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 #include <linux/blk-cgroup.h> 32 33 #include <trace/events/block.h> 34 #include "blk.h" 35 #include "blk-rq-qos.h" 36 37 /* 38 * Test patch to inline a certain number of bi_io_vec's inside the bio 39 * itself, to shrink a bio data allocation from two mempool calls to one 40 */ 41 #define BIO_INLINE_VECS 4 42 43 /* 44 * if you change this list, also change bvec_alloc or things will 45 * break badly! cannot be bigger than what you can fit into an 46 * unsigned short 47 */ 48 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 49 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 50 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 51 }; 52 #undef BV 53 54 /* 55 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 56 * IO code that does not need private memory pools. 57 */ 58 struct bio_set fs_bio_set; 59 EXPORT_SYMBOL(fs_bio_set); 60 61 /* 62 * Our slab pool management 63 */ 64 struct bio_slab { 65 struct kmem_cache *slab; 66 unsigned int slab_ref; 67 unsigned int slab_size; 68 char name[8]; 69 }; 70 static DEFINE_MUTEX(bio_slab_lock); 71 static struct bio_slab *bio_slabs; 72 static unsigned int bio_slab_nr, bio_slab_max; 73 74 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 75 { 76 unsigned int sz = sizeof(struct bio) + extra_size; 77 struct kmem_cache *slab = NULL; 78 struct bio_slab *bslab, *new_bio_slabs; 79 unsigned int new_bio_slab_max; 80 unsigned int i, entry = -1; 81 82 mutex_lock(&bio_slab_lock); 83 84 i = 0; 85 while (i < bio_slab_nr) { 86 bslab = &bio_slabs[i]; 87 88 if (!bslab->slab && entry == -1) 89 entry = i; 90 else if (bslab->slab_size == sz) { 91 slab = bslab->slab; 92 bslab->slab_ref++; 93 break; 94 } 95 i++; 96 } 97 98 if (slab) 99 goto out_unlock; 100 101 if (bio_slab_nr == bio_slab_max && entry == -1) { 102 new_bio_slab_max = bio_slab_max << 1; 103 new_bio_slabs = krealloc(bio_slabs, 104 new_bio_slab_max * sizeof(struct bio_slab), 105 GFP_KERNEL); 106 if (!new_bio_slabs) 107 goto out_unlock; 108 bio_slab_max = new_bio_slab_max; 109 bio_slabs = new_bio_slabs; 110 } 111 if (entry == -1) 112 entry = bio_slab_nr++; 113 114 bslab = &bio_slabs[entry]; 115 116 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 117 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 118 SLAB_HWCACHE_ALIGN, NULL); 119 if (!slab) 120 goto out_unlock; 121 122 bslab->slab = slab; 123 bslab->slab_ref = 1; 124 bslab->slab_size = sz; 125 out_unlock: 126 mutex_unlock(&bio_slab_lock); 127 return slab; 128 } 129 130 static void bio_put_slab(struct bio_set *bs) 131 { 132 struct bio_slab *bslab = NULL; 133 unsigned int i; 134 135 mutex_lock(&bio_slab_lock); 136 137 for (i = 0; i < bio_slab_nr; i++) { 138 if (bs->bio_slab == bio_slabs[i].slab) { 139 bslab = &bio_slabs[i]; 140 break; 141 } 142 } 143 144 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 145 goto out; 146 147 WARN_ON(!bslab->slab_ref); 148 149 if (--bslab->slab_ref) 150 goto out; 151 152 kmem_cache_destroy(bslab->slab); 153 bslab->slab = NULL; 154 155 out: 156 mutex_unlock(&bio_slab_lock); 157 } 158 159 unsigned int bvec_nr_vecs(unsigned short idx) 160 { 161 return bvec_slabs[--idx].nr_vecs; 162 } 163 164 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 165 { 166 if (!idx) 167 return; 168 idx--; 169 170 BIO_BUG_ON(idx >= BVEC_POOL_NR); 171 172 if (idx == BVEC_POOL_MAX) { 173 mempool_free(bv, pool); 174 } else { 175 struct biovec_slab *bvs = bvec_slabs + idx; 176 177 kmem_cache_free(bvs->slab, bv); 178 } 179 } 180 181 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 182 mempool_t *pool) 183 { 184 struct bio_vec *bvl; 185 186 /* 187 * see comment near bvec_array define! 188 */ 189 switch (nr) { 190 case 1: 191 *idx = 0; 192 break; 193 case 2 ... 4: 194 *idx = 1; 195 break; 196 case 5 ... 16: 197 *idx = 2; 198 break; 199 case 17 ... 64: 200 *idx = 3; 201 break; 202 case 65 ... 128: 203 *idx = 4; 204 break; 205 case 129 ... BIO_MAX_PAGES: 206 *idx = 5; 207 break; 208 default: 209 return NULL; 210 } 211 212 /* 213 * idx now points to the pool we want to allocate from. only the 214 * 1-vec entry pool is mempool backed. 215 */ 216 if (*idx == BVEC_POOL_MAX) { 217 fallback: 218 bvl = mempool_alloc(pool, gfp_mask); 219 } else { 220 struct biovec_slab *bvs = bvec_slabs + *idx; 221 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 222 223 /* 224 * Make this allocation restricted and don't dump info on 225 * allocation failures, since we'll fallback to the mempool 226 * in case of failure. 227 */ 228 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 229 230 /* 231 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 232 * is set, retry with the 1-entry mempool 233 */ 234 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 235 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 236 *idx = BVEC_POOL_MAX; 237 goto fallback; 238 } 239 } 240 241 (*idx)++; 242 return bvl; 243 } 244 245 void bio_uninit(struct bio *bio) 246 { 247 bio_disassociate_blkg(bio); 248 } 249 EXPORT_SYMBOL(bio_uninit); 250 251 static void bio_free(struct bio *bio) 252 { 253 struct bio_set *bs = bio->bi_pool; 254 void *p; 255 256 bio_uninit(bio); 257 258 if (bs) { 259 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 260 261 /* 262 * If we have front padding, adjust the bio pointer before freeing 263 */ 264 p = bio; 265 p -= bs->front_pad; 266 267 mempool_free(p, &bs->bio_pool); 268 } else { 269 /* Bio was allocated by bio_kmalloc() */ 270 kfree(bio); 271 } 272 } 273 274 /* 275 * Users of this function have their own bio allocation. Subsequently, 276 * they must remember to pair any call to bio_init() with bio_uninit() 277 * when IO has completed, or when the bio is released. 278 */ 279 void bio_init(struct bio *bio, struct bio_vec *table, 280 unsigned short max_vecs) 281 { 282 memset(bio, 0, sizeof(*bio)); 283 atomic_set(&bio->__bi_remaining, 1); 284 atomic_set(&bio->__bi_cnt, 1); 285 286 bio->bi_io_vec = table; 287 bio->bi_max_vecs = max_vecs; 288 } 289 EXPORT_SYMBOL(bio_init); 290 291 /** 292 * bio_reset - reinitialize a bio 293 * @bio: bio to reset 294 * 295 * Description: 296 * After calling bio_reset(), @bio will be in the same state as a freshly 297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are 298 * preserved are the ones that are initialized by bio_alloc_bioset(). See 299 * comment in struct bio. 300 */ 301 void bio_reset(struct bio *bio) 302 { 303 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 304 305 bio_uninit(bio); 306 307 memset(bio, 0, BIO_RESET_BYTES); 308 bio->bi_flags = flags; 309 atomic_set(&bio->__bi_remaining, 1); 310 } 311 EXPORT_SYMBOL(bio_reset); 312 313 static struct bio *__bio_chain_endio(struct bio *bio) 314 { 315 struct bio *parent = bio->bi_private; 316 317 if (!parent->bi_status) 318 parent->bi_status = bio->bi_status; 319 bio_put(bio); 320 return parent; 321 } 322 323 static void bio_chain_endio(struct bio *bio) 324 { 325 bio_endio(__bio_chain_endio(bio)); 326 } 327 328 /** 329 * bio_chain - chain bio completions 330 * @bio: the target bio 331 * @parent: the @bio's parent bio 332 * 333 * The caller won't have a bi_end_io called when @bio completes - instead, 334 * @parent's bi_end_io won't be called until both @parent and @bio have 335 * completed; the chained bio will also be freed when it completes. 