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 bs->cache = NULL; 669 } 670 671 /** 672 * bio_put - release a reference to a bio 673 * @bio: bio to release reference to 674 * 675 * Description: 676 * Put a reference to a &struct bio, either one you have gotten with 677 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 678 **/ 679 void bio_put(struct bio *bio) 680 { 681 if (unlikely(bio_flagged(bio, BIO_REFFED))) { 682 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 683 if (!atomic_dec_and_test(&bio->__bi_cnt)) 684 return; 685 } 686 687 if (bio_flagged(bio, BIO_PERCPU_CACHE)) { 688 struct bio_alloc_cache *cache; 689 690 bio_uninit(bio); 691 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 692 bio_list_add_head(&cache->free_list, bio); 693 if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 694 bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 695 put_cpu(); 696 } else { 697 bio_free(bio); 698 } 699 } 700 EXPORT_SYMBOL(bio_put); 701 702 /** 703 * __bio_clone_fast - clone a bio that shares the original bio's biovec 704 * @bio: destination bio 705 * @bio_src: bio to clone 706 * 707 * Clone a &bio. Caller will own the returned bio, but not 708 * the actual data it points to. Reference count of returned 709 * bio will be one. 710 * 711 * Caller must ensure that @bio_src is not freed before @bio. 712 */ 713 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 714 { 715 WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs); 716 717 /* 718 * most users will be overriding ->bi_bdev with a new target, 719 * so we don't set nor calculate new physical/hw segment counts here 720 */ 721 bio->bi_bdev = bio_src->bi_bdev; 722 bio_set_flag(bio, BIO_CLONED); 723 if (bio_flagged(bio_src, BIO_THROTTLED)) 724 bio_set_flag(bio, BIO_THROTTLED); 725 if (bio_flagged(bio_src, BIO_REMAPPED)) 726 bio_set_flag(bio, BIO_REMAPPED); 727 bio->bi_opf = bio_src->bi_opf; 728 bio->bi_ioprio = bio_src->bi_ioprio; 729 bio->bi_write_hint = bio_src->bi_write_hint; 730 bio->bi_iter = bio_src->bi_iter; 731 bio->bi_io_vec = bio_src->bi_io_vec; 732 733 bio_clone_blkg_association(bio, bio_src); 734 blkcg_bio_issue_init(bio); 735 } 736 EXPORT_SYMBOL(__bio_clone_fast); 737 738 /** 739 * bio_clone_fast - clone a bio that shares the original bio's biovec 740 * @bio: bio to clone 741 * @gfp_mask: allocation priority 742 * @bs: bio_set to allocate from 743 * 744 * Like __bio_clone_fast, only also allocates the returned bio 745 */ 746 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 747 { 748 struct bio *b; 749 750 b = bio_alloc_bioset(gfp_mask, 0, bs); 751 if (!b) 752 return NULL; 753 754 __bio_clone_fast(b, bio); 755 756 if (bio_crypt_clone(b, bio, gfp_mask) < 0) 757 goto err_put; 758 759 if (bio_integrity(bio) && 760 bio_integrity_clone(b, bio, gfp_mask) < 0) 761 goto err_put; 762 763 return b; 764 765 err_put: 766 bio_put(b); 767 return NULL; 768 } 769 EXPORT_SYMBOL(bio_clone_fast); 770 771 const char *bio_devname(struct bio *bio, char *buf) 772 { 773 return bdevname(bio->bi_bdev, buf); 774 } 775 EXPORT_SYMBOL(bio_devname); 776 777 static inline bool page_is_mergeable(const struct bio_vec *bv, 778 struct page *page, unsigned int len, unsigned int off, 779 bool *same_page) 780 { 781 size_t bv_end = bv->bv_offset + bv->bv_len; 782 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 783 phys_addr_t page_addr = page_to_phys(page); 784 785 if (vec_end_addr + 1 != page_addr + off) 786 return false; 787 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 788 return false; 789 790 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 791 if (*same_page) 792 return true; 793 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 794 } 795 796 /* 797 * Try to merge a page into a segment, while obeying the hardware segment 798 * size limit. This is not for normal read/write bios, but for passthrough 799 * or Zone Append operations that we can't split. 