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 bio_vec *table, 253 unsigned short max_vecs) 254 { 255 bio->bi_next = NULL; 256 bio->bi_bdev = NULL; 257 bio->bi_opf = 0; 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 #ifdef CONFIG_BLK_CGROUP_IOCOST 272 bio->bi_iocost_cost = 0; 273 #endif 274 #endif 275 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 276 bio->bi_crypt_context = NULL; 277 #endif 278 #ifdef CONFIG_BLK_DEV_INTEGRITY 279 bio->bi_integrity = NULL; 280 #endif 281 bio->bi_vcnt = 0; 282 283 atomic_set(&bio->__bi_remaining, 1); 284 atomic_set(&bio->__bi_cnt, 1); 285 bio->bi_cookie = BLK_QC_T_NONE; 286 287 bio->bi_max_vecs = max_vecs; 288 bio->bi_io_vec = table; 289 bio->bi_pool = NULL; 290 } 291 EXPORT_SYMBOL(bio_init); 292 293 /** 294 * bio_reset - reinitialize a bio 295 * @bio: bio to reset 296 * 297 * Description: 298 * After calling bio_reset(), @bio will be in the same state as a freshly 299 * allocated bio returned bio bio_alloc_bioset() - the only fields that are 300 * preserved are the ones that are initialized by bio_alloc_bioset(). See 301 * comment in struct bio. 302 */ 303 void bio_reset(struct bio *bio) 304 { 305 bio_uninit(bio); 306 memset(bio, 0, BIO_RESET_BYTES); 307 atomic_set(&bio->__bi_remaining, 1); 308 } 309 EXPORT_SYMBOL(bio_reset); 310 311 static struct bio *__bio_chain_endio(struct bio *bio) 312 { 313 struct bio *parent = bio->bi_private; 314 315 if (bio->bi_status && !parent->bi_status) 316 parent->bi_status = bio->bi_status; 317 bio_put(bio); 318 return parent; 319 } 320 321 static void bio_chain_endio(struct bio *bio) 322 { 323 bio_endio(__bio_chain_endio(bio)); 324 } 325 326 /** 327 * bio_chain - chain bio completions 328 * @bio: the target bio 329 * @parent: the parent bio of @bio 330 * 331 * The caller won't have a bi_end_io called when @bio completes - instead, 332 * @parent's bi_end_io won't be called until both @parent and @bio have 333 * completed; the chained bio will also be freed when it completes. 334 * 335 * The caller must not set bi_private or bi_end_io in @bio. 336 */ 337 void bio_chain(struct bio *bio, struct bio *parent) 338 { 339 BUG_ON(bio->bi_private || bio->bi_end_io); 340 341 bio->bi_private = parent; 342 bio->bi_end_io = bio_chain_endio; 343 bio_inc_remaining(parent); 344 } 345 EXPORT_SYMBOL(bio_chain); 346 347 static void bio_alloc_rescue(struct work_struct *work) 348 { 349 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 350 struct bio *bio; 351 352 while (1) { 353 spin_lock(&bs->rescue_lock); 354 bio = bio_list_pop(&bs->rescue_list); 355 spin_unlock(&bs->rescue_lock); 356 357 if (!bio) 358 break; 359 360 submit_bio_noacct(bio); 361 } 362 } 363 364 static void punt_bios_to_rescuer(struct bio_set *bs) 365 { 366 struct bio_list punt, nopunt; 367 struct bio *bio; 368 369 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 370 return; 371 /* 372 * In order to guarantee forward progress we must punt only bios that 373 * were allocated from this bio_set; otherwise, if there was a bio on 374 * there for a stacking driver higher up in the stack, processing it 375 * could require allocating bios from this bio_set, and doing that from 376 * our own rescuer would be bad. 377 * 378 * Since bio lists are singly linked, pop them all instead of trying to 379 * remove from the middle of the list: 380 */ 381 382 bio_list_init(&punt); 383 bio_list_init(&nopunt); 384 385 while ((bio = bio_list_pop(¤t->bio_list[0]))) 386 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 387 current->bio_list[0] = nopunt; 388 389 bio_list_init(&nopunt); 390 while ((bio = bio_list_pop(¤t->bio_list[1]))) 391 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 392 current->bio_list[1] = nopunt; 393 394 spin_lock(&bs->rescue_lock); 395 bio_list_merge(&bs->rescue_list, &punt); 396 spin_unlock(&bs->rescue_lock); 397 398 queue_work(bs->rescue_workqueue, &bs->rescue_work); 399 } 400 401 /** 402 * bio_alloc_bioset - allocate a bio for I/O 403 * @gfp_mask: the GFP_* mask given to the slab allocator 404 * @nr_iovecs: number of iovecs to pre-allocate 405 * @bs: the bio_set to allocate from. 406 * 407 * Allocate a bio from the mempools in @bs. 408 * 409 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 410 * allocate a bio. This is due to the mempool guarantees. To make this work, 411 * callers must never allocate more than 1 bio at a time from the general pool. 412 * Callers that need to allocate more than 1 bio must always submit the 413 * previously allocated bio for IO before attempting to allocate a new one. 414 * Failure to do so can cause deadlocks under memory pressure. 415 * 416 * Note that when running under submit_bio_noacct() (i.e. any block driver), 417 * bios are not submitted until after you return - see the code in 418 * submit_bio_noacct() that converts recursion into iteration, to prevent 419 * stack overflows. 420 * 421 * This would normally mean allocating multiple bios under submit_bio_noacct() 422 * would be susceptible to deadlocks, but we have 423 * deadlock avoidance code that resubmits any blocked bios from a rescuer 424 * thread. 425 * 426 * However, we do not guarantee forward progress for allocations from other 427 * mempools. Doing multiple allocations from the same mempool under 428 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 429 * for per bio allocations. 430 * 431 * Returns: Pointer to new bio on success, NULL on failure. 432 */ 433 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs, 434 struct bio_set *bs) 435 { 436 gfp_t saved_gfp = gfp_mask; 437 struct bio *bio; 438 void *p; 439 440 /* should not use nobvec bioset for nr_iovecs > 0 */ 441 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) 442 return NULL; 443 444 /* 445 * submit_bio_noacct() converts recursion to iteration; this means if 446 * we're running beneath it, any bios we allocate and submit will not be 447 * submitted (and thus freed) until after we return. 