336 * 337 * The caller must not set bi_private or bi_end_io in @bio. 338 */ 339 void bio_chain(struct bio *bio, struct bio *parent) 340 { 341 BUG_ON(bio->bi_private || bio->bi_end_io); 342 343 bio->bi_private = parent; 344 bio->bi_end_io = bio_chain_endio; 345 bio_inc_remaining(parent); 346 } 347 EXPORT_SYMBOL(bio_chain); 348 349 static void bio_alloc_rescue(struct work_struct *work) 350 { 351 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 352 struct bio *bio; 353 354 while (1) { 355 spin_lock(&bs->rescue_lock); 356 bio = bio_list_pop(&bs->rescue_list); 357 spin_unlock(&bs->rescue_lock); 358 359 if (!bio) 360 break; 361 362 generic_make_request(bio); 363 } 364 } 365 366 static void punt_bios_to_rescuer(struct bio_set *bs) 367 { 368 struct bio_list punt, nopunt; 369 struct bio *bio; 370 371 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 372 return; 373 /* 374 * In order to guarantee forward progress we must punt only bios that 375 * were allocated from this bio_set; otherwise, if there was a bio on 376 * there for a stacking driver higher up in the stack, processing it 377 * could require allocating bios from this bio_set, and doing that from 378 * our own rescuer would be bad. 379 * 380 * Since bio lists are singly linked, pop them all instead of trying to 381 * remove from the middle of the list: 382 */ 383 384 bio_list_init(&punt); 385 bio_list_init(&nopunt); 386 387 while ((bio = bio_list_pop(¤t->bio_list[0]))) 388 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 389 current->bio_list[0] = nopunt; 390 391 bio_list_init(&nopunt); 392 while ((bio = bio_list_pop(¤t->bio_list[1]))) 393 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 394 current->bio_list[1] = nopunt; 395 396 spin_lock(&bs->rescue_lock); 397 bio_list_merge(&bs->rescue_list, &punt); 398 spin_unlock(&bs->rescue_lock); 399 400 queue_work(bs->rescue_workqueue, &bs->rescue_work); 401 } 402 403 /** 404 * bio_alloc_bioset - allocate a bio for I/O 405 * @gfp_mask: the GFP_* mask given to the slab allocator 406 * @nr_iovecs: number of iovecs to pre-allocate 407 * @bs: the bio_set to allocate from. 408 * 409 * Description: 410 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 411 * backed by the @bs's mempool. 412 * 413 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 414 * always be able to allocate a bio. This is due to the mempool guarantees. 415 * To make this work, callers must never allocate more than 1 bio at a time 416 * from this pool. Callers that need to allocate more than 1 bio must always 417 * submit the previously allocated bio for IO before attempting to allocate 418 * a new one. Failure to do so can cause deadlocks under memory pressure. 419 * 420 * Note that when running under generic_make_request() (i.e. any block 421 * driver), bios are not submitted until after you return - see the code in 422 * generic_make_request() that converts recursion into iteration, to prevent 423 * stack overflows. 424 * 425 * This would normally mean allocating multiple bios under 426 * generic_make_request() would be susceptible to deadlocks, but we have 427 * deadlock avoidance code that resubmits any blocked bios from a rescuer 428 * thread. 429 * 430 * However, we do not guarantee forward progress for allocations from other 431 * mempools. Doing multiple allocations from the same mempool under 432 * generic_make_request() should be avoided - instead, use bio_set's front_pad 433 * for per bio allocations. 434 * 435 * RETURNS: 436 * Pointer to new bio on success, NULL on failure. 437 */ 438 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 439 struct bio_set *bs) 440 { 441 gfp_t saved_gfp = gfp_mask; 442 unsigned front_pad; 443 unsigned inline_vecs; 444 struct bio_vec *bvl = NULL; 445 struct bio *bio; 446 void *p; 447 448 if (!bs) { 449 if (nr_iovecs > UIO_MAXIOV) 450 return NULL; 451 452 p = kmalloc(sizeof(struct bio) + 453 nr_iovecs * sizeof(struct bio_vec), 454 gfp_mask); 455 front_pad = 0; 456 inline_vecs = nr_iovecs; 457 } else { 458 /* should not use nobvec bioset for nr_iovecs > 0 */ 459 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 460 nr_iovecs > 0)) 461 return NULL; 462 /* 463 * generic_make_request() converts recursion to iteration; this 464 * means if we're running beneath it, any bios we allocate and 465 * submit will not be submitted (and thus freed) until after we 466 * return. 467 * 468 * This exposes us to a potential deadlock if we allocate 469 * multiple bios from the same bio_set() while running 470 * underneath generic_make_request(). If we were to allocate 471 * multiple bios (say a stacking block driver that was splitting 472 * bios), we would deadlock if we exhausted the mempool's 473 * reserve. 474 * 475 * We solve this, and guarantee forward progress, with a rescuer 476 * workqueue per bio_set. If we go to allocate and there are 477 * bios on current->bio_list, we first try the allocation 478 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 479 * bios we would be blocking to the rescuer workqueue before 480 * we retry with the original gfp_flags. 481 */ 482 483 if (current->bio_list && 484 (!bio_list_empty(¤t->bio_list[0]) || 485 !bio_list_empty(¤t->bio_list[1])) && 486 bs->rescue_workqueue) 487 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 488 489 p = mempool_alloc(&bs->bio_pool, gfp_mask); 490 if (!p && gfp_mask != saved_gfp) { 491 punt_bios_to_rescuer(bs); 492 gfp_mask = saved_gfp; 493 p = mempool_alloc(&bs->bio_pool, gfp_mask); 494 } 495 496 front_pad = bs->front_pad; 497 inline_vecs = BIO_INLINE_VECS; 498 } 499 500 if (unlikely(!p)) 501 return NULL; 502 503 bio = p + front_pad; 504 bio_init(bio, NULL, 0); 505 506 if (nr_iovecs > inline_vecs) { 507 unsigned long idx = 0; 508 509 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 510 if (!bvl && gfp_mask != saved_gfp) { 511 punt_bios_to_rescuer(bs); 512 gfp_mask = saved_gfp; 513 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 514 } 515 516 if (unlikely(!bvl)) 517 goto err_free; 518 519 bio->bi_flags |= idx << BVEC_POOL_OFFSET; 520 } else if (nr_iovecs) { 521 bvl = bio->bi_inline_vecs; 522 } 523 524 bio->bi_pool = bs; 525 bio->bi_max_vecs = nr_iovecs; 526 bio->bi_io_vec = bvl; 527 return bio; 528 529 err_free: 530 mempool_free(p, &bs->bio_pool); 531 return NULL; 532 } 533 EXPORT_SYMBOL(bio_alloc_bioset); 534 535 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 536 { 537 unsigned long flags; 538 struct bio_vec bv; 539 struct bvec_iter iter; 540 541 __bio_for_each_segment(bv, bio, iter, start) { 542 char *data = bvec_kmap_irq(&bv, &flags); 543 memset(data, 0, bv.bv_len); 544 flush_dcache_page(bv.bv_page); 545 bvec_kunmap_irq(data, &flags); 546 } 547 } 548 EXPORT_SYMBOL(zero_fill_bio_iter); 549 550 /** 551 * bio_put - release a reference to a bio 552 * @bio: bio to release reference to 553 * 554 * Description: 555 * Put a reference to a &struct bio, either one you have gotten with 556 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 557 **/ 558 void bio_put(struct bio *bio) 559 { 560 if (!bio_flagged(bio, BIO_REFFED)) 561 bio_free(bio); 562 else { 563 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 564 565 /* 566 * last put frees it 567 */ 568 if (atomic_dec_and_test(&bio->__bi_cnt)) 569 bio_free(bio); 570 } 571 } 572 EXPORT_SYMBOL(bio_put); 573 574 int bio_phys_segments(struct request_queue *q, struct bio *bio) 575 { 576 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 577 blk_recount_segments(q, bio); 578 579 return bio->bi_phys_segments; 580 } 581 582 /** 583 * __bio_clone_fast - clone a bio that shares the original bio's biovec 584 * @bio: destination bio 585 * @bio_src: bio to clone 