개념 설명 전체 · v6.18.37 / block/bio.c
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-integrity.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/highmem.h> 19 #include <linux/blk-crypto.h> 20 #include <linux/xarray.h> 21 22 #include <trace/events/block.h> 23 #include "blk.h" 24 #include "blk-rq-qos.h" 25 #include "blk-cgroup.h" 26 27 #define ALLOC_CACHE_THRESHOLD 16 28 #define ALLOC_CACHE_MAX 256 29 30 struct bio_alloc_cache { 31 struct bio *free_list; 32 struct bio *free_list_irq; 33 unsigned int nr; 34 unsigned int nr_irq; 35 }; 36 37 static struct biovec_slab { 38 int nr_vecs; 39 char *name; 40 struct kmem_cache *slab; 41 } bvec_slabs[] __read_mostly = { 42 { .nr_vecs = 16, .name = "biovec-16" }, 43 { .nr_vecs = 64, .name = "biovec-64" }, 44 { .nr_vecs = 128, .name = "biovec-128" }, 45 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" }, 46 }; 47 48 static struct biovec_slab *biovec_slab(unsigned short nr_vecs) 49 { 50 switch (nr_vecs) { 51 /* smaller bios use inline vecs */ 52 case 5 ... 16: 53 return &bvec_slabs[0]; 54 case 17 ... 64: 55 return &bvec_slabs[1]; 56 case 65 ... 128: 57 return &bvec_slabs[2]; 58 case 129 ... BIO_MAX_VECS: 59 return &bvec_slabs[3]; 60 default: 61 BUG(); 62 return NULL; 63 } 64 } 65 66 /* 67 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 68 * IO code that does not need private memory pools. 69 */ 70 struct bio_set fs_bio_set; 71 EXPORT_SYMBOL(fs_bio_set); 72 73 /* 74 * Our slab pool management 75 */ 76 struct bio_slab { 77 struct kmem_cache *slab; 78 unsigned int slab_ref; 79 unsigned int slab_size; 80 char name[12]; 81 }; 82 static DEFINE_MUTEX(bio_slab_lock); 83 static DEFINE_XARRAY(bio_slabs); 84 85 static struct bio_slab *create_bio_slab(unsigned int size) 86 { 87 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); 88 89 if (!bslab) 90 return NULL; 91 92 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); 93 bslab->slab = kmem_cache_create(bslab->name, size, 94 ARCH_KMALLOC_MINALIGN, 95 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL); 96 if (!bslab->slab) 97 goto fail_alloc_slab; 98 99 bslab->slab_ref = 1; 100 bslab->slab_size = size; 101 102 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) 103 return bslab; 104 105 kmem_cache_destroy(bslab->slab); 106 107 fail_alloc_slab: 108 kfree(bslab); 109 return NULL; 110 } 111 112 static inline unsigned int bs_bio_slab_size(struct bio_set *bs) 113 { 114 return bs->front_pad + sizeof(struct bio) + bs->back_pad; 115 } 116 117 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) 118 { 119 unsigned int size = bs_bio_slab_size(bs); 120 struct bio_slab *bslab; 121 122 mutex_lock(&bio_slab_lock); 123 bslab = xa_load(&bio_slabs, size); 124 if (bslab) 125 bslab->slab_ref++; 126 else 127 bslab = create_bio_slab(size); 128 mutex_unlock(&bio_slab_lock); 129 130 if (bslab) 131 return bslab->slab; 132 return NULL; 133 } 134 135 static void bio_put_slab(struct bio_set *bs) 136 { 137 struct bio_slab *bslab = NULL; 138 unsigned int slab_size = bs_bio_slab_size(bs); 139 140 mutex_lock(&bio_slab_lock); 141 142 bslab = xa_load(&bio_slabs, slab_size); 143 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 144 goto out; 145 146 WARN_ON_ONCE(bslab->slab != bs->bio_slab); 147 148 WARN_ON(!bslab->slab_ref); 149 150 if (--bslab->slab_ref) 151 goto out; 152 153 xa_erase(&bio_slabs, slab_size); 154 155 kmem_cache_destroy(bslab->slab); 156 kfree(bslab); 157 158 out: 159 mutex_unlock(&bio_slab_lock); 160 } 161 162 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs) 163 { 164 BUG_ON(nr_vecs > BIO_MAX_VECS); 165 166 if (nr_vecs == BIO_MAX_VECS) 167 mempool_free(bv, pool); 168 else if (nr_vecs > BIO_INLINE_VECS) 169 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv); 170 } 171 172 /* 173 * Make the first allocation restricted and don't dump info on allocation 174 * failures, since we'll fall back to the mempool in case of failure. 175 */ 176 static inline gfp_t bvec_alloc_gfp(gfp_t gfp) 177 { 178 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | 179 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 180 } 181 182 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, 183 gfp_t gfp_mask) 184 { 185 struct biovec_slab *bvs = biovec_slab(*nr_vecs); 186 187 if (WARN_ON_ONCE(!bvs)) 188 return NULL; 189 190 /* 191 * Upgrade the nr_vecs request to take full advantage of the allocation. 192 * We also rely on this in the bvec_free path. 193 */ 194 *nr_vecs = bvs->nr_vecs; 195 196 /* 197 * Try a slab allocation first for all smaller allocations. If that 198 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. 199 * The mempool is sized to handle up to BIO_MAX_VECS entries. 200 */ 201 if (*nr_vecs < BIO_MAX_VECS) { 202 struct bio_vec *bvl; 203 204 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); 205 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) 206 return bvl; 207 *nr_vecs = BIO_MAX_VECS; 208 } 209 210 return mempool_alloc(pool, gfp_mask); 211 } 212 213 void bio_uninit(struct bio *bio) 214 { 215 #ifdef CONFIG_BLK_CGROUP 216 if (bio->bi_blkg) { 217 blkg_put(bio->bi_blkg); 218 bio->bi_blkg = NULL; 219 } 220 #endif 221 if (bio_integrity(bio)) 222 bio_integrity_free(bio); 223 224 bio_crypt_free_ctx(bio); 225 } 226 EXPORT_SYMBOL(bio_uninit); 227 228 static void bio_free(struct bio *bio) 229 { 230 struct bio_set *bs = bio->bi_pool; 231 void *p = bio; 232 233 WARN_ON_ONCE(!bs); 234 235 bio_uninit(bio); 236 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs); 237 mempool_free(p - bs->front_pad, &bs->bio_pool); 238 } 239 240 /* 241 * Users of this function have their own bio allocation. Subsequently, 242 * they must remember to pair any call to bio_init() with bio_uninit() 243 * when IO has completed, or when the bio is released. 244 */ 245 void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table, 246 unsigned short max_vecs, blk_opf_t opf) 247 { 248 bio->bi_next = NULL; 249 bio->bi_bdev = bdev; 250 bio->bi_opf = opf; 251 bio->bi_flags = 0; 252 bio->bi_ioprio = 0; 253 bio->bi_write_hint = 0; 254 bio->bi_write_stream = 0; 255 bio->bi_status = 0; 256 bio->bi_iter.bi_sector = 0; 257 bio->bi_iter.bi_size = 0; 258 bio->bi_iter.bi_idx = 0; 259 bio->bi_iter.bi_bvec_done = 0; 260 bio->bi_end_io = NULL; 261 bio->bi_private = NULL; 262 #ifdef CONFIG_BLK_CGROUP 263 bio->bi_blkg = NULL; 264 bio->issue_time_ns = 0; 265 if (bdev) 266 bio_associate_blkg(bio); 267 #ifdef CONFIG_BLK_CGROUP_IOCOST 268 bio->bi_iocost_cost = 0; 269 #endif 270 #endif 271 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 272 bio->bi_crypt_context = NULL; 273 #endif 274 #ifdef CONFIG_BLK_DEV_INTEGRITY 275 bio->bi_integrity = NULL; 276 #endif 277 bio->bi_vcnt = 0; 278 279 atomic_set(&bio->__bi_remaining, 1); 280 atomic_set(&bio->__bi_cnt, 1); 281 bio->bi_cookie = BLK_QC_T_NONE; 282 283 bio->bi_max_vecs = max_vecs; 284 bio->bi_io_vec = table; 285 bio->bi_pool = NULL; 286 } 287 EXPORT_SYMBOL(bio_init); 288 289 /** 290 * bio_reset - reinitialize a bio 291 * @bio: bio to reset 292 * @bdev: block device to use the bio for 293 * @opf: operation and flags for bio 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, struct block_device *bdev, blk_opf_t opf) 302 { 303 bio_uninit(bio); 304 memset(bio, 0, BIO_RESET_BYTES); 305 atomic_set(&bio->__bi_remaining, 1); 306 bio->bi_bdev = bdev; 307 if (bio->bi_bdev) 308 bio_associate_blkg(bio); 309 bio->bi_opf = opf; 310 } 311 EXPORT_SYMBOL(bio_reset); 312 313 static struct bio *__bio_chain_endio(struct bio *bio) 314 { 315 struct bio *parent = bio->bi_private; 316 317 if (bio->bi_status && !parent->bi_status) 318 parent->bi_status = bio->bi_status; 319 bio_put(bio); 320 return parent; 321 } 322 323 static void bio_chain_endio(struct bio *bio) 324 { 325 bio_endio(__bio_chain_endio(bio)); 326 } 327 328 /** 329 * bio_chain - chain bio completions 330 * @bio: the target bio 331 * @parent: the parent bio of @bio 332 * 333 * The caller won't have a bi_end_io called when @bio completes - instead, 334 * @parent's bi_end_io won't be called until both @parent and @bio have 335 * completed; the chained bio will also be freed when it completes. 336 * 337 * The caller must not set bi_private or bi_end_io in @bio. 338 */ 339 void bio_chain(struct bio *bio, struct bio *parent) 340 { 341 BUG_ON(bio->bi_private || bio->bi_end_io); 342 343 bio->bi_private = parent; 344 bio->bi_end_io = bio_chain_endio; 345 bio_inc_remaining(parent); 346 } 347 EXPORT_SYMBOL(bio_chain); 348 349 /** 350 * bio_chain_and_submit - submit a bio after chaining it to another one 351 * @prev: bio to chain and submit 352 * @new: bio to chain to 353 * 354 * If @prev is non-NULL, chain it to @new and submit it. 355 * 356 * Return: @new. 357 */ 358 struct bio *bio_chain_and_submit(struct bio *prev, struct bio *new) 359 { 360 if (prev) { 361 bio_chain(prev, new); 362 submit_bio(prev); 363 } 364 return new; 365 } 366 367 struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev, 368 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp) 369 { 370 return bio_chain_and_submit(bio, bio_alloc(bdev, nr_pages, opf, gfp)); 371 } 372 EXPORT_SYMBOL_GPL(blk_next_bio); 373 374 static void bio_alloc_rescue(struct work_struct *work) 375 { 376 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 377 struct bio *bio; 378 379 while (1) { 380 spin_lock(&bs->rescue_lock); 381 bio = bio_list_pop(&bs->rescue_list); 382 spin_unlock(&bs->rescue_lock); 383 384 if (!bio) 385 break; 386 387 submit_bio_noacct(bio); 388 } 389 } 390 391 static void punt_bios_to_rescuer(struct bio_set *bs) 392 { 393 struct bio_list punt, nopunt; 394 struct bio *bio; 395 396 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 397 return; 398 /* 399 * In order to guarantee forward progress we must punt only bios that 400 * were allocated from this bio_set; otherwise, if there was a bio on 401 * there for a stacking driver higher up in the stack, processing it 402 * could require allocating bios from this bio_set, and doing that from 403 * our own rescuer would be bad. 404 * 405 * Since bio lists are singly linked, pop them all instead of trying to 406 * remove from the middle of the list: 407 */ 408 409 bio_list_init(&punt); 410 bio_list_init(&nopunt); 411 412 while ((bio = bio_list_pop(¤t->bio_list[0]))) 413 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 414 current->bio_list[0] = nopunt; 415 416 bio_list_init(&nopunt); 417 while ((bio = bio_list_pop(¤t->bio_list[1]))) 418 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 419 current->bio_list[1] = nopunt; 420 421 spin_lock(&bs->rescue_lock); 422 bio_list_merge(&bs->rescue_list, &punt); 423 spin_unlock(&bs->rescue_lock); 424 425 queue_work(bs->rescue_workqueue, &bs->rescue_work); 426 } 427 428 static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache) 429 { 430 unsigned long flags; 431 432 /* cache->free_list must be empty */ 433 if (WARN_ON_ONCE(cache->free_list)) 434 return; 435 436 local_irq_save(flags); 437 cache->free_list = cache->free_list_irq; 438 cache->free_list_irq = NULL; 439 cache->nr += cache->nr_irq; 440 cache->nr_irq = 0; 441 local_irq_restore(flags); 442 } 443 444 static struct bio *bio_alloc_percpu_cache(struct block_device *bdev, 445 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp, 446 struct bio_set *bs) 447 { 448 struct bio_alloc_cache *cache; 449 struct bio *bio; 450 451 cache = per_cpu_ptr(bs->cache, get_cpu()); 452 if (!cache->free_list) { 453 if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD) 454 bio_alloc_irq_cache_splice(cache); 455 if (!cache->free_list) { 456 put_cpu(); 457 return NULL; 458 } 459 } 460 bio = cache->free_list; 461 cache->free_list = bio->bi_next; 462 cache->nr--; 463 put_cpu(); 464 465 if (nr_vecs) 466 bio_init_inline(bio, bdev, nr_vecs, opf); 467 else 468 bio_init(bio, bdev, NULL, nr_vecs, opf); 469 bio->bi_pool = bs; 470 return bio; 471 } 472 473 /** 474 * bio_alloc_bioset - allocate a bio for I/O 475 * @bdev: block device to allocate the bio for (can be %NULL) 476 * @nr_vecs: number of bvecs to pre-allocate 477 * @opf: operation and flags for bio 478 * @gfp_mask: the GFP_* mask given to the slab allocator 479 * @bs: the bio_set to allocate from. 480 * 481 * Allocate a bio from the mempools in @bs. 482 * 483 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 484 * allocate a bio. This is due to the mempool guarantees. To make this work, 485 * callers must never allocate more than 1 bio at a time from the general pool. 486 * Callers that need to allocate more than 1 bio must always submit the 487 * previously allocated bio for IO before attempting to allocate a new one. 488 * Failure to do so can cause deadlocks under memory pressure. 489 * 490 * Note that when running under submit_bio_noacct() (i.e. any block driver), 491 * bios are not submitted until after you return - see the code in 492 * submit_bio_noacct() that converts recursion into iteration, to prevent 493 * stack overflows. 494 * 495 * This would normally mean allocating multiple bios under submit_bio_noacct() 496 * would be susceptible to deadlocks, but we have 497 * deadlock avoidance code that resubmits any blocked bios from a rescuer 498 * thread. 499 * 500 * However, we do not guarantee forward progress for allocations from other 501 * mempools. Doing multiple allocations from the same mempool under 502 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 503 * for per bio allocations. 504 * 505 * Returns: Pointer to new bio on success, NULL on failure. 