biovec |
| 582 | * @bio: destination bio |
| 583 | * @bio_src: bio to clone |
| 584 | * |
| 585 | * Clone a &bio. Caller will own the returned bio, but not |
| 586 | * the actual data it points to. Reference count of returned |
| 587 | * bio will be one. |
| 588 | * |
| 589 | * Caller must ensure that @bio_src is not freed before @bio. |
| 590 | */ |
| 591 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) |
| 592 | { |
| 593 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); |
| 594 | |
| 595 | /* |
| 596 | * most users will be overriding ->bi_disk with a new target, |
| 597 | * so we don't set nor calculate new physical/hw segment counts here |
| 598 | */ |
| 599 | bio->bi_disk = bio_src->bi_disk; |
| 600 | bio->bi_partno = bio_src->bi_partno; |
| 601 | bio_set_flag(bio, BIO_CLONED); |
| 602 | bio->bi_opf = bio_src->bi_opf; |
| 603 | bio->bi_write_hint = bio_src->bi_write_hint; |
| 604 | bio->bi_iter = bio_src->bi_iter; |
| 605 | bio->bi_io_vec = bio_src->bi_io_vec; |
| 606 | |
| 607 | bio_clone_blkcg_association(bio, bio_src); |
| 608 | } |
| 609 | EXPORT_SYMBOL(__bio_clone_fast); |
| 610 | |
| 611 | /** |
| 612 | * bio_clone_fast - clone a bio that shares the original bio's biovec |
| 613 | * @bio: bio to clone |
| 614 | * @gfp_mask: allocation priority |
| 615 | * @bs: bio_set to allocate from |
| 616 | * |
| 617 | * Like __bio_clone_fast, only also allocates the returned bio |
| 618 | */ |
| 619 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) |
| 620 | { |
| 621 | struct bio *b; |
| 622 | |
| 623 | b = bio_alloc_bioset(gfp_mask, 0, bs); |
| 624 | if (!b) |
| 625 | return NULL; |
| 626 | |
| 627 | __bio_clone_fast(b, bio); |
| 628 | |
| 629 | if (bio_integrity(bio)) { |
| 630 | int ret; |
| 631 | |
| 632 | ret = bio_integrity_clone(b, bio, gfp_mask); |
| 633 | |
| 634 | if (ret < 0) { |
| 635 | bio_put(b); |
| 636 | return NULL; |
| 637 | } |
| 638 | } |
| 639 | |
| 640 | return b; |
| 641 | } |
| 642 | EXPORT_SYMBOL(bio_clone_fast); |
| 643 | |
| 644 | /** |
| 645 | * bio_clone_bioset - clone a bio |
| 646 | * @bio_src: bio to clone |
| 647 | * @gfp_mask: allocation priority |
| 648 | * @bs: bio_set to allocate from |
| 649 | * |
| 650 | * Clone bio. Caller will own the returned bio, but not the actual data it |
| 651 | * points to. Reference count of returned bio will be one. |
| 652 | */ |
| 653 | struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, |
| 654 | struct bio_set *bs) |
| 655 | { |
| 656 | struct bvec_iter iter; |
| 657 | struct bio_vec bv; |
| 658 | struct bio *bio; |
| 659 | |
| 660 | /* |
| 661 | * Pre immutable biovecs, __bio_clone() used to just do a memcpy from |
| 662 | * bio_src->bi_io_vec to bio->bi_io_vec. |
| 663 | * |
| 664 | * We can't do that anymore, because: |
| 665 | * |
| 666 | * - The point of cloning the biovec is to produce a bio with a biovec |
| 667 | * the caller can modify: bi_idx and bi_bvec_done should be 0. |
| 668 | * |
| 669 | * - The original bio could've had more than BIO_MAX_PAGES biovecs; if |
| 670 | * we tried to clone the whole thing bio_alloc_bioset() would fail. |
| 671 | * But the clone should succeed as long as the number of biovecs we |
| 672 | * actually need to allocate is fewer than BIO_MAX_PAGES. |
| 673 | * |
| 674 | * - Lastly, bi_vcnt should not be looked at or relied upon by code |
| 675 | * that does not own the bio - reason being drivers don't use it for |
| 676 | * iterating over the biovec anymore, so expecting it to be kept up |
| 677 | * to date (i.e. for clones that share the parent biovec) is just |
| 678 | * asking for trouble and would force extra work on |
| 679 | * __bio_clone_fast() anyways. |
| 680 | */ |
| 681 | |
| 682 | bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs); |
| 683 | if (!bio) |
| 684 | return NULL; |
| 685 | bio->bi_disk = bio_src->bi_disk; |
| 686 | bio->bi_opf = bio_src->bi_opf; |
| 687 | bio->bi_write_hint = bio_src->bi_write_hint; |
| 688 | bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; |
| 689 | bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; |
| 690 | |
| 691 | switch (bio_op(bio)) { |
| 692 | case REQ_OP_DISCARD: |
| 693 | case REQ_OP_SECURE_ERASE: |
| 694 | case REQ_OP_WRITE_ZEROES: |
| 695 | break; |
| 696 | case REQ_OP_WRITE_SAME: |
| 697 | bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; |
| 698 | break; |
| 699 | default: |
| 700 | bio_for_each_segment(bv, bio_src, iter) |
| 701 | bio->bi_io_vec[bio->bi_vcnt++] = bv; |
| 702 | break; |
| 703 | } |
| 704 | |
| 705 | if (bio_integrity(bio_src)) { |
| 706 | int ret; |
| 707 | |
| 708 | ret = bio_integrity_clone(bio, bio_src, gfp_mask); |
| 709 | if (ret < 0) { |
| 710 | bio_put(bio); |
| 711 | return NULL; |
| 712 | } |
| 713 | } |
| 714 | |
| 715 | bio_clone_blkcg_association(bio, bio_src); |
| 716 | |
| 717 | return bio; |
| 718 | } |
| 719 | EXPORT_SYMBOL(bio_clone_bioset); |
| 720 | |
| 721 | /** |
| 722 | * bio_add_pc_page - attempt to add page to bio |
| 723 | * @q: the target queue |
| 724 | * @bio: destination bio |
| 725 | * @page: page to add |
| 726 | * @len: vec entry length |
| 727 | * @offset: vec entry offset |
| 728 | * |
| 729 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
| 730 | * number of reasons, such as the bio being full or target block device |
| 731 | * limitations. The target block device must allow bio's up to PAGE_SIZE, |
| 732 | * so it is always possible to add a single page to an empty bio. |
| 733 | * |
| 734 | * This should only be used by REQ_PC bios. |
| 735 | */ |
| 736 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page |
| 737 | *page, unsigned int len, unsigned int offset) |
| 738 | { |
| 739 | int retried_segments = 0; |
| 740 | struct bio_vec *bvec; |
| 741 | |
| 742 | /* |
| 743 | * cloned bio must not modify vec list |
| 744 | */ |
| 745 | if (unlikely(bio_flagged(bio, BIO_CLONED))) |
| 746 | return 0; |
| 747 | |
| 748 | if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) |
| 749 | return 0; |
| 750 | |
| 751 | /* |
| 752 | * For filesystems with a blocksize smaller than the pagesize |
| 753 | * we will often be called with the same page as last time and |
