| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> |
| 4 | */ |
| 5 | #include <linux/mm.h> |
| 6 | #include <linux/swap.h> |
| 7 | #include <linux/bio.h> |
| 8 | #include <linux/blkdev.h> |
| 9 | #include <linux/uio.h> |
| 10 | #include <linux/iocontext.h> |
| 11 | #include <linux/slab.h> |
| 12 | #include <linux/init.h> |
| 13 | #include <linux/kernel.h> |
| 14 | #include <linux/export.h> |
| 15 | #include <linux/mempool.h> |
| 16 | #include <linux/workqueue.h> |
| 17 | #include <linux/cgroup.h> |
| 18 | #include <linux/blk-cgroup.h> |
| 19 | #include <linux/highmem.h> |
| 20 | |
| 21 | #include <trace/events/block.h> |
| 22 | #include "blk.h" |
| 23 | #include "blk-rq-qos.h" |
| 24 | |
| 25 | /* |
| 26 | * Test patch to inline a certain number of bi_io_vec's inside the bio |
| 27 | * itself, to shrink a bio data allocation from two mempool calls to one |
| 28 | */ |
| 29 | #define BIO_INLINE_VECS 4 |
| 30 | |
| 31 | /* |
| 32 | * if you change this list, also change bvec_alloc or things will |
| 33 | * break badly! cannot be bigger than what you can fit into an |
| 34 | * unsigned short |
| 35 | */ |
| 36 | #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } |
| 37 | static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { |
| 38 | BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), |
| 39 | }; |
| 40 | #undef BV |
| 41 | |
| 42 | /* |
| 43 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by |
| 44 | * IO code that does not need private memory pools. |
| 45 | */ |
| 46 | struct bio_set fs_bio_set; |
| 47 | EXPORT_SYMBOL(fs_bio_set); |
| 48 | |
| 49 | /* |
| 50 | * Our slab pool management |
| 51 | */ |
| 52 | struct bio_slab { |
| 53 | struct kmem_cache *slab; |
| 54 | unsigned int slab_ref; |
| 55 | unsigned int slab_size; |
| 56 | char name[8]; |
| 57 | }; |
| 58 | static DEFINE_MUTEX(bio_slab_lock); |
| 59 | static struct bio_slab *bio_slabs; |
| 60 | static unsigned int bio_slab_nr, bio_slab_max; |
| 61 | |
| 62 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) |
| 63 | { |
| 64 | unsigned int sz = sizeof(struct bio) + extra_size; |
| 65 | struct kmem_cache *slab = NULL; |
| 66 | struct bio_slab *bslab, *new_bio_slabs; |
| 67 | unsigned int new_bio_slab_max; |
| 68 | unsigned int i, entry = -1; |
| 69 | |
| 70 | mutex_lock(&bio_slab_lock); |
| 71 | |
| 72 | i = 0; |
| 73 | while (i < bio_slab_nr) { |
| 74 | bslab = &bio_slabs[i]; |
| 75 | |
| 76 | if (!bslab->slab && entry == -1) |
| 77 | entry = i; |
| 78 | else if (bslab->slab_size == sz) { |
| 79 | slab = bslab->slab; |
| 80 | bslab->slab_ref++; |
| 81 | break; |
| 82 | } |
| 83 | i++; |
| 84 | } |
| 85 | |
| 86 | if (slab) |
| 87 | goto out_unlock; |
| 88 | |
| 89 | if (bio_slab_nr == bio_slab_max && entry == -1) { |
| 90 | new_bio_slab_max = bio_slab_max << 1; |
| 91 | new_bio_slabs = krealloc(bio_slabs, |
| 92 | new_bio_slab_max * sizeof(struct bio_slab), |
| 93 | GFP_KERNEL); |
| 94 | if (!new_bio_slabs) |
| 95 | goto out_unlock; |
| 96 | bio_slab_max = new_bio_slab_max; |
| 97 | bio_slabs = new_bio_slabs; |
| 98 | } |
| 99 | if (entry == -1) |
| 100 | entry = bio_slab_nr++; |
| 101 | |
| 102 | bslab = &bio_slabs[entry]; |
| 103 | |
| 104 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); |
| 105 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, |
| 106 | SLAB_HWCACHE_ALIGN, NULL); |
| 107 | if (!slab) |
| 108 | goto out_unlock; |
| 109 | |
| 110 | bslab->slab = slab; |
| 111 | bslab->slab_ref = 1; |
| 112 | bslab->slab_size = sz; |
| 113 | out_unlock: |
| 114 | mutex_unlock(&bio_slab_lock); |
| 115 | return slab; |
| 116 | } |
| 117 | |
| 118 | static void bio_put_slab(struct bio_set *bs) |
| 119 | { |
| 120 | struct bio_slab *bslab = NULL; |
| 121 | unsigned int i; |
| 122 | |
| 123 | mutex_lock(&bio_slab_lock); |
| 124 | |
| 125 | for (i = 0; i < bio_slab_nr; i++) { |
| 126 | if (bs->bio_slab == bio_slabs[i].slab) { |
| 127 | bslab = &bio_slabs[i]; |
| 128 | break; |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
| 133 | goto out; |
| 134 | |
| 135 | WARN_ON(!bslab->slab_ref); |
| 136 | |
| 137 | if (--bslab->slab_ref) |
| 138 | goto out; |
| 139 | |
| 140 | kmem_cache_destroy(bslab->slab); |
| 141 | bslab->slab = NULL; |
| 142 | |
| 143 | out: |
| 144 | mutex_unlock(&bio_slab_lock); |
| 145 | } |
| 146 | |
| 147 | unsigned int bvec_nr_vecs(unsigned short idx) |
| 148 | { |
| 149 | return bvec_slabs[--idx].nr_vecs; |
| 150 | } |
| 151 | |
| 152 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) |
| 153 | { |
| 154 | if (!idx) |
| 155 | return; |
| 156 | idx--; |
| 157 | |
| 158 | BIO_BUG_ON(idx >= BVEC_POOL_NR); |
| 159 | |
| 160 | if (idx == BVEC_POOL_MAX) { |
| 161 | mempool_free(bv, pool); |
| 162 | } else { |
| 163 | struct biovec_slab *bvs = bvec_slabs + idx; |
| 164 | |
| 165 | kmem_cache_free(bvs->slab, bv); |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, |
| 170 | mempool_t *pool) |
| 171 | { |
| 172 | struct bio_vec *bvl; |
| 173 | |
| 174 | /* |
| 175 | * see comment near bvec_array define! |
| 176 | */ |
| 177 | switch (nr) { |
| 178 | case 1: |
| 179 | *idx = 0; |
| 180 | break; |
| 181 | case 2 ... 4: |
| 182 | *idx = 1; |
| 183 | break; |
| 184 | case 5 ... 16: |
| 185 | *idx = 2; |
| 186 | break; |
| 187 | case 17 ... 64: |
| 188 | *idx = 3; |
| 189 | break; |
| 190 | case 65 ... 128: |
| 191 | *idx = 4; |
| 192 | break; |
| 193 | case 129 ... BIO_MAX_PAGES: |
| 194 | *idx = 5; |
| 195 | break; |
| 196 | default: |
| 197 | return NULL; |
| 198 | } |
| 199 | |
| 200 | /* |
| 201 | * idx now points to the pool we want to allocate from. only the |
| 202 | * 1-vec entry pool is mempool backed. |
| 203 | */ |
| 204 | if (*idx == BVEC_POOL_MAX) { |
| 205 | fallback: |
| 206 | bvl = mempool_alloc(pool, gfp_mask); |
| 207 | } else { |
| 208 | struct biovec_slab *bvs = bvec_slabs + *idx; |
| 209 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); |
| 210 | |
| 211 | /* |
| 212 | * Make this allocation restricted and don't dump info on |
| 213 | * allocation failures, since we'll fallback to the mempool |
| 214 | * in case of failure. |
| 215 | */ |
| 216 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
| 217 | |
| 218 | /* |
| 219 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM |
| 220 | * is set, retry with the 1-entry mempool |
| 221 | */ |
| 222 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
| 223 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { |
| 224 | *idx = BVEC_POOL_MAX; |
| 225 | goto fallback; |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | (*idx)++; |
| 230 | return bvl; |
| 231 | } |
| 232 | |
| 233 | void bio_uninit(struct bio *bio) |
| 234 | { |
| 235 | bio_disassociate_blkg(bio); |
| 236 | } |
| 237 | EXPORT_SYMBOL(bio_uninit); |
| 238 | |
| 239 | static void bio_free(struct bio *bio) |
| 240 | { |
| 241 | struct bio_set *bs = bio->bi_pool; |
| 242 | void *p; |
| 243 | |
| 244 | bio_uninit(bio); |
| 245 | |
| 246 | if (bs) { |
| 247 | bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); |
| 248 | |
| 249 | /* |
| 250 | * If we have front padding, adjust the bio pointer before freeing |
| 251 | */ |
| 252 | p = bio; |
| 253 | p -= bs->front_pad; |
| 254 | |
| 255 | mempool_free(p, &bs->bio_pool); |
| 256 | } else { |
| 257 | /* Bio was allocated by bio_kmalloc() */ |
| 258 | kfree(bio); |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | /* |
| 263 | * Users of this function have their own bio allocation. Subsequently, |
| 264 | * they must remember to pair any call to bio_init() with bio_uninit() |
| 265 | * when IO has completed, or when the bio is released. |
| 266 | */ |
| 267 | void bio_init(struct bio *bio, struct bio_vec *table, |
| 268 | unsigned short max_vecs) |
| 269 | { |
| 270 | memset(bio, 0, sizeof(*bio)); |
| 271 | atomic_set(&bio->__bi_remaining, 1); |
| 272 | atomic_set(&bio->__bi_cnt, 1); |
| 273 | |
| 274 | bio->bi_io_vec = table; |
| 275 | bio->bi_max_vecs = max_vecs; |
| 276 | } |
| 277 | EXPORT_SYMBOL(bio_init); |
| 278 | |
| 279 | /** |
| 280 | * bio_reset - reinitialize a bio |
| 281 | * @bio: bio to reset |
| 282 | * |
| 283 | * Description: |
| 284 | * After calling bio_reset(), @bio will be in the same state as a freshly |
| 285 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are |
| 286 | * preserved are the ones that are initialized by bio_alloc_bioset(). See |
| 287 | * comment in struct bio. |
| 288 | */ |
| 289 | void bio_reset(struct bio *bio) |
| 290 | { |
| 291 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); |
| 292 | |
| 293 | bio_uninit(bio); |
| 294 | |
| 295 | memset(bio, 0, BIO_RESET_BYTES); |
| 296 | bio->bi_flags = flags; |
| 297 | atomic_set(&bio->__bi_remaining, 1); |
| 298 | } |
| 299 | EXPORT_SYMBOL(bio_reset); |
| 300 | |
| 301 | static struct bio *__bio_chain_endio(struct bio *bio) |
| 302 | { |
| 303 | struct bio *parent = bio->bi_private; |
| 304 | |
| 305 | if (!parent->bi_status) |
| 306 | parent->bi_status = bio->bi_status; |
| 307 | bio_put(bio); |
| 308 | return parent; |
| 309 | } |
| 310 | |
| 311 | static void bio_chain_endio(struct bio *bio) |
| 312 | { |
| 313 | bio_endio(__bio_chain_endio(bio)); |
| 314 | } |
| 315 | |
| 316 | /** |
| 317 | * bio_chain - chain bio completions |
| 318 | * @bio: the target bio |