800 */ 801 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 802 struct page *page, unsigned len, 803 unsigned offset, bool *same_page) 804 { 805 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 806 unsigned long mask = queue_segment_boundary(q); 807 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 808 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 809 810 if ((addr1 | mask) != (addr2 | mask)) 811 return false; 812 if (bv->bv_len + len > queue_max_segment_size(q)) 813 return false; 814 return __bio_try_merge_page(bio, page, len, offset, same_page); 815 } 816 817 /** 818 * bio_add_hw_page - attempt to add a page to a bio with hw constraints 819 * @q: the target queue 820 * @bio: destination bio 821 * @page: page to add 822 * @len: vec entry length 823 * @offset: vec entry offset 824 * @max_sectors: maximum number of sectors that can be added 825 * @same_page: return if the segment has been merged inside the same page 826 * 827 * Add a page to a bio while respecting the hardware max_sectors, max_segment 828 * and gap limitations. 829 */ 830 int bio_add_hw_page(struct request_queue *q, struct bio *bio, 831 struct page *page, unsigned int len, unsigned int offset, 832 unsigned int max_sectors, bool *same_page) 833 { 834 struct bio_vec *bvec; 835 836 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 837 return 0; 838 839 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 840 return 0; 841 842 if (bio->bi_vcnt > 0) { 843 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 844 return len; 845 846 /* 847 * If the queue doesn't support SG gaps and adding this segment 848 * would create a gap, disallow it. 849 */ 850 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 851 if (bvec_gap_to_prev(q, bvec, offset)) 852 return 0; 853 } 854 855 if (bio_full(bio, len)) 856 return 0; 857 858 if (bio->bi_vcnt >= queue_max_segments(q)) 859 return 0; 860 861 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 862 bvec->bv_page = page; 863 bvec->bv_len = len; 864 bvec->bv_offset = offset; 865 bio->bi_vcnt++; 866 bio->bi_iter.bi_size += len; 867 return len; 868 } 869 870 /** 871 * bio_add_pc_page - attempt to add page to passthrough bio 872 * @q: the target queue 873 * @bio: destination bio 874 * @page: page to add 875 * @len: vec entry length 876 * @offset: vec entry offset 877 * 878 * Attempt to add a page to the bio_vec maplist. This can fail for a 879 * number of reasons, such as the bio being full or target block device 880 * limitations. The target block device must allow bio's up to PAGE_SIZE, 881 * so it is always possible to add a single page to an empty bio. 882 * 883 * This should only be used by passthrough bios. 884 */ 885 int bio_add_pc_page(struct request_queue *q, struct bio *bio, 886 struct page *page, unsigned int len, unsigned int offset) 887 { 888 bool same_page = false; 889 return bio_add_hw_page(q, bio, page, len, offset, 890 queue_max_hw_sectors(q), &same_page); 891 } 892 EXPORT_SYMBOL(bio_add_pc_page); 893 894 /** 895 * bio_add_zone_append_page - attempt to add page to zone-append bio 896 * @bio: destination bio 897 * @page: page to add 898 * @len: vec entry length 899 * @offset: vec entry offset 900 * 901 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 902 * for a zone-append request. This can fail for a number of reasons, such as the 903 * bio being full or the target block device is not a zoned block device or 904 * other limitations of the target block device. The target block device must 905 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 906 * to an empty bio. 907 * 908 * Returns: number of bytes added to the bio, or 0 in case of a failure. 909 */ 910 int bio_add_zone_append_page(struct bio *bio, struct page *page, 911 unsigned int len, unsigned int offset) 912 { 913 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 914 bool same_page = false; 915 916 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 917 return 0; 918 919 if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 920 return 0; 921 922 return bio_add_hw_page(q, bio, page, len, offset, 923 queue_max_zone_append_sectors(q), &same_page); 924 } 925 EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 926 927 /** 928 * __bio_try_merge_page - try appending data to an existing bvec. 929 * @bio: destination bio 930 * @page: start page to add 931 * @len: length of the data to add 932 * @off: offset of the data relative to @page 933 * @same_page: return if the segment has been merged inside the same page 934 * 935 * Try to add the data at @page + @off to the last bvec of @bio. This is a 936 * useful optimisation for file systems with a block size smaller than the 937 * page size. 938 * 939 * Warn if (@len, @off) crosses pages in case that @same_page is true. 940 * 941 * Return %true on success or %false on failure. 