586 * 587 * Clone a &bio. Caller will own the returned bio, but not 588 * the actual data it points to. Reference count of returned 589 * bio will be one. 590 * 591 * Caller must ensure that @bio_src is not freed before @bio. 592 */ 593 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 594 { 595 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 596 597 /* 598 * most users will be overriding ->bi_disk with a new target, 599 * so we don't set nor calculate new physical/hw segment counts here 600 */ 601 bio->bi_disk = bio_src->bi_disk; 602 bio->bi_partno = bio_src->bi_partno; 603 bio_set_flag(bio, BIO_CLONED); 604 if (bio_flagged(bio_src, BIO_THROTTLED)) 605 bio_set_flag(bio, BIO_THROTTLED); 606 bio->bi_opf = bio_src->bi_opf; 607 bio->bi_ioprio = bio_src->bi_ioprio; 608 bio->bi_write_hint = bio_src->bi_write_hint; 609 bio->bi_iter = bio_src->bi_iter; 610 bio->bi_io_vec = bio_src->bi_io_vec; 611 612 bio_clone_blkg_association(bio, bio_src); 613 blkcg_bio_issue_init(bio); 614 } 615 EXPORT_SYMBOL(__bio_clone_fast); 616 617 /** 618 * bio_clone_fast - clone a bio that shares the original bio's biovec 619 * @bio: bio to clone 620 * @gfp_mask: allocation priority 621 * @bs: bio_set to allocate from 622 * 623 * Like __bio_clone_fast, only also allocates the returned bio 624 */ 625 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 626 { 627 struct bio *b; 628 629 b = bio_alloc_bioset(gfp_mask, 0, bs); 630 if (!b) 631 return NULL; 632 633 __bio_clone_fast(b, bio); 634 635 if (bio_integrity(bio)) { 636 int ret; 637 638 ret = bio_integrity_clone(b, bio, gfp_mask); 639 640 if (ret < 0) { 641 bio_put(b); 642 return NULL; 643 } 644 } 645 646 return b; 647 } 648 EXPORT_SYMBOL(bio_clone_fast); 649 650 /** 651 * bio_add_pc_page - attempt to add page to bio 652 * @q: the target queue 653 * @bio: destination bio 654 * @page: page to add 655 * @len: vec entry length 656 * @offset: vec entry offset 657 * 658 * Attempt to add a page to the bio_vec maplist. This can fail for a 659 * number of reasons, such as the bio being full or target block device 660 * limitations. The target block device must allow bio's up to PAGE_SIZE, 661 * so it is always possible to add a single page to an empty bio. 662 * 663 * This should only be used by REQ_PC bios. 664 */ 665 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page 666 *page, unsigned int len, unsigned int offset) 667 { 668 int retried_segments = 0; 669 struct bio_vec *bvec; 670 671 /* 672 * cloned bio must not modify vec list 673 */ 674 if (unlikely(bio_flagged(bio, BIO_CLONED))) 675 return 0; 676 677 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 678 return 0; 679 680 /* 681 * For filesystems with a blocksize smaller than the pagesize 682 * we will often be called with the same page as last time and 683 * a consecutive offset. Optimize this special case. 684 */ 685 if (bio->bi_vcnt > 0) { 686 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; 687 688 if (page == prev->bv_page && 689 offset == prev->bv_offset + prev->bv_len) { 690 prev->bv_len += len; 691 bio->bi_iter.bi_size += len; 692 goto done; 693 } 694 695 /* 696 * If the queue doesn't support SG gaps and adding this 697 * offset would create a gap, disallow it. 698 */ 699 if (bvec_gap_to_prev(q, prev, offset)) 700 return 0; 701 } 702 703 if (bio_full(bio)) 704 return 0; 705 706 /* 707 * setup the new entry, we might clear it again later if we 708 * cannot add the page 709 */ 710 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 711 bvec->bv_page = page; 712 bvec->bv_len = len; 713 bvec->bv_offset = offset; 714 bio->bi_vcnt++; 715 bio->bi_phys_segments++; 716 bio->bi_iter.bi_size += len; 717 718 /* 719 * Perform a recount if the number of segments is greater 720 * than queue_max_segments(q). 721 */ 722 723 while (bio->bi_phys_segments > queue_max_segments(q)) { 724 725 if (retried_segments) 726 goto failed; 727 728 retried_segments = 1; 729 blk_recount_segments(q, bio); 730 } 731 732 /* If we may be able to merge these biovecs, force a recount */ 733 if (bio->bi_vcnt > 1 && biovec_phys_mergeable(q, bvec - 1, bvec)) 734 bio_clear_flag(bio, BIO_SEG_VALID); 735 736 done: 737 return len; 738 739 failed: 740 bvec->bv_page = NULL; 741 bvec->bv_len = 0; 742 bvec->bv_offset = 0; 743 bio->bi_vcnt--; 744 bio->bi_iter.bi_size -= len; 745 blk_recount_segments(q, bio); 746 return 0; 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: page to add 754 * @len: length of the data to add 755 * @off: offset of the data in @page 756 * @same_page: if %true only merge if the new data is in the same physical 757 * page as the last segment of the bio. 758 * 759 * Try to add the data at @page + @off to the last bvec of @bio. This is a 760 * a useful optimisation for file systems with a block size smaller than the 761 * page size. 762 * 763 * Return %true on success or %false on failure. 764 */ 765 bool __bio_try_merge_page(struct bio *bio, struct page *page, 766 unsigned int len, unsigned int off, bool same_page) 767 { 768 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 769 return false; 770 771 if (bio->bi_vcnt > 0) { 772 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 773 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 774 bv->bv_offset + bv->bv_len - 1; 775 phys_addr_t page_addr = page_to_phys(page); 776 777 if (vec_end_addr + 1 != page_addr + off) 778 return false; 779 if (same_page && (vec_end_addr & PAGE_MASK) != page_addr) 780 return false; 781 782 bv->bv_len += len; 783 bio->bi_iter.bi_size += len; 784 return true; 785 } 786 return false; 787 } 788 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 789 790 /** 791 * __bio_add_page - add page to a bio in a new segment 792 * @bio: destination bio 793 * @page: page to add 794 * @len: length of the data to add 795 * @off: offset of the data in @page 796 * 797 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 798 * that @bio has space for another bvec. 799 */ 800 void __bio_add_page(struct bio *bio, struct page *page, 801 unsigned int len, unsigned int off) 802 { 803 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 804 805 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 806 WARN_ON_ONCE(bio_full(bio)); 807 808 bv->bv_page = page; 809 bv->bv_offset = off; 810 bv->bv_len = len; 811 812 bio->bi_iter.bi_size += len; 813 bio->bi_vcnt++; 814 } 815 EXPORT_SYMBOL_GPL(__bio_add_page); 816 817 /** 818 * bio_add_page - attempt to add page to bio 819 * @bio: destination bio 820 * @page: page to add 821 * @len: vec entry length 822 * @offset: vec entry offset 823 * 824 * Attempt to add a page to the bio_vec maplist. This will only fail 825 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 826 */ 827 int bio_add_page(struct bio *bio, struct page *page, 828 unsigned int len, unsigned int offset) 829 { 830 if (!