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 23 #include <trace/events/block.h> 24 #include "blk.h" 25 #include "blk-rq-qos.h" 26 27 /* 28 * Test patch to inline a certain number of bi_io_vec's inside the bio 29 * itself, to shrink a bio data allocation from two mempool calls to one 30 */ 31 #define BIO_INLINE_VECS 4 32 33 /* 34 * if you change this list, also change bvec_alloc or things will 35 * break badly! cannot be bigger than what you can fit into an 36 * unsigned short 37 */ 38 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 39 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 40 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 41 }; 42 #undef BV 43 44 /* 45 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 46 * IO code that does not need private memory pools. 47 */ 48 struct bio_set fs_bio_set; 49 EXPORT_SYMBOL(fs_bio_set); 50 51 /* 52 * Our slab pool management 53 */ 54 struct bio_slab { 55 struct kmem_cache *slab; 56 unsigned int slab_ref; 57 unsigned int slab_size; 58 char name[8]; 59 }; 60 static DEFINE_MUTEX(bio_slab_lock); 61 static struct bio_slab *bio_slabs; 62 static unsigned int bio_slab_nr, bio_slab_max; 63 64 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 65 { 66 unsigned int sz = sizeof(struct bio) + extra_size; 67 struct kmem_cache *slab = NULL; 68 struct bio_slab *bslab, *new_bio_slabs; 69 unsigned int new_bio_slab_max; 70 unsigned int i, entry = -1; 71 72 mutex_lock(&bio_slab_lock); 73 74 i = 0; 75 while (i < bio_slab_nr) { 76 bslab = &bio_slabs[i]; 77 78 if (!bslab->slab && entry == -1) 79 entry = i; 80 else if (bslab->slab_size == sz) { 81 slab = bslab->slab; 82 bslab->slab_ref++; 83 break; 84 } 85 i++; 86 } 87 88 if (slab) 89 goto out_unlock; 90 91 if (bio_slab_nr == bio_slab_max && entry == -1) { 92 new_bio_slab_max = bio_slab_max << 1; 93 new_bio_slabs = krealloc(bio_slabs, 94 new_bio_slab_max * sizeof(struct bio_slab), 95 GFP_KERNEL); 96 if (!new_bio_slabs) 97 goto out_unlock; 98 bio_slab_max = new_bio_slab_max; 99 bio_slabs = new_bio_slabs; 100 } 101 if (entry == -1) 102 entry = bio_slab_nr++; 103 104 bslab = &bio_slabs[entry]; 105 106 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 107 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 108 SLAB_HWCACHE_ALIGN, NULL); 109 if (!slab) 110 goto out_unlock; 111 112 bslab->slab = slab; 113 bslab->slab_ref = 1; 114 bslab->slab_size = sz; 115 out_unlock: 116 mutex_unlock(&bio_slab_lock); 117 return slab; 118 } 119 120 static void bio_put_slab(struct bio_set *bs) 121 { 122 struct bio_slab *bslab = NULL; 123 unsigned int i; 124 125 mutex_lock(&bio_slab_lock); 126 127 for (i = 0; i < bio_slab_nr; i++) { 128 if (bs->bio_slab == bio_slabs[i].slab) { 129 bslab = &bio_slabs[i]; 130 break; 131 } 132 } 133 134 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 135 goto out; 136 137 WARN_ON(!bslab->slab_ref); 138 139 if (--bslab->slab_ref) 140 goto out; 141 142 kmem_cache_destroy(bslab->slab); 143 bslab->slab = NULL; 144 145 out: 146 mutex_unlock(&bio_slab_lock); 147 } 148 149 unsigned int bvec_nr_vecs(unsigned short idx) 150 { 151 return bvec_slabs[--idx].nr_vecs; 152 } 153 154 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 155 { 156 if (!idx) 157 return; 158 idx--; 159 160 BIO_BUG_ON(idx >= BVEC_POOL_NR); 161 162 if (idx == BVEC_POOL_MAX) { 163 mempool_free(bv, pool); 164 } else { 165 struct biovec_slab *bvs = bvec_slabs + idx; 166 167 kmem_cache_free(bvs->slab, bv); 168 } 169 } 170 171 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 172 mempool_t *pool) 173 { 174 struct bio_vec *bvl; 175 176 /* 177 * see comment near bvec_array define! 178 */ 179 switch (nr) { 180 case 1: 181 *idx = 0; 182 break; 183 case 2 ... 4: 184 *idx = 1; 185 break; 186 case 5 ... 16: 187 *idx = 2; 188 break; 189 case 17 ... 64: 190 *idx = 3; 191 break; 192 case 65 ... 128: 193 *idx = 4; 194 break; 195 case 129 ... BIO_MAX_PAGES: 196 *idx = 5; 197 break; 198 default: 199 return NULL; 200 } 201 202 /* 203 * idx now points to the pool we want to allocate from. only the 204 * 1-vec entry pool is mempool backed. 