942 */ 943 bool __bio_try_merge_page(struct bio *bio, struct page *page, 944 unsigned int len, unsigned int off, bool *same_page) 945 { 946 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 947 return false; 948 949 if (bio->bi_vcnt > 0) { 950 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 951 952 if (page_is_mergeable(bv, page, len, off, same_page)) { 953 if (bio->bi_iter.bi_size > UINT_MAX - len) { 954 *same_page = false; 955 return false; 956 } 957 bv->bv_len += len; 958 bio->bi_iter.bi_size += len; 959 return true; 960 } 961 } 962 return false; 963 } 964 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 965 966 /** 967 * __bio_add_page - add page(s) to a bio in a new segment 968 * @bio: destination bio 969 * @page: start page to add 970 * @len: length of the data to add, may cross pages 971 * @off: offset of the data relative to @page, may cross pages 972 * 973 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 974 * that @bio has space for another bvec. 975 */ 976 void __bio_add_page(struct bio *bio, struct page *page, 977 unsigned int len, unsigned int off) 978 { 979 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 980 981 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 982 WARN_ON_ONCE(bio_full(bio, len)); 983 984 bv->bv_page = page; 985 bv->bv_offset = off; 986 bv->bv_len = len; 987 988 bio->bi_iter.bi_size += len; 989 bio->bi_vcnt++; 990 991 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 992 bio_set_flag(bio, BIO_WORKINGSET); 993 } 994 EXPORT_SYMBOL_GPL(__bio_add_page); 995 996 /** 997 * bio_add_page - attempt to add page(s) to bio 998 * @bio: destination bio 999 * @page: start page to add 1000 * @len: vec entry length, may cross pages 1001 * @offset: vec entry offset relative to @page, may cross pages 1002 * 1003 * Attempt to add page(s) to the bio_vec maplist. This will only fail 1004 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1005 */ 1006 int bio_add_page(struct bio *bio, struct page *page, 1007 unsigned int len, unsigned int offset) 1008 { 1009 bool same_page = false; 1010 1011 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1012 if (bio_full(bio, len)) 1013 return 0; 1014 __bio_add_page(bio, page, len, offset); 1015 } 1016 return len; 1017 } 1018 EXPORT_SYMBOL(bio_add_page); 1019 1020 void bio_release_pages(struct bio *bio, bool mark_dirty) 1021 { 1022 struct bvec_iter_all iter_all; 1023 struct bio_vec *bvec; 1024 1025 if (bio_flagged(bio, BIO_NO_PAGE_REF)) 1026 return; 1027 1028 bio_for_each_segment_all(bvec, bio, iter_all) { 1029 if (mark_dirty && !PageCompound(bvec->bv_page)) 1030 set_page_dirty_lock(bvec->bv_page); 1031 put_page(bvec->bv_page); 1032 } 1033 } 1034 EXPORT_SYMBOL_GPL(bio_release_pages); 1035 1036 static void __bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 1037 { 1038 WARN_ON_ONCE(bio->bi_max_vecs); 1039 1040 bio->bi_vcnt = iter->nr_segs; 1041 bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1042 bio->bi_iter.bi_bvec_done = iter->iov_offset; 1043 bio->bi_iter.bi_size = iter->count; 1044 bio_set_flag(bio, BIO_NO_PAGE_REF); 1045 bio_set_flag(bio, BIO_CLONED); 1046 } 1047 1048 static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 1049 { 1050 __bio_iov_bvec_set(bio, iter); 1051 iov_iter_advance(iter, iter->count); 1052 return 0; 1053 } 1054 1055 static int bio_iov_bvec_set_append(struct bio *bio, struct iov_iter *iter) 1056 { 1057 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1058 struct iov_iter i = *iter; 1059 1060 iov_iter_truncate(&i, queue_max_zone_append_sectors(q) << 9); 1061 __bio_iov_bvec_set(bio, &i); 1062 iov_iter_advance(iter, i.count); 1063 return 0; 1064 } 1065 1066 static void bio_put_pages(struct page **pages, size_t size, size_t off) 1067 { 1068 size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE); 1069 1070 for (i = 0; i < nr; i++) 1071 put_page(pages[i]); 1072 } 1073 1074 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1075 1076 /** 1077 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 1078 * @bio: bio to add pages to 1079 * @iter: iov iterator describing the region to be mapped 1080 * 1081 * Pins pages from *iter and appends them to @bio's bvec array. The 1082 * pages will have to be released using put_page() when done. 1083 * For multi-segment *iter, this function only adds pages from the 1084 * next non-empty segment of the iov iterator. 