| 754 | * a consecutive offset. Optimize this special case. |
| 755 | */ |
| 756 | if (bio->bi_vcnt > 0) { |
| 757 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
| 758 | |
| 759 | if (page == prev->bv_page && |
| 760 | offset == prev->bv_offset + prev->bv_len) { |
| 761 | prev->bv_len += len; |
| 762 | bio->bi_iter.bi_size += len; |
| 763 | goto done; |
| 764 | } |
| 765 | |
| 766 | /* |
| 767 | * If the queue doesn't support SG gaps and adding this |
| 768 | * offset would create a gap, disallow it. |
| 769 | */ |
| 770 | if (bvec_gap_to_prev(q, prev, offset)) |
| 771 | return 0; |
| 772 | } |
| 773 | |
| 774 | if (bio->bi_vcnt >= bio->bi_max_vecs) |
| 775 | return 0; |
| 776 | |
| 777 | /* |
| 778 | * setup the new entry, we might clear it again later if we |
| 779 | * cannot add the page |
| 780 | */ |
| 781 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
| 782 | bvec->bv_page = page; |
| 783 | bvec->bv_len = len; |
| 784 | bvec->bv_offset = offset; |
| 785 | bio->bi_vcnt++; |
| 786 | bio->bi_phys_segments++; |
| 787 | bio->bi_iter.bi_size += len; |
| 788 | |
| 789 | /* |
| 790 | * Perform a recount if the number of segments is greater |
| 791 | * than queue_max_segments(q). |
| 792 | */ |
| 793 | |
| 794 | while (bio->bi_phys_segments > queue_max_segments(q)) { |
| 795 | |
| 796 | if (retried_segments) |
| 797 | goto failed; |
| 798 | |
| 799 | retried_segments = 1; |
| 800 | blk_recount_segments(q, bio); |
| 801 | } |
| 802 | |
| 803 | /* If we may be able to merge these biovecs, force a recount */ |
| 804 | if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
| 805 | bio_clear_flag(bio, BIO_SEG_VALID); |
| 806 | |
| 807 | done: |
| 808 | return len; |
| 809 | |
| 810 | failed: |
| 811 | bvec->bv_page = NULL; |
| 812 | bvec->bv_len = 0; |
| 813 | bvec->bv_offset = 0; |
| 814 | bio->bi_vcnt--; |
| 815 | bio->bi_iter.bi_size -= len; |
| 816 | blk_recount_segments(q, bio); |
| 817 | return 0; |
| 818 | } |
| 819 | EXPORT_SYMBOL(bio_add_pc_page); |
| 820 | |
| 821 | /** |
| 822 | * bio_add_page - attempt to add page to bio |
| 823 | * @bio: destination bio |
| 824 | * @page: page to add |
| 825 | * @len: vec entry length |
| 826 | * @offset: vec entry offset |
| 827 | * |
| 828 | * Attempt to add a page to the bio_vec maplist. This will only fail |
| 829 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
| 830 | */ |
| 831 | int bio_add_page(struct bio *bio, struct page *page, |
| 832 | unsigned int len, unsigned int offset) |
| 833 | { |
| 834 | struct bio_vec *bv; |
| 835 | |
| 836 | /* |
| 837 | * cloned bio must not modify vec list |
| 838 | */ |
| 839 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
| 840 | return 0; |
| 841 | |
| 842 | /* |
| 843 | * For filesystems with a blocksize smaller than the pagesize |
| 844 | * we will often be called with the same page as last time and |
| 845 | * a consecutive offset. Optimize this special case. |
| 846 | */ |
| 847 | if (bio->bi_vcnt > 0) { |
| 848 | bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
| 849 | |
| 850 | if (page == bv->bv_page && |
| 851 | offset == bv->bv_offset + bv->bv_len) { |
| 852 | bv->bv_len += len; |
| 853 | goto done; |
| 854 | } |
| 855 | } |
| 856 | |
| 857 | if (bio->bi_vcnt >= bio->bi_max_vecs) |
| 858 | return 0; |
| 859 | |
| 860 | bv = &bio->bi_io_vec[bio->bi_vcnt]; |
| 861 | bv->bv_page = page; |
| 862 | bv->bv_len = len; |
| 863 | bv->bv_offset = offset; |
| 864 | |
| 865 | bio->bi_vcnt++; |
| 866 | done: |
| 867 | bio->bi_iter.bi_size += len; |
| 868 | return len; |
| 869 | } |
| 870 | EXPORT_SYMBOL(bio_add_page); |
| 871 | |
| 872 | /** |
| 873 | * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
| 874 | * @bio: bio to add pages to |
| 875 | * @iter: iov iterator describing the region to be mapped |
| 876 | * |
| 877 | * Pins as many pages from *iter and appends them to @bio's bvec array. The |
| 878 | * pages will have to be released using put_page() when done. |
| 879 | */ |
| 880 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
| 881 | { |
| 882 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
| 883 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
| 884 | struct page **pages = (struct page **)bv; |
| 885 | size_t offset, diff; |
| 886 | ssize_t size; |
| 887 | |
| 888 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); |
| 889 | if (unlikely(size <= 0)) |
| 890 | return size ? size : -EFAULT; |
| 891 | nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE; |
| 892 | |
| 893 | /* |
| 894 | * Deep magic below: We need to walk the pinned pages backwards |
| 895 | * because we are abusing the space allocated for the bio_vecs |
| 896 | * for the page array. Because the bio_vecs are larger than the |
| 897 | * page pointers by definition this will always work. But it also |
| 898 | * means we can't use bio_add_page, so any changes to it's semantics |
| 899 | * need to be reflected here as well. |
| 900 | */ |
| 901 | bio->bi_iter.bi_size += size; |
| 902 | bio->bi_vcnt += nr_pages; |
| 903 | |
| 904 | diff = (nr_pages * PAGE_SIZE - offset) - size; |
| 905 | while (nr_pages--) { |
| 906 | bv[nr_pages].bv_page = pages[nr_pages]; |
| 907 | bv[nr_pages].bv_len = PAGE_SIZE; |
| 908 | bv[nr_pages].bv_offset = 0; |
| 909 | } |
| 910 | |
| 911 | bv[0].bv_offset += offset; |
| 912 | bv[0].bv_len -= offset; |
| 913 | if (diff) |
| 914 | bv[bio->bi_vcnt - 1].bv_len -= diff; |
| 915 | |
| 916 | iov_iter_advance(iter, size); |
| 917 | return 0; |
| 918 | } |
| 919 | EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); |
| 920 | |
| 921 | static void submit_bio_wait_endio(struct bio *bio) |
| 922 | { |
| 923 | complete(bio->bi_private); |
| 924 | } |
| 925 | |
| 926 | /** |
| 927 | * submit_bio_wait - submit a bio, and wait until it completes |
| 928 | * @bio: The &struct bio which describes the I/O |
| 929 | * |
| 930 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from |
| 931 | * bio_endio() on failure. |
| 932 | * |
| 933 | * WARNING: Unlike to how submit_bio() is usually used, this function does not |
| 934 | * result in bio reference to be consumed. The caller must drop the reference |