205 */ 206 if (*idx == BVEC_POOL_MAX) { 207 fallback: 208 bvl = mempool_alloc(pool, gfp_mask); 209 } else { 210 struct biovec_slab *bvs = bvec_slabs + *idx; 211 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 212 213 /* 214 * Make this allocation restricted and don't dump info on 215 * allocation failures, since we'll fallback to the mempool 216 * in case of failure. 217 */ 218 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 219 220 /* 221 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 222 * is set, retry with the 1-entry mempool 223 */ 224 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 225 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 226 *idx = BVEC_POOL_MAX; 227 goto fallback; 228 } 229 } 230 231 (*idx)++; 232 return bvl; 233 } 234 235 void bio_uninit(struct bio *bio) 236 { 237 #ifdef CONFIG_BLK_CGROUP 238 if (bio->bi_blkg) { 239 blkg_put(bio->bi_blkg); 240 bio->bi_blkg = NULL; 241 } 242 #endif 243 if (bio_integrity(bio)) 244 bio_integrity_free(bio); 245 246 bio_crypt_free_ctx(bio); 247 } 248 EXPORT_SYMBOL(bio_uninit); 249 250 static void bio_free(struct bio *bio) 251 { 252 struct bio_set *bs = bio->bi_pool; 253 void *p; 254 255 bio_uninit(bio); 256 257 if (bs) { 258 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 259 260 /* 261 * If we have front padding, adjust the bio pointer before freeing 262 */ 263 p = bio; 264 p -= bs->front_pad; 265 266 mempool_free(p, &bs->bio_pool); 267 } else { 268 /* Bio was allocated by bio_kmalloc() */ 269 kfree(bio); 270 } 271 } 272 273 /* 274 * Users of this function have their own bio allocation. Subsequently, 275 * they must remember to pair any call to bio_init() with bio_uninit() 276 * when IO has completed, or when the bio is released. 277 */ 278 void bio_init(struct bio *bio, struct bio_vec *table, 279 unsigned short max_vecs) 280 { 281 memset(bio, 0, sizeof(*bio)); 282 atomic_set(&bio->__bi_remaining, 1); 283 atomic_set(&bio->__bi_cnt, 1); 284 285 bio->bi_io_vec = table; 286 bio->bi_max_vecs = max_vecs; 287 } 288 EXPORT_SYMBOL(bio_init); 289 290 /** 291 * bio_reset - reinitialize a bio 292 * @bio: bio to reset 293 * 294 * Description: 295 * After calling bio_reset(), @bio will be in the same state as a freshly 296 * allocated bio returned bio bio_alloc_bioset() - the only fields that are 297 * preserved are the ones that are initialized by bio_alloc_bioset(). See 298 * comment in struct bio. 299 */ 300 void bio_reset(struct bio *bio) 301 { 302 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 303 304 bio_uninit(bio); 305 306 memset(bio, 0, BIO_RESET_BYTES); 307 bio->bi_flags = flags; 308 atomic_set(&bio->__bi_remaining, 1); 309 } 310 EXPORT_SYMBOL(bio_reset); 311 312 static struct bio *__bio_chain_endio(struct bio *bio) 313 { 314 struct bio *parent = bio->bi_private; 315 316 if (!parent->bi_status) 317 parent->bi_status = bio->bi_status; 318 bio_put(bio); 319 return parent; 320 } 321 322 static void bio_chain_endio(struct bio *bio) 323 { 324 bio_endio(__bio_chain_endio(bio)); 325 } 326 327 /** 328 * bio_chain - chain bio completions 329 * @bio: the target bio 330 * @parent: the parent bio of @bio 331 * 332 * The caller won't have a bi_end_io called when @bio completes - instead, 333 * @parent's bi_end_io won't be called until both @parent and @bio have 334 * completed; the chained bio will also be freed when it completes. 335 * 336 * The caller must not set bi_private or bi_end_io in @bio. 337 */ 338 void bio_chain(struct bio *bio, struct bio *parent) 339 { 340 BUG_ON(bio->bi_private || bio->bi_end_io); 341 342 bio->bi_private = parent; 343 bio->bi_end_io = bio_chain_endio; 344 bio_inc_remaining(parent); 345 } 346 EXPORT_SYMBOL(bio_chain); 347 348 static void bio_alloc_rescue(struct work_struct *work) 349 { 350 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 351 struct bio *bio; 352 353 while (1) { 354 spin_lock(&bs->rescue_lock); 355 bio = bio_list_pop(&bs->rescue_list); 356 spin_unlock(&bs->rescue_lock); 357 358 if (!bio) 359 break; 360 361 submit_bio_noacct(bio); 362 } 363 } 364 365 static void punt_bios_to_rescuer(struct bio_set *bs) 366 { 367 struct bio_list punt, nopunt; 368 struct bio *bio; 369 370 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 371 return; 372 /* 373 * In order to guarantee forward progress we must punt only bios that 374 * were allocated from this bio_set; otherwise, if there was a bio on 375 * there for a stacking driver higher up in the stack, processing it 376 * could require allocating bios from this bio_set, and doing that from 377 * our own rescuer would be bad. 378 * 379 * Since bio lists are singly linked, pop them all instead of trying to 380 * remove from the middle of the list: 381 */ 382 383 bio_list_init(&punt); 384 bio_list_init(&nopunt); 385 386 while ((bio = bio_list_pop(¤t->bio_list[0]))) 387 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 388 current->bio_list[0] = nopunt; 389 390 bio_list_init(&nopunt); 391 while ((bio = bio_list_pop(¤t->bio_list[1]))) 392 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 393 current->bio_list[1] = nopunt; 394 395 spin_lock(&bs->rescue_lock); 396 bio_list_merge(&bs->rescue_list, &punt); 397 spin_unlock(&bs->rescue_lock); 398 399 queue_work(bs->rescue_workqueue, &bs->rescue_work); 400 } 401 402 /** 403 * bio_alloc_bioset - allocate a bio for I/O 404 * @gfp_mask: the GFP_* mask given to the slab allocator 405 * @nr_iovecs: number of iovecs to pre-allocate 406 * @bs: the bio_set to allocate from. 407 * 408 * Description: 409 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 410 * backed by the @bs's mempool. 411 * 412 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 413 * always be able to allocate a bio. This is due to the mempool guarantees. 414 * To make this work, callers must never allocate more than 1 bio at a time 415 * from this pool. Callers that need to allocate more than 1 bio must always 416 * submit the previously allocated bio for IO before attempting to allocate 417 * a new one. Failure to do so can cause deadlocks under memory pressure. 418 * 419 * Note that when running under submit_bio_noacct() (i.e. any block 420 * driver), bios are not submitted until after you return - see the code in 421 * submit_bio_noacct() that converts recursion into iteration, to prevent 422 * stack overflows. 423 * 424 * This would normally mean allocating multiple bios under 425 * submit_bio_noacct() would be susceptible to deadlocks, but we have 426 * deadlock avoidance code that resubmits any blocked bios from a rescuer 427 * thread. 428 * 429 * However, we do not guarantee forward progress for allocations from other 430 * mempools. Doing multiple allocations from the same mempool under 431 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 432 * for per bio allocations. 433 * 434 * RETURNS: 435 * Pointer to new bio on success, NULL on failure. 436 */ 437 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 438 struct bio_set *bs) 439 { 440 gfp_t saved_gfp = gfp_mask; 441 unsigned front_pad; 442 unsigned inline_vecs; 443 struct bio_vec *bvl = NULL; 444 struct bio *bio; 445 void *p; 446 447 if (!bs) { 448 if (nr_iovecs > UIO_MAXIOV) 449 return NULL; 450 451 p = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 452 front_pad = 0; 453 inline_vecs = nr_iovecs; 454 } else { 455 /* should not use nobvec bioset for nr_iovecs > 0 */ 456 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 457 nr_iovecs > 0)) 458 return NULL; 459 /* 460 * submit_bio_noacct() converts recursion to iteration; this 461 * means if we're running beneath it, any bios we allocate and 462 * submit will not be submitted (and thus freed) until after we 463 * return. 464 * 465 * This exposes us to a potential deadlock if we allocate 466 * multiple bios from the same bio_set() while running 467 * underneath submit_bio_noacct(). If we were to allocate 468 * multiple bios (say a stacking block driver that was splitting 469 * bios), we would deadlock if we exhausted the mempool's 470 * reserve. 471 * 472 * We solve this, and guarantee forward progress, with a rescuer 473 * workqueue per bio_set. If we go to allocate and there are 474 * bios on current->bio_list, we first try the allocation 475 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 476 * bios we would be blocking to the rescuer workqueue before 477 * we retry with the original gfp_flags. 