1085 */ 1086 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1087 { 1088 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1089 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1090 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 1091 struct page **pages = (struct page **)bv; 1092 bool same_page = false; 1093 ssize_t size, left; 1094 unsigned len, i; 1095 size_t offset; 1096 1097 /* 1098 * Move page array up in the allocated memory for the bio vecs as far as 1099 * possible so that we can start filling biovecs from the beginning 1100 * without overwriting the temporary page array. 1101 */ 1102 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1103 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1104 1105 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 1106 if (unlikely(size <= 0)) 1107 return size ? size : -EFAULT; 1108 1109 for (left = size, i = 0; left > 0; left -= len, i++) { 1110 struct page *page = pages[i]; 1111 1112 len = min_t(size_t, PAGE_SIZE - offset, left); 1113 1114 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1115 if (same_page) 1116 put_page(page); 1117 } else { 1118 if (WARN_ON_ONCE(bio_full(bio, len))) { 1119 bio_put_pages(pages + i, left, offset); 1120 return -EINVAL; 1121 } 1122 __bio_add_page(bio, page, len, offset); 1123 } 1124 offset = 0; 1125 } 1126 1127 iov_iter_advance(iter, size); 1128 return 0; 1129 } 1130 1131 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 1132 { 1133 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1134 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1135 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1136 unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 1137 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 1138 struct page **pages = (struct page **)bv; 1139 ssize_t size, left; 1140 unsigned len, i; 1141 size_t offset; 1142 int ret = 0; 1143 1144 if (WARN_ON_ONCE(!max_append_sectors)) 1145 return 0; 1146 1147 /* 1148 * Move page array up in the allocated memory for the bio vecs as far as 1149 * possible so that we can start filling biovecs from the beginning 1150 * without overwriting the temporary page array. 1151 */ 1152 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1153 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1154 1155 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 1156 if (unlikely(size <= 0)) 1157 return size ? size : -EFAULT; 1158 1159 for (left = size, i = 0; left > 0; left -= len, i++) { 1160 struct page *page = pages[i]; 1161 bool same_page = false; 1162 1163 len = min_t(size_t, PAGE_SIZE - offset, left); 1164 if (bio_add_hw_page(q, bio, page, len, offset, 1165 max_append_sectors, &same_page) != len) { 1166 bio_put_pages(pages + i, left, offset); 1167 ret = -EINVAL; 1168 break; 1169 } 1170 if (same_page) 1171 put_page(page); 1172 offset = 0; 1173 } 1174 1175 iov_iter_advance(iter, size - left); 1176 return ret; 1177 } 1178 1179 /** 1180 * bio_iov_iter_get_pages - add user or kernel pages to a bio 1181 * @bio: bio to add pages to 1182 * @iter: iov iterator describing the region to be added 1183 * 1184 * This takes either an iterator pointing to user memory, or one pointing to 1185 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 1186 * map them into the kernel. On IO completion, the caller should put those 1187 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1188 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1189 * to ensure the bvecs and pages stay referenced until the submitted I/O is 1190 * completed by a call to ->ki_complete() or returns with an error other than 1191 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1192 * on IO completion. If it isn't, then pages should be released. 1193 * 1194 * The function tries, but does not guarantee, to pin as many pages as 1195 * fit into the bio, or are requested in @iter, whatever is smaller. If 1196 * MM encounters an error pinning the requested pages, it stops. Error 1197 * is returned only if 0 pages could be pinned. 1198 * 1199 * It's intended for direct IO, so doesn't do PSI tracking, the caller is 1200 * responsible for setting BIO_WORKINGSET if necessary. 