__bio_try_merge_page(bio, page, len, offset, false)) { 831 if (bio_full(bio)) 832 return 0; 833 __bio_add_page(bio, page, len, offset); 834 } 835 return len; 836 } 837 EXPORT_SYMBOL(bio_add_page); 838 839 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 840 { 841 const struct bio_vec *bv = iter->bvec; 842 unsigned int len; 843 size_t size; 844 845 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 846 return -EINVAL; 847 848 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 849 size = bio_add_page(bio, bv->bv_page, len, 850 bv->bv_offset + iter->iov_offset); 851 if (size == len) { 852 struct page *page; 853 int i; 854 855 /* 856 * For the normal O_DIRECT case, we could skip grabbing this 857 * reference and then not have to put them again when IO 858 * completes. But this breaks some in-kernel users, like 859 * splicing to/from a loop device, where we release the pipe 860 * pages unconditionally. If we can fix that case, we can 861 * get rid of the get here and the need to call 862 * bio_release_pages() at IO completion time. 863 */ 864 mp_bvec_for_each_page(page, bv, i) 865 get_page(page); 866 iov_iter_advance(iter, size); 867 return 0; 868 } 869 870 return -EINVAL; 871 } 872 873 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 874 875 /** 876 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 877 * @bio: bio to add pages to 878 * @iter: iov iterator describing the region to be mapped 879 * 880 * Pins pages from *iter and appends them to @bio's bvec array. The 881 * pages will have to be released using put_page() when done. 882 * For multi-segment *iter, this function only adds pages from the 883 * the next non-empty segment of the iov iterator. 884 */ 885 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 886 { 887 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 888 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 889 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 890 struct page **pages = (struct page **)bv; 891 ssize_t size, left; 892 unsigned len, i; 893 size_t offset; 894 895 /* 896 * Move page array up in the allocated memory for the bio vecs as far as 897 * possible so that we can start filling biovecs from the beginning 898 * without overwriting the temporary page array. 899 */ 900 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 901 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 902 903 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 904 if (unlikely(size <= 0)) 905 return size ? size : -EFAULT; 906 907 for (left = size, i = 0; left > 0; left -= len, i++) { 908 struct page *page = pages[i]; 909 910 len = min_t(size_t, PAGE_SIZE - offset, left); 911 if (WARN_ON_ONCE(bio_add_page(bio, page, len, offset) != len)) 912 return -EINVAL; 913 offset = 0; 914 } 915 916 iov_iter_advance(iter, size); 917 return 0; 918 } 919 920 /** 921 * bio_iov_iter_get_pages - add user or kernel pages to a bio 922 * @bio: bio to add pages to 923 * @iter: iov iterator describing the region to be added 924 * 925 * This takes either an iterator pointing to user memory, or one pointing to 926 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 927 * map them into the kernel. On IO completion, the caller should put those 928 * pages. For now, when adding kernel pages, we still grab a reference to the 929 * page. This isn't strictly needed for the common case, but some call paths 930 * end up releasing pages from eg a pipe and we can't easily control these. 931 * See comment in __bio_iov_bvec_add_pages(). 932 * 933 * The function tries, but does not guarantee, to pin as many pages as 934 * fit into the bio, or are requested in *iter, whatever is smaller. If 935 * MM encounters an error pinning the requested pages, it stops. Error 936 * is returned only if 0 pages could be pinned. 937 */ 938 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 939 { 940 const bool is_bvec = iov_iter_is_bvec(iter); 941 unsigned short orig_vcnt = bio->bi_vcnt; 942 943 do { 944 int ret; 945 946 if (is_bvec) 947 ret = __bio_iov_bvec_add_pages(bio, iter); 948 else 949 ret = __bio_iov_iter_get_pages(bio, iter); 950 951 if (unlikely(ret)) 952 return bio->bi_vcnt > orig_vcnt ? 0 : ret; 953 954 } while (iov_iter_count(iter) && !bio_full(bio)); 955 956 return 0; 957 } 958 959 static void submit_bio_wait_endio(struct bio *bio) 960 { 961 complete(bio->bi_private); 962 } 963 964 /** 965 * submit_bio_wait - submit a bio, and wait until it completes 966 * @bio: The &struct bio which describes the I/O 967 * 968 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 969 * bio_endio() on failure. 970 * 971 * WARNING: Unlike to how submit_bio() is usually used, this function does not 972 * result in bio reference to be consumed. The caller must drop the reference 973 * on his own. 974 */ 975 int submit_bio_wait(struct bio *bio) 976 { 977 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 978 979 bio->bi_private = &done; 980 bio->bi_end_io = submit_bio_wait_endio; 981 bio->bi_opf |= REQ_SYNC; 982 submit_bio(bio); 983 wait_for_completion_io(&done); 984 985 return blk_status_to_errno(bio->bi_status); 986 } 987 EXPORT_SYMBOL(submit_bio_wait); 988 989 /** 990 * bio_advance - increment/complete a bio by some number of bytes 991 * @bio: bio to advance 992 * @bytes: number of bytes to complete 993 * 994 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 995 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 996 * be updated on the last bvec as well. 997 * 998 * @bio will then represent the remaining, uncompleted portion of the io. 999 */ 1000 void bio_advance(struct bio *bio, unsigned bytes) 1001 { 1002 if (bio_integrity(bio)) 1003 bio_integrity_advance(bio, bytes); 1004 1005 bio_advance_iter(bio, &bio->bi_iter, bytes); 1006 } 1007 EXPORT_SYMBOL(bio_advance); 1008 1009 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1010 struct bio *src, struct bvec_iter *src_iter) 1011 { 1012 struct bio_vec src_bv, dst_bv; 1013 void *src_p, *dst_p; 1014 unsigned bytes; 1015 1016 while (src_iter->bi_size && dst_iter->bi_size) { 1017 src_bv = bio_iter_iovec(src, *src_iter); 1018 dst_bv = bio_iter_iovec(dst, *dst_iter); 1019 1020 bytes = min(src_bv.bv_len, dst_bv.bv_len); 1021 1022 src_p = kmap_atomic(src_bv.bv_page); 1023 dst_p = kmap_atomic(dst_bv.bv_page); 1024 1025 memcpy(dst_p + dst_bv.bv_offset, 1026 src_p + src_bv.bv_offset, 1027 bytes); 1028 1029 kunmap_atomic(dst_p); 1030 kunmap_atomic(src_p); 1031 1032 flush_dcache_page(dst_bv.bv_page); 1033 1034 bio_advance_iter(src, src_iter, bytes); 1035 bio_advance_iter(dst, dst_iter, bytes); 1036 } 1037 } 1038 EXPORT_SYMBOL(bio_copy_data_iter); 1039 1040 /** 1041 * bio_copy_data - copy contents of data buffers from one bio to another 1042 * @src: source bio 1043 * @dst: destination bio 1044 * 1045 * Stops when it reaches the end of either @src or @dst - that is, copies 1046 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1047 */ 1048 void bio_copy_data(struct bio *dst, struct bio *src) 1049 { 1050 struct bvec_iter src_iter = src->bi_iter; 1051 struct bvec_iter dst_iter = dst->bi_iter; 1052 1053 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1054 } 1055 EXPORT_SYMBOL(bio_copy_data); 1056 1057 /** 1058 * bio_list_copy_data - copy contents of data buffers from one chain of bios to 1059 * another 1060 * @src: source bio list 1061 * @dst: destination bio list 1062 * 1063 * Stops when it reaches the end of either the @src list or @dst list - that is, 1064 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 1065 * bios). 