448 * 449 * This exposes us to a potential deadlock if we allocate multiple bios 450 * from the same bio_set() while running underneath submit_bio_noacct(). 451 * If we were to allocate multiple bios (say a stacking block driver 452 * that was splitting bios), we would deadlock if we exhausted the 453 * mempool's reserve. 454 * 455 * We solve this, and guarantee forward progress, with a rescuer 456 * workqueue per bio_set. If we go to allocate and there are bios on 457 * current->bio_list, we first try the allocation without 458 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 459 * blocking to the rescuer workqueue before we retry with the original 460 * gfp_flags. 461 */ 462 if (current->bio_list && 463 (!bio_list_empty(¤t->bio_list[0]) || 464 !bio_list_empty(¤t->bio_list[1])) && 465 bs->rescue_workqueue) 466 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 467 468 p = mempool_alloc(&bs->bio_pool, gfp_mask); 469 if (!p && gfp_mask != saved_gfp) { 470 punt_bios_to_rescuer(bs); 471 gfp_mask = saved_gfp; 472 p = mempool_alloc(&bs->bio_pool, gfp_mask); 473 } 474 if (unlikely(!p)) 475 return NULL; 476 477 bio = p + bs->front_pad; 478 if (nr_iovecs > BIO_INLINE_VECS) { 479 struct bio_vec *bvl = NULL; 480 481 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 482 if (!bvl && gfp_mask != saved_gfp) { 483 punt_bios_to_rescuer(bs); 484 gfp_mask = saved_gfp; 485 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 486 } 487 if (unlikely(!bvl)) 488 goto err_free; 489 490 bio_init(bio, bvl, nr_iovecs); 491 } else if (nr_iovecs) { 492 bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS); 493 } else { 494 bio_init(bio, NULL, 0); 495 } 496 497 bio->bi_pool = bs; 498 return bio; 499 500 err_free: 501 mempool_free(p, &bs->bio_pool); 502 return NULL; 503 } 504 EXPORT_SYMBOL(bio_alloc_bioset); 505 506 /** 507 * bio_kmalloc - kmalloc a bio for I/O 508 * @gfp_mask: the GFP_* mask given to the slab allocator 509 * @nr_iovecs: number of iovecs to pre-allocate 510 * 511 * Use kmalloc to allocate and initialize a bio. 512 * 513 * Returns: Pointer to new bio on success, NULL on failure. 514 */ 515 struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) 516 { 517 struct bio *bio; 518 519 if (nr_iovecs > UIO_MAXIOV) 520 return NULL; 521 522 bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 523 if (unlikely(!bio)) 524 return NULL; 525 bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs); 526 bio->bi_pool = NULL; 527 return bio; 528 } 529 EXPORT_SYMBOL(bio_kmalloc); 530 531 void zero_fill_bio(struct bio *bio) 532 { 533 struct bio_vec bv; 534 struct bvec_iter iter; 535 536 bio_for_each_segment(bv, bio, iter) 537 memzero_bvec(&bv); 538 } 539 EXPORT_SYMBOL(zero_fill_bio); 540 541 /** 542 * bio_truncate - truncate the bio to small size of @new_size 543 * @bio: the bio to be truncated 544 * @new_size: new size for truncating the bio 545 * 546 * Description: 547 * Truncate the bio to new size of @new_size. If bio_op(bio) is 548 * REQ_OP_READ, zero the truncated part. This function should only 549 * be used for handling corner cases, such as bio eod. 550 */ 551 static void bio_truncate(struct bio *bio, unsigned new_size) 552 { 553 struct bio_vec bv; 554 struct bvec_iter iter; 555 unsigned int done = 0; 556 bool truncated = false; 557 558 if (new_size >= bio->bi_iter.bi_size) 559 return; 560 561 if (bio_op(bio) != REQ_OP_READ) 562 goto exit; 563 564 bio_for_each_segment(bv, bio, iter) { 565 if (done + bv.bv_len > new_size) { 566 unsigned offset; 567 568 if (!truncated) 569 offset = new_size - done; 570 else 571 offset = 0; 572 zero_user(bv.bv_page, offset, bv.bv_len - offset); 573 truncated = true; 574 } 575 done += bv.bv_len; 576 } 577 578 exit: 579 /* 580 * Don't touch bvec table here and make it really immutable, since 581 * fs bio user has to retrieve all pages via bio_for_each_segment_all 582 * in its .end_bio() callback. 583 * 584 * It is enough to truncate bio by updating .bi_size since we can make 585 * correct bvec with the updated .bi_size for drivers. 586 */ 587 bio->bi_iter.bi_size = new_size; 588 } 589 590 /** 591 * guard_bio_eod - truncate a BIO to fit the block device 592 * @bio: bio to truncate 593 * 594 * This allows us to do IO even on the odd last sectors of a device, even if the 595 * block size is some multiple of the physical sector size. 596 * 597 * We'll just truncate the bio to the size of the device, and clear the end of 598 * the buffer head manually. Truly out-of-range accesses will turn into actual 599 * I/O errors, this only handles the "we need to be able to do I/O at the final 600 * sector" case. 601 */ 602 void guard_bio_eod(struct bio *bio) 603 { 604 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 605 606 if (!maxsector) 607 return; 608 609 /* 610 * If the *whole* IO is past the end of the device, 611 * let it through, and the IO layer will turn it into 612 * an EIO. 613 */ 614 if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 615 return; 616 617 maxsector -= bio->bi_iter.bi_sector; 618 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 619 return; 620 621 bio_truncate(bio, maxsector << 9); 622 } 623 624 #define ALLOC_CACHE_MAX 512 625 #define ALLOC_CACHE_SLACK 64 626 627 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 628 unsigned int nr) 629 { 630 unsigned int i = 0; 631 struct bio *bio; 632 633 while ((bio = cache->free_list) != NULL) { 634 cache->free_list = bio->bi_next; 635 cache->nr--; 636 bio_free(bio); 637 if (++i == nr) 638 break; 639 } 640 } 641 642 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 643 { 644 struct bio_set *bs; 645 646 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 647 if (bs->cache) { 648 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 649 650 bio_alloc_cache_prune(cache, -1U); 651 } 652 return 0; 653 } 654 655 static void