| 935 | * on his own. |
| 936 | */ |
| 937 | int submit_bio_wait(struct bio *bio) |
| 938 | { |
| 939 | DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); |
| 940 | |
| 941 | bio->bi_private = &done; |
| 942 | bio->bi_end_io = submit_bio_wait_endio; |
| 943 | bio->bi_opf |= REQ_SYNC; |
| 944 | submit_bio(bio); |
| 945 | wait_for_completion_io(&done); |
| 946 | |
| 947 | return blk_status_to_errno(bio->bi_status); |
| 948 | } |
| 949 | EXPORT_SYMBOL(submit_bio_wait); |
| 950 | |
| 951 | /** |
| 952 | * bio_advance - increment/complete a bio by some number of bytes |
| 953 | * @bio: bio to advance |
| 954 | * @bytes: number of bytes to complete |
| 955 | * |
| 956 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to |
| 957 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will |
| 958 | * be updated on the last bvec as well. |
| 959 | * |
| 960 | * @bio will then represent the remaining, uncompleted portion of the io. |
| 961 | */ |
| 962 | void bio_advance(struct bio *bio, unsigned bytes) |
| 963 | { |
| 964 | if (bio_integrity(bio)) |
| 965 | bio_integrity_advance(bio, bytes); |
| 966 | |
| 967 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
| 968 | } |
| 969 | EXPORT_SYMBOL(bio_advance); |
| 970 | |
| 971 | /** |
| 972 | * bio_alloc_pages - allocates a single page for each bvec in a bio |
| 973 | * @bio: bio to allocate pages for |
| 974 | * @gfp_mask: flags for allocation |
| 975 | * |
| 976 | * Allocates pages up to @bio->bi_vcnt. |
| 977 | * |
| 978 | * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are |
| 979 | * freed. |
| 980 | */ |
| 981 | int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) |
| 982 | { |
| 983 | int i; |
| 984 | struct bio_vec *bv; |
| 985 | |
| 986 | bio_for_each_segment_all(bv, bio, i) { |
| 987 | bv->bv_page = alloc_page(gfp_mask); |
| 988 | if (!bv->bv_page) { |
| 989 | while (--bv >= bio->bi_io_vec) |
| 990 | __free_page(bv->bv_page); |
| 991 | return -ENOMEM; |
| 992 | } |
| 993 | } |
| 994 | |
| 995 | return 0; |
| 996 | } |
| 997 | EXPORT_SYMBOL(bio_alloc_pages); |
| 998 | |
| 999 | /** |
| 1000 | * bio_copy_data - copy contents of data buffers from one chain of bios to |
| 1001 | * another |
| 1002 | * @src: source bio list |
| 1003 | * @dst: destination bio list |
| 1004 | * |
| 1005 | * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats |
| 1006 | * @src and @dst as linked lists of bios. |
| 1007 | * |
| 1008 | * Stops when it reaches the end of either @src or @dst - that is, copies |
| 1009 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). |
| 1010 | */ |
| 1011 | void bio_copy_data(struct bio *dst, struct bio *src) |
| 1012 | { |
| 1013 | struct bvec_iter src_iter, dst_iter; |
| 1014 | struct bio_vec src_bv, dst_bv; |
| 1015 | void *src_p, *dst_p; |
| 1016 | unsigned bytes; |
| 1017 | |
| 1018 | src_iter = src->bi_iter; |
| 1019 | dst_iter = dst->bi_iter; |
| 1020 | |
| 1021 | while (1) { |
| 1022 | if (!src_iter.bi_size) { |
| 1023 | src = src->bi_next; |
| 1024 | if (!src) |
| 1025 | break; |
| 1026 | |
| 1027 | src_iter = src->bi_iter; |
| 1028 | } |
| 1029 | |
| 1030 | if (!dst_iter.bi_size) { |
| 1031 | dst = dst->bi_next; |
| 1032 | if (!dst) |
| 1033 | break; |
| 1034 | |
| 1035 | dst_iter = dst->bi_iter; |
| 1036 | } |
| 1037 | |
| 1038 | src_bv = bio_iter_iovec(src, src_iter); |
| 1039 | dst_bv = bio_iter_iovec(dst, dst_iter); |
| 1040 | |
| 1041 | bytes = min(src_bv.bv_len, dst_bv.bv_len); |
| 1042 | |
| 1043 | src_p = kmap_atomic(src_bv.bv_page); |
| 1044 | dst_p = kmap_atomic(dst_bv.bv_page); |
| 1045 | |
| 1046 | memcpy(dst_p + dst_bv.bv_offset, |
| 1047 | src_p + src_bv.bv_offset, |
| 1048 | bytes); |
| 1049 | |
| 1050 | kunmap_atomic(dst_p); |
| 1051 | kunmap_atomic(src_p); |
| 1052 | |
| 1053 | bio_advance_iter(src, &src_iter, bytes); |
| 1054 | bio_advance_iter(dst, &dst_iter, bytes); |
| 1055 | } |
| 1056 | } |
| 1057 | EXPORT_SYMBOL(bio_copy_data); |
| 1058 | |
| 1059 | struct bio_map_data { |
| 1060 | int is_our_pages; |
| 1061 | struct iov_iter iter; |
| 1062 | struct iovec iov[]; |
| 1063 | }; |
| 1064 | |
| 1065 | static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, |
| 1066 | gfp_t gfp_mask) |
| 1067 | { |
| 1068 | struct bio_map_data *bmd; |
| 1069 | if (data->nr_segs > UIO_MAXIOV) |
| 1070 | return NULL; |
| 1071 | |
| 1072 | bmd = kmalloc(sizeof(struct bio_map_data) + |
| 1073 | sizeof(struct iovec) * data->nr_segs, gfp_mask); |
| 1074 | if (!bmd) |
| 1075 | return NULL; |
| 1076 | memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); |
| 1077 | bmd->iter = *data; |
| 1078 | bmd->iter.iov = bmd->iov; |
| 1079 | return bmd; |
| 1080 | } |
| 1081 | |
| 1082 | /** |
| 1083 | * bio_copy_from_iter - copy all pages from iov_iter to bio |
| 1084 | * @bio: The &struct bio which describes the I/O as destination |
| 1085 | * @iter: iov_iter as source |
| 1086 | * |
| 1087 | * Copy all pages from iov_iter to bio. |
| 1088 | * Returns 0 on success, or error on failure. |
| 1089 | */ |
| 1090 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) |
| 1091 | { |
| 1092 | int i; |
| 1093 | struct bio_vec *bvec; |
| 1094 | |
| 1095 | bio_for_each_segment_all(bvec, bio, i) { |
| 1096 | ssize_t ret; |
| 1097 | |
| 1098 | ret = copy_page_from_iter(bvec->bv_page, |
| 1099 | bvec->bv_offset, |
| 1100 | bvec->bv_len, |
| 1101 | iter); |
| 1102 | |
| 1103 | if (!iov_iter_count(iter)) |
| 1104 | break; |
| 1105 | |
| 1106 | if (ret < bvec->bv_len) |
| 1107 | return -EFAULT; |
| 1108 | } |
| 1109 | |
| 1110 | return 0; |
| 1111 | } |
| 1112 | |
| 1113 | /** |
| 1114 | * bio_copy_to_iter - copy all pages from bio to iov_iter |
| 1115 | * @bio: The &struct bio which describes the I/O as source |
| 1116 | * @iter: iov_iter as destination |
| 1117 | * |
| 1118 | * Copy all pages from bio to iov_iter. |
| 1119 | * Returns 0 on success, or error on failure. |
| 1120 | */ |
| 1121 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) |
| 1122 | { |
| 1123 | int i; |
| 1124 | struct bio_vec *bvec; |
| 1125 | |
| 1126 | bio_for_each_segment_all(bvec, bio, i) { |
| 1127 | ssize_t ret; |