478 */ 479 480 if (current->bio_list && 481 (!bio_list_empty(¤t->bio_list[0]) || 482 !bio_list_empty(¤t->bio_list[1])) && 483 bs->rescue_workqueue) 484 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 485 486 p = mempool_alloc(&bs->bio_pool, gfp_mask); 487 if (!p && gfp_mask != saved_gfp) { 488 punt_bios_to_rescuer(bs); 489 gfp_mask = saved_gfp; 490 p = mempool_alloc(&bs->bio_pool, gfp_mask); 491 } 492 493 front_pad = bs->front_pad; 494 inline_vecs = BIO_INLINE_VECS; 495 } 496 497 if (unlikely(!p)) 498 return NULL; 499 500 bio = p + front_pad; 501 bio_init(bio, NULL, 0); 502 503 if (nr_iovecs > inline_vecs) { 504 unsigned long idx = 0; 505 506 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 507 if (!bvl && gfp_mask != saved_gfp) { 508 punt_bios_to_rescuer(bs); 509 gfp_mask = saved_gfp; 510 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 511 } 512 513 if (unlikely(!bvl)) 514 goto err_free; 515 516 bio->bi_flags |= idx << BVEC_POOL_OFFSET; 517 } else if (nr_iovecs) { 518 bvl = bio->bi_inline_vecs; 519 } 520 521 bio->bi_pool = bs; 522 bio->bi_max_vecs = nr_iovecs; 523 bio->bi_io_vec = bvl; 524 return bio; 525 526 err_free: 527 mempool_free(p, &bs->bio_pool); 528 return NULL; 529 } 530 EXPORT_SYMBOL(bio_alloc_bioset); 531 532 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 533 { 534 unsigned long flags; 535 struct bio_vec bv; 536 struct bvec_iter iter; 537 538 __bio_for_each_segment(bv, bio, iter, start) { 539 char *data = bvec_kmap_irq(&bv, &flags); 540 memset(data, 0, bv.bv_len); 541 flush_dcache_page(bv.bv_page); 542 bvec_kunmap_irq(data, &flags); 543 } 544 } 545 EXPORT_SYMBOL(zero_fill_bio_iter); 546 547 /** 548 * bio_truncate - truncate the bio to small size of @new_size 549 * @bio: the bio to be truncated 550 * @new_size: new size for truncating the bio 551 * 552 * Description: 553 * Truncate the bio to new size of @new_size. If bio_op(bio) is 554 * REQ_OP_READ, zero the truncated part. This function should only 555 * be used for handling corner cases, such as bio eod. 556 */ 557 void bio_truncate(struct bio *bio, unsigned new_size) 558 { 559 struct bio_vec bv; 560 struct bvec_iter iter; 561 unsigned int done = 0; 562 bool truncated = false; 563 564 if (new_size >= bio->bi_iter.bi_size) 565 return; 566 567 if (bio_op(bio) != REQ_OP_READ) 568 goto exit; 569 570 bio_for_each_segment(bv, bio, iter) { 571 if (done + bv.bv_len > new_size) { 572 unsigned offset; 573 574 if (!truncated) 575 offset = new_size - done; 576 else 577 offset = 0; 578 zero_user(bv.bv_page, offset, bv.bv_len - offset); 579 truncated = true; 580 } 581 done += bv.bv_len; 582 } 583 584 exit: 585 /* 586 * Don't touch bvec table here and make it really immutable, since 587 * fs bio user has to retrieve all pages via bio_for_each_segment_all 588 * in its .end_bio() callback. 589 * 590 * It is enough to truncate bio by updating .bi_size since we can make 591 * correct bvec with the updated .bi_size for drivers. 592 */ 593 bio->bi_iter.bi_size = new_size; 594 } 595 596 /** 597 * guard_bio_eod - truncate a BIO to fit the block device 598 * @bio: bio to truncate 599 * 600 * This allows us to do IO even on the odd last sectors of a device, even if the 601 * block size is some multiple of the physical sector size. 602 * 603 * We'll just truncate the bio to the size of the device, and clear the end of 604 * the buffer head manually. Truly out-of-range accesses will turn into actual 605 * I/O errors, this only handles the "we need to be able to do I/O at the final 606 * sector" case. 607 */ 608 void guard_bio_eod(struct bio *bio) 609 { 610 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 611 612 if (!maxsector) 613 return; 614 615 /* 616 * If the *whole* IO is past the end of the device, 617 * let it through, and the IO layer will turn it into 618 * an EIO. 619 */ 620 if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 621 return; 622 623 maxsector -= bio->bi_iter.bi_sector; 