506 */ 507 struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs, 508 blk_opf_t opf, gfp_t gfp_mask, 509 struct bio_set *bs) 510 { 511 gfp_t saved_gfp = gfp_mask; 512 struct bio *bio; 513 void *p; 514 515 /* should not use nobvec bioset for nr_vecs > 0 */ 516 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0)) 517 return NULL; 518 519 if (opf & REQ_ALLOC_CACHE) { 520 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) { 521 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf, 522 gfp_mask, bs); 523 if (bio) 524 return bio; 525 /* 526 * No cached bio available, bio returned below marked with 527 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache. 528 */ 529 } else { 530 opf &= ~REQ_ALLOC_CACHE; 531 } 532 } 533 534 /* 535 * submit_bio_noacct() converts recursion to iteration; this means if 536 * we're running beneath it, any bios we allocate and submit will not be 537 * submitted (and thus freed) until after we return. 538 * 539 * This exposes us to a potential deadlock if we allocate multiple bios 540 * from the same bio_set() while running underneath submit_bio_noacct(). 541 * If we were to allocate multiple bios (say a stacking block driver 542 * that was splitting bios), we would deadlock if we exhausted the 543 * mempool's reserve. 544 * 545 * We solve this, and guarantee forward progress, with a rescuer 546 * workqueue per bio_set. If we go to allocate and there are bios on 547 * current->bio_list, we first try the allocation without 548 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 549 * blocking to the rescuer workqueue before we retry with the original 550 * gfp_flags. 551 */ 552 if (current->bio_list && 553 (!bio_list_empty(¤t->bio_list[0]) || 554 !bio_list_empty(¤t->bio_list[1])) && 555 bs->rescue_workqueue) 556 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 557 558 p = mempool_alloc(&bs->bio_pool, gfp_mask); 559 if (!p && gfp_mask != saved_gfp) { 560 punt_bios_to_rescuer(bs); 561 gfp_mask = saved_gfp; 562 p = mempool_alloc(&bs->bio_pool, gfp_mask); 563 } 564 if (unlikely(!p)) 565 return NULL; 566 if (!mempool_is_saturated(&bs->bio_pool)) 567 opf &= ~REQ_ALLOC_CACHE; 568 569 bio = p + bs->front_pad; 570 if (nr_vecs > BIO_INLINE_VECS) { 571 struct bio_vec *bvl = NULL; 572 573 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 574 if (!bvl && gfp_mask != saved_gfp) { 575 punt_bios_to_rescuer(bs); 576 gfp_mask = saved_gfp; 577 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 578 } 579 if (unlikely(!bvl)) 580 goto err_free; 581 582 bio_init(bio, bdev, bvl, nr_vecs, opf); 583 } else if (nr_vecs) { 584 bio_init_inline(bio, bdev, BIO_INLINE_VECS, opf); 585 } else { 586 bio_init(bio, bdev, NULL, 0, opf); 587 } 588 589 bio->bi_pool = bs; 590 return bio; 591 592 err_free: 593 mempool_free(p, &bs->bio_pool); 594 return NULL; 595 } 596 EXPORT_SYMBOL(bio_alloc_bioset); 597 598 /** 599 * bio_kmalloc - kmalloc a bio 600 * @nr_vecs: number of bio_vecs to allocate 601 * @gfp_mask: the GFP_* mask given to the slab allocator 602 * 603 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized 604 * using bio_init() before use. To free a bio returned from this function use 605 * kfree() after calling bio_uninit(). A bio returned from this function can 606 * be reused by calling bio_uninit() before calling bio_init() again. 607 * 608 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this 609 * function are not backed by a mempool can fail. Do not use this function 610 * for allocations in the file system I/O path. 611 * 612 * Returns: Pointer to new bio on success, NULL on failure. 613 */ 614 struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask) 615 { 616 struct bio *bio; 617 618 if (nr_vecs > BIO_MAX_INLINE_VECS) 619 return NULL; 620 return kmalloc(sizeof(*bio) + nr_vecs * sizeof(struct bio_vec), 621 gfp_mask); 622 } 623 EXPORT_SYMBOL(bio_kmalloc); 624 625 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 626 { 627 struct bio_vec bv; 628 struct bvec_iter iter; 629 630 __bio_for_each_segment(bv, bio, iter, start) 631 memzero_bvec(&bv); 632 } 633 EXPORT_SYMBOL(zero_fill_bio_iter); 634 635 /** 636 * bio_truncate - truncate the bio to small size of @new_size 637 * @bio: the bio to be truncated 638 * @new_size: new size for truncating the bio 639 * 640 * Description: 641 * Truncate the bio to new size of @new_size. If bio_op(bio) is 642 * REQ_OP_READ, zero the truncated part. This function should only 643 * be used for handling corner cases, such as bio eod. 644 */ 645 static void bio_truncate(struct bio *bio, unsigned new_size) 646 { 647 struct bio_vec bv; 648 struct bvec_iter iter; 649 unsigned int done = 0; 650 bool truncated = false; 651 652 if (new_size >= bio->bi_iter.bi_size) 653 return; 654 655 if (bio_op(bio) != REQ_OP_READ) 656 goto exit; 657 658 bio_for_each_segment(bv, bio, iter) { 659 if (done + bv.bv_len > new_size) { 660 size_t offset; 661 662 if (!truncated) 663 offset = new_size - done; 664 else 665 offset = 0; 666 memzero_page(bv.bv_page, bv.bv_offset + offset, 667 bv.bv_len - offset); 668 truncated = true; 669 } 670 done += bv.bv_len; 671 } 672 673 exit: 674 /* 675 * Don't touch bvec table here and make it really immutable, since 676 * fs bio user has to retrieve all pages via bio_for_each_segment_all 677 * in its .end_bio() callback. 678 * 679 * It is enough to truncate bio by updating .bi_size since we can make 680 * correct bvec with the updated .bi_size for drivers. 681 */ 682 bio->bi_iter.bi_size = new_size; 683 } 684 685 /** 686 * guard_bio_eod - truncate a BIO to fit the block device 687 * @bio: bio to truncate 688 * 689 * This allows us to do IO even on the odd last sectors of a device, even if the 690 * block size is some multiple of the physical sector size. 691 * 692 * We'll just truncate the bio to the size of the device, and clear the end of 693 * the buffer head manually. Truly out-of-range accesses will turn into actual 694 * I/O errors, this only handles the "we need to be able to do I/O at the final 695 * sector" case. 696 */ 697 void guard_bio_eod(struct bio *bio) 698 { 699 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 700 701 if (!maxsector) 702 return; 703 704 /* 705 * If the *whole* IO is past the end of the device, 706 * let it through, and the IO layer will turn it into 707 * an EIO. 708 */ 709 if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 710 return; 711 712 maxsector -= bio->bi_iter.bi_sector; 713 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 714 return; 715 716 bio_truncate(bio, maxsector << 9); 717 } 718 719 static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache, 720 unsigned int nr) 721 { 722 unsigned int i = 0; 723 struct bio *bio; 724 725 while ((bio = cache->free_list) != NULL) { 726 cache->free_list = bio->bi_next; 727 cache->nr--; 728 bio_free(bio); 729 if (++i == nr) 730 break; 731 } 732 return i; 733 } 734 735 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 736 unsigned int nr) 737 { 738 nr -= __bio_alloc_cache_prune(cache, nr); 739 if (!READ_ONCE(cache->free_list)) { 740 bio_alloc_irq_cache_splice(cache); 741 __bio_alloc_cache_prune(cache, nr); 742 } 743 } 744 745 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 746 { 747 struct bio_set *bs; 748 749 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 750 if (bs->cache) { 751 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 