1066 */ 1067 void bio_list_copy_data(struct bio *dst, struct bio *src) 1068 { 1069 struct bvec_iter src_iter = src->bi_iter; 1070 struct bvec_iter dst_iter = dst->bi_iter; 1071 1072 while (1) { 1073 if (!src_iter.bi_size) { 1074 src = src->bi_next; 1075 if (!src) 1076 break; 1077 1078 src_iter = src->bi_iter; 1079 } 1080 1081 if (!dst_iter.bi_size) { 1082 dst = dst->bi_next; 1083 if (!dst) 1084 break; 1085 1086 dst_iter = dst->bi_iter; 1087 } 1088 1089 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1090 } 1091 } 1092 EXPORT_SYMBOL(bio_list_copy_data); 1093 1094 struct bio_map_data { 1095 int is_our_pages; 1096 struct iov_iter iter; 1097 struct iovec iov[]; 1098 }; 1099 1100 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1101 gfp_t gfp_mask) 1102 { 1103 struct bio_map_data *bmd; 1104 if (data->nr_segs > UIO_MAXIOV) 1105 return NULL; 1106 1107 bmd = kmalloc(sizeof(struct bio_map_data) + 1108 sizeof(struct iovec) * data->nr_segs, gfp_mask); 1109 if (!bmd) 1110 return NULL; 1111 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 1112 bmd->iter = *data; 1113 bmd->iter.iov = bmd->iov; 1114 return bmd; 1115 } 1116 1117 /** 1118 * bio_copy_from_iter - copy all pages from iov_iter to bio 1119 * @bio: The &struct bio which describes the I/O as destination 1120 * @iter: iov_iter as source 1121 * 1122 * Copy all pages from iov_iter to bio. 1123 * Returns 0 on success, or error on failure. 1124 */ 1125 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1126 { 1127 int i; 1128 struct bio_vec *bvec; 1129 struct bvec_iter_all iter_all; 1130 1131 bio_for_each_segment_all(bvec, bio, i, iter_all) { 1132 ssize_t ret; 1133 1134 ret = copy_page_from_iter(bvec->bv_page, 1135 bvec->bv_offset, 1136 bvec->bv_len, 1137 iter); 1138 1139 if (!iov_iter_count(iter)) 1140 break; 1141 1142 if (ret < bvec->bv_len) 1143 return -EFAULT; 1144 } 1145 1146 return 0; 1147 } 1148 1149 /** 1150 * bio_copy_to_iter - copy all pages from bio to iov_iter 1151 * @bio: The &struct bio which describes the I/O as source 1152 * @iter: iov_iter as destination 1153 * 1154 * Copy all pages from bio to iov_iter. 1155 * Returns 0 on success, or error on failure. 1156 */ 1157 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 1158 { 1159 int i; 1160 struct bio_vec *bvec; 1161 struct bvec_iter_all iter_all; 1162 1163 bio_for_each_segment_all(bvec, bio, i, iter_all) { 1164 ssize_t ret; 1165 1166 ret = copy_page_to_iter(bvec->bv_page, 1167 bvec->bv_offset, 1168 bvec->bv_len, 1169 &iter); 1170 1171 if (!iov_iter_count(&iter)) 1172 break; 1173 1174 if (ret < bvec->bv_len) 1175 return -EFAULT; 1176 } 1177 1178 return 0; 1179 } 1180 1181 void bio_free_pages(struct bio *bio) 1182 { 1183 struct bio_vec *bvec; 1184 int i; 1185 struct bvec_iter_all iter_all; 1186 1187 bio_for_each_segment_all(bvec, bio, i, iter_all) 1188 __free_page(bvec->bv_page); 1189 } 1190 EXPORT_SYMBOL(bio_free_pages); 1191 1192 /** 1193 * bio_uncopy_user - finish previously mapped bio 1194 * @bio: bio being terminated 1195 * 1196 * Free pages allocated from bio_copy_user_iov() and write back data 1197 * to user space in case of a read. 1198 */ 1199 int bio_uncopy_user(struct bio *bio) 1200 { 1201 struct bio_map_data *bmd = bio->bi_private; 1202 int ret = 0; 1203 1204 if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1205 /* 1206 * if we're in a workqueue, the request is orphaned, so 1207 * don't copy into a random user address space, just free 1208 * and return -EINTR so user space doesn't expect any data. 1209 */ 1210 if (!current->mm) 1211 ret = -EINTR; 1212 else if (bio_data_dir(bio) == READ) 1213 ret = bio_copy_to_iter(bio, bmd->iter); 1214 if (bmd->is_our_pages) 1215 bio_free_pages(bio); 1216 } 1217 kfree(bmd); 1218 bio_put(bio); 1219 return ret; 1220 } 1221 1222 /** 1223 * bio_copy_user_iov - copy user data to bio 1224 * @q: destination block queue 1225 * @map_data: pointer to the rq_map_data holding pages (if necessary) 1226 * @iter: iovec iterator 1227 * @gfp_mask: memory allocation flags 1228 * 1229 * Prepares and returns a bio for indirect user io, bouncing data 1230 * to/from kernel pages as necessary. Must be paired with 1231 * call bio_uncopy_user() on io completion. 1232 */ 1233 struct bio *bio_copy_user_iov(struct request_queue *q, 1234 struct rq_map_data *map_data, 1235 struct iov_iter *iter, 1236 gfp_t gfp_mask) 1237 { 1238 struct bio_map_data *bmd; 1239 struct page *page; 1240 struct bio *bio; 1241 int i = 0, ret; 1242 int nr_pages; 1243 unsigned int len = iter->count; 1244 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1245 1246 bmd = bio_alloc_map_data(iter, gfp_mask); 1247 if (!bmd) 1248 return ERR_PTR(-ENOMEM); 1249 1250 /* 1251 * We need to do a deep copy of the iov_iter including the iovecs. 1252 * The caller provided iov might point to an on-stack or otherwise 1253 * shortlived one. 1254 */ 1255 bmd->is_our_pages = map_data ? 0 : 1; 1256 1257 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1258 if (nr_pages > BIO_MAX_PAGES) 1259 nr_pages = BIO_MAX_PAGES; 1260 1261 ret = -ENOMEM; 1262 bio = bio_kmalloc(gfp_mask, nr_pages); 1263 if (!bio) 1264 goto out_bmd; 1265 1266 ret = 0; 1267 1268 if (map_data) { 1269 nr_pages = 1 << map_data->page_order; 1270 i = map_data->offset / PAGE_SIZE; 1271 } 1272 while (len) { 1273 unsigned int bytes = PAGE_SIZE; 1274 1275 bytes -= offset; 1276 1277 if (bytes > len) 1278 bytes = len; 1279 1280 if (map_data) { 1281 if (i == map_data->nr_entries * nr_pages) { 1282 ret = -ENOMEM; 1283 break; 1284 } 1285 1286 page = map_data->pages[i / nr_pages]; 1287 page += (i % nr_pages); 1288 1289 i++; 1290 } else { 1291 page = alloc_page(q->bounce_gfp | gfp_mask); 1292 if (!page) { 1293 ret = -ENOMEM; 1294 break; 1295 } 1296 } 1297 1298 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) 1299 break; 1300 1301 len -= bytes; 1302 offset = 0; 1303 } 1304 1305 if (ret) 1306 goto cleanup; 1307 1308 if (map_data) 1309 map_data->offset += bio->bi_iter.bi_size; 1310 1311 /* 1312 * success 1313 */ 1314 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1315 (map_data && map_data->from_user)) { 1316 ret = bio_copy_from_iter(bio, iter); 1317 if (ret) 1318 goto cleanup; 1319 } else { 1320 if (bmd->is_our_pages) 1321 zero_fill_bio(bio); 1322 iov_iter_advance(iter, bio->bi_iter.bi_size); 1323 } 1324 1325 bio->bi_private = bmd; 1326 if (map_data && map_data->null_mapped) 1327 bio_set_flag(bio, BIO_NULL_MAPPED); 1328 return bio; 1329 cleanup: 1330 if (!map_data) 1331 bio_free_pages(bio); 1332 bio_put(bio); 1333 out_bmd: 1334 kfree(bmd); 1335 return ERR_PTR(ret); 1336 } 1337 1338 /** 1339 * bio_map_user_iov - map user iovec into bio 1340 * @q: the struct request_queue for the bio 1341 * @iter: iovec iterator 1342 * @gfp_mask: memory allocation flags 1343 * 1344 * Map the user space address into a bio suitable for io to a block 1345 * device. Returns an error pointer in case of error. 1346 */ 1347 struct bio *bio_map_user_iov(struct request_queue *q, 1348 struct iov_iter *iter, 1349 gfp_t gfp_mask) 1350 { 1351 int j; 1352 struct bio *bio; 1353 int ret; 1354 struct bio_vec *bvec; 1355 struct bvec_iter_all iter_all; 1356 1357 if (!iov_iter_count(iter)) 1358 return ERR_PTR(-EINVAL); 1359 1360 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1361 if (!bio) 1362 return ERR_PTR(-ENOMEM); 1363 1364 while (iov_iter_count(iter)) { 1365 struct page **pages; 1366 ssize_t bytes; 1367 size_t offs, added = 0; 1368 int npages; 1369 1370 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1371 if (unlikely(bytes <= 0)) { 1372 ret = bytes ? bytes : -EFAULT; 1373 goto out_unmap; 1374 } 1375 1376 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1377 1378 if (unlikely(offs & queue_dma_alignment(q))) { 1379 ret = -EINVAL; 1380 j = 0; 1381 } else { 1382 for (j = 0; j < npages; j++) { 1383 struct page *page = pages[j]; 1384 unsigned int n = PAGE_SIZE - offs; 1385 unsigned short prev_bi_vcnt = bio->bi_vcnt; 1386 1387 if (n > bytes) 1388 n = bytes; 1389 1390 if (!bio_add_pc_page(q, bio, page, n, offs)) 1391 break; 1392 1393 /* 1394 * check if vector was merged with previous 1395 * drop page reference