bio_alloc_cache_destroy(struct bio_set *bs) 656 { 657 int cpu; 658 659 if (!bs->cache) 660 return; 661 662 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 663 for_each_possible_cpu(cpu) { 664 struct bio_alloc_cache *cache; 665 666 cache = per_cpu_ptr(bs->cache, cpu); 667 bio_alloc_cache_prune(cache, -1U); 668 } 669 free_percpu(bs->cache); 670 } 671 672 /** 673 * bio_put - release a reference to a bio 674 * @bio: bio to release reference to 675 * 676 * Description: 677 * Put a reference to a &struct bio, either one you have gotten with 678 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 679 **/ 680 void bio_put(struct bio *bio) 681 { 682 if (unlikely(bio_flagged(bio, BIO_REFFED))) { 683 BUG_ON(!atomic_read(&bio->__bi_cnt)); 684 if (!atomic_dec_and_test(&bio->__bi_cnt)) 685 return; 686 } 687 688 if (bio_flagged(bio, BIO_PERCPU_CACHE)) { 689 struct bio_alloc_cache *cache; 690 691 bio_uninit(bio); 692 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 693 bio->bi_next = cache->free_list; 694 cache->free_list = bio; 695 if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 696 bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 697 put_cpu(); 698 } else { 699 bio_free(bio); 700 } 701 } 702 EXPORT_SYMBOL(bio_put); 703 704 /** 705 * __bio_clone_fast - clone a bio that shares the original bio's biovec 706 * @bio: destination bio 707 * @bio_src: bio to clone 708 * 709 * Clone a &bio. Caller will own the returned bio, but not 710 * the actual data it points to. Reference count of returned 711 * bio will be one. 712 * 713 * Caller must ensure that @bio_src is not freed before @bio. 714 */ 715 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 716 { 717 WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs); 718 719 /* 720 * most users will be overriding ->bi_bdev with a new target, 721 * so we don't set nor calculate new physical/hw segment counts here 722 */ 723 bio->bi_bdev = bio_src->bi_bdev; 724 bio_set_flag(bio, BIO_CLONED); 725 if (bio_flagged(bio_src, BIO_THROTTLED)) 726 bio_set_flag(bio, BIO_THROTTLED); 727 if (bio_flagged(bio_src, BIO_REMAPPED)) 728 bio_set_flag(bio, BIO_REMAPPED); 729 bio->bi_opf = bio_src->bi_opf; 730 bio->bi_ioprio = bio_src->bi_ioprio; 731 bio->bi_write_hint = bio_src->bi_write_hint; 732 bio->bi_iter = bio_src->bi_iter; 733 bio->bi_io_vec = bio_src->bi_io_vec; 734 735 bio_clone_blkg_association(bio, bio_src); 736 blkcg_bio_issue_init(bio); 737 } 738 EXPORT_SYMBOL(__bio_clone_fast); 739 740 /** 741 * bio_clone_fast - clone a bio that shares the original bio's biovec 742 * @bio: bio to clone 743 * @gfp_mask: allocation priority 744 * @bs: bio_set to allocate from 745 * 746 * Like __bio_clone_fast, only also allocates the returned bio 747 */ 748 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 749 { 750 struct bio *b; 751 752 b = bio_alloc_bioset(gfp_mask, 0, bs); 753 if (!b) 754 return NULL; 755 756 __bio_clone_fast(b, bio); 757 758 if (bio_crypt_clone(b, bio, gfp_mask) < 0) 759 goto err_put; 760 761 if (bio_integrity(bio) && 762 bio_integrity_clone(b, bio, gfp_mask) < 0) 763 goto err_put; 764 765 return b; 766 767 err_put: 768 bio_put(b); 769 return NULL; 770 } 771 EXPORT_SYMBOL(bio_clone_fast); 772 773 const char *bio_devname(struct bio *bio, char *buf) 774 { 775 return bdevname(bio->bi_bdev, buf); 776 } 777 EXPORT_SYMBOL(bio_devname); 778 779 /** 780 * bio_full - check if the bio is full 781 * @bio: bio to check 782 * @len: length of one segment to be added 783 * 784 * Return true if @bio is full and one segment with @len bytes can't be 785 * added to the bio, otherwise return false 786 */ 787 static inline bool bio_full(struct bio *bio, unsigned len) 788 { 789 if (bio->bi_vcnt >= bio->bi_max_vecs) 790 return true; 791 if (bio->bi_iter.bi_size > UINT_MAX - len) 792 return true; 793 return false; 794 } 795 796 static inline bool page_is_mergeable(const struct bio_vec *bv, 797 struct page *page, unsigned int len, unsigned int off, 798 bool *same_page) 799 { 800 size_t bv_end = bv->bv_offset + bv->bv_len; 801 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 802 phys_addr_t page_addr = page_to_phys(page); 803 804 if (vec_end_addr + 1 != page_addr + off) 805 return false; 806 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 807 return false; 808 809 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 810 if (*same_page) 811 return true; 812 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 813 } 814 815 /** 816 * __bio_try_merge_page - try appending data to an existing bvec. 817 * @bio: destination bio 818 * @page: start page to add 819 * @len: length of the data to add 820 * @off: offset of the data relative to @page 821 * @same_page: return if the segment has been merged inside the same page 822 * 823 * Try to add the data at @page + @off to the last bvec of @bio. This is a 824 * useful optimisation for file systems with a block size smaller than the 825 * page size. 826 * 827 * Warn if (@len, @off) crosses pages in case that @same_page is true. 828 * 829 * Return %true on success or %false on failure. 830 */ 831 static bool __bio_try_merge_page(struct bio *bio, struct page *page, 832 unsigned int len, unsigned int off, bool *same_page) 833 { 834 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 835 return false; 836 837 if (bio->bi_vcnt > 0) { 838 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 839 840 if (page_is_mergeable(bv, page, len, off, same_page)) { 841 if (bio->bi_iter.bi_size > UINT_MAX - len) { 842 *same_page = false; 843 return false; 844 } 845 bv->bv_len += len; 846 bio->bi_iter.bi_size += len; 847 return true; 848 } 849 } 850 return false; 851 } 852 853 /* 854 * Try to merge a page into a segment, while obeying the hardware segment 855 * size limit. This is not for normal read/write bios, but for passthrough 856 * or Zone Append operations that we can't split. 