752 753 bio_alloc_cache_prune(cache, -1U); 754 } 755 return 0; 756 } 757 758 static void bio_alloc_cache_destroy(struct bio_set *bs) 759 { 760 int cpu; 761 762 if (!bs->cache) 763 return; 764 765 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 766 for_each_possible_cpu(cpu) { 767 struct bio_alloc_cache *cache; 768 769 cache = per_cpu_ptr(bs->cache, cpu); 770 bio_alloc_cache_prune(cache, -1U); 771 } 772 free_percpu(bs->cache); 773 bs->cache = NULL; 774 } 775 776 static inline void bio_put_percpu_cache(struct bio *bio) 777 { 778 struct bio_alloc_cache *cache; 779 780 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 781 if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) 782 goto out_free; 783 784 if (in_task()) { 785 bio_uninit(bio); 786 bio->bi_next = cache->free_list; 787 /* Not necessary but helps not to iopoll already freed bios */ 788 bio->bi_bdev = NULL; 789 cache->free_list = bio; 790 cache->nr++; 791 } else if (in_hardirq()) { 792 lockdep_assert_irqs_disabled(); 793 794 bio_uninit(bio); 795 bio->bi_next = cache->free_list_irq; 796 cache->free_list_irq = bio; 797 cache->nr_irq++; 798 } else { 799 goto out_free; 800 } 801 put_cpu(); 802 return; 803 out_free: 804 put_cpu(); 805 bio_free(bio); 806 } 807 808 /** 809 * bio_put - release a reference to a bio 810 * @bio: bio to release reference to 811 * 812 * Description: 813 * Put a reference to a &struct bio, either one you have gotten with 814 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 815 **/ 816 void bio_put(struct bio *bio) 817 { 818 if (unlikely(bio_flagged(bio, BIO_REFFED))) { 819 BUG_ON(!atomic_read(&bio->__bi_cnt)); 820 if (!atomic_dec_and_test(&bio->__bi_cnt)) 821 return; 822 } 823 if (bio->bi_opf & REQ_ALLOC_CACHE) 824 bio_put_percpu_cache(bio); 825 else 826 bio_free(bio); 827 } 828 EXPORT_SYMBOL(bio_put); 829 830 static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp) 831 { 832 bio_set_flag(bio, BIO_CLONED); 833 bio->bi_ioprio = bio_src->bi_ioprio; 834 bio->bi_write_hint = bio_src->bi_write_hint; 835 bio->bi_write_stream = bio_src->bi_write_stream; 836 bio->bi_iter = bio_src->bi_iter; 837 838 if (bio->bi_bdev) { 839 if (bio->bi_bdev == bio_src->bi_bdev && 840 bio_flagged(bio_src, BIO_REMAPPED)) 841 bio_set_flag(bio, BIO_REMAPPED); 842 bio_clone_blkg_association(bio, bio_src); 843 } 844 845 if (bio_crypt_clone(bio, bio_src, gfp) < 0) 846 return -ENOMEM; 847 if (bio_integrity(bio_src) && 848 bio_integrity_clone(bio, bio_src, gfp) < 0) 849 return -ENOMEM; 850 return 0; 851 } 852 853 /** 854 * bio_alloc_clone - clone a bio that shares the original bio's biovec 855 * @bdev: block_device to clone onto 856 * @bio_src: bio to clone from 857 * @gfp: allocation priority 858 * @bs: bio_set to allocate from 859 * 860 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned 861 * bio, but not the actual data it points to. 862 * 863 * The caller must ensure that the return bio is not freed before @bio_src. 864 */ 865 struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src, 866 gfp_t gfp, struct bio_set *bs) 867 { 868 struct bio *bio; 869 870 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs); 871 if (!bio) 872 return NULL; 873 874 if (__bio_clone(bio, bio_src, gfp) < 0) { 875 bio_put(bio); 876 return NULL; 877 } 878 bio->bi_io_vec = bio_src->bi_io_vec; 879 880 return bio; 881 } 882 EXPORT_SYMBOL(bio_alloc_clone); 883 884 /** 885 * bio_init_clone - clone a bio that shares the original bio's biovec 886 * @bdev: block_device to clone onto 887 * @bio: bio to clone into 888 * @bio_src: bio to clone from 889 * @gfp: allocation priority 890 * 891 * Initialize a new bio in caller provided memory that is a clone of @bio_src. 892 * The caller owns the returned bio, but not the actual data it points to. 893 * 894 * The caller must ensure that @bio_src is not freed before @bio. 895 */ 896 int bio_init_clone(struct block_device *bdev, struct bio *bio, 897 struct bio *bio_src, gfp_t gfp) 898 { 899 int ret; 900 901 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf); 902 ret = __bio_clone(bio, bio_src, gfp); 903 if (ret) 904 bio_uninit(bio); 905 return ret; 906 } 907 EXPORT_SYMBOL(bio_init_clone); 908 909 /** 910 * bio_full - check if the bio is full 911 * @bio: bio to check 912 * @len: length of one segment to be added 913 * 914 * Return true if @bio is full and one segment with @len bytes can't be 915 * added to the bio, otherwise return false 916 */ 917 static inline bool bio_full(struct bio *bio, unsigned len) 918 { 919 if (bio->bi_vcnt >= bio->bi_max_vecs) 920 return true; 921 if (bio->bi_iter.bi_size > UINT_MAX - len) 922 return true; 923 return false; 924 } 925 926 static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page, 927 unsigned int len, unsigned int off) 928 { 929 size_t bv_end = bv->bv_offset + bv->bv_len; 930 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 931 phys_addr_t page_addr = page_to_phys(page); 932 933 if (vec_end_addr + 1 != page_addr + off) 934 return false; 935 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 936 return false; 937 938 if ((vec_end_addr & PAGE_MASK) != ((page_addr + off) & PAGE_MASK)) { 939 if (IS_ENABLED(CONFIG_KMSAN)) 940 return false; 941 if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE) 942 return false; 943 } 944 945 bv->bv_len += len; 946 return true; 947 } 948 949 /* 950 * Try to merge a page into a segment, while obeying the hardware segment 951 * size limit. 952 * 953 * This is kept around for the integrity metadata, which is still tries 954 * to build the initial bio to the hardware limit and doesn't have proper 955 * helpers to split. Hopefully this will go away soon. 956 */ 957 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv, 958 struct page *page, unsigned len, unsigned offset) 959 { 960 unsigned long mask = queue_segment_boundary(q); 961 phys_addr_t addr1 = bvec_phys(bv); 962 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 963 964 if ((addr1 | mask) != (addr2 | mask)) 965 return false; 966 if (len > queue_max_segment_size(q) - bv->bv_len) 967 return false; 968 return bvec_try_merge_page(bv, page, len, offset); 969 } 970 971 /** 972 * __bio_add_page - add page(s) to a bio in a new segment 973 * @bio: destination bio 974 * @page: start page to add 975 * @len: length of the data to add, may cross pages 976 * @off: offset of the data relative to @page, may cross pages 977 * 978 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 979 * that @bio has space for another bvec. 980 */ 981 void __bio_add_page(struct bio *bio, struct page *page, 982 unsigned int len, unsigned int off) 983 { 984 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 985 WARN_ON_ONCE(bio_full(bio, len)); 986 987 if (is_pci_p2pdma_page(page)) 988 bio->bi_opf |= REQ_NOMERGE; 989 990 bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off); 991 bio->bi_iter.bi_size += len; 992 bio->bi_vcnt++; 993 } 994 EXPORT_SYMBOL_GPL(__bio_add_page); 995 996 /** 997 * bio_add_virt_nofail - add data in the direct kernel mapping to a bio 998 * @bio: destination bio 999 * @vaddr: data to add 1000 * @len: length of the data to add, may cross pages 1001 * 1002 * Add the data at @vaddr to @bio. The caller must have ensure a segment 1003 * is available for the added data. No merging into an existing segment 1004 * will be performed. 