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