857 */ 858 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 859 struct page *page, unsigned len, 860 unsigned offset, bool *same_page) 861 { 862 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 863 unsigned long mask = queue_segment_boundary(q); 864 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 865 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 866 867 if ((addr1 | mask) != (addr2 | mask)) 868 return false; 869 if (bv->bv_len + len > queue_max_segment_size(q)) 870 return false; 871 return __bio_try_merge_page(bio, page, len, offset, same_page); 872 } 873 874 /** 875 * bio_add_hw_page - attempt to add a page to a bio with hw constraints 876 * @q: the target queue 877 * @bio: destination bio 878 * @page: page to add 879 * @len: vec entry length 880 * @offset: vec entry offset 881 * @max_sectors: maximum number of sectors that can be added 882 * @same_page: return if the segment has been merged inside the same page 883 * 884 * Add a page to a bio while respecting the hardware max_sectors, max_segment 885 * and gap limitations. 886 */ 887 int bio_add_hw_page(struct request_queue *q, struct bio *bio, 888 struct page *page, unsigned int len, unsigned int offset, 889 unsigned int max_sectors, bool *same_page) 890 { 891 struct bio_vec *bvec; 892 893 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 894 return 0; 895 896 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 897 return 0; 898 899 if (bio->bi_vcnt > 0) { 900 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 901 return len; 902 903 /* 904 * If the queue doesn't support SG gaps and adding this segment 905 * would create a gap, disallow it. 906 */ 907 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 908 if (bvec_gap_to_prev(q, bvec, offset)) 909 return 0; 910 } 911 912 if (bio_full(bio, len)) 913 return 0; 914 915 if (bio->bi_vcnt >= queue_max_segments(q)) 916 return 0; 917 918 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 919 bvec->bv_page = page; 920 bvec->bv_len = len; 921 bvec->bv_offset = offset; 922 bio->bi_vcnt++; 923 bio->bi_iter.bi_size += len; 924 return len; 925 } 926 927 /** 928 * bio_add_pc_page - attempt to add page to passthrough bio 929 * @q: the target queue 930 * @bio: destination bio 931 * @page: page to add 932 * @len: vec entry length 933 * @offset: vec entry offset 934 * 935 * Attempt to add a page to the bio_vec maplist. This can fail for a 936 * number of reasons, such as the bio being full or target block device 937 * limitations. The target block device must allow bio's up to PAGE_SIZE, 938 * so it is always possible to add a single page to an empty bio. 939 * 940 * This should only be used by passthrough bios. 941 */ 942 int bio_add_pc_page(struct request_queue *q, struct bio *bio, 943 struct page *page, unsigned int len, unsigned int offset) 944 { 945 bool same_page = false; 946 return bio_add_hw_page(q, bio, page, len, offset, 947 queue_max_hw_sectors(q), &same_page); 948 } 949 EXPORT_SYMBOL(bio_add_pc_page); 950 951 /** 952 * bio_add_zone_append_page - attempt to add page to zone-append bio 953 * @bio: destination bio 954 * @page: page to add 955 * @len: vec entry length 956 * @offset: vec entry offset 957 * 958 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 959 * for a zone-append request. This can fail for a number of reasons, such as the 960 * bio being full or the target block device is not a zoned block device or 961 * other limitations of the target block device. The target block device must 962 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 963 * to an empty bio. 964 * 965 * Returns: number of bytes added to the bio, or 0 in case of a failure. 966 */ 967 int bio_add_zone_append_page(struct bio *bio, struct page *page, 968 unsigned int len, unsigned int offset) 969 { 970 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 971 bool same_page = false; 972 973 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 974 return 0; 975 976 if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 977 return 0; 978 979 return bio_add_hw_page(q, bio, page, len, offset, 980 queue_max_zone_append_sectors(q), &same_page); 981 } 982 EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 983 984 /** 985 * __bio_add_page - add page(s) to a bio in a new segment 986 * @bio: destination bio 987 * @page: start page to add 988 * @len: length of the data to add, may cross pages 989 * @off: offset of the data relative to @page, may cross pages 990 * 991 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 992 * that @bio has space for another bvec. 993 */ 994 void __bio_add_page(struct bio *bio, struct page *page, 995 unsigned int len, unsigned int off) 996 { 997 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 998 999 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 1000 WARN_ON_ONCE(bio_full(bio, len)); 1001 1002 bv->bv_page = page; 1003 bv->bv_offset = off; 1004 bv->bv_len = len; 1005 1006 bio->bi_iter.bi_size += len; 1007 bio->bi_vcnt++; 1008 1009 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 1010 bio_set_flag(bio, BIO_WORKINGSET); 1011 } 1012 EXPORT_SYMBOL_GPL(__bio_add_page); 1013 1014 /** 1015 * bio_add_page - attempt to add page(s) to bio 1016 * @bio: destination bio 1017 * @page: start page to add 1018 * @len: vec entry length, may cross pages 1019 * @offset: vec entry offset relative to @page, may cross pages 1020 * 1021 * Attempt to add page(s) to the bio_vec maplist. This will only fail 1022 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1023 */ 1024 int bio_add_page(struct bio *bio, struct page *page, 1025 unsigned int len, unsigned int offset) 1026 { 1027 bool same_page = false; 1028 1029 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1030 if (bio_full(bio, len)) 1031 return 0; 1032 __bio_add_page(bio, page, len, offset); 1033 } 1034 return len; 1035 } 1036 EXPORT_SYMBOL(bio_add_page); 1037 1038 /** 1039 * bio_add_folio - Attempt to add part of a folio to a bio. 1040 * @bio: BIO to add to. 1041 * @folio: Folio to add. 1042 * @len: How many bytes from the folio to add. 1043 * @off: First byte in this folio to add. 1044 * 1045 * Filesystems that use folios can call this function instead of calling 1046 * bio_add_page() for each page in the folio. If @off is bigger than 1047 * PAGE_SIZE, this function can create a bio_vec that starts in a page 1048 * after the bv_page. BIOs do not support folios that are 4GiB or larger. 