| 319 | * @parent: the @bio's parent bio |
| 320 | * |
| 321 | * The caller won't have a bi_end_io called when @bio completes - instead, |
| 322 | * @parent's bi_end_io won't be called until both @parent and @bio have |
| 323 | * completed; the chained bio will also be freed when it completes. |
| 324 | * |
| 325 | * The caller must not set bi_private or bi_end_io in @bio. |
| 326 | */ |
| 327 | void bio_chain(struct bio *bio, struct bio *parent) |
| 328 | { |
| 329 | BUG_ON(bio->bi_private || bio->bi_end_io); |
| 330 | |
| 331 | bio->bi_private = parent; |
| 332 | bio->bi_end_io = bio_chain_endio; |
| 333 | bio_inc_remaining(parent); |
| 334 | } |
| 335 | EXPORT_SYMBOL(bio_chain); |
| 336 | |
| 337 | static void bio_alloc_rescue(struct work_struct *work) |
| 338 | { |
| 339 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); |
| 340 | struct bio *bio; |
| 341 | |
| 342 | while (1) { |
| 343 | spin_lock(&bs->rescue_lock); |
| 344 | bio = bio_list_pop(&bs->rescue_list); |
| 345 | spin_unlock(&bs->rescue_lock); |
| 346 | |
| 347 | if (!bio) |
| 348 | break; |
| 349 | |
| 350 | generic_make_request(bio); |
| 351 | } |
| 352 | } |
| 353 | |
| 354 | static void punt_bios_to_rescuer(struct bio_set *bs) |
| 355 | { |
| 356 | struct bio_list punt, nopunt; |
| 357 | struct bio *bio; |
| 358 | |
| 359 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) |
| 360 | return; |
| 361 | /* |
| 362 | * In order to guarantee forward progress we must punt only bios that |
| 363 | * were allocated from this bio_set; otherwise, if there was a bio on |
| 364 | * there for a stacking driver higher up in the stack, processing it |
| 365 | * could require allocating bios from this bio_set, and doing that from |
| 366 | * our own rescuer would be bad. |
| 367 | * |
| 368 | * Since bio lists are singly linked, pop them all instead of trying to |
| 369 | * remove from the middle of the list: |
| 370 | */ |
| 371 | |
| 372 | bio_list_init(&punt); |
| 373 | bio_list_init(&nopunt); |
| 374 | |
| 375 | while ((bio = bio_list_pop(¤t->bio_list[0]))) |
| 376 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
| 377 | current->bio_list[0] = nopunt; |
| 378 | |
| 379 | bio_list_init(&nopunt); |
| 380 | while ((bio = bio_list_pop(¤t->bio_list[1]))) |
| 381 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); |
| 382 | current->bio_list[1] = nopunt; |
| 383 | |
| 384 | spin_lock(&bs->rescue_lock); |
| 385 | bio_list_merge(&bs->rescue_list, &punt); |
| 386 | spin_unlock(&bs->rescue_lock); |
| 387 | |
| 388 | queue_work(bs->rescue_workqueue, &bs->rescue_work); |
| 389 | } |
| 390 | |
| 391 | /** |
| 392 | * bio_alloc_bioset - allocate a bio for I/O |
| 393 | * @gfp_mask: the GFP_* mask given to the slab allocator |
| 394 | * @nr_iovecs: number of iovecs to pre-allocate |
| 395 | * @bs: the bio_set to allocate from. |
| 396 | * |
| 397 | * Description: |
| 398 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
| 399 | * backed by the @bs's mempool. |
| 400 | * |
| 401 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will |
| 402 | * always be able to allocate a bio. This is due to the mempool guarantees. |
| 403 | * To make this work, callers must never allocate more than 1 bio at a time |
| 404 | * from this pool. Callers that need to allocate more than 1 bio must always |
| 405 | * submit the previously allocated bio for IO before attempting to allocate |
| 406 | * a new one. Failure to do so can cause deadlocks under memory pressure. |
| 407 | * |
| 408 | * Note that when running under generic_make_request() (i.e. any block |
| 409 | * driver), bios are not submitted until after you return - see the code in |
| 410 | * generic_make_request() that converts recursion into iteration, to prevent |
| 411 | * stack overflows. |
| 412 | * |
| 413 | * This would normally mean allocating multiple bios under |
| 414 | * generic_make_request() would be susceptible to deadlocks, but we have |
| 415 | * deadlock avoidance code that resubmits any blocked bios from a rescuer |
| 416 | * thread. |
| 417 | * |
| 418 | * However, we do not guarantee forward progress for allocations from other |
| 419 | * mempools. Doing multiple allocations from the same mempool under |
| 420 | * generic_make_request() should be avoided - instead, use bio_set's front_pad |
| 421 | * for per bio allocations. |
| 422 | * |
| 423 | * RETURNS: |
| 424 | * Pointer to new bio on success, NULL on failure. |
| 425 | */ |
| 426 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, |
| 427 | struct bio_set *bs) |
| 428 | { |
| 429 | gfp_t saved_gfp = gfp_mask; |
| 430 | unsigned front_pad; |
| 431 | unsigned inline_vecs; |
| 432 | struct bio_vec *bvl = NULL; |
| 433 | struct bio *bio; |
| 434 | void *p; |
| 435 | |
| 436 | if (!bs) { |
| 437 | if (nr_iovecs > UIO_MAXIOV) |
| 438 | return NULL; |
| 439 | |
| 440 | p = kmalloc(sizeof(struct bio) + |
| 441 | nr_iovecs * sizeof(struct bio_vec), |
| 442 | gfp_mask); |
| 443 | front_pad = 0; |
| 444 | inline_vecs = nr_iovecs; |
| 445 | } else { |
| 446 | /* should not use nobvec bioset for nr_iovecs > 0 */ |
| 447 | if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && |
| 448 | nr_iovecs > 0)) |
| 449 | return NULL; |
| 450 | /* |
| 451 | * generic_make_request() converts recursion to iteration; this |
| 452 | * means if we're running beneath it, any bios we allocate and |
| 453 | * submit will not be submitted (and thus freed) until after we |
| 454 | * return. |
| 455 | * |
| 456 | * This exposes us to a potential deadlock if we allocate |
| 457 | * multiple bios from the same bio_set() while running |
| 458 | * underneath generic_make_request(). If we were to allocate |
| 459 | * multiple bios (say a stacking block driver that was splitting |
| 460 | * bios), we would deadlock if we exhausted the mempool's |
| 461 | * reserve. |
| 462 | * |
| 463 | * We solve this, and guarantee forward progress, with a rescuer |
| 464 | * workqueue per bio_set. If we go to allocate and there are |
| 465 | * bios on current->bio_list, we first try the allocation |
| 466 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those |
| 467 | * bios we would be blocking to the rescuer workqueue before |
| 468 | * we retry with the original gfp_flags. |
| 469 | */ |
| 470 | |
| 471 | if (current->bio_list && |
| 472 | (!bio_list_empty(¤t->bio_list[0]) || |
| 473 | !bio_list_empty(¤t->bio_list[1])) && |
| 474 | bs->rescue_workqueue) |
| 475 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; |
| 476 | |
| 477 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
| 478 | if (!p && gfp_mask != saved_gfp) { |
| 479 | punt_bios_to_rescuer(bs); |
| 480 | gfp_mask = saved_gfp; |
| 481 | p = mempool_alloc(&bs->bio_pool, gfp_mask); |
| 482 | } |
| 483 | |
| 484 | front_pad = bs->front_pad; |
| 485 | inline_vecs = BIO_INLINE_VECS; |
| 486 | } |
| 487 | |
| 488 | if (unlikely(!p)) |
| 489 | return NULL; |
| 490 | |
| 491 | bio = p + front_pad; |
| 492 | bio_init(bio, NULL, 0); |
| 493 | |
| 494 | if (nr_iovecs > inline_vecs) { |
| 495 | unsigned long idx = 0; |
| 496 | |
| 497 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
| 498 | if (!bvl && gfp_mask != saved_gfp) { |
| 499 | punt_bios_to_rescuer(bs); |
| 500 | gfp_mask = saved_gfp; |
| 501 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); |
| 502 | } |
| 503 | |
| 504 | if (unlikely(!bvl)) |
| 505 | goto err_free; |
| 506 | |
| 507 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; |
| 508 | } else if (nr_iovecs) { |
| 509 | bvl = bio->bi_inline_vecs; |
| 510 | } |
| 511 | |
| 512 | bio->bi_pool = bs; |
| 513 | bio->bi_max_vecs = nr_iovecs; |
| 514 | bio->bi_io_vec = bvl; |
| 515 | return bio; |
| 516 | |
| 517 | err_free: |
| 518 | mempool_free(p, &bs->bio_pool); |
| 519 | return NULL; |
| 520 | } |
| 521 | EXPORT_SYMBOL(bio_alloc_bioset); |
| 522 | |
| 523 | void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) |
| 524 | { |
| 525 | unsigned long flags; |
| 526 | struct bio_vec bv; |
| 527 | struct bvec_iter iter; |
| 528 | |
| 529 | __bio_for_each_segment(bv, bio, iter, start) { |
| 530 | char *data = bvec_kmap_irq(&bv, &flags); |
| 531 | memset(data, 0, bv.bv_len); |
| 532 | flush_dcache_page(bv.bv_page); |
| 533 | bvec_kunmap_irq(data, &flags); |
| 534 | } |
| 535 | } |
| 536 | EXPORT_SYMBOL(zero_fill_bio_iter); |
| 537 | |
| 538 | /** |
| 539 | * bio_put - release a reference to a bio |
| 540 | * @bio: bio to release reference to |
| 541 | * |
| 542 | * Description: |
| 543 | * Put a reference to a &struct bio, either one you have gotten with |
| 544 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. |
| 545 | **/ |
| 546 | void bio_put(struct bio *bio) |
| 547 | { |
| 548 | if (!bio_flagged(bio, BIO_REFFED)) |
| 549 | bio_free(bio); |
| 550 | else { |
| 551 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); |
| 552 | |
| 553 | /* |
| 554 | * last put frees it |
| 555 | */ |
| 556 | if (atomic_dec_and_test(&bio->__bi_cnt)) |
| 557 | bio_free(bio); |
| 558 | } |
| 559 | } |
| 560 | EXPORT_SYMBOL(bio_put); |
| 561 | |
| 562 | /** |
| 563 | * __bio_clone_fast - clone a bio that shares the original bio's biovec |
| 564 | * @bio: destination bio |
| 565 | * @bio_src: bio to clone |
| 566 | * |
| 567 | * Clone a &bio. Caller will own the returned bio, but not |