| 1128 | |
| 1129 | ret = copy_page_to_iter(bvec->bv_page, |
| 1130 | bvec->bv_offset, |
| 1131 | bvec->bv_len, |
| 1132 | &iter); |
| 1133 | |
| 1134 | if (!iov_iter_count(&iter)) |
| 1135 | break; |
| 1136 | |
| 1137 | if (ret < bvec->bv_len) |
| 1138 | return -EFAULT; |
| 1139 | } |
| 1140 | |
| 1141 | return 0; |
| 1142 | } |
| 1143 | |
| 1144 | void bio_free_pages(struct bio *bio) |
| 1145 | { |
| 1146 | struct bio_vec *bvec; |
| 1147 | int i; |
| 1148 | |
| 1149 | bio_for_each_segment_all(bvec, bio, i) |
| 1150 | __free_page(bvec->bv_page); |
| 1151 | } |
| 1152 | EXPORT_SYMBOL(bio_free_pages); |
| 1153 | |
| 1154 | /** |
| 1155 | * bio_uncopy_user - finish previously mapped bio |
| 1156 | * @bio: bio being terminated |
| 1157 | * |
| 1158 | * Free pages allocated from bio_copy_user_iov() and write back data |
| 1159 | * to user space in case of a read. |
| 1160 | */ |
| 1161 | int bio_uncopy_user(struct bio *bio) |
| 1162 | { |
| 1163 | struct bio_map_data *bmd = bio->bi_private; |
| 1164 | int ret = 0; |
| 1165 | |
| 1166 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { |
| 1167 | /* |
| 1168 | * if we're in a workqueue, the request is orphaned, so |
| 1169 | * don't copy into a random user address space, just free |
| 1170 | * and return -EINTR so user space doesn't expect any data. |
| 1171 | */ |
| 1172 | if (!current->mm) |
| 1173 | ret = -EINTR; |
| 1174 | else if (bio_data_dir(bio) == READ) |
| 1175 | ret = bio_copy_to_iter(bio, bmd->iter); |
| 1176 | if (bmd->is_our_pages) |
| 1177 | bio_free_pages(bio); |
| 1178 | } |
| 1179 | kfree(bmd); |
| 1180 | bio_put(bio); |
| 1181 | return ret; |
| 1182 | } |
| 1183 | |
| 1184 | /** |
| 1185 | * bio_copy_user_iov - copy user data to bio |
| 1186 | * @q: destination block queue |
| 1187 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
| 1188 | * @iter: iovec iterator |
| 1189 | * @gfp_mask: memory allocation flags |
| 1190 | * |
| 1191 | * Prepares and returns a bio for indirect user io, bouncing data |
| 1192 | * to/from kernel pages as necessary. Must be paired with |
| 1193 | * call bio_uncopy_user() on io completion. |
| 1194 | */ |
| 1195 | struct bio *bio_copy_user_iov(struct request_queue *q, |
| 1196 | struct rq_map_data *map_data, |
| 1197 | struct iov_iter *iter, |
| 1198 | gfp_t gfp_mask) |
| 1199 | { |
| 1200 | struct bio_map_data *bmd; |
| 1201 | struct page *page; |
| 1202 | struct bio *bio; |
| 1203 | int i = 0, ret; |
| 1204 | int nr_pages; |
| 1205 | unsigned int len = iter->count; |
| 1206 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; |
| 1207 | |
| 1208 | bmd = bio_alloc_map_data(iter, gfp_mask); |
| 1209 | if (!bmd) |
| 1210 | return ERR_PTR(-ENOMEM); |
| 1211 | |
| 1212 | /* |
| 1213 | * We need to do a deep copy of the iov_iter including the iovecs. |
| 1214 | * The caller provided iov might point to an on-stack or otherwise |
| 1215 | * shortlived one. |
| 1216 | */ |
| 1217 | bmd->is_our_pages = map_data ? 0 : 1; |
| 1218 | |
| 1219 | nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); |
| 1220 | if (nr_pages > BIO_MAX_PAGES) |
| 1221 | nr_pages = BIO_MAX_PAGES; |
| 1222 | |
| 1223 | ret = -ENOMEM; |
| 1224 | bio = bio_kmalloc(gfp_mask, nr_pages); |
| 1225 | if (!bio) |
| 1226 | goto out_bmd; |
| 1227 | |
| 1228 | ret = 0; |
| 1229 | |
| 1230 | if (map_data) { |
| 1231 | nr_pages = 1 << map_data->page_order; |
| 1232 | i = map_data->offset / PAGE_SIZE; |
| 1233 | } |
| 1234 | while (len) { |
| 1235 | unsigned int bytes = PAGE_SIZE; |
| 1236 | |
| 1237 | bytes -= offset; |
| 1238 | |
| 1239 | if (bytes > len) |
| 1240 | bytes = len; |
| 1241 | |
| 1242 | if (map_data) { |
| 1243 | if (i == map_data->nr_entries * nr_pages) { |
| 1244 | ret = -ENOMEM; |
| 1245 | break; |
| 1246 | } |
| 1247 | |
| 1248 | page = map_data->pages[i / nr_pages]; |
| 1249 | page += (i % nr_pages); |
| 1250 | |
| 1251 | i++; |
| 1252 | } else { |
| 1253 | page = alloc_page(q->bounce_gfp | gfp_mask); |
| 1254 | if (!page) { |
| 1255 | ret = -ENOMEM; |
| 1256 | break; |
| 1257 | } |
| 1258 | } |
| 1259 | |
| 1260 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) |
| 1261 | break; |
| 1262 | |
| 1263 | len -= bytes; |
| 1264 | offset = 0; |
| 1265 | } |
| 1266 | |
| 1267 | if (ret) |
| 1268 | goto cleanup; |
| 1269 | |
| 1270 | if (map_data) |
| 1271 | map_data->offset += bio->bi_iter.bi_size; |
| 1272 | |
| 1273 | /* |
| 1274 | * success |
| 1275 | */ |
| 1276 | if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) || |
| 1277 | (map_data && map_data->from_user)) { |
| 1278 | ret = bio_copy_from_iter(bio, iter); |
| 1279 | if (ret) |
| 1280 | goto cleanup; |
| 1281 | } else { |
| 1282 | iov_iter_advance(iter, bio->bi_iter.bi_size); |
| 1283 | } |
| 1284 | |
| 1285 | bio->bi_private = bmd; |
| 1286 | if (map_data && map_data->null_mapped) |
| 1287 | bio_set_flag(bio, BIO_NULL_MAPPED); |
| 1288 | return bio; |
| 1289 | cleanup: |
| 1290 | if (!map_data) |
| 1291 | bio_free_pages(bio); |
| 1292 | bio_put(bio); |
| 1293 | out_bmd: |
| 1294 | kfree(bmd); |
| 1295 | return ERR_PTR(ret); |
| 1296 | } |
| 1297 | |
| 1298 | /** |
| 1299 | * bio_map_user_iov - map user iovec into bio |
| 1300 | * @q: the struct request_queue for the bio |
| 1301 | * @iter: iovec iterator |
| 1302 | * @gfp_mask: memory allocation flags |
| 1303 | * |
| 1304 | * Map the user space address into a bio suitable for io to a block |
| 1305 | * device. Returns an error pointer in case of error. |
| 1306 | */ |
| 1307 | struct bio *bio_map_user_iov(struct request_queue *q, |
| 1308 | struct iov_iter *iter, |
| 1309 | gfp_t gfp_mask) |
| 1310 | { |
| 1311 | int j; |
| 1312 | struct bio *bio; |
| 1313 | int ret; |
| 1314 | struct bio_vec *bvec; |
| 1315 | |
| 1316 | if (!iov_iter_count(iter)) |
| 1317 | return ERR_PTR(-EINVAL); |
| 1318 | |
| 1319 | bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); |
| 1320 | if (!bio) |
| 1321 | return ERR_PTR(-ENOMEM); |
| 1322 | |
| 1323 | while (iov_iter_count(iter)) { |
| 1324 | struct page **pages; |
| 1325 | ssize_t bytes; |
| 1326 | size_t offs, added = 0; |
| 1327 | int npages; |
| 1328 | |