1201 */ 1202 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1203 { 1204 int ret = 0; 1205 1206 if (iov_iter_is_bvec(iter)) { 1207 if (bio_op(bio) == REQ_OP_ZONE_APPEND) 1208 return bio_iov_bvec_set_append(bio, iter); 1209 return bio_iov_bvec_set(bio, iter); 1210 } 1211 1212 do { 1213 if (bio_op(bio) == REQ_OP_ZONE_APPEND) 1214 ret = __bio_iov_append_get_pages(bio, iter); 1215 else 1216 ret = __bio_iov_iter_get_pages(bio, iter); 1217 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1218 1219 /* don't account direct I/O as memory stall */ 1220 bio_clear_flag(bio, BIO_WORKINGSET); 1221 return bio->bi_vcnt ? 0 : ret; 1222 } 1223 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 1224 1225 static void submit_bio_wait_endio(struct bio *bio) 1226 { 1227 complete(bio->bi_private); 1228 } 1229 1230 /** 1231 * submit_bio_wait - submit a bio, and wait until it completes 1232 * @bio: The &struct bio which describes the I/O 1233 * 1234 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1235 * bio_endio() on failure. 1236 * 1237 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1238 * result in bio reference to be consumed. The caller must drop the reference 1239 * on his own. 1240 */ 1241 int submit_bio_wait(struct bio *bio) 1242 { 1243 DECLARE_COMPLETION_ONSTACK_MAP(done, 1244 bio->bi_bdev->bd_disk->lockdep_map); 1245 unsigned long hang_check; 1246 1247 bio->bi_private = &done; 1248 bio->bi_end_io = submit_bio_wait_endio; 1249 bio->bi_opf |= REQ_SYNC; 1250 submit_bio(bio); 1251 1252 /* Prevent hang_check timer from firing at us during very long I/O */ 1253 hang_check = sysctl_hung_task_timeout_secs; 1254 if (hang_check) 1255 while (!wait_for_completion_io_timeout(&done, 1256 hang_check * (HZ/2))) 1257 ; 1258 else 1259 wait_for_completion_io(&done); 1260 1261 return blk_status_to_errno(bio->bi_status); 1262 } 1263 EXPORT_SYMBOL(submit_bio_wait); 1264 1265 /** 1266 * bio_advance - increment/complete a bio by some number of bytes 1267 * @bio: bio to advance 1268 * @bytes: number of bytes to complete 1269 * 1270 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1271 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1272 * be updated on the last bvec as well. 1273 * 1274 * @bio will then represent the remaining, uncompleted portion of the io. 1275 */ 1276 void bio_advance(struct bio *bio, unsigned bytes) 1277 { 1278 if (bio_integrity(bio)) 1279 bio_integrity_advance(bio, bytes); 1280 1281 bio_crypt_advance(bio, bytes); 1282 bio_advance_iter(bio, &bio->bi_iter, bytes); 1283 } 1284 EXPORT_SYMBOL(bio_advance); 1285 1286 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1287 struct bio *src, struct bvec_iter *src_iter) 1288 { 1289 while (src_iter->bi_size && dst_iter->bi_size) { 1290 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1291 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1292 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1293 void *src_buf = bvec_kmap_local(&src_bv); 1294 void *dst_buf = bvec_kmap_local(&dst_bv); 1295 1296 memcpy(dst_buf, src_buf, bytes); 1297 1298 kunmap_local(dst_buf); 1299 kunmap_local(src_buf); 1300 1301 bio_advance_iter_single(src, src_iter, bytes); 1302 bio_advance_iter_single(dst, dst_iter, bytes); 1303 } 1304 } 1305 EXPORT_SYMBOL(bio_copy_data_iter); 1306 1307 /** 1308 * bio_copy_data - copy contents of data buffers from one bio to another 1309 * @src: source bio 1310 * @dst: destination bio 1311 * 1312 * Stops when it reaches the end of either @src or @dst - that is, copies 1313 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1314 */ 1315 void bio_copy_data(struct bio *dst, struct bio *src) 1316 { 1317 struct bvec_iter src_iter = src->bi_iter; 1318 struct bvec_iter dst_iter = dst->bi_iter; 1319 1320 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1321 } 1322 EXPORT_SYMBOL(bio_copy_data); 1323 1324 void bio_free_pages(struct bio *bio) 1325 { 1326 struct bio_vec *bvec; 1327 struct bvec_iter_all iter_all; 1328 1329 bio_for_each_segment_all(bvec, bio, iter_all) 1330 __free_page(bvec->bv_page); 1331 } 1332 EXPORT_SYMBOL(bio_free_pages); 1333 1334 /* 1335 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1336 * for performing direct-IO in BIOs. 1337 * 1338 * The problem is that we cannot run set_page_dirty() from interrupt context 1339 * because the required locks are not interrupt-safe. So what we can do is to 1340 * mark the pages dirty _before_ performing IO. And in interrupt context, 1341 * check that the pages are still dirty. If so, fine. If not, redirty them 1342 * in process context. 