if needed 1396 */ 1397 if (bio->bi_vcnt == prev_bi_vcnt) 1398 put_page(page); 1399 1400 added += n; 1401 bytes -= n; 1402 offs = 0; 1403 } 1404 iov_iter_advance(iter, added); 1405 } 1406 /* 1407 * release the pages we didn't map into the bio, if any 1408 */ 1409 while (j < npages) 1410 put_page(pages[j++]); 1411 kvfree(pages); 1412 /* couldn't stuff something into bio? */ 1413 if (bytes) 1414 break; 1415 } 1416 1417 bio_set_flag(bio, BIO_USER_MAPPED); 1418 1419 /* 1420 * subtle -- if bio_map_user_iov() ended up bouncing a bio, 1421 * it would normally disappear when its bi_end_io is run. 1422 * however, we need it for the unmap, so grab an extra 1423 * reference to it 1424 */ 1425 bio_get(bio); 1426 return bio; 1427 1428 out_unmap: 1429 bio_for_each_segment_all(bvec, bio, j, iter_all) { 1430 put_page(bvec->bv_page); 1431 } 1432 bio_put(bio); 1433 return ERR_PTR(ret); 1434 } 1435 1436 static void __bio_unmap_user(struct bio *bio) 1437 { 1438 struct bio_vec *bvec; 1439 int i; 1440 struct bvec_iter_all iter_all; 1441 1442 /* 1443 * make sure we dirty pages we wrote to 1444 */ 1445 bio_for_each_segment_all(bvec, bio, i, iter_all) { 1446 if (bio_data_dir(bio) == READ) 1447 set_page_dirty_lock(bvec->bv_page); 1448 1449 put_page(bvec->bv_page); 1450 } 1451 1452 bio_put(bio); 1453 } 1454 1455 /** 1456 * bio_unmap_user - unmap a bio 1457 * @bio: the bio being unmapped 1458 * 1459 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 1460 * process context. 1461 * 1462 * bio_unmap_user() may sleep. 1463 */ 1464 void bio_unmap_user(struct bio *bio) 1465 { 1466 __bio_unmap_user(bio); 1467 bio_put(bio); 1468 } 1469 1470 static void bio_map_kern_endio(struct bio *bio) 1471 { 1472 bio_put(bio); 1473 } 1474 1475 /** 1476 * bio_map_kern - map kernel address into bio 1477 * @q: the struct request_queue for the bio 1478 * @data: pointer to buffer to map 1479 * @len: length in bytes 1480 * @gfp_mask: allocation flags for bio allocation 1481 * 1482 * Map the kernel address into a bio suitable for io to a block 1483 * device. Returns an error pointer in case of error. 1484 */ 1485 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 1486 gfp_t gfp_mask) 1487 { 1488 unsigned long kaddr = (unsigned long)data; 1489 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1490 unsigned long start = kaddr >> PAGE_SHIFT; 1491 const int nr_pages = end - start; 1492 int offset, i; 1493 struct bio *bio; 1494 1495 bio = bio_kmalloc(gfp_mask, nr_pages); 1496 if (!bio) 1497 return ERR_PTR(-ENOMEM); 1498 1499 offset = offset_in_page(kaddr); 1500 for (i = 0; i < nr_pages; i++) { 1501 unsigned int bytes = PAGE_SIZE - offset; 1502 1503 if (len <= 0) 1504 break; 1505 1506 if (bytes > len) 1507 bytes = len; 1508 1509 if (bio_add_pc_page(q, bio, virt_to_page(data), 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 int i; 1537 struct bvec_iter_all iter_all; 1538 1539 bio_for_each_segment_all(bvec, bio, i, iter_all) { 1540 memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1541 p += bvec->bv_len; 1542 } 1543 1544 bio_copy_kern_endio(bio); 1545 } 1546 1547 /** 1548 * bio_copy_kern - copy kernel address into bio 1549 * @q: the struct request_queue for the bio 1550 * @data: pointer to buffer to copy 1551 * @len: length in bytes 1552 * @gfp_mask: allocation flags for bio and page allocation 1553 * @reading: data direction is READ 1554 * 1555 * copy the kernel address into a bio suitable for io to a block 1556 * device. Returns an error pointer in case of error. 1557 */ 1558 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1559 gfp_t gfp_mask, int reading) 1560 { 1561 unsigned long kaddr = (unsigned long)data; 1562 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1563 unsigned long start = kaddr >> PAGE_SHIFT; 1564 struct bio *bio; 1565 void *p = data; 1566 int nr_pages = 0; 1567 1568 /* 1569 * Overflow, abort 1570 */ 1571 if (end < start) 1572 return ERR_PTR(-EINVAL); 1573 1574 nr_pages = end - start; 1575 bio = bio_kmalloc(gfp_mask, nr_pages); 1576 if (!bio) 1577 return ERR_PTR(-ENOMEM); 1578 1579 while (len) { 1580 struct page *page; 1581 unsigned int bytes = PAGE_SIZE; 1582 1583 if (bytes > len) 1584 bytes = len; 1585 1586 page = alloc_page(q->bounce_gfp | gfp_mask); 1587 if (!page) 1588 goto cleanup; 1589 1590 if (!reading) 1591 memcpy(page_address(page), p, bytes); 1592 1593 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 1594 break; 1595 1596 len -= bytes; 1597 p += bytes; 1598 } 1599 1600 if (reading) { 1601 bio->bi_end_io = bio_copy_kern_endio_read; 1602 bio->bi_private = data; 1603 } else { 1604 bio->bi_end_io = bio_copy_kern_endio; 1605 } 1606 1607 return bio; 1608 1609 cleanup: 1610 bio_free_pages(bio); 1611 bio_put(bio); 1612 return ERR_PTR(-ENOMEM); 1613 } 1614 1615 /* 1616 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1617 * for performing direct-IO in BIOs. 1618 * 1619 * The problem is that we cannot run set_page_dirty() from interrupt context 1620 * because the required locks are not interrupt-safe. So what we can do is to 1621 * mark the pages dirty _before_ performing IO. And in interrupt context, 1622 * check that the pages are still dirty. If so, fine. If not, redirty them 1623 * in process context. 1624 * 1625 * We special-case compound pages here: normally this means reads into hugetlb 1626 * pages. The logic in here doesn't really work right for compound pages 1627 * because the VM does not uniformly chase down the head page in all cases. 1628 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1629 * handle them at all. So we skip compound pages here at an early stage. 1630 * 1631 * Note that this code is very hard to test under normal circumstances because 1632 * direct-io pins the pages with get_user_pages(). This makes 1633 * is_page_cache_freeable return false, and the VM will not clean the pages. 1634 * But other code (eg, flusher threads) could clean the pages if they are mapped 1635 * pagecache. 1636 * 1637 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1638 * deferred bio dirtying paths. 1639 */ 1640 1641 /* 1642 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1643 */ 1644 void bio_set_pages_dirty(struct bio *bio) 1645 { 1646 struct bio_vec *bvec; 1647 int i; 1648 struct bvec_iter_all iter_all; 1649 1650 bio_for_each_segment_all(bvec, bio, i, iter_all) { 1651 if (!PageCompound(bvec->bv_page)) 1652 set_page_dirty_lock(bvec->bv_page); 1653 } 1654 } 1655 1656 static void bio_release_pages(struct bio *bio) 1657 { 1658 struct bio_vec *bvec; 1659 int i; 1660 struct bvec_iter_all iter_all; 1661 1662 bio_for_each_segment_all(bvec, bio, i, iter_all) 1663 put_page(bvec->bv_page); 1664 } 1665 1666 /* 1667 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1668 * If they are, then fine. If, however, some pages are clean then they must 1669 * have been written out during the direct-IO read. So we take another ref on 1670 * the BIO and re-dirty the pages in process context. 