1049 * 1050 * Return: Whether the addition was successful. 1051 */ 1052 bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, 1053 size_t off) 1054 { 1055 if (len > UINT_MAX || off > UINT_MAX) 1056 return 0; 1057 return bio_add_page(bio, &folio->page, len, off) > 0; 1058 } 1059 1060 void __bio_release_pages(struct bio *bio, bool mark_dirty) 1061 { 1062 struct bvec_iter_all iter_all; 1063 struct bio_vec *bvec; 1064 1065 bio_for_each_segment_all(bvec, bio, iter_all) { 1066 if (mark_dirty && !PageCompound(bvec->bv_page)) 1067 set_page_dirty_lock(bvec->bv_page); 1068 put_page(bvec->bv_page); 1069 } 1070 } 1071 EXPORT_SYMBOL_GPL(__bio_release_pages); 1072 1073 void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 1074 { 1075 size_t size = iov_iter_count(iter); 1076 1077 WARN_ON_ONCE(bio->bi_max_vecs); 1078 1079 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1080 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1081 size_t max_sectors = queue_max_zone_append_sectors(q); 1082 1083 size = min(size, max_sectors << SECTOR_SHIFT); 1084 } 1085 1086 bio->bi_vcnt = iter->nr_segs; 1087 bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1088 bio->bi_iter.bi_bvec_done = iter->iov_offset; 1089 bio->bi_iter.bi_size = size; 1090 bio_set_flag(bio, BIO_NO_PAGE_REF); 1091 bio_set_flag(bio, BIO_CLONED); 1092 } 1093 1094 static void bio_put_pages(struct page **pages, size_t size, size_t off) 1095 { 1096 size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE); 1097 1098 for (i = 0; i < nr; i++) 1099 put_page(pages[i]); 1100 } 1101 1102 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1103 1104 /** 1105 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 1106 * @bio: bio to add pages to 1107 * @iter: iov iterator describing the region to be mapped 1108 * 1109 * Pins pages from *iter and appends them to @bio's bvec array. The 1110 * pages will have to be released using put_page() when done. 1111 * For multi-segment *iter, this function only adds pages from the 1112 * next non-empty segment of the iov iterator. 1113 */ 1114 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1115 { 1116 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1117 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1118 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 1119 struct page **pages = (struct page **)bv; 1120 bool same_page = false; 1121 ssize_t size, left; 1122 unsigned len, i; 1123 size_t offset; 1124 1125 /* 1126 * Move page array up in the allocated memory for the bio vecs as far as 1127 * possible so that we can start filling biovecs from the beginning 1128 * without overwriting the temporary page array. 1129 */ 1130 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1131 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1132 1133 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 1134 if (unlikely(size <= 0)) 1135 return size ? size : -EFAULT; 1136 1137 for (left = size, i = 0; left > 0; left -= len, i++) { 1138 struct page *page = pages[i]; 1139 1140 len = min_t(size_t, PAGE_SIZE - offset, left); 1141 1142 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1143 if (same_page) 1144 put_page(page); 1145 } else { 1146 if (WARN_ON_ONCE(bio_full(bio, len))) { 1147 bio_put_pages(pages + i, left, offset); 1148 return -EINVAL; 1149 } 1150 __bio_add_page(bio, page, len, offset); 1151 } 1152 offset = 0; 1153 } 1154 1155 iov_iter_advance(iter, size); 1156 return 0; 1157 } 1158 1159 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 1160 { 1161 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1162 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1163 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1164 unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 1165 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 1166 struct page **pages = (struct page **)bv; 1167 ssize_t size, left; 1168 unsigned len, i; 1169 size_t offset; 1170 int ret = 0; 1171 1172 if (WARN_ON_ONCE(!max_append_sectors)) 1173 return 0; 1174 1175 /* 1176 * Move page array up in the allocated memory for the bio vecs as far as 1177 * possible so that we can start filling biovecs from the beginning 1178 * without overwriting the temporary page array. 1179 */ 1180 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1181 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1182 1183 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 1184 if (unlikely(size <= 0)) 1185 return size ? size : -EFAULT; 1186 1187 for (left = size, i = 0; left > 0; left -= len, i++) { 1188 struct page *page = pages[i]; 1189 bool same_page = false; 1190 1191 len = min_t(size_t, PAGE_SIZE - offset, left); 1192 if (bio_add_hw_page(q, bio, page, len, offset, 1193 max_append_sectors, &same_page) != len) { 1194 bio_put_pages(pages + i, left, offset); 1195 ret = -EINVAL; 1196 break; 1197 } 1198 if (same_page) 1199 put_page(page); 1200 offset = 0; 1201 } 1202 1203 iov_iter_advance(iter, size - left); 1204 return ret; 1205 } 1206 1207 /** 1208 * bio_iov_iter_get_pages - add user or kernel pages to a bio 1209 * @bio: bio to add pages to 1210 * @iter: iov iterator describing the region to be added 1211 * 1212 * This takes either an iterator pointing to user memory, or one pointing to 1213 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 1214 * map them into the kernel. On IO completion, the caller should put those 1215 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1216 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1217 * to ensure the bvecs and pages stay referenced until the submitted I/O is 1218 * completed by a call to ->ki_complete() or returns with an error other than 1219 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1220 * on IO completion. If it isn't, then pages should be released. 