1343 * 1344 * We special-case compound pages here: normally this means reads into hugetlb 1345 * pages. The logic in here doesn't really work right for compound pages 1346 * because the VM does not uniformly chase down the head page in all cases. 1347 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1348 * handle them at all. So we skip compound pages here at an early stage. 1349 * 1350 * Note that this code is very hard to test under normal circumstances because 1351 * direct-io pins the pages with get_user_pages(). This makes 1352 * is_page_cache_freeable return false, and the VM will not clean the pages. 1353 * But other code (eg, flusher threads) could clean the pages if they are mapped 1354 * pagecache. 1355 * 1356 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1357 * deferred bio dirtying paths. 1358 */ 1359 1360 /* 1361 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1362 */ 1363 void bio_set_pages_dirty(struct bio *bio) 1364 { 1365 struct bio_vec *bvec; 1366 struct bvec_iter_all iter_all; 1367 1368 bio_for_each_segment_all(bvec, bio, iter_all) { 1369 if (!PageCompound(bvec->bv_page)) 1370 set_page_dirty_lock(bvec->bv_page); 1371 } 1372 } 1373 1374 /* 1375 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1376 * If they are, then fine. If, however, some pages are clean then they must 1377 * have been written out during the direct-IO read. So we take another ref on 1378 * the BIO and re-dirty the pages in process context. 1379 * 1380 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1381 * here on. It will run one put_page() against each page and will run one 1382 * bio_put() against the BIO. 1383 */ 1384 1385 static void bio_dirty_fn(struct work_struct *work); 1386 1387 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1388 static DEFINE_SPINLOCK(bio_dirty_lock); 1389 static struct bio *bio_dirty_list; 1390 1391 /* 1392 * This runs in process context 1393 */ 1394 static void bio_dirty_fn(struct work_struct *work) 1395 { 1396 struct bio *bio, *next; 1397 1398 spin_lock_irq(&bio_dirty_lock); 1399 next = bio_dirty_list; 1400 bio_dirty_list = NULL; 1401 spin_unlock_irq(&bio_dirty_lock); 1402 1403 while ((bio = next) != NULL) { 1404 next = bio->bi_private; 1405 1406 bio_release_pages(bio, true); 1407 bio_put(bio); 1408 } 1409 } 1410 1411 void bio_check_pages_dirty(struct bio *bio) 1412 { 1413 struct bio_vec *bvec; 1414 unsigned long flags; 1415 struct bvec_iter_all iter_all; 1416 1417 bio_for_each_segment_all(bvec, bio, iter_all) { 1418 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1419 goto defer; 1420 } 1421 1422 bio_release_pages(bio, false); 1423 bio_put(bio); 1424 return; 1425 defer: 1426 spin_lock_irqsave(&bio_dirty_lock, flags); 1427 bio->bi_private = bio_dirty_list; 1428 bio_dirty_list = bio; 1429 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1430 schedule_work(&bio_dirty_work); 1431 } 1432 1433 static inline bool bio_remaining_done(struct bio *bio) 1434 { 1435 /* 1436 * If we're not chaining, then ->__bi_remaining is always 1 and 1437 * we always end io on the first invocation. 1438 */ 1439 if (!bio_flagged(bio, BIO_CHAIN)) 1440 return true; 1441 1442 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1443 1444 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1445 bio_clear_flag(bio, BIO_CHAIN); 1446 return true; 1447 } 1448 1449 return false; 1450 } 1451 1452 /** 1453 * bio_endio - end I/O on a bio 1454 * @bio: bio 1455 * 1456 * Description: 1457 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1458 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1459 * bio unless they own it and thus know that it has an end_io function. 