1671 * 1672 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1673 * here on. It will run one put_page() against each page and will run one 1674 * bio_put() against the BIO. 1675 */ 1676 1677 static void bio_dirty_fn(struct work_struct *work); 1678 1679 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1680 static DEFINE_SPINLOCK(bio_dirty_lock); 1681 static struct bio *bio_dirty_list; 1682 1683 /* 1684 * This runs in process context 1685 */ 1686 static void bio_dirty_fn(struct work_struct *work) 1687 { 1688 struct bio *bio, *next; 1689 1690 spin_lock_irq(&bio_dirty_lock); 1691 next = bio_dirty_list; 1692 bio_dirty_list = NULL; 1693 spin_unlock_irq(&bio_dirty_lock); 1694 1695 while ((bio = next) != NULL) { 1696 next = bio->bi_private; 1697 1698 bio_set_pages_dirty(bio); 1699 bio_release_pages(bio); 1700 bio_put(bio); 1701 } 1702 } 1703 1704 void bio_check_pages_dirty(struct bio *bio) 1705 { 1706 struct bio_vec *bvec; 1707 unsigned long flags; 1708 int i; 1709 struct bvec_iter_all iter_all; 1710 1711 bio_for_each_segment_all(bvec, bio, i, iter_all) { 1712 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1713 goto defer; 1714 } 1715 1716 bio_release_pages(bio); 1717 bio_put(bio); 1718 return; 1719 defer: 1720 spin_lock_irqsave(&bio_dirty_lock, flags); 1721 bio->bi_private = bio_dirty_list; 1722 bio_dirty_list = bio; 1723 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1724 schedule_work(&bio_dirty_work); 1725 } 1726 1727 void update_io_ticks(struct hd_struct *part, unsigned long now) 1728 { 1729 unsigned long stamp; 1730 again: 1731 stamp = READ_ONCE(part->stamp); 1732 if (unlikely(stamp != now)) { 1733 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 1734 __part_stat_add(part, io_ticks, 1); 1735 } 1736 } 1737 if (part->partno) { 1738 part = &part_to_disk(part)->part0; 1739 goto again; 1740 } 1741 } 1742 1743 void generic_start_io_acct(struct request_queue *q, int op, 1744 unsigned long sectors, struct hd_struct *part) 1745 { 1746 const int sgrp = op_stat_group(op); 1747 1748 part_stat_lock(); 1749 1750 update_io_ticks(part, jiffies); 1751 part_stat_inc(part, ios[sgrp]); 1752 part_stat_add(part, sectors[sgrp], sectors); 1753 part_inc_in_flight(q, part, op_is_write(op)); 1754 1755 part_stat_unlock(); 1756 } 1757 EXPORT_SYMBOL(generic_start_io_acct); 1758 1759 void generic_end_io_acct(struct request_queue *q, int req_op, 1760 struct hd_struct *part, unsigned long start_time) 1761 { 1762 unsigned long now = jiffies; 1763 unsigned long duration = now - start_time; 1764 const int sgrp = op_stat_group(req_op); 1765 1766 part_stat_lock(); 1767 1768 update_io_ticks(part, now); 1769 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 1770 part_stat_add(part, time_in_queue, duration); 1771 part_dec_in_flight(q, part, op_is_write(req_op)); 1772 1773 part_stat_unlock(); 1774 } 1775 EXPORT_SYMBOL(generic_end_io_acct); 1776 1777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1778 void bio_flush_dcache_pages(struct bio *bi) 1779 { 1780 struct bio_vec bvec; 1781 struct bvec_iter iter; 1782 1783 bio_for_each_segment(bvec, bi, iter) 1784 flush_dcache_page(bvec.bv_page); 1785 } 1786 EXPORT_SYMBOL(bio_flush_dcache_pages); 1787 #endif 1788 1789 static inline bool bio_remaining_done(struct bio *bio) 1790 { 1791 /* 1792 * If we're not chaining, then ->__bi_remaining is always 1 and 1793 * we always end io on the first invocation. 1794 */ 1795 if (!bio_flagged(bio, BIO_CHAIN)) 1796 return true; 1797 1798 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1799 1800 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1801 bio_clear_flag(bio, BIO_CHAIN); 1802 return true; 1803 } 1804 1805 return false; 1806 } 1807 1808 /** 1809 * bio_endio - end I/O on a bio 1810 * @bio: bio 1811 * 1812 * Description: 1813 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1814 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1815 * bio unless they own it and thus know that it has an end_io function. 1816 * 1817 * bio_endio() can be called several times on a bio that has been chained 1818 * using bio_chain(). The ->bi_end_io() function will only be called the 1819 * last time. At this point the BLK_TA_COMPLETE tracing event will be 1820 * generated if BIO_TRACE_COMPLETION is set. 1821 **/ 1822 void bio_endio(struct bio *bio) 1823 { 1824 again: 1825 if (!bio_remaining_done(bio)) 1826 return; 1827 if (!bio_integrity_endio(bio)) 1828 return; 1829 1830 if (bio->bi_disk) 1831 rq_qos_done_bio(bio->bi_disk->queue, bio); 1832 1833 /* 1834 * Need to have a real endio function for chained bios, otherwise 1835 * various corner cases will break (like stacking block devices that 1836 * save/restore bi_end_io) - however, we want to avoid unbounded 1837 * recursion and blowing the stack. Tail call optimization would 1838 * handle this, but compiling with frame pointers also disables 1839 * gcc's sibling call optimization. 1840 */ 1841 if (bio->bi_end_io == bio_chain_endio) { 1842 bio = __bio_chain_endio(bio); 1843 goto again; 1844 } 1845 1846 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1847 trace_block_bio_complete(bio->bi_disk->queue, bio, 1848 blk_status_to_errno(bio->bi_status)); 1849 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1850 } 1851 1852 blk_throtl_bio_endio(bio); 1853 /* release cgroup info */ 1854 bio_uninit(bio); 1855 if (bio->bi_end_io) 1856 bio->bi_end_io(bio); 1857 } 1858 EXPORT_SYMBOL(bio_endio); 1859 1860 /** 1861 * bio_split - split a bio 1862 * @bio: bio to split 1863 * @sectors: number of sectors to split from the front of @bio 1864 * @gfp: gfp mask 1865 * @bs: bio set to allocate from 1866 * 1867 * Allocates and returns a new bio which represents @sectors from the start of 1868 * @bio, and updates @bio to represent the remaining sectors. 1869 * 1870 * Unless this is a discard request the newly allocated bio will point 1871 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that 1872 * @bio is not freed before the split. 1873 */ 1874 struct bio *bio_split(struct bio *bio, int sectors, 1875 gfp_t gfp, struct bio_set *bs) 1876 { 1877 struct bio *split; 1878 1879 BUG_ON(sectors <= 0); 1880 BUG_ON(sectors >= bio_sectors(bio)); 1881 1882 split = bio_clone_fast(bio, gfp, bs); 1883 if (!split) 1884 return NULL; 1885 1886 split->bi_iter.bi_size = sectors << 9; 1887 1888 if (bio_integrity(split)) 1889 bio_integrity_trim(split); 1890 1891 bio_advance(bio, split->bi_iter.bi_size); 1892 1893 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1894 bio_set_flag(split, BIO_TRACE_COMPLETION); 1895 1896 return split; 1897 } 1898 EXPORT_SYMBOL(bio_split); 1899 1900 /** 1901 * bio_trim - trim a bio 1902 * @bio: bio to trim 1903 * @offset: number of sectors to trim from the front of @bio 1904 * @size: size we want to trim @bio to, in sectors 1905 */ 1906 void bio_trim(struct bio *bio, int offset, int size) 1907 { 1908 /* 'bio' is a cloned bio which we need to trim to match 1909 * the given offset and size. 1910 */ 1911 1912 size <<= 9; 1913 if (offset == 0 && size == bio->bi_iter.bi_size) 1914 return; 1915 1916 bio_clear_flag(bio, BIO_SEG_VALID); 1917 1918 bio_advance(bio, offset << 9); 1919 1920 bio->bi_iter.bi_size = size; 1921 1922 if (bio_integrity(bio)) 1923 bio_integrity_trim(bio); 1924 1925 } 1926 EXPORT_SYMBOL_GPL(bio_trim); 1927 1928 /* 1929 * create memory pools for biovec's in a bio_set. 1930 * use the global biovec slabs created for general use. 1931 */ 1932 int biovec_init_pool(mempool_t *pool, int pool_entries) 1933 { 1934 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1935 1936 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1937 } 1938 1939 /* 1940 * bioset_exit - exit a bioset initialized with bioset_init() 1941 * 1942 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1943 * kzalloc()). 