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