| 1329 | bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); |
| 1330 | if (unlikely(bytes <= 0)) { |
| 1331 | ret = bytes ? bytes : -EFAULT; |
| 1332 | goto out_unmap; |
| 1333 | } |
| 1334 | |
| 1335 | npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); |
| 1336 | |
| 1337 | if (unlikely(offs & queue_dma_alignment(q))) { |
| 1338 | ret = -EINVAL; |
| 1339 | j = 0; |
| 1340 | } else { |
| 1341 | for (j = 0; j < npages; j++) { |
| 1342 | struct page *page = pages[j]; |
| 1343 | unsigned int n = PAGE_SIZE - offs; |
| 1344 | unsigned short prev_bi_vcnt = bio->bi_vcnt; |
| 1345 | |
| 1346 | if (n > bytes) |
| 1347 | n = bytes; |
| 1348 | |
| 1349 | if (!bio_add_pc_page(q, bio, page, n, offs)) |
| 1350 | break; |
| 1351 | |
| 1352 | /* |
| 1353 | * check if vector was merged with previous |
| 1354 | * drop page reference if needed |
| 1355 | */ |
| 1356 | if (bio->bi_vcnt == prev_bi_vcnt) |
| 1357 | put_page(page); |
| 1358 | |
| 1359 | added += n; |
| 1360 | bytes -= n; |
| 1361 | offs = 0; |
| 1362 | } |
| 1363 | iov_iter_advance(iter, added); |
| 1364 | } |
| 1365 | /* |
| 1366 | * release the pages we didn't map into the bio, if any |
| 1367 | */ |
| 1368 | while (j < npages) |
| 1369 | put_page(pages[j++]); |
| 1370 | kvfree(pages); |
| 1371 | /* couldn't stuff something into bio? */ |
| 1372 | if (bytes) |
| 1373 | break; |
| 1374 | } |
| 1375 | |
| 1376 | bio_set_flag(bio, BIO_USER_MAPPED); |
| 1377 | |
| 1378 | /* |
| 1379 | * subtle -- if bio_map_user_iov() ended up bouncing a bio, |
| 1380 | * it would normally disappear when its bi_end_io is run. |
| 1381 | * however, we need it for the unmap, so grab an extra |
| 1382 | * reference to it |
| 1383 | */ |
| 1384 | bio_get(bio); |
| 1385 | return bio; |
| 1386 | |
| 1387 | out_unmap: |
| 1388 | bio_for_each_segment_all(bvec, bio, j) { |
| 1389 | put_page(bvec->bv_page); |
| 1390 | } |
| 1391 | bio_put(bio); |
| 1392 | return ERR_PTR(ret); |
| 1393 | } |
| 1394 | |
| 1395 | static void __bio_unmap_user(struct bio *bio) |
| 1396 | { |
| 1397 | struct bio_vec *bvec; |
| 1398 | int i; |
| 1399 | |
| 1400 | /* |
| 1401 | * make sure we dirty pages we wrote to |
| 1402 | */ |
| 1403 | bio_for_each_segment_all(bvec, bio, i) { |
| 1404 | if (bio_data_dir(bio) == READ) |
| 1405 | set_page_dirty_lock(bvec->bv_page); |
| 1406 | |
| 1407 | put_page(bvec->bv_page); |
| 1408 | } |
| 1409 | |
| 1410 | bio_put(bio); |
| 1411 | } |
| 1412 | |
| 1413 | /** |
| 1414 | * bio_unmap_user - unmap a bio |
| 1415 | * @bio: the bio being unmapped |
| 1416 | * |
| 1417 | * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from |
| 1418 | * process context. |
| 1419 | * |
| 1420 | * bio_unmap_user() may sleep. |
| 1421 | */ |
| 1422 | void bio_unmap_user(struct bio *bio) |
| 1423 | { |
| 1424 | __bio_unmap_user(bio); |
| 1425 | bio_put(bio); |
| 1426 | } |
| 1427 | |
| 1428 | static void bio_map_kern_endio(struct bio *bio) |
| 1429 | { |
| 1430 | bio_put(bio); |
| 1431 | } |
| 1432 | |
| 1433 | /** |
| 1434 | * bio_map_kern - map kernel address into bio |
| 1435 | * @q: the struct request_queue for the bio |
| 1436 | * @data: pointer to buffer to map |
| 1437 | * @len: length in bytes |
| 1438 | * @gfp_mask: allocation flags for bio allocation |
| 1439 | * |
| 1440 | * Map the kernel address into a bio suitable for io to a block |
| 1441 | * device. Returns an error pointer in case of error. |
| 1442 | */ |
| 1443 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
| 1444 | gfp_t gfp_mask) |
| 1445 | { |
| 1446 | unsigned long kaddr = (unsigned long)data; |
| 1447 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1448 | unsigned long start = kaddr >> PAGE_SHIFT; |
| 1449 | const int nr_pages = end - start; |
| 1450 | int offset, i; |
| 1451 | struct bio *bio; |
| 1452 | |
| 1453 | bio = bio_kmalloc(gfp_mask, nr_pages); |
| 1454 | if (!bio) |
| 1455 | return ERR_PTR(-ENOMEM); |
| 1456 | |
| 1457 | offset = offset_in_page(kaddr); |
| 1458 | for (i = 0; i < nr_pages; i++) { |
| 1459 | unsigned int bytes = PAGE_SIZE - offset; |
| 1460 | |
| 1461 | if (len <= 0) |
| 1462 | break; |
| 1463 | |
| 1464 | if (bytes > len) |
| 1465 | bytes = len; |
| 1466 | |
| 1467 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
| 1468 | offset) < bytes) { |
| 1469 | /* we don't support partial mappings */ |
| 1470 | bio_put(bio); |
| 1471 | return ERR_PTR(-EINVAL); |
| 1472 | } |
| 1473 | |
| 1474 | data += bytes; |
| 1475 | len -= bytes; |
| 1476 | offset = 0; |
| 1477 | } |
| 1478 | |
| 1479 | bio->bi_end_io = bio_map_kern_endio; |
| 1480 | return bio; |
| 1481 | } |
| 1482 | EXPORT_SYMBOL(bio_map_kern); |
| 1483 | |
| 1484 | static void bio_copy_kern_endio(struct bio *bio) |
| 1485 | { |
| 1486 | bio_free_pages(bio); |
| 1487 | bio_put(bio); |
| 1488 | } |
| 1489 | |
| 1490 | static void bio_copy_kern_endio_read(struct bio *bio) |
| 1491 | { |
| 1492 | char *p = bio->bi_private; |
| 1493 | struct bio_vec *bvec; |
| 1494 | int i; |
| 1495 | |
| 1496 | bio_for_each_segment_all(bvec, bio, i) { |
| 1497 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); |
| 1498 | p += bvec->bv_len; |
| 1499 | } |
| 1500 | |
| 1501 | bio_copy_kern_endio(bio); |
| 1502 | } |
| 1503 | |
| 1504 | /** |
| 1505 | * bio_copy_kern - copy kernel address into bio |
| 1506 | * @q: the struct request_queue for the bio |
| 1507 | * @data: pointer to buffer to copy |
| 1508 | * @len: length in bytes |
| 1509 | * @gfp_mask: allocation flags for bio and page allocation |
| 1510 | * @reading: data direction is READ |
| 1511 | * |
| 1512 | * copy the kernel address into a bio suitable for io to a block |
| 1513 | * device. Returns an error pointer in case of error. |
| 1514 | */ |
| 1515 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, |
| 1516 | gfp_t gfp_mask, int reading) |
| 1517 | { |
| 1518 | unsigned long kaddr = (unsigned long)data; |
| 1519 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1520 | unsigned long start = kaddr >> PAGE_SHIFT; |
| 1521 | struct bio *bio; |
| 1522 | void *p = data; |
| 1523 | int nr_pages = 0; |
| 1524 | |
| 1525 | /* |
| 1526 | * Overflow, abort |
| 1527 | */ |