1005 */ 1006 void bio_add_virt_nofail(struct bio *bio, void *vaddr, unsigned len) 1007 { 1008 __bio_add_page(bio, virt_to_page(vaddr), len, offset_in_page(vaddr)); 1009 } 1010 EXPORT_SYMBOL_GPL(bio_add_virt_nofail); 1011 1012 /** 1013 * bio_add_page - attempt to add page(s) to bio 1014 * @bio: destination bio 1015 * @page: start page to add 1016 * @len: vec entry length, may cross pages 1017 * @offset: vec entry offset relative to @page, may cross pages 1018 * 1019 * Attempt to add page(s) to the bio_vec maplist. This will only fail 1020 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1021 */ 1022 int bio_add_page(struct bio *bio, struct page *page, 1023 unsigned int len, unsigned int offset) 1024 { 1025 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 1026 return 0; 1027 if (bio->bi_iter.bi_size > UINT_MAX - len) 1028 return 0; 1029 1030 if (bio->bi_vcnt > 0) { 1031 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 1032 1033 if (!zone_device_pages_compatible(bv->bv_page, page)) 1034 return 0; 1035 if (zone_device_pages_have_same_pgmap(bv->bv_page, page) && 1036 bvec_try_merge_page(bv, page, len, offset)) { 1037 bio->bi_iter.bi_size += len; 1038 return len; 1039 } 1040 } 1041 1042 if (bio->bi_vcnt >= bio->bi_max_vecs) 1043 return 0; 1044 __bio_add_page(bio, page, len, offset); 1045 return len; 1046 } 1047 EXPORT_SYMBOL(bio_add_page); 1048 1049 void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len, 1050 size_t off) 1051 { 1052 unsigned long nr = off / PAGE_SIZE; 1053 1054 WARN_ON_ONCE(len > UINT_MAX); 1055 __bio_add_page(bio, folio_page(folio, nr), len, off % PAGE_SIZE); 1056 } 1057 EXPORT_SYMBOL_GPL(bio_add_folio_nofail); 1058 1059 /** 1060 * bio_add_folio - Attempt to add part of a folio to a bio. 1061 * @bio: BIO to add to. 1062 * @folio: Folio to add. 1063 * @len: How many bytes from the folio to add. 1064 * @off: First byte in this folio to add. 1065 * 1066 * Filesystems that use folios can call this function instead of calling 1067 * bio_add_page() for each page in the folio. If @off is bigger than 1068 * PAGE_SIZE, this function can create a bio_vec that starts in a page 1069 * after the bv_page. BIOs do not support folios that are 4GiB or larger. 1070 * 1071 * Return: Whether the addition was successful. 1072 */ 1073 bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, 1074 size_t off) 1075 { 1076 unsigned long nr = off / PAGE_SIZE; 1077 1078 if (len > UINT_MAX) 1079 return false; 1080 return bio_add_page(bio, folio_page(folio, nr), len, off % PAGE_SIZE) > 0; 1081 } 1082 EXPORT_SYMBOL(bio_add_folio); 1083 1084 /** 1085 * bio_add_vmalloc_chunk - add a vmalloc chunk to a bio 1086 * @bio: destination bio 1087 * @vaddr: vmalloc address to add 1088 * @len: total length in bytes of the data to add 1089 * 1090 * Add data starting at @vaddr to @bio and return how many bytes were added. 1091 * This may be less than the amount originally asked. Returns 0 if no data 1092 * could be added to @bio. 1093 * 1094 * This helper calls flush_kernel_vmap_range() for the range added. For reads 1095 * the caller still needs to manually call invalidate_kernel_vmap_range() in 1096 * the completion handler. 1097 */ 1098 unsigned int bio_add_vmalloc_chunk(struct bio *bio, void *vaddr, unsigned len) 1099 { 1100 unsigned int offset = offset_in_page(vaddr); 1101 1102 len = min(len, PAGE_SIZE - offset); 1103 if (bio_add_page(bio, vmalloc_to_page(vaddr), len, offset) < len) 1104 return 0; 1105 if (op_is_write(bio_op(bio))) 1106 flush_kernel_vmap_range(vaddr, len); 1107 return len; 1108 } 1109 EXPORT_SYMBOL_GPL(bio_add_vmalloc_chunk); 1110 1111 /** 1112 * bio_add_vmalloc - add a vmalloc region to a bio 1113 * @bio: destination bio 1114 * @vaddr: vmalloc address to add 1115 * @len: total length in bytes of the data to add 1116 * 1117 * Add data starting at @vaddr to @bio. Return %true on success or %false if 1118 * @bio does not have enough space for the payload. 1119 * 1120 * This helper calls flush_kernel_vmap_range() for the range added. For reads 1121 * the caller still needs to manually call invalidate_kernel_vmap_range() in 1122 * the completion handler. 1123 */ 1124 bool bio_add_vmalloc(struct bio *bio, void *vaddr, unsigned int len) 1125 { 1126 do { 1127 unsigned int added = bio_add_vmalloc_chunk(bio, vaddr, len); 1128 1129 if (!added) 1130 return false; 1131 vaddr += added; 1132 len -= added; 1133 } while (len); 1134 1135 return true; 1136 } 1137 EXPORT_SYMBOL_GPL(bio_add_vmalloc); 1138 1139 void __bio_release_pages(struct bio *bio, bool mark_dirty) 1140 { 1141 struct folio_iter fi; 1142 1143 bio_for_each_folio_all(fi, bio) { 1144 size_t nr_pages; 1145 1146 if (mark_dirty) { 1147 folio_lock(fi.folio); 1148 folio_mark_dirty(fi.folio); 1149 folio_unlock(fi.folio); 1150 } 1151 nr_pages = (fi.offset + fi.length - 1) / PAGE_SIZE - 1152 fi.offset / PAGE_SIZE + 1; 1153 unpin_user_folio(fi.folio, nr_pages); 1154 } 1155 } 1156 EXPORT_SYMBOL_GPL(__bio_release_pages); 1157 1158 void bio_iov_bvec_set(struct bio *bio, const struct iov_iter *iter) 1159 { 1160 WARN_ON_ONCE(bio->bi_max_vecs); 1161 1162 bio->bi_vcnt = iter->nr_segs; 1163 bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1164 bio->bi_iter.bi_bvec_done = iter->iov_offset; 1165 bio->bi_iter.bi_size = iov_iter_count(iter); 1166 bio_set_flag(bio, BIO_CLONED); 1167 } 1168 1169 static unsigned int get_contig_folio_len(unsigned int *num_pages, 1170 struct page **pages, unsigned int i, 1171 struct folio *folio, size_t left, 1172 size_t offset) 1173 { 1174 size_t bytes = left; 1175 size_t contig_sz = min_t(size_t, PAGE_SIZE - offset, bytes); 1176 unsigned int j; 1177 1178 /* 1179 * We might COW a single page in the middle of 1180 * a large folio, so we have to check that all 1181 * pages belong to the same folio. 1182 */ 1183 bytes -= contig_sz; 1184 for (j = i + 1; j < i + *num_pages; j++) { 1185 size_t next = min_t(size_t, PAGE_SIZE, bytes); 1186 1187 if (page_folio(pages[j]) != folio || 1188 pages[j] != pages[j - 1] + 1) { 1189 break; 1190 } 1191 contig_sz += next; 1192 bytes -= next; 1193 } 1194 *num_pages = j - i; 1195 1196 return contig_sz; 1197 } 1198 1199 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1200 1201 /** 1202 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 1203 * @bio: bio to add pages to 1204 * @iter: iov iterator describing the region to be mapped 1205 * 1206 * Extracts pages from *iter and appends them to @bio's bvec array. The pages 1207 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag. 1208 * For a multi-segment *iter, this function only adds pages from the next 1209 * non-empty segment of the iov iterator. 