| 568 | * the actual data it points to. Reference count of returned |
| 569 | * bio will be one. |
| 570 | * |
| 571 | * Caller must ensure that @bio_src is not freed before @bio. |
| 572 | */ |
| 573 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) |
| 574 | { |
| 575 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); |
| 576 | |
| 577 | /* |
| 578 | * most users will be overriding ->bi_disk with a new target, |
| 579 | * so we don't set nor calculate new physical/hw segment counts here |
| 580 | */ |
| 581 | bio->bi_disk = bio_src->bi_disk; |
| 582 | bio->bi_partno = bio_src->bi_partno; |
| 583 | bio_set_flag(bio, BIO_CLONED); |
| 584 | if (bio_flagged(bio_src, BIO_THROTTLED)) |
| 585 | bio_set_flag(bio, BIO_THROTTLED); |
| 586 | bio->bi_opf = bio_src->bi_opf; |
| 587 | bio->bi_ioprio = bio_src->bi_ioprio; |
| 588 | bio->bi_write_hint = bio_src->bi_write_hint; |
| 589 | bio->bi_iter = bio_src->bi_iter; |
| 590 | bio->bi_io_vec = bio_src->bi_io_vec; |
| 591 | |
| 592 | bio_clone_blkg_association(bio, bio_src); |
| 593 | blkcg_bio_issue_init(bio); |
| 594 | } |
| 595 | EXPORT_SYMBOL(__bio_clone_fast); |
| 596 | |
| 597 | /** |
| 598 | * bio_clone_fast - clone a bio that shares the original bio's biovec |
| 599 | * @bio: bio to clone |
| 600 | * @gfp_mask: allocation priority |
| 601 | * @bs: bio_set to allocate from |
| 602 | * |
| 603 | * Like __bio_clone_fast, only also allocates the returned bio |
| 604 | */ |
| 605 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) |
| 606 | { |
| 607 | struct bio *b; |
| 608 | |
| 609 | b = bio_alloc_bioset(gfp_mask, 0, bs); |
| 610 | if (!b) |
| 611 | return NULL; |
| 612 | |
| 613 | __bio_clone_fast(b, bio); |
| 614 | |
| 615 | if (bio_integrity(bio)) { |
| 616 | int ret; |
| 617 | |
| 618 | ret = bio_integrity_clone(b, bio, gfp_mask); |
| 619 | |
| 620 | if (ret < 0) { |
| 621 | bio_put(b); |
| 622 | return NULL; |
| 623 | } |
| 624 | } |
| 625 | |
| 626 | return b; |
| 627 | } |
| 628 | EXPORT_SYMBOL(bio_clone_fast); |
| 629 | |
| 630 | static inline bool page_is_mergeable(const struct bio_vec *bv, |
| 631 | struct page *page, unsigned int len, unsigned int off, |
| 632 | bool *same_page) |
| 633 | { |
| 634 | phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + |
| 635 | bv->bv_offset + bv->bv_len - 1; |
| 636 | phys_addr_t page_addr = page_to_phys(page); |
| 637 | |
| 638 | if (vec_end_addr + 1 != page_addr + off) |
| 639 | return false; |
| 640 | if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) |
| 641 | return false; |
| 642 | |
| 643 | *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); |
| 644 | if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) |
| 645 | return false; |
| 646 | return true; |
| 647 | } |
| 648 | |
| 649 | /* |
| 650 | * Check if the @page can be added to the current segment(@bv), and make |
| 651 | * sure to call it only if page_is_mergeable(@bv, @page) is true |
| 652 | */ |
| 653 | static bool can_add_page_to_seg(struct request_queue *q, |
| 654 | struct bio_vec *bv, struct page *page, unsigned len, |
| 655 | unsigned offset) |
| 656 | { |
| 657 | unsigned long mask = queue_segment_boundary(q); |
| 658 | phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; |
| 659 | phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; |
| 660 | |
| 661 | if ((addr1 | mask) != (addr2 | mask)) |
| 662 | return false; |
| 663 | |
| 664 | if (bv->bv_len + len > queue_max_segment_size(q)) |
| 665 | return false; |
| 666 | |
| 667 | return true; |
| 668 | } |
| 669 | |
| 670 | /** |
| 671 | * __bio_add_pc_page - attempt to add page to passthrough bio |
| 672 | * @q: the target queue |
| 673 | * @bio: destination bio |
| 674 | * @page: page to add |
| 675 | * @len: vec entry length |
| 676 | * @offset: vec entry offset |
| 677 | * @put_same_page: put the page if it is same with last added page |
| 678 | * |
| 679 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
| 680 | * number of reasons, such as the bio being full or target block device |
| 681 | * limitations. The target block device must allow bio's up to PAGE_SIZE, |
| 682 | * so it is always possible to add a single page to an empty bio. |
| 683 | * |
| 684 | * This should only be used by passthrough bios. |
| 685 | */ |
| 686 | static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, |
| 687 | struct page *page, unsigned int len, unsigned int offset, |
| 688 | bool put_same_page) |
| 689 | { |
| 690 | struct bio_vec *bvec; |
| 691 | bool same_page = false; |
| 692 | |
| 693 | /* |
| 694 | * cloned bio must not modify vec list |
| 695 | */ |
| 696 | if (unlikely(bio_flagged(bio, BIO_CLONED))) |
| 697 | return 0; |
| 698 | |
| 699 | if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) |
| 700 | return 0; |
| 701 | |
| 702 | if (bio->bi_vcnt > 0) { |
| 703 | bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
| 704 | |
| 705 | if (page == bvec->bv_page && |
| 706 | offset == bvec->bv_offset + bvec->bv_len) { |
| 707 | if (put_same_page) |
| 708 | put_page(page); |
| 709 | bvec->bv_len += len; |
| 710 | goto done; |
| 711 | } |
| 712 | |
| 713 | /* |
| 714 | * If the queue doesn't support SG gaps and adding this |
| 715 | * offset would create a gap, disallow it. |
| 716 | */ |
| 717 | if (bvec_gap_to_prev(q, bvec, offset)) |
| 718 | return 0; |
| 719 | |
| 720 | if (page_is_mergeable(bvec, page, len, offset, &same_page) && |
| 721 | can_add_page_to_seg(q, bvec, page, len, offset)) { |
| 722 | bvec->bv_len += len; |
| 723 | goto done; |
| 724 | } |
| 725 | } |
| 726 | |
| 727 | if (bio_full(bio, len)) |
| 728 | return 0; |
| 729 | |
| 730 | if (bio->bi_vcnt >= queue_max_segments(q)) |
| 731 | return 0; |
| 732 | |
| 733 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
| 734 | bvec->bv_page = page; |
| 735 | bvec->bv_len = len; |
| 736 | bvec->bv_offset = offset; |
| 737 | bio->bi_vcnt++; |
| 738 | done: |
| 739 | bio->bi_iter.bi_size += len; |
| 740 | return len; |
| 741 | } |
| 742 | |
| 743 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, |
| 744 | struct page *page, unsigned int len, unsigned int offset) |
| 745 | { |
| 746 | return __bio_add_pc_page(q, bio, page, len, offset, false); |
| 747 | } |
| 748 | EXPORT_SYMBOL(bio_add_pc_page); |
| 749 | |
| 750 | /** |
| 751 | * __bio_try_merge_page - try appending data to an existing bvec. |
| 752 | * @bio: destination bio |
| 753 | * @page: start page to add |
| 754 | * @len: length of the data to add |
| 755 | * @off: offset of the data relative to @page |
| 756 | * @same_page: return if the segment has been merged inside the same page |
| 757 | * |
| 758 | * Try to add the data at @page + @off to the last bvec of @bio. This is a |
| 759 | * a useful optimisation for file systems with a block size smaller than the |
| 760 | * page size. |
| 761 | * |
| 762 | * Warn if (@len, @off) crosses pages in case that @same_page is true. |
| 763 | * |
| 764 | * Return %true on success or %false on failure. |
| 765 | */ |
| 766 | bool __bio_try_merge_page(struct bio *bio, struct page *page, |
| 767 | unsigned int len, unsigned int off, bool *same_page) |
| 768 | { |
| 769 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) |
| 770 | return false; |
| 771 | |
| 772 | if (bio->bi_vcnt > 0) { |
| 773 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
| 774 | |
| 775 | if (page_is_mergeable(bv, page, len, off, same_page)) { |
| 776 | bv->bv_len += len; |
| 777 | bio->bi_iter.bi_size += len; |
| 778 | return true; |
| 779 | } |
| 780 | } |
| 781 | return false; |
| 782 | } |
| 783 | EXPORT_SYMBOL_GPL(__bio_try_merge_page); |
| 784 | |
| 785 | /** |
| 786 | * __bio_add_page - add page(s) to a bio in a new segment |
| 787 | * @bio: destination bio |
| 788 | * @page: start page to add |
| 789 | * @len: length of the data to add, may cross pages |
| 790 | * @off: offset of the data relative to @page, may cross pages |
| 791 | * |
| 792 | * Add the data at @page + @off to @bio as a new bvec. The caller must ensure |
| 793 | * that @bio has space for another bvec. |
| 794 | */ |
| 795 | void __bio_add_page(struct bio *bio, struct page *page, |
| 796 | unsigned int len, unsigned int off) |
| 797 | { |
| 798 | struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; |
| 799 | |
| 800 | WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); |
| 801 | WARN_ON_ONCE(bio_full(bio, len)); |
| 802 | |
| 803 | bv->bv_page = page; |
| 804 | bv->bv_offset = off; |
| 805 | bv->bv_len = len; |
| 806 | |
| 807 | bio->bi_iter.bi_size += len; |
| 808 | bio->bi_vcnt++; |
| 809 | } |
| 810 | EXPORT_SYMBOL_GPL(__bio_add_page); |
| 811 | |
| 812 | /** |
| 813 | * bio_add_page - attempt to add page(s) to bio |
| 814 | * @bio: destination bio |
| 815 | * @page: start page to add |
| 816 | * @len: vec entry length, may cross pages |
| 817 | * @offset: vec entry offset relative to @page, may cross pages |
| 818 | * |
| 819 | * Attempt to add page(s) to the bio_vec maplist. This will only fail |
| 820 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. |
| 821 | */ |
| 822 | int bio_add_page(struct bio *bio, struct page *page, |
| 823 | unsigned int len, unsigned int offset) |
| 824 | { |
| 825 | bool same_page = false; |
| 826 | |
| 827 | if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { |
| 828 | if (bio_full(bio, len)) |
| 829 | return 0; |
| 830 | __bio_add_page(bio, page, len, offset); |
| 831 | } |
| 832 | return len; |
| 833 | } |
| 834 | EXPORT_SYMBOL(bio_add_page); |
| 835 | |
| 836 | void bio_release_pages(struct bio *bio, bool mark_dirty) |
| 837 | { |
| 838 | struct bvec_iter_all iter_all; |
| 839 | struct bio_vec *bvec; |
| 840 | |
| 841 | if (bio_flagged(bio, BIO_NO_PAGE_REF)) |
| 842 | return; |
| 843 | |
| 844 | bio_for_each_segment_all(bvec, bio, iter_all) { |
| 845 | if (mark_dirty && !PageCompound(bvec->bv_page)) |
| 846 | set_page_dirty_lock(bvec->bv_page); |
| 847 | put_page(bvec->bv_page); |
| 848 | } |
| 849 | } |
| 850 | |
| 851 | static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) |
| 852 | { |
| 853 | const struct bio_vec *bv = iter->bvec; |
| 854 | unsigned int len; |
| 855 | size_t size; |
| 856 | |
| 857 | if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) |
| 858 | return -EINVAL; |
| 859 | |
| 860 | len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); |
| 861 | size = bio_add_page(bio, bv->bv_page, len, |
| 862 | bv->bv_offset + iter->iov_offset); |
| 863 | if (unlikely(size != len)) |
| 864 | return -EINVAL; |
| 865 | iov_iter_advance(iter, size); |
| 866 | return 0; |
| 867 | } |
| 868 | |
| 869 | #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) |
| 870 | |
| 871 | /** |
| 872 | * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio |
| 873 | * @bio: bio to add pages to |
| 874 | * @iter: iov iterator describing the region to be mapped |
| 875 | * |
| 876 | * Pins pages from *iter and appends them to @bio's bvec array. The |
| 877 | * pages will have to be released using put_page() when done. |
| 878 | * For multi-segment *iter, this function only adds pages from the |
| 879 | * the next non-empty segment of the iov iterator. |
| 880 | */ |
| 881 | static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
| 882 | { |
| 883 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; |
| 884 | unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; |
| 885 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; |
| 886 | struct page **pages = (struct page **)bv; |
| 887 | bool same_page = false; |
| 888 | ssize_t size, left; |
| 889 | unsigned len, i; |
| 890 | size_t offset; |
| 891 | |
| 892 | /* |
| 893 | * Move page array up in the allocated memory for the bio vecs as far as |
| 894 | * possible so that we can start filling biovecs from the beginning |
| 895 | * without overwriting the temporary page array. |
| 896 | */ |
| 897 | BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); |
| 898 | pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); |
| 899 | |
| 900 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); |
| 901 | if (unlikely(size <= 0)) |
| 902 | return size ? size : -EFAULT; |
| 903 | |
| 904 | for (left = size, i = 0; left > 0; left -= len, i++) { |
| 905 | struct page *page = pages[i]; |
| 906 | |
| 907 | len = min_t(size_t, PAGE_SIZE - offset, left); |
| 908 | |
| 909 | if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { |
| 910 | if (same_page) |
| 911 | put_page(page); |
| 912 | } else { |
| 913 | if (WARN_ON_ONCE(bio_full(bio, len))) |
| 914 | return -EINVAL; |
| 915 | __bio_add_page(bio, page, len, offset); |
| 916 | } |
| 917 | offset = 0; |
| 918 | } |
| 919 | |
| 920 | iov_iter_advance(iter, size); |
| 921 | return 0; |
| 922 | } |
| 923 | |
| 924 | /** |
| 925 | * bio_iov_iter_get_pages - add user or kernel pages to a bio |
| 926 | * @bio: bio to add pages to |
| 927 | * @iter: iov iterator describing the region to be added |
| 928 | * |
| 929 | * This takes either an iterator pointing to user memory, or one pointing to |
| 930 | * kernel pages (BVEC iterator). If we're adding user pages, we pin them and |
| 931 | * map them into the kernel. On IO completion, the caller should put those |
| 932 | * pages. If we're adding kernel pages, and the caller told us it's safe to |
| 933 | * do so, we just have to add the pages to the bio directly. We don't grab an |
| 934 | * extra reference to those pages (the user should already have that), and we |
| 935 | * don't put the page on IO completion. The caller needs to check if the bio is |
| 936 | * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be |
| 937 | * released. |
| 938 | * |
| 939 | * The function tries, but does not guarantee, to pin as many pages as |
| 940 | * fit into the bio, or are requested in *iter, whatever is smaller. If |
| 941 | * MM encounters an error pinning the requested pages, it stops. Error |
| 942 | * is returned only if 0 pages could be pinned. |
| 943 | */ |
| 944 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) |
| 945 | { |
| 946 | const bool is_bvec = iov_iter_is_bvec(iter); |
| 947 | int ret; |
| 948 | |
| 949 | if (WARN_ON_ONCE(bio->bi_vcnt)) |
| 950 | return -EINVAL; |
| 951 | |
| 952 | do { |
| 953 | if (is_bvec) |
| 954 | ret = __bio_iov_bvec_add_pages(bio, iter); |
| 955 | else |
| 956 | ret = __bio_iov_iter_get_pages(bio, iter); |
| 957 | } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); |
| 958 | |
| 959 | if (is_bvec) |
| 960 | bio_set_flag(bio, BIO_NO_PAGE_REF); |
| 961 | return bio->bi_vcnt ? 0 : ret; |
| 962 | } |
| 963 | |
| 964 | static void submit_bio_wait_endio(struct bio *bio) |
| 965 | { |
| 966 | complete(bio->bi_private); |
| 967 | } |
| 968 | |
| 969 | /** |
| 970 | * submit_bio_wait - submit a bio, and wait until it completes |
| 971 | * @bio: The &struct bio which describes the I/O |
| 972 | * |
| 973 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from |
| 974 | * bio_endio() on failure. |
| 975 | * |
| 976 | * WARNING: Unlike to how submit_bio() is usually used, this function does not |
| 977 | * result in bio reference to be consumed. The caller must drop the reference |
| 978 | * on his own. |
| 979 | */ |
| 980 | int submit_bio_wait(struct bio *bio) |
| 981 | { |
| 982 | DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); |
| 983 | |
| 984 | bio->bi_private = &done; |
| 985 | bio->bi_end_io = submit_bio_wait_endio; |
| 986 | bio->bi_opf |= REQ_SYNC; |
| 987 | submit_bio(bio); |
| 988 | wait_for_completion_io(&done); |
| 989 | |
| 990 | return blk_status_to_errno(bio->bi_status); |
| 991 | } |
| 992 | EXPORT_SYMBOL(submit_bio_wait); |
| 993 | |
| 994 | /** |
| 995 | * bio_advance - increment/complete a bio by some number of bytes |
| 996 | * @bio: bio to advance |
| 997 | * @bytes: number of bytes to complete |
| 998 | * |
| 999 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to |
| 1000 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will |
| 1001 | * be updated on the last bvec as well. |
| 1002 | * |
| 1003 | * @bio will then represent the remaining, uncompleted portion of the io. |
| 1004 | */ |
| 1005 | void bio_advance(struct bio *bio, unsigned bytes) |
| 1006 | { |
| 1007 | if (bio_integrity(bio)) |
| 1008 | bio_integrity_advance(bio, bytes); |
| 1009 | |
| 1010 | bio_advance_iter(bio, &bio->bi_iter, bytes); |
| 1011 | } |
| 1012 | EXPORT_SYMBOL(bio_advance); |
| 1013 | |
| 1014 | void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, |
| 1015 | struct bio *src, struct bvec_iter *src_iter) |
| 1016 | { |
| 1017 | struct bio_vec src_bv, dst_bv; |
| 1018 | void *src_p, *dst_p; |
| 1019 | unsigned bytes; |
| 1020 | |
| 1021 | while (src_iter->bi_size && dst_iter->bi_size) { |
| 1022 | src_bv = bio_iter_iovec(src, *src_iter); |
| 1023 | dst_bv = bio_iter_iovec(dst, *dst_iter); |
| 1024 | |
| 1025 | bytes = min(src_bv.bv_len, dst_bv.bv_len); |
| 1026 | |
| 1027 | src_p = kmap_atomic(src_bv.bv_page); |
| 1028 | dst_p = kmap_atomic(dst_bv.bv_page); |
| 1029 | |
| 1030 | memcpy(dst_p + dst_bv.bv_offset, |
| 1031 | src_p + src_bv.bv_offset, |
| 1032 | bytes); |
| 1033 | |
| 1034 | kunmap_atomic(dst_p); |
| 1035 | kunmap_atomic(src_p); |
| 1036 | |
| 1037 | flush_dcache_page(dst_bv.bv_page); |
| 1038 | |
| 1039 | bio_advance_iter(src, src_iter, bytes); |
| 1040 | bio_advance_iter(dst, dst_iter, bytes); |
| 1041 | } |
| 1042 | } |
| 1043 | EXPORT_SYMBOL(bio_copy_data_iter); |
| 1044 | |
| 1045 | /** |
| 1046 | * bio_copy_data - copy contents of data buffers from one bio to another |
| 1047 | * @src: source bio |
| 1048 | * @dst: destination bio |
| 1049 | * |
| 1050 | * Stops when it reaches the end of either @src or @dst - that is, copies |
| 1051 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). |
| 1052 | */ |
| 1053 | void bio_copy_data(struct bio *dst, struct bio *src) |
| 1054 | { |
| 1055 | struct bvec_iter src_iter = src->bi_iter; |
| 1056 | struct bvec_iter dst_iter = dst->bi_iter; |
| 1057 | |
| 1058 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); |
| 1059 | } |
| 1060 | EXPORT_SYMBOL(bio_copy_data); |
| 1061 | |
| 1062 | /** |
| 1063 | * bio_list_copy_data - copy contents of data buffers from one chain of bios to |
| 1064 | * another |
| 1065 | * @src: source bio list |
| 1066 | * @dst: destination bio list |
| 1067 | * |
| 1068 | * Stops when it reaches the end of either the @src list or @dst list - that is, |
| 1069 | * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of |
| 1070 | * bios). |
| 1071 | */ |
| 1072 | void bio_list_copy_data(struct bio *dst, struct bio *src) |
| 1073 | { |