| 1528 | if (end < start) |
| 1529 | return ERR_PTR(-EINVAL); |
| 1530 | |
| 1531 | nr_pages = end - start; |
| 1532 | bio = bio_kmalloc(gfp_mask, nr_pages); |
| 1533 | if (!bio) |
| 1534 | return ERR_PTR(-ENOMEM); |
| 1535 | |
| 1536 | while (len) { |
| 1537 | struct page *page; |
| 1538 | unsigned int bytes = PAGE_SIZE; |
| 1539 | |
| 1540 | if (bytes > len) |
| 1541 | bytes = len; |
| 1542 | |
| 1543 | page = alloc_page(q->bounce_gfp | gfp_mask); |
| 1544 | if (!page) |
| 1545 | goto cleanup; |
| 1546 | |
| 1547 | if (!reading) |
| 1548 | memcpy(page_address(page), p, bytes); |
| 1549 | |
| 1550 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
| 1551 | break; |
| 1552 | |
| 1553 | len -= bytes; |
| 1554 | p += bytes; |
| 1555 | } |
| 1556 | |
| 1557 | if (reading) { |
| 1558 | bio->bi_end_io = bio_copy_kern_endio_read; |
| 1559 | bio->bi_private = data; |
| 1560 | } else { |
| 1561 | bio->bi_end_io = bio_copy_kern_endio; |
| 1562 | } |
| 1563 | |
| 1564 | return bio; |
| 1565 | |
| 1566 | cleanup: |
| 1567 | bio_free_pages(bio); |
| 1568 | bio_put(bio); |
| 1569 | return ERR_PTR(-ENOMEM); |
| 1570 | } |
| 1571 | |
| 1572 | /* |
| 1573 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions |
| 1574 | * for performing direct-IO in BIOs. |
| 1575 | * |
| 1576 | * The problem is that we cannot run set_page_dirty() from interrupt context |
| 1577 | * because the required locks are not interrupt-safe. So what we can do is to |
| 1578 | * mark the pages dirty _before_ performing IO. And in interrupt context, |
| 1579 | * check that the pages are still dirty. If so, fine. If not, redirty them |
| 1580 | * in process context. |
| 1581 | * |
| 1582 | * We special-case compound pages here: normally this means reads into hugetlb |
| 1583 | * pages. The logic in here doesn't really work right for compound pages |
| 1584 | * because the VM does not uniformly chase down the head page in all cases. |
| 1585 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't |
| 1586 | * handle them at all. So we skip compound pages here at an early stage. |
| 1587 | * |
| 1588 | * Note that this code is very hard to test under normal circumstances because |
| 1589 | * direct-io pins the pages with get_user_pages(). This makes |
| 1590 | * is_page_cache_freeable return false, and the VM will not clean the pages. |
| 1591 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
| 1592 | * pagecache. |
| 1593 | * |
| 1594 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the |
| 1595 | * deferred bio dirtying paths. |
| 1596 | */ |
| 1597 | |
| 1598 | /* |
| 1599 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. |
| 1600 | */ |
| 1601 | void bio_set_pages_dirty(struct bio *bio) |
| 1602 | { |
| 1603 | struct bio_vec *bvec; |
| 1604 | int i; |
| 1605 | |
| 1606 | bio_for_each_segment_all(bvec, bio, i) { |
| 1607 | struct page *page = bvec->bv_page; |
| 1608 | |
| 1609 | if (page && !PageCompound(page)) |
| 1610 | set_page_dirty_lock(page); |
| 1611 | } |
| 1612 | } |
| 1613 | |
| 1614 | static void bio_release_pages(struct bio *bio) |
| 1615 | { |
| 1616 | struct bio_vec *bvec; |
| 1617 | int i; |
| 1618 | |
| 1619 | bio_for_each_segment_all(bvec, bio, i) { |
| 1620 | struct page *page = bvec->bv_page; |
| 1621 | |
| 1622 | if (page) |
| 1623 | put_page(page); |
| 1624 | } |
| 1625 | } |
| 1626 | |
| 1627 | /* |
| 1628 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. |
| 1629 | * If they are, then fine. If, however, some pages are clean then they must |
| 1630 | * have been written out during the direct-IO read. So we take another ref on |
| 1631 | * the BIO and the offending pages and re-dirty the pages in process context. |
| 1632 | * |
| 1633 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from |
| 1634 | * here on. It will run one put_page() against each page and will run one |
| 1635 | * bio_put() against the BIO. |
| 1636 | */ |
| 1637 | |
| 1638 | static void bio_dirty_fn(struct work_struct *work); |
| 1639 | |
| 1640 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
| 1641 | static DEFINE_SPINLOCK(bio_dirty_lock); |
| 1642 | static struct bio *bio_dirty_list; |
| 1643 | |
| 1644 | /* |
| 1645 | * This runs in process context |
| 1646 | */ |
| 1647 | static void bio_dirty_fn(struct work_struct *work) |
| 1648 | { |
| 1649 | unsigned long flags; |
| 1650 | struct bio *bio; |
| 1651 | |
| 1652 | spin_lock_irqsave(&bio_dirty_lock, flags); |
| 1653 | bio = bio_dirty_list; |
| 1654 | bio_dirty_list = NULL; |
| 1655 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
| 1656 | |
| 1657 | while (bio) { |
| 1658 | struct bio *next = bio->bi_private; |
| 1659 | |
| 1660 | bio_set_pages_dirty(bio); |
| 1661 | bio_release_pages(bio); |
| 1662 | bio_put(bio); |
| 1663 | bio = next; |
| 1664 | } |
| 1665 | } |
| 1666 | |
| 1667 | void bio_check_pages_dirty(struct bio *bio) |
| 1668 | { |
| 1669 | struct bio_vec *bvec; |
| 1670 | int nr_clean_pages = 0; |
| 1671 | int i; |
| 1672 | |
| 1673 | bio_for_each_segment_all(bvec, bio, i) { |
| 1674 | struct page *page = bvec->bv_page; |
| 1675 | |
| 1676 | if (PageDirty(page) || PageCompound(page)) { |
| 1677 | put_page(page); |
| 1678 | bvec->bv_page = NULL; |
| 1679 | } else { |
| 1680 | nr_clean_pages++; |
| 1681 | } |
| 1682 | } |
| 1683 | |
| 1684 | if (nr_clean_pages) { |
| 1685 | unsigned long flags; |
| 1686 | |
| 1687 | spin_lock_irqsave(&bio_dirty_lock, flags); |
| 1688 | bio->bi_private = bio_dirty_list; |
| 1689 | bio_dirty_list = bio; |
| 1690 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
| 1691 | schedule_work(&bio_dirty_work); |
| 1692 | } else { |
| 1693 | bio_put(bio); |
| 1694 | } |
| 1695 | } |
| 1696 | |
| 1697 | void generic_start_io_acct(struct request_queue *q, int rw, |
| 1698 | unsigned long sectors, struct hd_struct *part) |
| 1699 | { |
| 1700 | int cpu = part_stat_lock(); |
| 1701 | |
| 1702 | part_round_stats(q, cpu, part); |
| 1703 | part_stat_inc(cpu, part, ios[rw]); |
| 1704 | part_stat_add(cpu, part, sectors[rw], sectors); |
| 1705 | part_inc_in_flight(q, part, rw); |
| 1706 | |
| 1707 | part_stat_unlock(); |
| 1708 | } |
