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