| 1074 | struct bvec_iter src_iter = src->bi_iter; |
| 1075 | struct bvec_iter dst_iter = dst->bi_iter; |
| 1076 | |
| 1077 | while (1) { |
| 1078 | if (!src_iter.bi_size) { |
| 1079 | src = src->bi_next; |
| 1080 | if (!src) |
| 1081 | break; |
| 1082 | |
| 1083 | src_iter = src->bi_iter; |
| 1084 | } |
| 1085 | |
| 1086 | if (!dst_iter.bi_size) { |
| 1087 | dst = dst->bi_next; |
| 1088 | if (!dst) |
| 1089 | break; |
| 1090 | |
| 1091 | dst_iter = dst->bi_iter; |
| 1092 | } |
| 1093 | |
| 1094 | bio_copy_data_iter(dst, &dst_iter, src, &src_iter); |
| 1095 | } |
| 1096 | } |
| 1097 | EXPORT_SYMBOL(bio_list_copy_data); |
| 1098 | |
| 1099 | struct bio_map_data { |
| 1100 | int is_our_pages; |
| 1101 | struct iov_iter iter; |
| 1102 | struct iovec iov[]; |
| 1103 | }; |
| 1104 | |
| 1105 | static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, |
| 1106 | gfp_t gfp_mask) |
| 1107 | { |
| 1108 | struct bio_map_data *bmd; |
| 1109 | if (data->nr_segs > UIO_MAXIOV) |
| 1110 | return NULL; |
| 1111 | |
| 1112 | bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); |
| 1113 | if (!bmd) |
| 1114 | return NULL; |
| 1115 | memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); |
| 1116 | bmd->iter = *data; |
| 1117 | bmd->iter.iov = bmd->iov; |
| 1118 | return bmd; |
| 1119 | } |
| 1120 | |
| 1121 | /** |
| 1122 | * bio_copy_from_iter - copy all pages from iov_iter to bio |
| 1123 | * @bio: The &struct bio which describes the I/O as destination |
| 1124 | * @iter: iov_iter as source |
| 1125 | * |
| 1126 | * Copy all pages from iov_iter to bio. |
| 1127 | * Returns 0 on success, or error on failure. |
| 1128 | */ |
| 1129 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) |
| 1130 | { |
| 1131 | struct bio_vec *bvec; |
| 1132 | struct bvec_iter_all iter_all; |
| 1133 | |
| 1134 | bio_for_each_segment_all(bvec, bio, iter_all) { |
| 1135 | ssize_t ret; |
| 1136 | |
| 1137 | ret = copy_page_from_iter(bvec->bv_page, |
| 1138 | bvec->bv_offset, |
| 1139 | bvec->bv_len, |
| 1140 | iter); |
| 1141 | |
| 1142 | if (!iov_iter_count(iter)) |
| 1143 | break; |
| 1144 | |
| 1145 | if (ret < bvec->bv_len) |
| 1146 | return -EFAULT; |
| 1147 | } |
| 1148 | |
| 1149 | return 0; |
| 1150 | } |
| 1151 | |
| 1152 | /** |
| 1153 | * bio_copy_to_iter - copy all pages from bio to iov_iter |
| 1154 | * @bio: The &struct bio which describes the I/O as source |
| 1155 | * @iter: iov_iter as destination |
| 1156 | * |
| 1157 | * Copy all pages from bio to iov_iter. |
| 1158 | * Returns 0 on success, or error on failure. |
| 1159 | */ |
| 1160 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) |
| 1161 | { |
| 1162 | struct bio_vec *bvec; |
| 1163 | struct bvec_iter_all iter_all; |
| 1164 | |
| 1165 | bio_for_each_segment_all(bvec, bio, iter_all) { |
| 1166 | ssize_t ret; |
| 1167 | |
| 1168 | ret = copy_page_to_iter(bvec->bv_page, |
| 1169 | bvec->bv_offset, |
| 1170 | bvec->bv_len, |
| 1171 | &iter); |
| 1172 | |
| 1173 | if (!iov_iter_count(&iter)) |
| 1174 | break; |
| 1175 | |
| 1176 | if (ret < bvec->bv_len) |
| 1177 | return -EFAULT; |
| 1178 | } |
| 1179 | |
| 1180 | return 0; |
| 1181 | } |
| 1182 | |
| 1183 | void bio_free_pages(struct bio *bio) |
| 1184 | { |
| 1185 | struct bio_vec *bvec; |
| 1186 | struct bvec_iter_all iter_all; |
| 1187 | |
| 1188 | bio_for_each_segment_all(bvec, bio, iter_all) |
| 1189 | __free_page(bvec->bv_page); |
| 1190 | } |
| 1191 | EXPORT_SYMBOL(bio_free_pages); |
| 1192 | |
| 1193 | /** |
| 1194 | * bio_uncopy_user - finish previously mapped bio |
| 1195 | * @bio: bio being terminated |
| 1196 | * |
| 1197 | * Free pages allocated from bio_copy_user_iov() and write back data |
| 1198 | * to user space in case of a read. |
| 1199 | */ |
| 1200 | int bio_uncopy_user(struct bio *bio) |
| 1201 | { |
| 1202 | struct bio_map_data *bmd = bio->bi_private; |
| 1203 | int ret = 0; |
| 1204 | |
| 1205 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { |
| 1206 | /* |
| 1207 | * if we're in a workqueue, the request is orphaned, so |
| 1208 | * don't copy into a random user address space, just free |
| 1209 | * and return -EINTR so user space doesn't expect any data. |
| 1210 | */ |
| 1211 | if (!current->mm) |
| 1212 | ret = -EINTR; |
| 1213 | else if (bio_data_dir(bio) == READ) |
| 1214 | ret = bio_copy_to_iter(bio, bmd->iter); |
| 1215 | if (bmd->is_our_pages) |
| 1216 | bio_free_pages(bio); |
| 1217 | } |
| 1218 | kfree(bmd); |
| 1219 | bio_put(bio); |
| 1220 | return ret; |
| 1221 | } |
| 1222 | |
| 1223 | /** |
| 1224 | * bio_copy_user_iov - copy user data to bio |
| 1225 | * @q: destination block queue |
| 1226 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
| 1227 | * @iter: iovec iterator |
| 1228 | * @gfp_mask: memory allocation flags |
| 1229 | * |
| 1230 | * Prepares and returns a bio for indirect user io, bouncing data |
| 1231 | * to/from kernel pages as necessary. Must be paired with |
| 1232 | * call bio_uncopy_user() on io completion. |
| 1233 | */ |
| 1234 | struct bio *bio_copy_user_iov(struct request_queue *q, |
| 1235 | struct rq_map_data *map_data, |
| 1236 | struct iov_iter *iter, |
| 1237 | gfp_t gfp_mask) |
| 1238 | { |
| 1239 | struct bio_map_data *bmd; |
| 1240 | struct page *page; |
| 1241 | struct bio *bio; |
| 1242 | int i = 0, ret; |
| 1243 | int nr_pages; |
| 1244 | unsigned int len = iter->count; |
| 1245 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; |
| 1246 | |
| 1247 | bmd = bio_alloc_map_data(iter, gfp_mask); |
| 1248 | if (!bmd) |
| 1249 | return ERR_PTR(-ENOMEM); |
| 1250 | |
| 1251 | /* |
| 1252 | * We need to do a deep copy of the iov_iter including the iovecs. |
| 1253 | * The caller provided iov might point to an on-stack or otherwise |
| 1254 | * shortlived one. |
| 1255 | */ |
| 1256 | bmd->is_our_pages = map_data ? 0 : 1; |
| 1257 | |
| 1258 | nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); |
| 1259 | if (nr_pages > BIO_MAX_PAGES) |
| 1260 | nr_pages = BIO_MAX_PAGES; |
| 1261 | |
| 1262 | ret = -ENOMEM; |
| 1263 | bio = bio_kmalloc(gfp_mask, nr_pages); |
| 1264 | if (!bio) |
| 1265 | goto out_bmd; |
| 1266 | |
| 1267 | ret = 0; |
| 1268 | |
| 1269 | if (map_data) { |
| 1270 | nr_pages = 1 << map_data->page_order; |
| 1271 | i = map_data->offset / PAGE_SIZE; |
| 1272 | } |
| 1273 | while (len) { |
| 1274 | unsigned int bytes = PAGE_SIZE; |
| 1275 | |
| 1276 | bytes -= offset; |
| 1277 | |
| 1278 | if (bytes > len) |
| 1279 | bytes = len; |
| 1280 | |
| 1281 | if (map_data) { |
| 1282 | if (i == map_data->nr_entries * nr_pages) { |
| 1283 | ret = -ENOMEM; |
| 1284 | break; |
| 1285 | } |
| 1286 | |
| 1287 | page = map_data->pages[i / nr_pages]; |
| 1288 | page += (i % nr_pages); |
| 1289 | |
| 1290 | i++; |
| 1291 | } else { |
| 1292 | page = alloc_page(q->bounce_gfp | gfp_mask); |
| 1293 | if (!page) { |
| 1294 | ret = -ENOMEM; |
| 1295 | break; |
| 1296 | } |
| 1297 | } |
| 1298 | |
| 1299 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { |
| 1300 | if (!map_data) |
| 1301 | __free_page(page); |
| 1302 | break; |
| 1303 | } |
| 1304 | |
| 1305 | len -= bytes; |
| 1306 | offset = 0; |
| 1307 | } |
| 1308 | |
| 1309 | if (ret) |
| 1310 | goto cleanup; |
| 1311 | |
| 1312 | if (map_data) |
| 1313 | map_data->offset += bio->bi_iter.bi_size; |
| 1314 | |
| 1315 | /* |
| 1316 | * success |
| 1317 | */ |
| 1318 | if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || |
| 1319 | (map_data && map_data->from_user)) { |
| 1320 | ret = bio_copy_from_iter(bio, iter); |
| 1321 | if (ret) |
| 1322 | goto cleanup; |
| 1323 | } else { |
| 1324 | if (bmd->is_our_pages) |
| 1325 | zero_fill_bio(bio); |
| 1326 | iov_iter_advance(iter, bio->bi_iter.bi_size); |
| 1327 | } |
| 1328 | |
| 1329 | bio->bi_private = bmd; |
| 1330 | if (map_data && map_data->null_mapped) |
| 1331 | bio_set_flag(bio, BIO_NULL_MAPPED); |
| 1332 | return bio; |
| 1333 | cleanup: |
| 1334 | if (!map_data) |
| 1335 | bio_free_pages(bio); |
| 1336 | bio_put(bio); |
| 1337 | out_bmd: |
| 1338 | kfree(bmd); |
| 1339 | return ERR_PTR(ret); |
| 1340 | } |
| 1341 | |
| 1342 | /** |
| 1343 | * bio_map_user_iov - map user iovec into bio |
| 1344 | * @q: the struct request_queue for the bio |
| 1345 | * @iter: iovec iterator |
| 1346 | * @gfp_mask: memory allocation flags |
| 1347 | * |
| 1348 | * Map the user space address into a bio suitable for io to a block |
| 1349 | * device. Returns an error pointer in case of error. |
| 1350 | */ |
| 1351 | struct bio *bio_map_user_iov(struct request_queue *q, |
| 1352 | struct iov_iter *iter, |
| 1353 | gfp_t gfp_mask) |
| 1354 | { |
| 1355 | int j; |
| 1356 | struct bio *bio; |
| 1357 | int ret; |
| 1358 | |
| 1359 | if (!iov_iter_count(iter)) |
| 1360 | return ERR_PTR(-EINVAL); |
| 1361 | |
| 1362 | bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); |
| 1363 | if (!bio) |
| 1364 | return ERR_PTR(-ENOMEM); |
| 1365 | |
| 1366 | while (iov_iter_count(iter)) { |
| 1367 | struct page **pages; |
| 1368 | ssize_t bytes; |
| 1369 | size_t offs, added = 0; |
| 1370 | int npages; |
| 1371 | |
| 1372 | bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); |
| 1373 | if (unlikely(bytes <= 0)) { |
| 1374 | ret = bytes ? bytes : -EFAULT; |
| 1375 | goto out_unmap; |
| 1376 | } |
| 1377 | |