1221 * 1222 * The function tries, but does not guarantee, to pin as many pages as 1223 * fit into the bio, or are requested in @iter, whatever is smaller. If 1224 * MM encounters an error pinning the requested pages, it stops. Error 1225 * is returned only if 0 pages could be pinned. 1226 * 1227 * It's intended for direct IO, so doesn't do PSI tracking, the caller is 1228 * responsible for setting BIO_WORKINGSET if necessary. 1229 */ 1230 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1231 { 1232 int ret = 0; 1233 1234 if (iov_iter_is_bvec(iter)) { 1235 bio_iov_bvec_set(bio, iter); 1236 iov_iter_advance(iter, bio->bi_iter.bi_size); 1237 return 0; 1238 } 1239 1240 do { 1241 if (bio_op(bio) == REQ_OP_ZONE_APPEND) 1242 ret = __bio_iov_append_get_pages(bio, iter); 1243 else 1244 ret = __bio_iov_iter_get_pages(bio, iter); 1245 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1246 1247 /* don't account direct I/O as memory stall */ 1248 bio_clear_flag(bio, BIO_WORKINGSET); 1249 return bio->bi_vcnt ? 0 : ret; 1250 } 1251 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 1252 1253 static void submit_bio_wait_endio(struct bio *bio) 1254 { 1255 complete(bio->bi_private); 1256 } 1257 1258 /** 1259 * submit_bio_wait - submit a bio, and wait until it completes 1260 * @bio: The &struct bio which describes the I/O 1261 * 1262 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1263 * bio_endio() on failure. 1264 * 1265 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1266 * result in bio reference to be consumed. The caller must drop the reference 1267 * on his own. 1268 */ 1269 int submit_bio_wait(struct bio *bio) 1270 { 1271 DECLARE_COMPLETION_ONSTACK_MAP(done, 1272 bio->bi_bdev->bd_disk->lockdep_map); 1273 unsigned long hang_check; 1274 1275 bio->bi_private = &done; 1276 bio->bi_end_io = submit_bio_wait_endio; 1277 bio->bi_opf |= REQ_SYNC; 1278 submit_bio(bio); 1279 1280 /* Prevent hang_check timer from firing at us during very long I/O */ 1281 hang_check = sysctl_hung_task_timeout_secs; 1282 if (hang_check) 1283 while (!wait_for_completion_io_timeout(&done, 1284 hang_check * (HZ/2))) 1285 ; 1286 else 1287 wait_for_completion_io(&done); 1288 1289 return blk_status_to_errno(bio->bi_status); 1290 } 1291 EXPORT_SYMBOL(submit_bio_wait); 1292 1293 void __bio_advance(struct bio *bio, unsigned bytes) 1294 { 1295 if (bio_integrity(bio)) 1296 bio_integrity_advance(bio, bytes); 1297 1298 bio_crypt_advance(bio, bytes); 1299 bio_advance_iter(bio, &bio->bi_iter, bytes); 1300 } 1301 EXPORT_SYMBOL(__bio_advance); 1302 1303 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1304 struct bio *src, struct bvec_iter *src_iter) 1305 { 1306 while (src_iter->bi_size && dst_iter->bi_size) { 1307 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1308 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1309 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1310 void *src_buf; 1311 1312 src_buf = bvec_kmap_local(&src_bv); 1313 memcpy_to_bvec(&dst_bv, src_buf); 1314 kunmap_local(src_buf); 1315 1316 bio_advance_iter_single(src, src_iter, bytes); 1317 bio_advance_iter_single(dst, dst_iter, bytes); 1318 } 1319 } 1320 EXPORT_SYMBOL(bio_copy_data_iter); 1321 1322 /** 1323 * bio_copy_data - copy contents of data buffers from one bio to another 1324 * @src: source bio 1325 * @dst: destination bio 1326 * 1327 * Stops when it reaches the end of either @src or @dst - that is, copies 1328 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1329 */ 1330 void bio_copy_data(struct bio *dst, struct bio *src) 1331 { 1332 struct bvec_iter src_iter = src->bi_iter; 1333 struct bvec_iter dst_iter = dst->bi_iter; 1334 1335 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1336 } 1337 EXPORT_SYMBOL(bio_copy_data); 1338 1339 void bio_free_pages(struct bio *bio) 1340 { 1341 struct bio_vec *bvec; 1342 struct bvec_iter_all iter_all; 1343 1344 bio_for_each_segment_all(bvec, bio, iter_all) 1345 __free_page(bvec->bv_page); 1346 } 1347 EXPORT_SYMBOL(bio_free_pages); 1348 1349 /* 1350 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1351 * for performing direct-IO in BIOs. 1352 * 1353 * The problem is that we cannot run set_page_dirty() from interrupt context 1354 * because the required locks are not interrupt-safe. So what we can do is to 1355 * mark the pages dirty _before_ performing IO. And in interrupt context, 1356 * check that the pages are still dirty. If so, fine. If not, redirty them 1357 * in process context. 1358 * 1359 * We special-case compound pages here: normally this means reads into hugetlb 1360 * pages. The logic in here doesn't really work right for compound pages 1361 * because the VM does not uniformly chase down the head page in all cases. 1362 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1363 * handle them at all. So we skip compound pages here at an early stage. 1364 * 1365 * Note that this code is very hard to test under normal circumstances because 1366 * direct-io pins the pages with get_user_pages(). This makes 1367 * is_page_cache_freeable return false, and the VM will not clean the pages. 1368 * But other code (eg, flusher threads) could clean the pages if they are mapped 1369 * pagecache. 1370 * 1371 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1372 * deferred bio dirtying paths. 