1944 */ 1945 void bioset_exit(struct bio_set *bs) 1946 { 1947 if (bs->rescue_workqueue) 1948 destroy_workqueue(bs->rescue_workqueue); 1949 bs->rescue_workqueue = NULL; 1950 1951 mempool_exit(&bs->bio_pool); 1952 mempool_exit(&bs->bvec_pool); 1953 1954 bioset_integrity_free(bs); 1955 if (bs->bio_slab) 1956 bio_put_slab(bs); 1957 bs->bio_slab = NULL; 1958 } 1959 EXPORT_SYMBOL(bioset_exit); 1960 1961 /** 1962 * bioset_init - Initialize a bio_set 1963 * @bs: pool to initialize 1964 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1965 * @front_pad: Number of bytes to allocate in front of the returned bio 1966 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1967 * and %BIOSET_NEED_RESCUER 1968 * 1969 * Description: 1970 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1971 * to ask for a number of bytes to be allocated in front of the bio. 1972 * Front pad allocation is useful for embedding the bio inside 1973 * another structure, to avoid allocating extra data to go with the bio. 1974 * Note that the bio must be embedded at the END of that structure always, 1975 * or things will break badly. 1976 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1977 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1978 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1979 * dispatch queued requests when the mempool runs out of space. 1980 * 1981 */ 1982 int bioset_init(struct bio_set *bs, 1983 unsigned int pool_size, 1984 unsigned int front_pad, 1985 int flags) 1986 { 1987 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1988 1989 bs->front_pad = front_pad; 1990 1991 spin_lock_init(&bs->rescue_lock); 1992 bio_list_init(&bs->rescue_list); 1993 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1994 1995 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1996 if (!bs->bio_slab) 1997 return -ENOMEM; 1998 1999 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2000 goto bad; 2001 2002 if ((flags & BIOSET_NEED_BVECS) && 2003 biovec_init_pool(&bs->bvec_pool, pool_size)) 2004 goto bad; 2005 2006 if (!(flags & BIOSET_NEED_RESCUER)) 2007 return 0; 2008 2009 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2010 if (!bs->rescue_workqueue) 2011 goto bad; 2012 2013 return 0; 2014 bad: 2015 bioset_exit(bs); 2016 return -ENOMEM; 2017 } 2018 EXPORT_SYMBOL(bioset_init); 2019 2020 /* 2021 * Initialize and setup a new bio_set, based on the settings from 2022 * another bio_set. 2023 */ 2024 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 2025 { 2026 int flags; 2027 2028 flags = 0; 2029 if (src->bvec_pool.min_nr) 2030 flags |= BIOSET_NEED_BVECS; 2031 if (src->rescue_workqueue) 2032 flags |= BIOSET_NEED_RESCUER; 2033 2034 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 2035 } 2036 EXPORT_SYMBOL(bioset_init_from_src); 2037 2038 #ifdef CONFIG_BLK_CGROUP 2039 2040 /** 2041 * bio_disassociate_blkg - puts back the blkg reference if associated 2042 * @bio: target bio 2043 * 2044 * Helper to disassociate the blkg from @bio if a blkg is associated. 2045 */ 2046 void bio_disassociate_blkg(struct bio *bio) 2047 { 2048 if (bio->bi_blkg) { 2049 blkg_put(bio->bi_blkg); 2050 bio->bi_blkg = NULL; 2051 } 2052 } 2053 EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2054 2055 /** 2056 * __bio_associate_blkg - associate a bio with the a blkg 2057 * @bio: target bio 2058 * @blkg: the blkg to associate 2059 * 2060 * This tries to associate @bio with the specified @blkg. Association failure 2061 * is handled by walking up the blkg tree. Therefore, the blkg associated can 2062 * be anything between @blkg and the root_blkg. This situation only happens 2063 * when a cgroup is dying and then the remaining bios will spill to the closest 2064 * alive blkg. 2065 * 2066 * A reference will be taken on the @blkg and will be released when @bio is 2067 * freed. 2068 */ 2069 static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2070 { 2071 bio_disassociate_blkg(bio); 2072 2073 bio->bi_blkg = blkg_tryget_closest(blkg); 2074 } 2075 2076 /** 2077 * bio_associate_blkg_from_css - associate a bio with a specified css 2078 * @bio: target bio 2079 * @css: target css 2080 * 2081 * Associate @bio with the blkg found by combining the css's blkg and the 2082 * request_queue of the @bio. This falls back to the queue's root_blkg if 2083 * the association fails with the css. 2084 */ 2085 void bio_associate_blkg_from_css(struct bio *bio, 2086 struct cgroup_subsys_state *css) 2087 { 2088 struct request_queue *q = bio->bi_disk->queue; 2089 struct blkcg_gq *blkg; 2090 2091 rcu_read_lock(); 2092 2093 if (!css || !css->parent) 2094 blkg = q->root_blkg; 2095 else 2096 blkg = blkg_lookup_create(css_to_blkcg(css), q); 2097 2098 __bio_associate_blkg(bio, blkg); 2099 2100 rcu_read_unlock(); 2101 } 2102 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2103 2104 #ifdef CONFIG_MEMCG 2105 /** 2106 * bio_associate_blkg_from_page - associate a bio with the page's blkg 2107 * @bio: target bio 2108 * @page: the page to lookup the blkcg from 2109 * 2110 * Associate @bio with the blkg from @page's owning memcg and the respective 2111 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2112 * root_blkg. 2113 */ 2114 void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 2115 { 2116 struct cgroup_subsys_state *css; 2117 2118 if (!page->mem_cgroup) 2119 return; 2120 2121 rcu_read_lock(); 2122 2123 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2124 bio_associate_blkg_from_css(bio, css); 2125 2126 rcu_read_unlock(); 2127 } 2128 #endif /* CONFIG_MEMCG */ 2129 2130 /** 2131 * bio_associate_blkg - associate a bio with a blkg 2132 * @bio: target bio 2133 * 2134 * Associate @bio with the blkg found from the bio's css and request_queue. 2135 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 2136 * already associated, the css is reused and association redone as the 2137 * request_queue may have changed. 2138 */ 2139 void bio_associate_blkg(struct bio *bio) 2140 { 2141 struct cgroup_subsys_state *css; 2142 2143 rcu_read_lock(); 2144 2145 if (bio->bi_blkg) 2146 css = &bio_blkcg(bio)->css; 2147 else 2148 css = blkcg_css(); 2149 2150 bio_associate_blkg_from_css(bio, css); 2151 2152 rcu_read_unlock(); 2153 } 2154 EXPORT_SYMBOL_GPL(bio_associate_blkg); 2155 2156 /** 2157 * bio_clone_blkg_association - clone blkg association from src to dst bio 2158 * @dst: destination bio 2159 * @src: source bio 2160 */ 2161 void bio_clone_blkg_association(struct bio *dst, struct bio *src) 2162 { 2163 rcu_read_lock(); 2164 2165 if (src->bi_blkg) 2166 __bio_associate_blkg(dst, src->bi_blkg); 2167 2168 rcu_read_unlock(); 2169 } 2170 EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2171 #endif /* CONFIG_BLK_CGROUP */ 2172 2173 static void __init biovec_init_slabs(void) 2174 { 2175 int i; 2176 2177 for (i = 0; i < BVEC_POOL_NR; i++) { 2178 int size; 2179 struct biovec_slab *bvs = bvec_slabs + i; 2180 2181 if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2182 bvs->slab = NULL; 2183 continue; 2184 } 2185 2186 size = bvs->nr_vecs * sizeof(struct bio_vec); 2187 bvs->slab = kmem_cache_create(bvs->name, size, 0, 2188 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2189 } 2190 } 2191 2192 static int __init init_bio(void) 2193 { 2194 bio_slab_max = 2; 2195 bio_slab_nr = 0; 2196 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 2197 GFP_KERNEL); 2198 if (!bio_slabs) 2199 panic("bio: can't allocate bios\n"); 2200 2201 bio_integrity_init(); 2202 biovec_init_slabs(); 2203 2204 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2205 panic("bio: can't allocate bios\n"); 2206 2207 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2208 panic("bio: can't create integrity pool\n"); 2209 2210 return 0; 2211 } 2212 subsys_initcall(init_bio); 2213