| 1378 | npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); |
| 1379 | |
| 1380 | if (unlikely(offs & queue_dma_alignment(q))) { |
| 1381 | ret = -EINVAL; |
| 1382 | j = 0; |
| 1383 | } else { |
| 1384 | for (j = 0; j < npages; j++) { |
| 1385 | struct page *page = pages[j]; |
| 1386 | unsigned int n = PAGE_SIZE - offs; |
| 1387 | |
| 1388 | if (n > bytes) |
| 1389 | n = bytes; |
| 1390 | |
| 1391 | if (!__bio_add_pc_page(q, bio, page, n, offs, |
| 1392 | true)) |
| 1393 | break; |
| 1394 | |
| 1395 | added += n; |
| 1396 | bytes -= n; |
| 1397 | offs = 0; |
| 1398 | } |
| 1399 | iov_iter_advance(iter, added); |
| 1400 | } |
| 1401 | /* |
| 1402 | * release the pages we didn't map into the bio, if any |
| 1403 | */ |
| 1404 | while (j < npages) |
| 1405 | put_page(pages[j++]); |
| 1406 | kvfree(pages); |
| 1407 | /* couldn't stuff something into bio? */ |
| 1408 | if (bytes) |
| 1409 | break; |
| 1410 | } |
| 1411 | |
| 1412 | bio_set_flag(bio, BIO_USER_MAPPED); |
| 1413 | |
| 1414 | /* |
| 1415 | * subtle -- if bio_map_user_iov() ended up bouncing a bio, |
| 1416 | * it would normally disappear when its bi_end_io is run. |
| 1417 | * however, we need it for the unmap, so grab an extra |
| 1418 | * reference to it |
| 1419 | */ |
| 1420 | bio_get(bio); |
| 1421 | return bio; |
| 1422 | |
| 1423 | out_unmap: |
| 1424 | bio_release_pages(bio, false); |
| 1425 | bio_put(bio); |
| 1426 | return ERR_PTR(ret); |
| 1427 | } |
| 1428 | |
| 1429 | /** |
| 1430 | * bio_unmap_user - unmap a bio |
| 1431 | * @bio: the bio being unmapped |
| 1432 | * |
| 1433 | * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from |
| 1434 | * process context. |
| 1435 | * |
| 1436 | * bio_unmap_user() may sleep. |
| 1437 | */ |
| 1438 | void bio_unmap_user(struct bio *bio) |
| 1439 | { |
| 1440 | bio_release_pages(bio, bio_data_dir(bio) == READ); |
| 1441 | bio_put(bio); |
| 1442 | bio_put(bio); |
| 1443 | } |
| 1444 | |
| 1445 | static void bio_invalidate_vmalloc_pages(struct bio *bio) |
| 1446 | { |
| 1447 | #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE |
| 1448 | if (bio->bi_private && !op_is_write(bio_op(bio))) { |
| 1449 | unsigned long i, len = 0; |
| 1450 | |
| 1451 | for (i = 0; i < bio->bi_vcnt; i++) |
| 1452 | len += bio->bi_io_vec[i].bv_len; |
| 1453 | invalidate_kernel_vmap_range(bio->bi_private, len); |
| 1454 | } |
| 1455 | #endif |
| 1456 | } |
| 1457 | |
| 1458 | static void bio_map_kern_endio(struct bio *bio) |
| 1459 | { |
| 1460 | bio_invalidate_vmalloc_pages(bio); |
| 1461 | bio_put(bio); |
| 1462 | } |
| 1463 | |
| 1464 | /** |
| 1465 | * bio_map_kern - map kernel address into bio |
| 1466 | * @q: the struct request_queue for the bio |
| 1467 | * @data: pointer to buffer to map |
| 1468 | * @len: length in bytes |
| 1469 | * @gfp_mask: allocation flags for bio allocation |
| 1470 | * |
| 1471 | * Map the kernel address into a bio suitable for io to a block |
| 1472 | * device. Returns an error pointer in case of error. |
| 1473 | */ |
| 1474 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
| 1475 | gfp_t gfp_mask) |
| 1476 | { |
| 1477 | unsigned long kaddr = (unsigned long)data; |
| 1478 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1479 | unsigned long start = kaddr >> PAGE_SHIFT; |
| 1480 | const int nr_pages = end - start; |
| 1481 | bool is_vmalloc = is_vmalloc_addr(data); |
| 1482 | struct page *page; |
| 1483 | int offset, i; |
| 1484 | struct bio *bio; |
| 1485 | |
| 1486 | bio = bio_kmalloc(gfp_mask, nr_pages); |
| 1487 | if (!bio) |
| 1488 | return ERR_PTR(-ENOMEM); |
| 1489 | |
| 1490 | if (is_vmalloc) { |
| 1491 | flush_kernel_vmap_range(data, len); |
| 1492 | bio->bi_private = data; |
| 1493 | } |
| 1494 | |
| 1495 | offset = offset_in_page(kaddr); |
| 1496 | for (i = 0; i < nr_pages; i++) { |
| 1497 | unsigned int bytes = PAGE_SIZE - offset; |
| 1498 | |
| 1499 | if (len <= 0) |
| 1500 | break; |
| 1501 | |
| 1502 | if (bytes > len) |
| 1503 | bytes = len; |
| 1504 | |
| 1505 | if (!is_vmalloc) |
| 1506 | page = virt_to_page(data); |
| 1507 | else |
| 1508 | page = vmalloc_to_page(data); |
| 1509 | if (bio_add_pc_page(q, bio, page, bytes, |
| 1510 | offset) < bytes) { |
| 1511 | /* we don't support partial mappings */ |
| 1512 | bio_put(bio); |
| 1513 | return ERR_PTR(-EINVAL); |
| 1514 | } |
| 1515 | |
| 1516 | data += bytes; |
| 1517 | len -= bytes; |
| 1518 | offset = 0; |
| 1519 | } |
| 1520 | |
| 1521 | bio->bi_end_io = bio_map_kern_endio; |
| 1522 | return bio; |
| 1523 | } |
| 1524 | EXPORT_SYMBOL(bio_map_kern); |
| 1525 | |
| 1526 | static void bio_copy_kern_endio(struct bio *bio) |
| 1527 | { |
| 1528 | bio_free_pages(bio); |
| 1529 | bio_put(bio); |
| 1530 | } |
| 1531 | |
| 1532 | static void bio_copy_kern_endio_read(struct bio *bio) |
| 1533 | { |
| 1534 | char *p = bio->bi_private; |
| 1535 | struct bio_vec *bvec; |
| 1536 | struct bvec_iter_all iter_all; |
| 1537 | |
| 1538 | bio_for_each_segment_all(bvec, bio, iter_all) { |
| 1539 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); |
| 1540 | p += bvec->bv_len; |
| 1541 | } |
| 1542 | |
| 1543 | bio_copy_kern_endio(bio); |
| 1544 | } |
| 1545 | |
| 1546 | /** |
| 1547 | * bio_copy_kern - copy kernel address into bio |
| 1548 | * @q: the struct request_queue for the bio |
| 1549 | * @data: pointer to buffer to copy |
| 1550 | * @len: length in bytes |
| 1551 | * @gfp_mask: allocation flags for bio and page allocation |
| 1552 | * @reading: data direction is READ |
| 1553 | * |
| 1554 | * copy the kernel address into a bio suitable for io to a block |
| 1555 | * device. Returns an error pointer in case of error. |
| 1556 | */ |
| 1557 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, |
| 1558 | gfp_t gfp_mask, int reading) |
| 1559 | { |
| 1560 | unsigned long kaddr = (unsigned long)data; |
| 1561 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1562 | unsigned long start = kaddr >> PAGE_SHIFT; |
| 1563 | struct bio *bio; |
| 1564 | void *p = data; |
| 1565 | int nr_pages = 0; |
| 1566 | |
| 1567 | /* |
| 1568 | * Overflow, abort |
| 1569 | */ |
| 1570 | if (end < start) |
| 1571 | return ERR_PTR(-EINVAL); |
| 1572 | |
| 1573 | nr_pages = end - start; |
| 1574 | bio = bio_kmalloc(gfp_mask, nr_pages); |
| 1575 | if (!bio) |
| 1576 | return ERR_PTR(-ENOMEM); |
| 1577 | |
| 1578 | while (len) { |
| 1579 | struct page *page; |
| 1580 | unsigned int bytes = PAGE_SIZE; |
| 1581 | |
| 1582 | if (bytes > len) |
| 1583 | bytes = len; |
| 1584 | |
| 1585 | page = alloc_page(q->bounce_gfp | gfp_mask); |
| 1586 | if (!page) |
| 1587 | goto cleanup; |
| 1588 | |
| 1589 | if (!reading) |
| 1590 | memcpy(page_address(page), p, bytes); |
| 1591 | |
| 1592 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
| 1593 | break; |
| 1594 | |
| 1595 | len -= bytes; |
| 1596 | p += bytes; |
| 1597 | } |
| 1598 | |
| 1599 | if (reading) { |
| 1600 | bio->bi_end_io = bio_copy_kern_endio_read; |
| 1601 | bio->bi_private = data; |
| 1602 | } else { |
| 1603 | bio->bi_end_io = bio_copy_kern_endio; |
| 1604 | } |
| 1605 | |
| 1606 | return bio; |
| 1607 | |
| 1608 | cleanup: |
| 1609 | bio_free_pages(bio); |
| 1610 | bio_put(bio); |
| 1611 | return ERR_PTR(-ENOMEM); |
| 1612 | } |
| 1613 | |
| 1614 | /* |
| 1615 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions |
| 1616 | * for performing direct-IO in BIOs. |
| 1617 | * |
| 1618 | * The problem is that we cannot run set_page_dirty() from interrupt context |
| 1619 | * because the required locks are not interrupt-safe. So what we can do is to |
| 1620 | * mark the pages dirty _before_ performing IO. And in interrupt context, |
| 1621 | * check that the pages are still dirty. If so, fine. If not, redirty them |
| 1622 | * in process context. |
| 1623 | * |
| 1624 | * We special-case compound pages here: normally this means reads into hugetlb |
| 1625 | * pages. The logic in here doesn't really work right for compound pages |
| 1626 | * because the VM does not uniformly chase down the head page in all cases. |
| 1627 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't |
| 1628 | * handle them at all. So we skip compound pages here at an early stage. |
| 1629 | * |
| 1630 | * Note that this code is very hard to test under normal circumstances because |
| 1631 | * direct-io pins the pages with get_user_pages(). This makes |
| 1632 | * is_page_cache_freeable return false, and the VM will not clean the pages. |
| 1633 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
| 1634 | * pagecache. |
| 1635 | * |
| 1636 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the |
| 1637 | * deferred bio dirtying paths. |
| 1638 | */ |
| 1639 | |
| 1640 | /* |
| 1641 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. |
| 1642 | */ |
| 1643 | void bio_set_pages_dirty(struct bio *bio) |
| 1644 | { |
| 1645 | struct bio_vec *bvec; |
| 1646 | struct bvec_iter_all iter_all; |
| 1647 | |
| 1648 | bio_for_each_segment_all(bvec, bio, iter_all) { |
| 1649 | if (!PageCompound(bvec->bv_page)) |
| 1650 | set_page_dirty_lock(bvec->bv_page); |
| 1651 | } |
| 1652 | } |
| 1653 | |
| 1654 | /* |
| 1655 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. |
| 1656 | * If they are, then fine. If, however, some pages are clean then they must |
| 1657 | * have been written out during the direct-IO read. So we take another ref on |