1460 * 1461 * bio_endio() can be called several times on a bio that has been chained 1462 * using bio_chain(). The ->bi_end_io() function will only be called the 1463 * last time. 1464 **/ 1465 void bio_endio(struct bio *bio) 1466 { 1467 again: 1468 if (!bio_remaining_done(bio)) 1469 return; 1470 if (!bio_integrity_endio(bio)) 1471 return; 1472 1473 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED)) 1474 rq_qos_done_bio(bdev_get_queue(bio->bi_bdev), bio); 1475 1476 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1477 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 1478 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1479 } 1480 1481 /* 1482 * Need to have a real endio function for chained bios, otherwise 1483 * various corner cases will break (like stacking block devices that 1484 * save/restore bi_end_io) - however, we want to avoid unbounded 1485 * recursion and blowing the stack. Tail call optimization would 1486 * handle this, but compiling with frame pointers also disables 1487 * gcc's sibling call optimization. 1488 */ 1489 if (bio->bi_end_io == bio_chain_endio) { 1490 bio = __bio_chain_endio(bio); 1491 goto again; 1492 } 1493 1494 blk_throtl_bio_endio(bio); 1495 /* release cgroup info */ 1496 bio_uninit(bio); 1497 if (bio->bi_end_io) 1498 bio->bi_end_io(bio); 1499 } 1500 EXPORT_SYMBOL(bio_endio); 1501 1502 /** 1503 * bio_split - split a bio 1504 * @bio: bio to split 1505 * @sectors: number of sectors to split from the front of @bio 1506 * @gfp: gfp mask 1507 * @bs: bio set to allocate from 1508 * 1509 * Allocates and returns a new bio which represents @sectors from the start of 1510 * @bio, and updates @bio to represent the remaining sectors. 1511 * 1512 * Unless this is a discard request the newly allocated bio will point 1513 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1514 * neither @bio nor @bs are freed before the split bio. 1515 */ 1516 struct bio *bio_split(struct bio *bio, int sectors, 1517 gfp_t gfp, struct bio_set *bs) 1518 { 1519 struct bio *split; 1520 1521 BUG_ON(sectors <= 0); 1522 BUG_ON(sectors >= bio_sectors(bio)); 1523 1524 /* Zone append commands cannot be split */ 1525 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 1526 return NULL; 1527 1528 split = bio_clone_fast(bio, gfp, bs); 1529 if (!split) 1530 return NULL; 1531 1532 split->bi_iter.bi_size = sectors << 9; 1533 1534 if (bio_integrity(split)) 1535 bio_integrity_trim(split); 1536 1537 bio_advance(bio, split->bi_iter.bi_size); 1538 1539 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1540 bio_set_flag(split, BIO_TRACE_COMPLETION); 1541 1542 return split; 1543 } 1544 EXPORT_SYMBOL(bio_split); 1545 1546 /** 1547 * bio_trim - trim a bio 1548 * @bio: bio to trim 1549 * @offset: number of sectors to trim from the front of @bio 1550 * @size: size we want to trim @bio to, in sectors 1551 * 1552 * This function is typically used for bios that are cloned and submitted 1553 * to the underlying device in parts. 1554 */ 1555 void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1556 { 1557 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 1558 offset + size > bio_sectors(bio))) 1559 return; 1560 1561 size <<= 9; 1562 if (offset == 0 && size == bio->bi_iter.bi_size) 1563 return; 1564 1565 bio_advance(bio, offset << 9); 1566 bio->bi_iter.bi_size = size; 1567 1568 if (bio_integrity(bio)) 1569 bio_integrity_trim(bio); 1570 } 1571 EXPORT_SYMBOL_GPL(bio_trim); 1572 1573 /* 1574 * create memory pools for biovec's in a bio_set. 1575 * use the global biovec slabs created for general use. 1576 */ 1577 int biovec_init_pool(mempool_t *pool, int pool_entries) 1578 { 1579 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1580 1581 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1582 } 1583 1584 /* 1585 * bioset_exit - exit a bioset initialized with bioset_init() 1586 * 1587 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1588 * kzalloc()). 1589 */ 1590 void bioset_exit(struct bio_set *bs) 1591 { 1592 bio_alloc_cache_destroy(bs); 1593 if (bs->rescue_workqueue) 1594 destroy_workqueue(bs->rescue_workqueue); 1595 bs->rescue_workqueue = NULL; 1596 1597 mempool_exit(&bs->bio_pool); 1598 mempool_exit(&bs->bvec_pool); 1599 1600 bioset_integrity_free(bs); 1601 if (bs->bio_slab) 1602 bio_put_slab(bs); 1603 bs->bio_slab = NULL; 1604 } 1605 EXPORT_SYMBOL(bioset_exit); 1606 1607 /** 1608 * bioset_init - Initialize a bio_set 1609 * @bs: pool to initialize 1610 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1611 * @front_pad: Number of bytes to allocate in front of the returned bio 1612 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1613 * and %BIOSET_NEED_RESCUER 1614 * 1615 * Description: 1616 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1617 * to ask for a number of bytes to be allocated in front of the bio. 1618 * Front pad allocation is useful for embedding the bio inside 1619 * another structure, to avoid allocating extra data to go with the bio. 1620 * Note that the bio must be embedded at the END of that structure always, 1621 * or things will break badly. 