1373 */ 1374 1375 /* 1376 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1377 */ 1378 void bio_set_pages_dirty(struct bio *bio) 1379 { 1380 struct bio_vec *bvec; 1381 struct bvec_iter_all iter_all; 1382 1383 bio_for_each_segment_all(bvec, bio, iter_all) { 1384 if (!PageCompound(bvec->bv_page)) 1385 set_page_dirty_lock(bvec->bv_page); 1386 } 1387 } 1388 1389 /* 1390 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1391 * If they are, then fine. If, however, some pages are clean then they must 1392 * have been written out during the direct-IO read. So we take another ref on 1393 * the BIO and re-dirty the pages in process context. 1394 * 1395 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1396 * here on. It will run one put_page() against each page and will run one 1397 * bio_put() against the BIO. 1398 */ 1399 1400 static void bio_dirty_fn(struct work_struct *work); 1401 1402 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1403 static DEFINE_SPINLOCK(bio_dirty_lock); 1404 static struct bio *bio_dirty_list; 1405 1406 /* 1407 * This runs in process context 1408 */ 1409 static void bio_dirty_fn(struct work_struct *work) 1410 { 1411 struct bio *bio, *next; 1412 1413 spin_lock_irq(&bio_dirty_lock); 1414 next = bio_dirty_list; 1415 bio_dirty_list = NULL; 1416 spin_unlock_irq(&bio_dirty_lock); 1417 1418 while ((bio = next) != NULL) { 1419 next = bio->bi_private; 1420 1421 bio_release_pages(bio, true); 1422 bio_put(bio); 1423 } 1424 } 1425 1426 void bio_check_pages_dirty(struct bio *bio) 1427 { 1428 struct bio_vec *bvec; 1429 unsigned long flags; 1430 struct bvec_iter_all iter_all; 1431 1432 bio_for_each_segment_all(bvec, bio, iter_all) { 1433 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1434 goto defer; 1435 } 1436 1437 bio_release_pages(bio, false); 1438 bio_put(bio); 1439 return; 1440 defer: 1441 spin_lock_irqsave(&bio_dirty_lock, flags); 1442 bio->bi_private = bio_dirty_list; 1443 bio_dirty_list = bio; 1444 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1445 schedule_work(&bio_dirty_work); 1446 } 1447 1448 static inline bool bio_remaining_done(struct bio *bio) 1449 { 1450 /* 1451 * If we're not chaining, then ->__bi_remaining is always 1 and 1452 * we always end io on the first invocation. 1453 */ 1454 if (!bio_flagged(bio, BIO_CHAIN)) 1455 return true; 1456 1457 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1458 1459 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1460 bio_clear_flag(bio, BIO_CHAIN); 1461 return true; 1462 } 1463 1464 return false; 1465 } 1466 1467 /** 1468 * bio_endio - end I/O on a bio 1469 * @bio: bio 1470 * 1471 * Description: 1472 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1473 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1474 * bio unless they own it and thus know that it has an end_io function. 1475 * 1476 * bio_endio() can be called several times on a bio that has been chained 1477 * using bio_chain(). The ->bi_end_io() function will only be called the 1478 * last time. 1479 **/ 1480 void bio_endio(struct bio *bio) 1481 { 1482 again: 1483 if (!bio_remaining_done(bio)) 1484 return; 1485 if (!bio_integrity_endio(bio)) 1486 return; 1487 1488 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED)) 1489 rq_qos_done_bio(bdev_get_queue(bio->bi_bdev), bio); 1490 1491 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1492 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 1493 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1494 } 1495 1496 /* 1497 * Need to have a real endio function for chained bios, otherwise 1498 * various corner cases will break (like stacking block devices that 1499 * save/restore bi_end_io) - however, we want to avoid unbounded 1500 * recursion and blowing the stack. Tail call optimization would 1501 * handle this, but compiling with frame pointers also disables 1502 * gcc's sibling call optimization. 1503 */ 1504 if (bio->bi_end_io == bio_chain_endio) { 1505 bio = __bio_chain_endio(bio); 1506 goto again; 1507 } 1508 1509 blk_throtl_bio_endio(bio); 1510 /* release cgroup info */ 1511 bio_uninit(bio); 1512 if (bio->bi_end_io) 1513 bio->bi_end_io(bio); 1514 } 1515 EXPORT_SYMBOL(bio_endio); 1516 1517 /** 1518 * bio_split - split a bio 1519 * @bio: bio to split 1520 * @sectors: number of sectors to split from the front of @bio 1521 * @gfp: gfp mask 1522 * @bs: bio set to allocate from 1523 * 1524 * Allocates and returns a new bio which represents @sectors from the start of 1525 * @bio, and updates @bio to represent the remaining sectors. 1526 * 1527 * Unless this is a discard request the newly allocated bio will point 1528 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1529 * neither @bio nor @bs are freed before the split bio. 1530 */ 1531 struct bio *bio_split(struct bio *bio, int sectors, 1532 gfp_t gfp, struct bio_set *bs) 1533 { 1534 struct bio *split; 1535 1536 BUG_ON(sectors <= 0); 1537 BUG_ON(sectors >= bio_sectors(bio)); 1538 1539 /* Zone append commands cannot be split */ 1540 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 1541 return NULL; 1542 1543 split = bio_clone_fast(bio, gfp, bs); 1544 if (!split) 1545 return NULL; 1546 1547 split->bi_iter.bi_size = sectors << 9; 1548 1549 if (bio_integrity(split)) 1550 bio_integrity_trim(split); 1551 1552 bio_advance(bio, split->bi_iter.bi_size); 1553 1554 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1555 bio_set_flag(split, BIO_TRACE_COMPLETION); 1556 1557 return split; 1558 } 1559 EXPORT_SYMBOL(bio_split); 1560 1561 /** 1562 * bio_trim - trim a bio 1563 * @bio: bio to trim 1564 * @offset: number of sectors to trim from the front of @bio 1565 * @size: size we want to trim @bio to, in sectors 1566 * 1567 * This function is typically used for bios that are cloned and submitted 1568 * to the underlying device in parts. 