| 1658 | * the BIO and re-dirty the pages in process context. |
| 1659 | * |
| 1660 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from |
| 1661 | * here on. It will run one put_page() against each page and will run one |
| 1662 | * bio_put() against the BIO. |
| 1663 | */ |
| 1664 | |
| 1665 | static void bio_dirty_fn(struct work_struct *work); |
| 1666 | |
| 1667 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
| 1668 | static DEFINE_SPINLOCK(bio_dirty_lock); |
| 1669 | static struct bio *bio_dirty_list; |
| 1670 | |
| 1671 | /* |
| 1672 | * This runs in process context |
| 1673 | */ |
| 1674 | static void bio_dirty_fn(struct work_struct *work) |
| 1675 | { |
| 1676 | struct bio *bio, *next; |
| 1677 | |
| 1678 | spin_lock_irq(&bio_dirty_lock); |
| 1679 | next = bio_dirty_list; |
| 1680 | bio_dirty_list = NULL; |
| 1681 | spin_unlock_irq(&bio_dirty_lock); |
| 1682 | |
| 1683 | while ((bio = next) != NULL) { |
| 1684 | next = bio->bi_private; |
| 1685 | |
| 1686 | bio_release_pages(bio, true); |
| 1687 | bio_put(bio); |
| 1688 | } |
| 1689 | } |
| 1690 | |
| 1691 | void bio_check_pages_dirty(struct bio *bio) |
| 1692 | { |
| 1693 | struct bio_vec *bvec; |
| 1694 | unsigned long flags; |
| 1695 | struct bvec_iter_all iter_all; |
| 1696 | |
| 1697 | bio_for_each_segment_all(bvec, bio, iter_all) { |
| 1698 | if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) |
| 1699 | goto defer; |
| 1700 | } |
| 1701 | |
| 1702 | bio_release_pages(bio, false); |
| 1703 | bio_put(bio); |
| 1704 | return; |
| 1705 | defer: |
| 1706 | spin_lock_irqsave(&bio_dirty_lock, flags); |
| 1707 | bio->bi_private = bio_dirty_list; |
| 1708 | bio_dirty_list = bio; |
| 1709 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
| 1710 | schedule_work(&bio_dirty_work); |
| 1711 | } |
| 1712 | |
| 1713 | void update_io_ticks(struct hd_struct *part, unsigned long now) |
| 1714 | { |
| 1715 | unsigned long stamp; |
| 1716 | again: |
| 1717 | stamp = READ_ONCE(part->stamp); |
| 1718 | if (unlikely(stamp != now)) { |
| 1719 | if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { |
| 1720 | __part_stat_add(part, io_ticks, 1); |
| 1721 | } |
| 1722 | } |
| 1723 | if (part->partno) { |
| 1724 | part = &part_to_disk(part)->part0; |
| 1725 | goto again; |
| 1726 | } |
| 1727 | } |
| 1728 | |
| 1729 | void generic_start_io_acct(struct request_queue *q, int op, |
| 1730 | unsigned long sectors, struct hd_struct *part) |
| 1731 | { |
| 1732 | const int sgrp = op_stat_group(op); |
| 1733 | |
| 1734 | part_stat_lock(); |
| 1735 | |
| 1736 | update_io_ticks(part, jiffies); |
| 1737 | part_stat_inc(part, ios[sgrp]); |
| 1738 | part_stat_add(part, sectors[sgrp], sectors); |
| 1739 | part_inc_in_flight(q, part, op_is_write(op)); |
| 1740 | |
| 1741 | part_stat_unlock(); |
| 1742 | } |
| 1743 | EXPORT_SYMBOL(generic_start_io_acct); |
| 1744 | |
| 1745 | void generic_end_io_acct(struct request_queue *q, int req_op, |
| 1746 | struct hd_struct *part, unsigned long start_time) |
| 1747 | { |
| 1748 | unsigned long now = jiffies; |
| 1749 | unsigned long duration = now - start_time; |
| 1750 | const int sgrp = op_stat_group(req_op); |
| 1751 | |
| 1752 | part_stat_lock(); |
| 1753 | |
| 1754 | update_io_ticks(part, now); |
| 1755 | part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); |
| 1756 | part_stat_add(part, time_in_queue, duration); |
| 1757 | part_dec_in_flight(q, part, op_is_write(req_op)); |
| 1758 | |
| 1759 | part_stat_unlock(); |
| 1760 | } |
| 1761 | EXPORT_SYMBOL(generic_end_io_acct); |
| 1762 | |
| 1763 | static inline bool bio_remaining_done(struct bio *bio) |
| 1764 | { |
| 1765 | /* |
| 1766 | * If we're not chaining, then ->__bi_remaining is always 1 and |
| 1767 | * we always end io on the first invocation. |
| 1768 | */ |
| 1769 | if (!bio_flagged(bio, BIO_CHAIN)) |
| 1770 | return true; |
| 1771 | |
| 1772 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); |
| 1773 | |
| 1774 | if (atomic_dec_and_test(&bio->__bi_remaining)) { |
| 1775 | bio_clear_flag(bio, BIO_CHAIN); |
| 1776 | return true; |
| 1777 | } |
| 1778 | |
| 1779 | return false; |
| 1780 | } |
| 1781 | |
| 1782 | /** |
| 1783 | * bio_endio - end I/O on a bio |
| 1784 | * @bio: bio |
| 1785 | * |
| 1786 | * Description: |
| 1787 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred |
| 1788 | * way to end I/O on a bio. No one should call bi_end_io() directly on a |
| 1789 | * bio unless they own it and thus know that it has an end_io function. |
| 1790 | * |
| 1791 | * bio_endio() can be called several times on a bio that has been chained |
| 1792 | * using bio_chain(). The ->bi_end_io() function will only be called the |
| 1793 | * last time. At this point the BLK_TA_COMPLETE tracing event will be |
| 1794 | * generated if BIO_TRACE_COMPLETION is set. |
| 1795 | **/ |
| 1796 | void bio_endio(struct bio *bio) |
| 1797 | { |
| 1798 | again: |
| 1799 | if (!bio_remaining_done(bio)) |
| 1800 | return; |
| 1801 | if (!bio_integrity_endio(bio)) |
| 1802 | return; |
| 1803 | |
| 1804 | if (bio->bi_disk) |
| 1805 | rq_qos_done_bio(bio->bi_disk->queue, bio); |
| 1806 | |
| 1807 | /* |
| 1808 | * Need to have a real endio function for chained bios, otherwise |
| 1809 | * various corner cases will break (like stacking block devices that |
| 1810 | * save/restore bi_end_io) - however, we want to avoid unbounded |
| 1811 | * recursion and blowing the stack. Tail call optimization would |
| 1812 | * handle this, but compiling with frame pointers also disables |
| 1813 | * gcc's sibling call optimization. |
| 1814 | */ |
| 1815 | if (bio->bi_end_io == bio_chain_endio) { |
| 1816 | bio = __bio_chain_endio(bio); |
| 1817 | goto again; |
| 1818 | } |
| 1819 | |
| 1820 | if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
| 1821 | trace_block_bio_complete(bio->bi_disk->queue, bio, |
| 1822 | blk_status_to_errno(bio->bi_status)); |
| 1823 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
| 1824 | } |
| 1825 | |
| 1826 | blk_throtl_bio_endio(bio); |
| 1827 | /* release cgroup info */ |
| 1828 | bio_uninit(bio); |
| 1829 | if (bio->bi_end_io) |
| 1830 | bio->bi_end_io(bio); |
| 1831 | } |
| 1832 | EXPORT_SYMBOL(bio_endio); |
| 1833 | |
| 1834 | /** |
| 1835 | * bio_split - split a bio |
| 1836 | * @bio: bio to split |
| 1837 | * @sectors: number of sectors to split from the front of @bio |
| 1838 | * @gfp: gfp mask |
| 1839 | * @bs: bio set to allocate from |
| 1840 | * |
| 1841 | * Allocates and returns a new bio which represents @sectors from the start of |
| 1842 | * @bio, and updates @bio to represent the remaining sectors. |
| 1843 | * |
| 1844 | * Unless this is a discard request the newly allocated bio will point |
| 1845 | * to @bio's bi_io_vec; it is the caller's responsibility to ensure that |
| 1846 | * @bio is not freed before the split. |
| 1847 | */ |
| 1848 | struct bio *bio_split(struct bio *bio, int sectors, |
| 1849 | gfp_t gfp, struct bio_set *bs) |
| 1850 | { |
| 1851 | struct bio *split; |
| 1852 | |
| 1853 | BUG_ON(sectors <= 0); |
| 1854 | BUG_ON(sectors >= bio_sectors(bio)); |
| 1855 | |
| 1856 | split = bio_clone_fast(bio, gfp, bs); |
| 1857 | if (!split) |
| 1858 | return NULL; |
| 1859 | |
| 1860 | split->bi_iter.bi_size = sectors << 9; |
| 1861 | |
| 1862 | if (bio_integrity(split)) |
| 1863 | bio_integrity_trim(split); |
| 1864 | |
| 1865 | bio_advance(bio, split->bi_iter.bi_size); |
| 1866 | |
| 1867 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) |
| 1868 | bio_set_flag(split, BIO_TRACE_COMPLETION); |
| 1869 | |
| 1870 | return split; |
| 1871 | } |
| 1872 | EXPORT_SYMBOL(bio_split); |
| 1873 | |
| 1874 | /** |
| 1875 | * bio_trim - trim a bio |
| 1876 | * @bio: bio to trim |
| 1877 | * @offset: number of sectors to trim from the front of @bio |
| 1878 | * @size: size we want to trim @bio to, in sectors |
| 1879 | */ |
| 1880 | void bio_trim(struct bio *bio, int offset, int size) |
| 1881 | { |
| 1882 | /* 'bio' is a cloned bio which we need to trim to match |
| 1883 | * the given offset and size. |
| 1884 | */ |
| 1885 | |
| 1886 | size <<= 9; |
| 1887 | if (offset == 0 && size == bio->bi_iter.bi_size) |
| 1888 | return; |
| 1889 | |
| 1890 | bio_advance(bio, offset << 9); |
| 1891 | bio->bi_iter.bi_size = size; |
| 1892 | |
| 1893 | if (bio_integrity(bio)) |
| 1894 | bio_integrity_trim(bio); |
| 1895 | |
| 1896 | } |
| 1897 | EXPORT_SYMBOL_GPL(bio_trim); |
| 1898 | |
| 1899 | /* |
| 1900 | * create memory pools for biovec's in a bio_set. |
| 1901 | * use the global biovec slabs created for general use. |
| 1902 | */ |
| 1903 | int biovec_init_pool(mempool_t *pool, int pool_entries) |
| 1904 | { |
| 1905 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; |
| 1906 | |
| 1907 | return mempool_init_slab_pool(pool, pool_entries, bp->slab); |
| 1908 | } |
| 1909 | |
| 1910 | /* |
| 1911 | * bioset_exit - exit a bioset initialized with bioset_init() |
| 1912 | * |
| 1913 | * May be called on a zeroed but uninitialized bioset (i.e. allocated with |
| 1914 | * kzalloc()). |
| 1915 | */ |
| 1916 | void bioset_exit(struct bio_set *bs) |
| 1917 | { |
| 1918 | if (bs->rescue_workqueue) |
| 1919 | destroy_workqueue(bs->rescue_workqueue); |
| 1920 | bs->rescue_workqueue = NULL; |