624 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 625 return; 626 627 bio_truncate(bio, maxsector << 9); 628 } 629 630 /** 631 * bio_put - release a reference to a bio 632 * @bio: bio to release reference to 633 * 634 * Description: 635 * Put a reference to a &struct bio, either one you have gotten with 636 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 637 **/ 638 void bio_put(struct bio *bio) 639 { 640 if (!bio_flagged(bio, BIO_REFFED)) 641 bio_free(bio); 642 else { 643 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 644 645 /* 646 * last put frees it 647 */ 648 if (atomic_dec_and_test(&bio->__bi_cnt)) 649 bio_free(bio); 650 } 651 } 652 EXPORT_SYMBOL(bio_put); 653 654 /** 655 * __bio_clone_fast - clone a bio that shares the original bio's biovec 656 * @bio: destination bio 657 * @bio_src: bio to clone 658 * 659 * Clone a &bio. Caller will own the returned bio, but not 660 * the actual data it points to. Reference count of returned 661 * bio will be one. 662 * 663 * Caller must ensure that @bio_src is not freed before @bio. 664 */ 665 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 666 { 667 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 668 669 /* 670 * most users will be overriding ->bi_bdev with a new target, 671 * so we don't set nor calculate new physical/hw segment counts here 672 */ 673 bio->bi_bdev = bio_src->bi_bdev; 674 bio_set_flag(bio, BIO_CLONED); 675 if (bio_flagged(bio_src, BIO_THROTTLED)) 676 bio_set_flag(bio, BIO_THROTTLED); 677 bio->bi_opf = bio_src->bi_opf; 678 bio->bi_ioprio = bio_src->bi_ioprio; 679 bio->bi_write_hint = bio_src->bi_write_hint; 680 bio->bi_iter = bio_src->bi_iter; 681 bio->bi_io_vec = bio_src->bi_io_vec; 682 683 bio_clone_blkg_association(bio, bio_src); 684 blkcg_bio_issue_init(bio); 685 } 686 EXPORT_SYMBOL(__bio_clone_fast); 687 688 /** 689 * bio_clone_fast - clone a bio that shares the original bio's biovec 690 * @bio: bio to clone 691 * @gfp_mask: allocation priority 692 * @bs: bio_set to allocate from 693 * 694 * Like __bio_clone_fast, only also allocates the returned bio 695 */ 696 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 697 { 698 struct bio *b; 699 700 b = bio_alloc_bioset(gfp_mask, 0, bs); 701 if (!b) 702 return NULL; 703 704 __bio_clone_fast(b, bio); 705 706 if (bio_crypt_clone(b, bio, gfp_mask) < 0) 707 goto err_put; 708 709 if (bio_integrity(bio) && 710 bio_integrity_clone(b, bio, gfp_mask) < 0) 711 goto err_put; 712 713 return b; 714 715 err_put: 716 bio_put(b); 717 return NULL; 718 } 719 EXPORT_SYMBOL(bio_clone_fast); 720 721 const char *bio_devname(struct bio *bio, char *buf) 722 { 723 return bdevname(bio->bi_bdev, buf); 724 } 725 EXPORT_SYMBOL(bio_devname); 726 727 static inline bool page_is_mergeable(const struct bio_vec *bv, 728 struct page *page, unsigned int len, unsigned int off, 729 bool *same_page) 730 { 731 size_t bv_end = bv->bv_offset + bv->bv_len; 732 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 733 phys_addr_t page_addr = page_to_phys(page); 734 735 if (vec_end_addr + 1 != page_addr + off) 736 return false; 737 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 738 return false; 739 740 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 741 if (*same_page) 742 return true; 743 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 744 } 745 746 /* 747 * Try to merge a page into a segment, while obeying the hardware segment 748 * size limit. This is not for normal read/write bios, but for passthrough 749 * or Zone Append operations that we can't split. 750 */ 751 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 752 struct page *page, unsigned len, 753 unsigned offset, bool *same_page) 754 { 755 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 756 unsigned long mask = queue_segment_boundary(q); 757 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 758 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 759 760 if ((addr1 | mask) != (addr2 | mask)) 761 return false; 762 if (bv->bv_len + len > queue_max_segment_size(q)) 763 return false; 764 return __bio_try_merge_page(bio, page, len, offset, same_page); 765 } 766 767 /** 768 * bio_add_hw_page - attempt to add a page to a bio with hw constraints 769 * @q: the target queue 770 * @bio: destination bio 771 * @page: page to add 772 * @len: vec entry length 773 * @offset: vec entry offset 774 * @max_sectors: maximum number of sectors that can be added 775 * @same_page: return if the segment has been merged inside the same page 776 * 777 * Add a page to a bio while respecting the hardware max_sectors, max_segment 778 * and gap limitations. 