| 1709 | EXPORT_SYMBOL(generic_start_io_acct); |
| 1710 | |
| 1711 | void generic_end_io_acct(struct request_queue *q, int rw, |
| 1712 | struct hd_struct *part, unsigned long start_time) |
| 1713 | { |
| 1714 | unsigned long duration = jiffies - start_time; |
| 1715 | int cpu = part_stat_lock(); |
| 1716 | |
| 1717 | part_stat_add(cpu, part, ticks[rw], duration); |
| 1718 | part_round_stats(q, cpu, part); |
| 1719 | part_dec_in_flight(q, part, rw); |
| 1720 | |
| 1721 | part_stat_unlock(); |
| 1722 | } |
| 1723 | EXPORT_SYMBOL(generic_end_io_acct); |
| 1724 | |
| 1725 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
| 1726 | void bio_flush_dcache_pages(struct bio *bi) |
| 1727 | { |
| 1728 | struct bio_vec bvec; |
| 1729 | struct bvec_iter iter; |
| 1730 | |
| 1731 | bio_for_each_segment(bvec, bi, iter) |
| 1732 | flush_dcache_page(bvec.bv_page); |
| 1733 | } |
| 1734 | EXPORT_SYMBOL(bio_flush_dcache_pages); |
| 1735 | #endif |
| 1736 | |
| 1737 | static inline bool bio_remaining_done(struct bio *bio) |
| 1738 | { |
| 1739 | /* |
| 1740 | * If we're not chaining, then ->__bi_remaining is always 1 and |
| 1741 | * we always end io on the first invocation. |
| 1742 | */ |
| 1743 | if (!bio_flagged(bio, BIO_CHAIN)) |
| 1744 | return true; |
| 1745 | |
| 1746 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); |
| 1747 | |
| 1748 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
| 1749 | bio_clear_flag(bio, BIO_CHAIN); |
| 1750 | return true; |
| 1751 | } |
| 1752 | |
| 1753 | return false; |
| 1754 | } |
| 1755 | |
| 1756 | /** |
| 1757 | * bio_endio - end I/O on a bio |
| 1758 | * @bio: bio |
| 1759 | * |
| 1760 | * Description: |
| 1761 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
| 1762 | * way to end I/O on a bio. No one should call bi_end_io() directly on a |
| 1763 | * bio unless they own it and thus know that it has an end_io function. |
| 1764 | * |
| 1765 | * bio_endio() can be called several times on a bio that has been chained |
| 1766 | * using bio_chain(). The ->bi_end_io() function will only be called the |
| 1767 | * last time. At this point the BLK_TA_COMPLETE tracing event will be |
| 1768 | * generated if BIO_TRACE_COMPLETION is set. |
| 1769 | **/ |
| 1770 | void bio_endio(struct bio *bio) |
| 1771 | { |
| 1772 | again: |
| 1773 | if (!bio_remaining_done(bio)) |
| 1774 | return; |
| 1775 | if (!bio_integrity_endio(bio)) |
| 1776 | return; |
| 1777 | |
| 1778 | /* |
| 1779 | * Need to have a real endio function for chained bios, otherwise |
| 1780 | * various corner cases will break (like stacking block devices that |
| 1781 | * save/restore bi_end_io) - however, we want to avoid unbounded |
| 1782 | * recursion and blowing the stack. Tail call optimization would |
| 1783 | * handle this, but compiling with frame pointers also disables |
| 1784 | * gcc's sibling call optimization. |
| 1785 | */ |
| 1786 | if (bio->bi_end_io == bio_chain_endio) { |
| 1787 | bio = __bio_chain_endio(bio); |
| 1788 | goto again; |
| 1789 | } |
| 1790 | |
| 1791 | if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
| 1792 | trace_block_bio_complete(bio->bi_disk->queue, bio, |
| 1793 | blk_status_to_errno(bio->bi_status)); |
| 1794 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
| 1795 | } |
| 1796 | |
| 1797 | blk_throtl_bio_endio(bio); |
| 1798 | /* release cgroup info */ |
| 1799 | bio_uninit(bio); |
| 1800 | if (bio->bi_end_io) |
| 1801 | bio->bi_end_io(bio); |
| 1802 | } |
| 1803 | EXPORT_SYMBOL(bio_endio); |
| 1804 | |
| 1805 | /** |
| 1806 | * bio_split - split a bio |
| 1807 | * @bio: bio to split |
| 1808 | * @sectors: number of sectors to split from the front of @bio |
| 1809 | * @gfp: gfp mask |
| 1810 | * @bs: bio set to allocate from |
| 1811 | * |
| 1812 | * Allocates and returns a new bio which represents @sectors from the start of |
| 1813 | * @bio, and updates @bio to represent the remaining sectors. |
| 1814 | * |
| 1815 | * Unless this is a discard request the newly allocated bio will point |
| 1816 | * to @bio's bi_io_vec; it is the caller's responsibility to ensure that |
| 1817 | * @bio is not freed before the split. |
| 1818 | */ |
| 1819 | struct bio *bio_split(struct bio *bio, int sectors, |
| 1820 | gfp_t gfp, struct bio_set *bs) |
| 1821 | { |
| 1822 | struct bio *split = NULL; |
| 1823 | |
| 1824 | BUG_ON(sectors <= 0); |
| 1825 | BUG_ON(sectors >= bio_sectors(bio)); |
| 1826 | |
| 1827 | split = bio_clone_fast(bio, gfp, bs); |
| 1828 | if (!split) |
| 1829 | return NULL; |
| 1830 | |
| 1831 | split->bi_iter.bi_size = sectors << 9; |
| 1832 | |
| 1833 | if (bio_integrity(split)) |
| 1834 | bio_integrity_trim(split); |
| 1835 | |
| 1836 | bio_advance(bio, split->bi_iter.bi_size); |
| 1837 | |
| 1838 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
| 1839 | bio_set_flag(bio, BIO_TRACE_COMPLETION); |
| 1840 | |
| 1841 | return split; |
| 1842 | } |
| 1843 | EXPORT_SYMBOL(bio_split); |
| 1844 | |
| 1845 | /** |
| 1846 | * bio_trim - trim a bio |
| 1847 | * @bio: bio to trim |
| 1848 | * @offset: number of sectors to trim from the front of @bio |
| 1849 | * @size: size we want to trim @bio to, in sectors |
| 1850 | */ |
| 1851 | void bio_trim(struct bio *bio, int offset, int size) |
| 1852 | { |
| 1853 | /* 'bio' is a cloned bio which we need to trim to match |
| 1854 | * the given offset and size. |
| 1855 | */ |
| 1856 | |
| 1857 | size <<= 9; |
| 1858 | if (offset == 0 && size == bio->bi_iter.bi_size) |
| 1859 | return; |
| 1860 | |
| 1861 | bio_clear_flag(bio, BIO_SEG_VALID); |
| 1862 | |
| 1863 | bio_advance(bio, offset << 9); |
| 1864 | |
| 1865 | bio->bi_iter.bi_size = size; |
| 1866 | |
| 1867 | if (bio_integrity(bio)) |
| 1868 | bio_integrity_trim(bio); |
| 1869 | |
| 1870 | } |
| 1871 | EXPORT_SYMBOL_GPL(bio_trim); |
| 1872 | |
| 1873 | /* |
| 1874 | * create memory pools for biovec's in a bio_set. |
| 1875 | * use the global biovec slabs created for general use. |
| 1876 | */ |
| 1877 | mempool_t *biovec_create_pool(int pool_entries) |
| 1878 | { |
| 1879 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; |
| 1880 | |
| 1881 | return mempool_create_slab_pool(pool_entries, bp->slab); |