1210 */ 1211 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1212 { 1213 iov_iter_extraction_t extraction_flags = 0; 1214 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1215 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1216 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 1217 struct page **pages = (struct page **)bv; 1218 ssize_t size; 1219 unsigned int num_pages, i = 0; 1220 size_t offset, folio_offset, left, len; 1221 int ret = 0; 1222 1223 /* 1224 * Move page array up in the allocated memory for the bio vecs as far as 1225 * possible so that we can start filling biovecs from the beginning 1226 * without overwriting the temporary page array. 1227 */ 1228 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1229 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1230 1231 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue)) 1232 extraction_flags |= ITER_ALLOW_P2PDMA; 1233 1234 size = iov_iter_extract_pages(iter, &pages, 1235 UINT_MAX - bio->bi_iter.bi_size, 1236 nr_pages, extraction_flags, &offset); 1237 if (unlikely(size <= 0)) 1238 return size ? size : -EFAULT; 1239 1240 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE); 1241 for (left = size, i = 0; left > 0; left -= len, i += num_pages) { 1242 struct page *page = pages[i]; 1243 struct folio *folio = page_folio(page); 1244 unsigned int old_vcnt = bio->bi_vcnt; 1245 1246 folio_offset = ((size_t)folio_page_idx(folio, page) << 1247 PAGE_SHIFT) + offset; 1248 1249 len = min(folio_size(folio) - folio_offset, left); 1250 1251 num_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1252 1253 if (num_pages > 1) 1254 len = get_contig_folio_len(&num_pages, pages, i, 1255 folio, left, offset); 1256 1257 if (!bio_add_folio(bio, folio, len, folio_offset)) { 1258 WARN_ON_ONCE(1); 1259 ret = -EINVAL; 1260 goto out; 1261 } 1262 1263 if (bio_flagged(bio, BIO_PAGE_PINNED)) { 1264 /* 1265 * We're adding another fragment of a page that already 1266 * was part of the last segment. Undo our pin as the 1267 * page was pinned when an earlier fragment of it was 1268 * added to the bio and __bio_release_pages expects a 1269 * single pin per page. 1270 */ 1271 if (offset && bio->bi_vcnt == old_vcnt) 1272 unpin_user_folio(folio, 1); 1273 } 1274 offset = 0; 1275 } 1276 1277 iov_iter_revert(iter, left); 1278 out: 1279 while (i < nr_pages) 1280 bio_release_page(bio, pages[i++]); 1281 1282 return ret; 1283 } 1284 1285 /* 1286 * Aligns the bio size to the len_align_mask, releasing excessive bio vecs that 1287 * __bio_iov_iter_get_pages may have inserted, and reverts the trimmed length 1288 * for the next iteration. 1289 */ 1290 static int bio_iov_iter_align_down(struct bio *bio, struct iov_iter *iter, 1291 unsigned len_align_mask) 1292 { 1293 size_t nbytes = bio->bi_iter.bi_size & len_align_mask; 1294 1295 if (!nbytes) 1296 return 0; 1297 1298 iov_iter_revert(iter, nbytes); 1299 bio->bi_iter.bi_size -= nbytes; 1300 do { 1301 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 1302 1303 if (nbytes < bv->bv_len) { 1304 bv->bv_len -= nbytes; 1305 break; 1306 } 1307 1308 bio_release_page(bio, bv->bv_page); 1309 bio->bi_vcnt--; 1310 nbytes -= bv->bv_len; 1311 } while (nbytes); 1312 1313 if (!bio->bi_vcnt) 1314 return -EFAULT; 1315 return 0; 1316 } 1317 1318 /** 1319 * bio_iov_iter_get_pages - add user or kernel pages to a bio 1320 * @bio: bio to add pages to 1321 * @iter: iov iterator describing the region to be added 1322 * @len_align_mask: the mask to align the total size to, 0 for any length 1323 * 1324 * This takes either an iterator pointing to user memory, or one pointing to 1325 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 1326 * map them into the kernel. On IO completion, the caller should put those 1327 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1328 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1329 * to ensure the bvecs and pages stay referenced until the submitted I/O is 1330 * completed by a call to ->ki_complete() or returns with an error other than 1331 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1332 * on IO completion. If it isn't, then pages should be released. 1333 * 1334 * The function tries, but does not guarantee, to pin as many pages as 1335 * fit into the bio, or are requested in @iter, whatever is smaller. If 1336 * MM encounters an error pinning the requested pages, it stops. Error 1337 * is returned only if 0 pages could be pinned. 1338 */ 1339 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter, 1340 unsigned len_align_mask) 1341 { 1342 int ret = 0; 1343 1344 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 1345 return -EIO; 1346 1347 if (iov_iter_is_bvec(iter)) { 1348 bio_iov_bvec_set(bio, iter); 1349 iov_iter_advance(iter, bio->bi_iter.bi_size); 1350 return 0; 1351 } 1352 1353 if (iov_iter_extract_will_pin(iter)) 1354 bio_set_flag(bio, BIO_PAGE_PINNED); 1355 do { 1356 ret = __bio_iov_iter_get_pages(bio, iter); 1357 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1358 1359 if (bio->bi_vcnt) 1360 return bio_iov_iter_align_down(bio, iter, len_align_mask); 1361 return ret; 1362 } 1363 1364 static void submit_bio_wait_endio(struct bio *bio) 1365 { 1366 complete(bio->bi_private); 1367 } 1368 1369 /** 1370 * submit_bio_wait - submit a bio, and wait until it completes 1371 * @bio: The &struct bio which describes the I/O 1372 * 1373 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1374 * bio_endio() on failure. 1375 * 1376 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1377 * result in bio reference to be consumed. The caller must drop the reference 1378 * on his own. 1379 */ 1380 int submit_bio_wait(struct bio *bio) 1381 { 1382 DECLARE_COMPLETION_ONSTACK_MAP(done, 1383 bio->bi_bdev->bd_disk->lockdep_map); 1384 1385 bio->bi_private = &done; 1386 bio->bi_end_io = submit_bio_wait_endio; 1387 bio->bi_opf |= REQ_SYNC; 1388 submit_bio(bio); 1389 blk_wait_io(&done); 1390 1391 return blk_status_to_errno(bio->bi_status); 1392 } 1393 EXPORT_SYMBOL(submit_bio_wait); 1394 1395 /** 1396 * bdev_rw_virt - synchronously read into / write from kernel mapping 1397 * @bdev: block device to access 1398 * @sector: sector to access 1399 * @data: data to read/write 1400 * @len: length in byte to read/write 1401 * @op: operation (e.g. REQ_OP_READ/REQ_OP_WRITE) 1402 * 1403 * Performs synchronous I/O to @bdev for @data/@len. @data must be in 1404 * the kernel direct mapping and not a vmalloc address. 