779 */ 780 int bio_add_hw_page(struct request_queue *q, struct bio *bio, 781 struct page *page, unsigned int len, unsigned int offset, 782 unsigned int max_sectors, bool *same_page) 783 { 784 struct bio_vec *bvec; 785 786 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 787 return 0; 788 789 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 790 return 0; 791 792 if (bio->bi_vcnt > 0) { 793 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 794 return len; 795 796 /* 797 * If the queue doesn't support SG gaps and adding this segment 798 * would create a gap, disallow it. 799 */ 800 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 801 if (bvec_gap_to_prev(q, bvec, offset)) 802 return 0; 803 } 804 805 if (bio_full(bio, len)) 806 return 0; 807 808 if (bio->bi_vcnt >= queue_max_segments(q)) 809 return 0; 810 811 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 812 bvec->bv_page = page; 813 bvec->bv_len = len; 814 bvec->bv_offset = offset; 815 bio->bi_vcnt++; 816 bio->bi_iter.bi_size += len; 817 return len; 818 } 819 820 /** 821 * bio_add_pc_page - attempt to add page to passthrough bio 822 * @q: the target queue 823 * @bio: destination bio 824 * @page: page to add 825 * @len: vec entry length 826 * @offset: vec entry offset 827 * 828 * Attempt to add a page to the bio_vec maplist. This can fail for a 829 * number of reasons, such as the bio being full or target block device 830 * limitations. The target block device must allow bio's up to PAGE_SIZE, 831 * so it is always possible to add a single page to an empty bio. 832 * 833 * This should only be used by passthrough bios. 834 */ 835 int bio_add_pc_page(struct request_queue *q, struct bio *bio, 836 struct page *page, unsigned int len, unsigned int offset) 837 { 838 bool same_page = false; 839 return bio_add_hw_page(q, bio, page, len, offset, 840 queue_max_hw_sectors(q), &same_page); 841 } 842 EXPORT_SYMBOL(bio_add_pc_page); 843 844 /** 845 * __bio_try_merge_page - try appending data to an existing bvec. 846 * @bio: destination bio 847 * @page: start page to add 848 * @len: length of the data to add 849 * @off: offset of the data relative to @page 850 * @same_page: return if the segment has been merged inside the same page 851 * 852 * Try to add the data at @page + @off to the last bvec of @bio. This is a 853 * useful optimisation for file systems with a block size smaller than the 854 * page size. 855 * 856 * Warn if (@len, @off) crosses pages in case that @same_page is true. 857 * 858 * Return %true on success or %false on failure. 859 */ 860 bool __bio_try_merge_page(struct bio *bio, struct page *page, 861 unsigned int len, unsigned int off, bool *same_page) 862 { 863 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 864 return false; 865 866 if (bio->bi_vcnt > 0) { 867 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 868 869 if (page_is_mergeable(bv, page, len, off, same_page)) { 870 if (bio->bi_iter.bi_size > UINT_MAX - len) { 871 *same_page = false; 872 return false; 873 } 874 bv->bv_len += len; 875 bio->bi_iter.bi_size += len; 876 return true; 877 } 878 } 879 return false; 880 } 881 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 882 883 /** 884 * __bio_add_page - add page(s) to a bio in a new segment 885 * @bio: destination bio 886 * @page: start page to add 887 * @len: length of the data to add, may cross pages 888 * @off: offset of the data relative to @page, may cross pages 889 * 890 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 891 * that @bio has space for another bvec. 892 */ 893 void __bio_add_page(struct bio *bio, struct page *page, 894 unsigned int len, unsigned int off) 895 { 896 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 897 898 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 899 WARN_ON_ONCE(bio_full(bio, len)); 900 901 bv->bv_page = page; 902 bv->bv_offset = off; 903 bv->bv_len = len; 904 905 bio->bi_iter.bi_size += len; 906 bio->bi_vcnt++; 907 908 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 909 bio_set_flag(bio, BIO_WORKINGSET); 910 } 911 EXPORT_SYMBOL_GPL(__bio_add_page); 912 913 /** 914 * bio_add_page - attempt to add page(s) to bio 915 * @bio: destination bio 916 * @page: start page to add 917 * @len: vec entry length, may cross pages 918 * @offset: vec entry offset relative to @page, may cross pages 919 * 920 * Attempt to add page(s) to the bio_vec maplist. This will only fail 921 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 922 */ 923 int bio_add_page(struct bio *bio, struct page *page, 924 unsigned int len, unsigned int offset) 925 { 926 bool same_page = false; 927 928 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 929 if (bio_full(bio, len)) 930 return 0; 931 __bio_add_page(bio, page, len, offset); 932 } 933 return len; 934 } 935 EXPORT_SYMBOL(bio_add_page); 936 937 void bio_release_pages(struct bio *bio, bool mark_dirty) 938 { 939 struct bvec_iter_all iter_all; 940 struct bio_vec *bvec; 941 942 if (bio_flagged(bio, BIO_NO_PAGE_REF)) 943 return; 944 945 bio_for_each_segment_all(bvec, bio, iter_all) { 946 if (mark_dirty && !PageCompound(bvec->bv_page)) 947 set_page_dirty_lock(bvec->bv_page); 948 put_page(bvec->bv_page); 949 } 950 } 951 EXPORT_SYMBOL_GPL(bio_release_pages); 952 953 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 954 { 955 const struct bio_vec *bv = iter->bvec; 956 unsigned int len; 957 size_t size; 958 959 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 960 return -EINVAL; 961 962 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 963 size = bio_add_page(bio, bv->bv_page, len, 964 bv->bv_offset + iter->iov_offset); 965 if (unlikely(size != len)) 966 return -EINVAL; 967 iov_iter_advance(iter, size); 968 return 0; 969 } 970 971 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 972 973 /** 974 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 975 * @bio: bio to add pages to 976 * @iter: iov iterator describing the region to be mapped 977 * 978 * Pins pages from *iter and appends them to @bio's bvec array. The 979 * pages will have to be released using put_page() when done. 980 * For multi-segment *iter, this function only adds pages from the 981 * next non-empty segment of the iov iterator. 982 */ 983 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 984 { 985 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 986 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 987 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 988 struct page **pages = (struct page **)bv; 989 bool same_page = false; 990 ssize_t size, left; 991 unsigned len, i; 992 size_t offset; 993 994 /* 995 * Move page array up in the allocated memory for the bio vecs as far as 996 * possible so that we can start filling biovecs from the beginning 997 * without overwriting the temporary page array. 998 */ 999 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1000 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1001 1002 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 1003 if (unlikely(size <= 0)) 1004 return size ? size : -EFAULT; 1005 1006 for (left = size, i = 0; left > 0; left -= len, i++) { 1007 struct page *page = pages[i]; 1008 1009 len = min_t(size_t, PAGE_SIZE - offset, left); 1010 1011 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1012 if (same_page) 1013 put_page(page); 1014 } else { 1015 if (WARN_ON_ONCE(bio_full(bio, len))) 1016 return -EINVAL; 1017 __bio_add_page(bio, page, len, offset); 1018 } 1019 offset = 0; 1020 } 1021 1022 iov_iter_advance(iter, size); 1023 return 0; 1024 } 1025 1026 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 1027 { 1028 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1029 