| 1921 | |
| 1922 | mempool_exit(&bs->bio_pool); |
| 1923 | mempool_exit(&bs->bvec_pool); |
| 1924 | |
| 1925 | bioset_integrity_free(bs); |
| 1926 | if (bs->bio_slab) |
| 1927 | bio_put_slab(bs); |
| 1928 | bs->bio_slab = NULL; |
| 1929 | } |
| 1930 | EXPORT_SYMBOL(bioset_exit); |
| 1931 | |
| 1932 | /** |
| 1933 | * bioset_init - Initialize a bio_set |
| 1934 | * @bs: pool to initialize |
| 1935 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
| 1936 | * @front_pad: Number of bytes to allocate in front of the returned bio |
| 1937 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS |
| 1938 | * and %BIOSET_NEED_RESCUER |
| 1939 | * |
| 1940 | * Description: |
| 1941 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller |
| 1942 | * to ask for a number of bytes to be allocated in front of the bio. |
| 1943 | * Front pad allocation is useful for embedding the bio inside |
| 1944 | * another structure, to avoid allocating extra data to go with the bio. |
| 1945 | * Note that the bio must be embedded at the END of that structure always, |
| 1946 | * or things will break badly. |
| 1947 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated |
| 1948 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). |
| 1949 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to |
| 1950 | * dispatch queued requests when the mempool runs out of space. |
| 1951 | * |
| 1952 | */ |
| 1953 | int bioset_init(struct bio_set *bs, |
| 1954 | unsigned int pool_size, |
| 1955 | unsigned int front_pad, |
| 1956 | int flags) |
| 1957 | { |
| 1958 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); |
| 1959 | |
| 1960 | bs->front_pad = front_pad; |
| 1961 | |
| 1962 | spin_lock_init(&bs->rescue_lock); |
| 1963 | bio_list_init(&bs->rescue_list); |
| 1964 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); |
| 1965 | |
| 1966 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); |
| 1967 | if (!bs->bio_slab) |
| 1968 | return -ENOMEM; |
| 1969 | |
| 1970 | if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) |
| 1971 | goto bad; |
| 1972 | |
| 1973 | if ((flags & BIOSET_NEED_BVECS) && |
| 1974 | biovec_init_pool(&bs->bvec_pool, pool_size)) |
| 1975 | goto bad; |
| 1976 | |
| 1977 | if (!(flags & BIOSET_NEED_RESCUER)) |
| 1978 | return 0; |
| 1979 | |
| 1980 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); |
| 1981 | if (!bs->rescue_workqueue) |
| 1982 | goto bad; |
| 1983 | |
| 1984 | return 0; |
| 1985 | bad: |
| 1986 | bioset_exit(bs); |
| 1987 | return -ENOMEM; |
| 1988 | } |
| 1989 | EXPORT_SYMBOL(bioset_init); |
| 1990 | |
| 1991 | /* |
| 1992 | * Initialize and setup a new bio_set, based on the settings from |
| 1993 | * another bio_set. |
| 1994 | */ |
| 1995 | int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) |
| 1996 | { |
| 1997 | int flags; |
| 1998 | |
| 1999 | flags = 0; |
| 2000 | if (src->bvec_pool.min_nr) |
| 2001 | flags |= BIOSET_NEED_BVECS; |
| 2002 | if (src->rescue_workqueue) |
| 2003 | flags |= BIOSET_NEED_RESCUER; |
| 2004 | |
| 2005 | return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); |
| 2006 | } |
| 2007 | EXPORT_SYMBOL(bioset_init_from_src); |
| 2008 | |
| 2009 | #ifdef CONFIG_BLK_CGROUP |
| 2010 | |
| 2011 | /** |
| 2012 | * bio_disassociate_blkg - puts back the blkg reference if associated |
| 2013 | * @bio: target bio |
| 2014 | * |
| 2015 | * Helper to disassociate the blkg from @bio if a blkg is associated. |
| 2016 | */ |
| 2017 | void bio_disassociate_blkg(struct bio *bio) |
| 2018 | { |
| 2019 | if (bio->bi_blkg) { |
| 2020 | blkg_put(bio->bi_blkg); |
| 2021 | bio->bi_blkg = NULL; |
| 2022 | } |
| 2023 | } |
| 2024 | EXPORT_SYMBOL_GPL(bio_disassociate_blkg); |
| 2025 | |
| 2026 | /** |
| 2027 | * __bio_associate_blkg - associate a bio with the a blkg |
| 2028 | * @bio: target bio |
| 2029 | * @blkg: the blkg to associate |
| 2030 | * |
| 2031 | * This tries to associate @bio with the specified @blkg. Association failure |
| 2032 | * is handled by walking up the blkg tree. Therefore, the blkg associated can |
| 2033 | * be anything between @blkg and the root_blkg. This situation only happens |
| 2034 | * when a cgroup is dying and then the remaining bios will spill to the closest |
| 2035 | * alive blkg. |
| 2036 | * |
| 2037 | * A reference will be taken on the @blkg and will be released when @bio is |
| 2038 | * freed. |
| 2039 | */ |
| 2040 | static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) |
| 2041 | { |
| 2042 | bio_disassociate_blkg(bio); |
| 2043 | |
| 2044 | bio->bi_blkg = blkg_tryget_closest(blkg); |
| 2045 | } |
| 2046 | |
| 2047 | /** |
| 2048 | * bio_associate_blkg_from_css - associate a bio with a specified css |
| 2049 | * @bio: target bio |
| 2050 | * @css: target css |
| 2051 | * |
| 2052 | * Associate @bio with the blkg found by combining the css's blkg and the |
| 2053 | * request_queue of the @bio. This falls back to the queue's root_blkg if |
| 2054 | * the association fails with the css. |
| 2055 | */ |
| 2056 | void bio_associate_blkg_from_css(struct bio *bio, |
| 2057 | struct cgroup_subsys_state *css) |
| 2058 | { |
| 2059 | struct request_queue *q = bio->bi_disk->queue; |
| 2060 | struct blkcg_gq *blkg; |
| 2061 | |
| 2062 | rcu_read_lock(); |
| 2063 | |
| 2064 | if (!css || !css->parent) |
| 2065 | blkg = q->root_blkg; |
| 2066 | else |
| 2067 | blkg = blkg_lookup_create(css_to_blkcg(css), q); |
| 2068 | |
| 2069 | __bio_associate_blkg(bio, blkg); |
| 2070 | |
| 2071 | rcu_read_unlock(); |
| 2072 | } |
| 2073 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); |
| 2074 | |
| 2075 | #ifdef CONFIG_MEMCG |
| 2076 | /** |
| 2077 | * bio_associate_blkg_from_page - associate a bio with the page's blkg |
| 2078 | * @bio: target bio |
| 2079 | * @page: the page to lookup the blkcg from |
| 2080 | * |
| 2081 | * Associate @bio with the blkg from @page's owning memcg and the respective |
| 2082 | * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's |
| 2083 | * root_blkg. |
| 2084 | */ |
| 2085 | void bio_associate_blkg_from_page(struct bio *bio, struct page *page) |
| 2086 | { |
| 2087 | struct cgroup_subsys_state *css; |
| 2088 | |
| 2089 | if (!page->mem_cgroup) |
| 2090 | return; |
| 2091 | |
| 2092 | rcu_read_lock(); |
| 2093 | |
| 2094 | css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); |
| 2095 | bio_associate_blkg_from_css(bio, css); |
| 2096 | |
| 2097 | rcu_read_unlock(); |
| 2098 | } |
| 2099 | #endif /* CONFIG_MEMCG */ |
| 2100 | |
| 2101 | /** |
| 2102 | * bio_associate_blkg - associate a bio with a blkg |
| 2103 | * @bio: target bio |
| 2104 | * |
| 2105 | * Associate @bio with the blkg found from the bio's css and request_queue. |
| 2106 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is |
| 2107 | * already associated, the css is reused and association redone as the |
| 2108 | * request_queue may have changed. |
| 2109 | */ |
| 2110 | void bio_associate_blkg(struct bio *bio) |
| 2111 | { |
| 2112 | struct cgroup_subsys_state *css; |
| 2113 | |
| 2114 | rcu_read_lock(); |
| 2115 | |
| 2116 | if (bio->bi_blkg) |
| 2117 | css = &bio_blkcg(bio)->css; |
| 2118 | else |
| 2119 | css = blkcg_css(); |
| 2120 | |
| 2121 | bio_associate_blkg_from_css(bio, css); |
| 2122 | |
| 2123 | rcu_read_unlock(); |
| 2124 | } |
| 2125 | EXPORT_SYMBOL_GPL(bio_associate_blkg); |
| 2126 | |
| 2127 | /** |
| 2128 | * bio_clone_blkg_association - clone blkg association from src to dst bio |
| 2129 | * @dst: destination bio |
| 2130 | * @src: source bio |
| 2131 | */ |
| 2132 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) |
| 2133 | { |
| 2134 | rcu_read_lock(); |
| 2135 | |
| 2136 | if (src->bi_blkg) |
| 2137 | __bio_associate_blkg(dst, src->bi_blkg); |
| 2138 | |
| 2139 | rcu_read_unlock(); |
| 2140 | } |
| 2141 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); |
| 2142 | #endif /* CONFIG_BLK_CGROUP */ |
| 2143 | |
| 2144 | static void __init biovec_init_slabs(void) |
| 2145 | { |
| 2146 | int i; |
| 2147 | |
| 2148 | for (i = 0; i < BVEC_POOL_NR; i++) { |
| 2149 | int size; |
| 2150 | struct biovec_slab *bvs = bvec_slabs + i; |
| 2151 | |
| 2152 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
| 2153 | bvs->slab = NULL; |
| 2154 | continue; |
| 2155 | } |
| 2156 | |
| 2157 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
| 2158 | bvs->slab = kmem_cache_create(bvs->name, size, 0, |
| 2159 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
| 2160 | } |
| 2161 | } |
| 2162 | |
| 2163 | static int __init init_bio(void) |
| 2164 | { |
| 2165 | bio_slab_max = 2; |
| 2166 | bio_slab_nr = 0; |
| 2167 | bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), |
| 2168 | GFP_KERNEL); |
| 2169 | |
| 2170 | BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); |
| 2171 | |
| 2172 | if (!bio_slabs) |
| 2173 | panic("bio: can't allocate bios\n"); |
| 2174 | |
| 2175 | bio_integrity_init(); |
| 2176 | biovec_init_slabs(); |
| 2177 | |
| 2178 | if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) |
| 2179 | panic("bio: can't allocate bios\n"); |
| 2180 | |
| 2181 | if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) |
| 2182 | panic("bio: can't create integrity pool\n"); |
| 2183 | |
| 2184 | return 0; |
| 2185 | } |
| 2186 | subsys_initcall(init_bio); |