| 1882 | } |
| 1883 | |
| 1884 | void bioset_free(struct bio_set *bs) |
| 1885 | { |
| 1886 | if (bs->rescue_workqueue) |
| 1887 | destroy_workqueue(bs->rescue_workqueue); |
| 1888 | |
| 1889 | mempool_destroy(bs->bio_pool); |
| 1890 | mempool_destroy(bs->bvec_pool); |
| 1891 | |
| 1892 | bioset_integrity_free(bs); |
| 1893 | bio_put_slab(bs); |
| 1894 | |
| 1895 | kfree(bs); |
| 1896 | } |
| 1897 | EXPORT_SYMBOL(bioset_free); |
| 1898 | |
| 1899 | /** |
| 1900 | * bioset_create - Create a bio_set |
| 1901 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
| 1902 | * @front_pad: Number of bytes to allocate in front of the returned bio |
| 1903 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS |
| 1904 | * and %BIOSET_NEED_RESCUER |
| 1905 | * |
| 1906 | * Description: |
| 1907 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller |
| 1908 | * to ask for a number of bytes to be allocated in front of the bio. |
| 1909 | * Front pad allocation is useful for embedding the bio inside |
| 1910 | * another structure, to avoid allocating extra data to go with the bio. |
| 1911 | * Note that the bio must be embedded at the END of that structure always, |
| 1912 | * or things will break badly. |
| 1913 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated |
| 1914 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). |
| 1915 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to |
| 1916 | * dispatch queued requests when the mempool runs out of space. |
| 1917 | * |
| 1918 | */ |
| 1919 | struct bio_set *bioset_create(unsigned int pool_size, |
| 1920 | unsigned int front_pad, |
| 1921 | int flags) |
| 1922 | { |
| 1923 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); |
| 1924 | struct bio_set *bs; |
| 1925 | |
| 1926 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
| 1927 | if (!bs) |
| 1928 | return NULL; |
| 1929 | |
| 1930 | bs->front_pad = front_pad; |
| 1931 | |
| 1932 | spin_lock_init(&bs->rescue_lock); |
| 1933 | bio_list_init(&bs->rescue_list); |
| 1934 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); |
| 1935 | |
| 1936 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); |
| 1937 | if (!bs->bio_slab) { |
| 1938 | kfree(bs); |
| 1939 | return NULL; |
| 1940 | } |
| 1941 | |
| 1942 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); |
| 1943 | if (!bs->bio_pool) |
| 1944 | goto bad; |
| 1945 | |
| 1946 | if (flags & BIOSET_NEED_BVECS) { |
| 1947 | bs->bvec_pool = biovec_create_pool(pool_size); |
| 1948 | if (!bs->bvec_pool) |
| 1949 | goto bad; |
| 1950 | } |
| 1951 | |
| 1952 | if (!(flags & BIOSET_NEED_RESCUER)) |
| 1953 | return bs; |
| 1954 | |
| 1955 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); |
| 1956 | if (!bs->rescue_workqueue) |
| 1957 | goto bad; |
| 1958 | |
| 1959 | return bs; |
| 1960 | bad: |
| 1961 | bioset_free(bs); |
| 1962 | return NULL; |
| 1963 | } |
| 1964 | EXPORT_SYMBOL(bioset_create); |
| 1965 | |
| 1966 | #ifdef CONFIG_BLK_CGROUP |
| 1967 | |
| 1968 | /** |
| 1969 | * bio_associate_blkcg - associate a bio with the specified blkcg |
| 1970 | * @bio: target bio |
| 1971 | * @blkcg_css: css of the blkcg to associate |
| 1972 | * |
| 1973 | * Associate @bio with the blkcg specified by @blkcg_css. Block layer will |
| 1974 | * treat @bio as if it were issued by a task which belongs to the blkcg. |
| 1975 | * |
| 1976 | * This function takes an extra reference of @blkcg_css which will be put |
| 1977 | * when @bio is released. The caller must own @bio and is responsible for |
| 1978 | * synchronizing calls to this function. |
| 1979 | */ |
| 1980 | int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css) |
| 1981 | { |
| 1982 | if (unlikely(bio->bi_css)) |
| 1983 | return -EBUSY; |
| 1984 | css_get(blkcg_css); |
| 1985 | bio->bi_css = blkcg_css; |
| 1986 | return 0; |
| 1987 | } |
| 1988 | EXPORT_SYMBOL_GPL(bio_associate_blkcg); |
| 1989 | |
| 1990 | /** |
| 1991 | * bio_disassociate_task - undo bio_associate_current() |
| 1992 | * @bio: target bio |
| 1993 | */ |
| 1994 | void bio_disassociate_task(struct bio *bio) |
| 1995 | { |
| 1996 | if (bio->bi_ioc) { |
| 1997 | put_io_context(bio->bi_ioc); |
| 1998 | bio->bi_ioc = NULL; |
| 1999 | } |
| 2000 | if (bio->bi_css) { |
| 2001 | css_put(bio->bi_css); |
| 2002 | bio->bi_css = NULL; |
| 2003 | } |
| 2004 | } |
| 2005 | |
| 2006 | /** |
| 2007 | * bio_clone_blkcg_association - clone blkcg association from src to dst bio |
| 2008 | * @dst: destination bio |
| 2009 | * @src: source bio |
| 2010 | */ |
| 2011 | void bio_clone_blkcg_association(struct bio *dst, struct bio *src) |
| 2012 | { |
| 2013 | if (src->bi_css) |
| 2014 | WARN_ON(bio_associate_blkcg(dst, src->bi_css)); |
| 2015 | } |
| 2016 | EXPORT_SYMBOL_GPL(bio_clone_blkcg_association); |
| 2017 | #endif /* CONFIG_BLK_CGROUP */ |
| 2018 | |
| 2019 | static void __init biovec_init_slabs(void) |
| 2020 | { |
| 2021 | int i; |
| 2022 | |
| 2023 | for (i = 0; i < BVEC_POOL_NR; i++) { |
| 2024 | int size; |
| 2025 | struct biovec_slab *bvs = bvec_slabs + i; |
| 2026 | |
| 2027 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
| 2028 | bvs->slab = NULL; |
| 2029 | continue; |
| 2030 | } |
| 2031 | |
| 2032 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
| 2033 | bvs->slab = kmem_cache_create(bvs->name, size, 0, |
| 2034 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
| 2035 | } |
| 2036 | } |
| 2037 | |
| 2038 | static int __init init_bio(void) |
| 2039 | { |
| 2040 | bio_slab_max = 2; |
| 2041 | bio_slab_nr = 0; |
| 2042 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); |
| 2043 | if (!bio_slabs) |
| 2044 | panic("bio: can't allocate bios\n"); |
| 2045 | |
| 2046 | bio_integrity_init(); |
| 2047 | biovec_init_slabs(); |
| 2048 | |
| 2049 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
| 2050 | if (!fs_bio_set) |
| 2051 | panic("bio: can't allocate bios\n"); |
| 2052 | |
| 2053 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) |
| 2054 | panic("bio: can't create integrity pool\n"); |
| 2055 | |
| 2056 | return 0; |
| 2057 | } |
| 2058 | subsys_initcall(init_bio); |