1405 */ 1406 int bdev_rw_virt(struct block_device *bdev, sector_t sector, void *data, 1407 size_t len, enum req_op op) 1408 { 1409 struct bio_vec bv; 1410 struct bio bio; 1411 int error; 1412 1413 if (WARN_ON_ONCE(is_vmalloc_addr(data))) 1414 return -EIO; 1415 1416 bio_init(&bio, bdev, &bv, 1, op); 1417 bio.bi_iter.bi_sector = sector; 1418 bio_add_virt_nofail(&bio, data, len); 1419 error = submit_bio_wait(&bio); 1420 bio_uninit(&bio); 1421 return error; 1422 } 1423 EXPORT_SYMBOL_GPL(bdev_rw_virt); 1424 1425 static void bio_wait_end_io(struct bio *bio) 1426 { 1427 complete(bio->bi_private); 1428 bio_put(bio); 1429 } 1430 1431 /* 1432 * bio_await_chain - ends @bio and waits for every chained bio to complete 1433 */ 1434 void bio_await_chain(struct bio *bio) 1435 { 1436 DECLARE_COMPLETION_ONSTACK_MAP(done, 1437 bio->bi_bdev->bd_disk->lockdep_map); 1438 1439 bio->bi_private = &done; 1440 bio->bi_end_io = bio_wait_end_io; 1441 bio_endio(bio); 1442 blk_wait_io(&done); 1443 } 1444 1445 void __bio_advance(struct bio *bio, unsigned bytes) 1446 { 1447 if (bio_integrity(bio)) 1448 bio_integrity_advance(bio, bytes); 1449 1450 bio_crypt_advance(bio, bytes); 1451 bio_advance_iter(bio, &bio->bi_iter, bytes); 1452 } 1453 EXPORT_SYMBOL(__bio_advance); 1454 1455 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1456 struct bio *src, struct bvec_iter *src_iter) 1457 { 1458 while (src_iter->bi_size && dst_iter->bi_size) { 1459 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1460 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1461 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1462 void *src_buf = bvec_kmap_local(&src_bv); 1463 void *dst_buf = bvec_kmap_local(&dst_bv); 1464 1465 memcpy(dst_buf, src_buf, bytes); 1466 1467 kunmap_local(dst_buf); 1468 kunmap_local(src_buf); 1469 1470 bio_advance_iter_single(src, src_iter, bytes); 1471 bio_advance_iter_single(dst, dst_iter, bytes); 1472 } 1473 } 1474 EXPORT_SYMBOL(bio_copy_data_iter); 1475 1476 /** 1477 * bio_copy_data - copy contents of data buffers from one bio to another 1478 * @src: source bio 1479 * @dst: destination bio 1480 * 1481 * Stops when it reaches the end of either @src or @dst - that is, copies 1482 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1483 */ 1484 void bio_copy_data(struct bio *dst, struct bio *src) 1485 { 1486 struct bvec_iter src_iter = src->bi_iter; 1487 struct bvec_iter dst_iter = dst->bi_iter; 1488 1489 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1490 } 1491 EXPORT_SYMBOL(bio_copy_data); 1492 1493 void bio_free_pages(struct bio *bio) 1494 { 1495 struct bio_vec *bvec; 1496 struct bvec_iter_all iter_all; 1497 1498 bio_for_each_segment_all(bvec, bio, iter_all) 1499 __free_page(bvec->bv_page); 1500 } 1501 EXPORT_SYMBOL(bio_free_pages); 1502 1503 /* 1504 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1505 * for performing direct-IO in BIOs. 1506 * 1507 * The problem is that we cannot run folio_mark_dirty() from interrupt context 1508 * because the required locks are not interrupt-safe. So what we can do is to 1509 * mark the pages dirty _before_ performing IO. And in interrupt context, 1510 * check that the pages are still dirty. If so, fine. If not, redirty them 1511 * in process context. 1512 * 1513 * Note that this code is very hard to test under normal circumstances because 1514 * direct-io pins the pages with get_user_pages(). This makes 1515 * is_page_cache_freeable return false, and the VM will not clean the pages. 1516 * But other code (eg, flusher threads) could clean the pages if they are mapped 1517 * pagecache. 1518 * 1519 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1520 * deferred bio dirtying paths. 1521 */ 1522 1523 /* 1524 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1525 */ 1526 void bio_set_pages_dirty(struct bio *bio) 1527 { 1528 struct folio_iter fi; 1529 1530 bio_for_each_folio_all(fi, bio) { 1531 folio_lock(fi.folio); 1532 folio_mark_dirty(fi.folio); 1533 folio_unlock(fi.folio); 1534 } 1535 } 1536 EXPORT_SYMBOL_GPL(bio_set_pages_dirty); 1537 1538 /* 1539 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1540 * If they are, then fine. If, however, some pages are clean then they must 1541 * have been written out during the direct-IO read. So we take another ref on 1542 * the BIO and re-dirty the pages in process context. 1543 * 1544 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1545 * here on. It will unpin each page and will run one bio_put() against the 1546 * BIO. 1547 */ 1548 1549 static void bio_dirty_fn(struct work_struct *work); 1550 1551 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1552 static DEFINE_SPINLOCK(bio_dirty_lock); 1553 static struct bio *bio_dirty_list; 1554 1555 /* 1556 * This runs in process context 1557 */ 1558 static void bio_dirty_fn(struct work_struct *work) 1559 { 1560 struct bio *bio, *next; 1561 1562 spin_lock_irq(&bio_dirty_lock); 1563 next = bio_dirty_list; 1564 bio_dirty_list = NULL; 1565 spin_unlock_irq(&bio_dirty_lock); 1566 1567 while ((bio = next) != NULL) { 1568 next = bio->bi_private; 1569 1570 bio_release_pages(bio, true); 1571 bio_put(bio); 1572 } 1573 } 1574 1575 void bio_check_pages_dirty(struct bio *bio) 1576 { 1577 struct folio_iter fi; 1578 unsigned long flags; 1579 1580 bio_for_each_folio_all(fi, bio) { 1581 if (!folio_test_dirty(fi.folio)) 1582 goto defer; 1583 } 1584 1585 bio_release_pages(bio, false); 1586 bio_put(bio); 1587 return; 1588 defer: 1589 spin_lock_irqsave(&bio_dirty_lock, flags); 1590 bio->bi_private = bio_dirty_list; 1591 bio_dirty_list = bio; 1592 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1593 schedule_work(&bio_dirty_work); 1594 } 1595 EXPORT_SYMBOL_GPL(bio_check_pages_dirty); 1596 1597 static inline bool bio_remaining_done(struct bio *bio) 1598 { 1599 /* 1600 * If we're not chaining, then ->__bi_remaining is always 1 and 1601 * we always end io on the first invocation. 1602 */ 1603 if (!bio_flagged(bio, BIO_CHAIN)) 1604 return true; 1605 1606 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1607 1608 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1609 bio_clear_flag(bio, BIO_CHAIN); 1610 return true; 1611 } 1612 1613 return false; 1614 } 1615 1616 /** 1617 * bio_endio - end I/O on a bio 1618 * @bio: bio 1619 * 1620 * Description: 1621 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1622 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1623 * bio unless they own it and thus know that it has an end_io function. 1624 * 1625 * bio_endio() can be called several times on a bio that has been chained 1626 * using bio_chain(). The ->bi_end_io() function will only be called the 1627 * last time. 1628 **/ 1629 void bio_endio(struct bio *bio) 1630 { 1631 again: 1632 if (!bio_remaining_done(bio)) 1633 return; 1634 if (!bio_integrity_endio(bio)) 1635 return; 1636 1637 blk_zone_bio_endio(bio); 1638 1639 rq_qos_done_bio(bio); 1640 1641 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1642 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 1643 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1644 } 1645 1646 /* 1647 * Need to have a real endio function for chained bios, otherwise 1648 * various corner cases will break (like stacking block devices that 1649 * save/restore bi_end_io) - however, we want to avoid unbounded 1650 * recursion and blowing the stack. Tail call optimization would 1651 * handle this, but compiling with frame pointers also disables 1652 * gcc's sibling call optimization. 