1622 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1623 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1624 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1625 * dispatch queued requests when the mempool runs out of space. 1626 * 1627 */ 1628 int bioset_init(struct bio_set *bs, 1629 unsigned int pool_size, 1630 unsigned int front_pad, 1631 int flags) 1632 { 1633 bs->front_pad = front_pad; 1634 if (flags & BIOSET_NEED_BVECS) 1635 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1636 else 1637 bs->back_pad = 0; 1638 1639 spin_lock_init(&bs->rescue_lock); 1640 bio_list_init(&bs->rescue_list); 1641 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1642 1643 bs->bio_slab = bio_find_or_create_slab(bs); 1644 if (!bs->bio_slab) 1645 return -ENOMEM; 1646 1647 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1648 goto bad; 1649 1650 if ((flags & BIOSET_NEED_BVECS) && 1651 biovec_init_pool(&bs->bvec_pool, pool_size)) 1652 goto bad; 1653 1654 if (flags & BIOSET_NEED_RESCUER) { 1655 bs->rescue_workqueue = alloc_workqueue("bioset", 1656 WQ_MEM_RECLAIM, 0); 1657 if (!bs->rescue_workqueue) 1658 goto bad; 1659 } 1660 if (flags & BIOSET_PERCPU_CACHE) { 1661 bs->cache = alloc_percpu(struct bio_alloc_cache); 1662 if (!bs->cache) 1663 goto bad; 1664 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1665 } 1666 1667 return 0; 1668 bad: 1669 bioset_exit(bs); 1670 return -ENOMEM; 1671 } 1672 EXPORT_SYMBOL(bioset_init); 1673 1674 /* 1675 * Initialize and setup a new bio_set, based on the settings from 1676 * another bio_set. 1677 */ 1678 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 1679 { 1680 int flags; 1681 1682 flags = 0; 1683 if (src->bvec_pool.min_nr) 1684 flags |= BIOSET_NEED_BVECS; 1685 if (src->rescue_workqueue) 1686 flags |= BIOSET_NEED_RESCUER; 1687 1688 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 1689 } 1690 EXPORT_SYMBOL(bioset_init_from_src); 1691 1692 /** 1693 * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb 1694 * @kiocb: kiocb describing the IO 1695 * @nr_vecs: number of iovecs to pre-allocate 1696 * @bs: bio_set to allocate from 1697 * 1698 * Description: 1699 * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only 1700 * used to check if we should dip into the per-cpu bio_set allocation 1701 * cache. The allocation uses GFP_KERNEL internally. On return, the 1702 * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio 1703 * MUST be done from process context, not hard/soft IRQ. 1704 * 1705 */ 1706 struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs, 1707 struct bio_set *bs) 1708 { 1709 struct bio_alloc_cache *cache; 1710 struct bio *bio; 1711 1712 if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS) 1713 return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1714 1715 cache = per_cpu_ptr(bs->cache, get_cpu()); 1716 bio = bio_list_pop(&cache->free_list); 1717 if (bio) { 1718 cache->nr--; 1719 put_cpu(); 1720 bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs); 1721 bio->bi_pool = bs; 1722 bio_set_flag(bio, BIO_PERCPU_CACHE); 1723 return bio; 1724 } 1725 put_cpu(); 1726 bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1727 bio_set_flag(bio, BIO_PERCPU_CACHE); 1728 return bio; 1729 } 1730 EXPORT_SYMBOL_GPL(bio_alloc_kiocb); 1731 1732 static int __init init_bio(void) 1733 { 1734 int i; 1735 1736 bio_integrity_init(); 1737 1738 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1739 struct biovec_slab *bvs = bvec_slabs + i; 1740 1741 bvs->slab = kmem_cache_create(bvs->name, 1742 bvs->nr_vecs * sizeof(struct bio_vec), 0, 1743 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1744 } 1745 1746 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1747 bio_cpu_dead); 1748 1749 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1750 panic("bio: can't allocate bios\n"); 1751 1752 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1753 panic("bio: can't create integrity pool\n"); 1754 1755 return 0; 1756 } 1757 subsys_initcall(init_bio); 1758