1569 */ 1570 void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1571 { 1572 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 1573 offset + size > bio->bi_iter.bi_size)) 1574 return; 1575 1576 size <<= 9; 1577 if (offset == 0 && size == bio->bi_iter.bi_size) 1578 return; 1579 1580 bio_advance(bio, offset << 9); 1581 bio->bi_iter.bi_size = size; 1582 1583 if (bio_integrity(bio)) 1584 bio_integrity_trim(bio); 1585 } 1586 EXPORT_SYMBOL_GPL(bio_trim); 1587 1588 /* 1589 * create memory pools for biovec's in a bio_set. 1590 * use the global biovec slabs created for general use. 1591 */ 1592 int biovec_init_pool(mempool_t *pool, int pool_entries) 1593 { 1594 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1595 1596 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1597 } 1598 1599 /* 1600 * bioset_exit - exit a bioset initialized with bioset_init() 1601 * 1602 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1603 * kzalloc()). 1604 */ 1605 void bioset_exit(struct bio_set *bs) 1606 { 1607 bio_alloc_cache_destroy(bs); 1608 if (bs->rescue_workqueue) 1609 destroy_workqueue(bs->rescue_workqueue); 1610 bs->rescue_workqueue = NULL; 1611 1612 mempool_exit(&bs->bio_pool); 1613 mempool_exit(&bs->bvec_pool); 1614 1615 bioset_integrity_free(bs); 1616 if (bs->bio_slab) 1617 bio_put_slab(bs); 1618 bs->bio_slab = NULL; 1619 } 1620 EXPORT_SYMBOL(bioset_exit); 1621 1622 /** 1623 * bioset_init - Initialize a bio_set 1624 * @bs: pool to initialize 1625 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1626 * @front_pad: Number of bytes to allocate in front of the returned bio 1627 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1628 * and %BIOSET_NEED_RESCUER 1629 * 1630 * Description: 1631 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1632 * to ask for a number of bytes to be allocated in front of the bio. 1633 * Front pad allocation is useful for embedding the bio inside 1634 * another structure, to avoid allocating extra data to go with the bio. 1635 * Note that the bio must be embedded at the END of that structure always, 1636 * or things will break badly. 1637 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1638 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1639 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1640 * dispatch queued requests when the mempool runs out of space. 1641 * 1642 */ 1643 int bioset_init(struct bio_set *bs, 1644 unsigned int pool_size, 1645 unsigned int front_pad, 1646 int flags) 1647 { 1648 bs->front_pad = front_pad; 1649 if (flags & BIOSET_NEED_BVECS) 1650 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1651 else 1652 bs->back_pad = 0; 1653 1654 spin_lock_init(&bs->rescue_lock); 1655 bio_list_init(&bs->rescue_list); 1656 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1657 1658 bs->bio_slab = bio_find_or_create_slab(bs); 1659 if (!bs->bio_slab) 1660 return -ENOMEM; 1661 1662 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1663 goto bad; 1664 1665 if ((flags & BIOSET_NEED_BVECS) && 1666 biovec_init_pool(&bs->bvec_pool, pool_size)) 1667 goto bad; 1668 1669 if (flags & BIOSET_NEED_RESCUER) { 1670 bs->rescue_workqueue = alloc_workqueue("bioset", 1671 WQ_MEM_RECLAIM, 0); 1672 if (!bs->rescue_workqueue) 1673 goto bad; 1674 } 1675 if (flags & BIOSET_PERCPU_CACHE) { 1676 bs->cache = alloc_percpu(struct bio_alloc_cache); 1677 if (!bs->cache) 1678 goto bad; 1679 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1680 } 1681 1682 return 0; 1683 bad: 1684 bioset_exit(bs); 1685 return -ENOMEM; 1686 } 1687 EXPORT_SYMBOL(bioset_init); 1688 1689 /* 1690 * Initialize and setup a new bio_set, based on the settings from 1691 * another bio_set. 1692 */ 1693 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 1694 { 1695 int flags; 1696 1697 flags = 0; 1698 if (src->bvec_pool.min_nr) 1699 flags |= BIOSET_NEED_BVECS; 1700 if (src->rescue_workqueue) 1701 flags |= BIOSET_NEED_RESCUER; 1702 1703 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 1704 } 1705 EXPORT_SYMBOL(bioset_init_from_src); 1706 1707 /** 1708 * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb 1709 * @kiocb: kiocb describing the IO 1710 * @nr_vecs: number of iovecs to pre-allocate 1711 * @bs: bio_set to allocate from 1712 * 1713 * Description: 1714 * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only 1715 * used to check if we should dip into the per-cpu bio_set allocation 1716 * cache. The allocation uses GFP_KERNEL internally. On return, the 1717 * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio 1718 * MUST be done from process context, not hard/soft IRQ. 1719 * 1720 */ 1721 struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs, 1722 struct bio_set *bs) 1723 { 1724 struct bio_alloc_cache *cache; 1725 struct bio *bio; 1726 1727 if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS) 1728 return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1729 1730 cache = per_cpu_ptr(bs->cache, get_cpu()); 1731 if (cache->free_list) { 1732 bio = cache->free_list; 1733 cache->free_list = bio->bi_next; 1734 cache->nr--; 1735 put_cpu(); 1736 bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs); 1737 bio->bi_pool = bs; 1738 bio_set_flag(bio, BIO_PERCPU_CACHE); 1739 return bio; 1740 } 1741 put_cpu(); 1742 bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1743 bio_set_flag(bio, BIO_PERCPU_CACHE); 1744 return bio; 1745 } 1746 EXPORT_SYMBOL_GPL(bio_alloc_kiocb); 1747 1748 static int __init init_bio(void) 1749 { 1750 int i; 1751 1752 bio_integrity_init(); 1753 1754 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1755 struct biovec_slab *bvs = bvec_slabs + i; 1756 1757 bvs->slab = kmem_cache_create(bvs->name, 1758 bvs->nr_vecs * sizeof(struct bio_vec), 0, 1759 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1760 } 1761 1762 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1763 bio_cpu_dead); 1764 1765 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1766 panic("bio: can't allocate bios\n"); 1767 1768 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1769 panic("bio: can't create integrity pool\n"); 1770 1771 return 0; 1772 } 1773 subsys_initcall(init_bio); 1774