1653 */ 1654 if (bio->bi_end_io == bio_chain_endio) { 1655 bio = __bio_chain_endio(bio); 1656 goto again; 1657 } 1658 1659 #ifdef CONFIG_BLK_CGROUP 1660 /* 1661 * Release cgroup info. We shouldn't have to do this here, but quite 1662 * a few callers of bio_init fail to call bio_uninit, so we cover up 1663 * for that here at least for now. 1664 */ 1665 if (bio->bi_blkg) { 1666 blkg_put(bio->bi_blkg); 1667 bio->bi_blkg = NULL; 1668 } 1669 #endif 1670 1671 if (bio->bi_end_io) 1672 bio->bi_end_io(bio); 1673 } 1674 EXPORT_SYMBOL(bio_endio); 1675 1676 /** 1677 * bio_split - split a bio 1678 * @bio: bio to split 1679 * @sectors: number of sectors to split from the front of @bio 1680 * @gfp: gfp mask 1681 * @bs: bio set to allocate from 1682 * 1683 * Allocates and returns a new bio which represents @sectors from the start of 1684 * @bio, and updates @bio to represent the remaining sectors. 1685 * 1686 * Unless this is a discard request the newly allocated bio will point 1687 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1688 * neither @bio nor @bs are freed before the split bio. 1689 */ 1690 struct bio *bio_split(struct bio *bio, int sectors, 1691 gfp_t gfp, struct bio_set *bs) 1692 { 1693 struct bio *split; 1694 1695 if (WARN_ON_ONCE(sectors <= 0)) 1696 return ERR_PTR(-EINVAL); 1697 if (WARN_ON_ONCE(sectors >= bio_sectors(bio))) 1698 return ERR_PTR(-EINVAL); 1699 1700 /* Zone append commands cannot be split */ 1701 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 1702 return ERR_PTR(-EINVAL); 1703 1704 /* atomic writes cannot be split */ 1705 if (bio->bi_opf & REQ_ATOMIC) 1706 return ERR_PTR(-EINVAL); 1707 1708 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs); 1709 if (!split) 1710 return ERR_PTR(-ENOMEM); 1711 1712 split->bi_iter.bi_size = sectors << 9; 1713 1714 if (bio_integrity(split)) 1715 bio_integrity_trim(split); 1716 1717 bio_advance(bio, split->bi_iter.bi_size); 1718 1719 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1720 bio_set_flag(split, BIO_TRACE_COMPLETION); 1721 1722 return split; 1723 } 1724 EXPORT_SYMBOL(bio_split); 1725 1726 /** 1727 * bio_trim - trim a bio 1728 * @bio: bio to trim 1729 * @offset: number of sectors to trim from the front of @bio 1730 * @size: size we want to trim @bio to, in sectors 1731 * 1732 * This function is typically used for bios that are cloned and submitted 1733 * to the underlying device in parts. 1734 */ 1735 void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1736 { 1737 /* We should never trim an atomic write */ 1738 if (WARN_ON_ONCE(bio->bi_opf & REQ_ATOMIC && size)) 1739 return; 1740 1741 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 1742 offset + size > bio_sectors(bio))) 1743 return; 1744 1745 size <<= 9; 1746 if (offset == 0 && size == bio->bi_iter.bi_size) 1747 return; 1748 1749 bio_advance(bio, offset << 9); 1750 bio->bi_iter.bi_size = size; 1751 1752 if (bio_integrity(bio)) 1753 bio_integrity_trim(bio); 1754 } 1755 EXPORT_SYMBOL_GPL(bio_trim); 1756 1757 /* 1758 * create memory pools for biovec's in a bio_set. 1759 * use the global biovec slabs created for general use. 1760 */ 1761 int biovec_init_pool(mempool_t *pool, int pool_entries) 1762 { 1763 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1764 1765 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1766 } 1767 1768 /* 1769 * bioset_exit - exit a bioset initialized with bioset_init() 1770 * 1771 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1772 * kzalloc()). 1773 */ 1774 void bioset_exit(struct bio_set *bs) 1775 { 1776 bio_alloc_cache_destroy(bs); 1777 if (bs->rescue_workqueue) 1778 destroy_workqueue(bs->rescue_workqueue); 1779 bs->rescue_workqueue = NULL; 1780 1781 mempool_exit(&bs->bio_pool); 1782 mempool_exit(&bs->bvec_pool); 1783 1784 if (bs->bio_slab) 1785 bio_put_slab(bs); 1786 bs->bio_slab = NULL; 1787 } 1788 EXPORT_SYMBOL(bioset_exit); 1789 1790 /** 1791 * bioset_init - Initialize a bio_set 1792 * @bs: pool to initialize 1793 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1794 * @front_pad: Number of bytes to allocate in front of the returned bio 1795 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1796 * and %BIOSET_NEED_RESCUER 1797 * 1798 * Description: 1799 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1800 * to ask for a number of bytes to be allocated in front of the bio. 1801 * Front pad allocation is useful for embedding the bio inside 1802 * another structure, to avoid allocating extra data to go with the bio. 1803 * Note that the bio must be embedded at the END of that structure always, 1804 * or things will break badly. 1805 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1806 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone(). 1807 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used 1808 * to dispatch queued requests when the mempool runs out of space. 1809 * 1810 */ 1811 int bioset_init(struct bio_set *bs, 1812 unsigned int pool_size, 1813 unsigned int front_pad, 1814 int flags) 1815 { 1816 bs->front_pad = front_pad; 1817 if (flags & BIOSET_NEED_BVECS) 1818 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1819 else 1820 bs->back_pad = 0; 1821 1822 spin_lock_init(&bs->rescue_lock); 1823 bio_list_init(&bs->rescue_list); 1824 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1825 1826 bs->bio_slab = bio_find_or_create_slab(bs); 1827 if (!bs->bio_slab) 1828 return -ENOMEM; 1829 1830 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1831 goto bad; 1832 1833 if ((flags & BIOSET_NEED_BVECS) && 1834 biovec_init_pool(&bs->bvec_pool, pool_size)) 1835 goto bad; 1836 1837 if (flags & BIOSET_NEED_RESCUER) { 1838 bs->rescue_workqueue = alloc_workqueue("bioset", 1839 WQ_MEM_RECLAIM, 0); 1840 if (!bs->rescue_workqueue) 1841 goto bad; 1842 } 1843 if (flags & BIOSET_PERCPU_CACHE) { 1844 bs->cache = alloc_percpu(struct bio_alloc_cache); 1845 if (!bs->cache) 1846 goto bad; 1847 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1848 } 1849 1850 return 0; 1851 bad: 1852 bioset_exit(bs); 1853 return -ENOMEM; 1854 } 1855 EXPORT_SYMBOL(bioset_init); 1856 1857 static int __init init_bio(void) 1858 { 1859 int i; 1860 1861 BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags)); 1862 1863 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1864 struct biovec_slab *bvs = bvec_slabs + i; 1865 1866 bvs->slab = kmem_cache_create(bvs->name, 1867 bvs->nr_vecs * sizeof(struct bio_vec), 0, 1868 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1869 } 1870 1871 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1872 bio_cpu_dead); 1873 1874 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, 1875 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE)) 1876 panic("bio: can't allocate bios\n"); 1877 1878 return 0; 1879 } 1880 subsys_initcall(init_bio);