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1030 struct request_queue *q = bio->bi_bdev->bd_disk->queue; 1031 unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 1032 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 1033 struct page **pages = (struct page **)bv; 1034 ssize_t size, left; 1035 unsigned len, i; 1036 size_t offset; 1037 int ret = 0; 1038 1039 if (WARN_ON_ONCE(!max_append_sectors)) 1040 return 0; 1041 1042 /* 1043 * Move page array up in the allocated memory for the bio vecs as far as 1044 * possible so that we can start filling biovecs from the beginning 1045 * without overwriting the temporary page array. 1046 */ 1047 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1048 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1049 1050 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 1051 if (unlikely(size <= 0)) 1052 return size ? size : -EFAULT; 1053 1054 for (left = size, i = 0; left > 0; left -= len, i++) { 1055 struct page *page = pages[i]; 1056 bool same_page = false; 1057 1058 len = min_t(size_t, PAGE_SIZE - offset, left); 1059 if (bio_add_hw_page(q, bio, page, len, offset, 1060 max_append_sectors, &same_page) != len) { 1061 ret = -EINVAL; 1062 break; 1063 } 1064 if (same_page) 1065 put_page(page); 1066 offset = 0; 1067 } 1068 1069 iov_iter_advance(iter, size - left); 1070 return ret; 1071 } 1072 1073 /** 1074 * bio_iov_iter_get_pages - add user or kernel pages to a bio 1075 * @bio: bio to add pages to 1076 * @iter: iov iterator describing the region to be added 1077 * 1078 * This takes either an iterator pointing to user memory, or one pointing to 1079 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 1080 * map them into the kernel. On IO completion, the caller should put those 1081 * pages. If we're adding kernel pages, and the caller told us it's safe to 1082 * do so, we just have to add the pages to the bio directly. We don't grab an 1083 * extra reference to those pages (the user should already have that), and we 1084 * don't put the page on IO completion. The caller needs to check if the bio is 1085 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 1086 * released. 1087 * 1088 * The function tries, but does not guarantee, to pin as many pages as 1089 * fit into the bio, or are requested in @iter, whatever is smaller. If 1090 * MM encounters an error pinning the requested pages, it stops. Error 1091 * is returned only if 0 pages could be pinned. 1092 */ 1093 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1094 { 1095 const bool is_bvec = iov_iter_is_bvec(iter); 1096 int ret; 1097 1098 if (WARN_ON_ONCE(bio->bi_vcnt)) 1099 return -EINVAL; 1100 1101 do { 1102 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1103 if (WARN_ON_ONCE(is_bvec)) 1104 return -EINVAL; 1105 ret = __bio_iov_append_get_pages(bio, iter); 1106 } else { 1107 if (is_bvec) 1108 ret = __bio_iov_bvec_add_pages(bio, iter); 1109 else 1110 ret = __bio_iov_iter_get_pages(bio, iter); 1111 } 1112 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1113 1114 if (is_bvec) 1115 bio_set_flag(bio, BIO_NO_PAGE_REF); 1116 return bio->bi_vcnt ? 0 : ret; 1117 } 1118 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 1119 1120 static void submit_bio_wait_endio(struct bio *bio) 1121 { 1122 complete(bio->bi_private); 1123 } 1124 1125 /** 1126 * submit_bio_wait - submit a bio, and wait until it completes 1127 * @bio: The &struct bio which describes the I/O 1128 * 1129 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1130 * bio_endio() on failure. 1131 * 1132 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1133 * result in bio reference to be consumed. The caller must drop the reference 1134 * on his own. 1135 */ 1136 int submit_bio_wait(struct bio *bio) 1137 { 1138 DECLARE_COMPLETION_ONSTACK_MAP(done, 1139 bio->bi_bdev->bd_disk->lockdep_map); 1140 unsigned long hang_check; 1141 1142 bio->bi_private = &done; 1143 bio->bi_end_io = submit_bio_wait_endio; 1144 bio->bi_opf |= REQ_SYNC; 1145 submit_bio(bio); 1146 1147 /* Prevent hang_check timer from firing at us during very long I/O */ 1148 hang_check = sysctl_hung_task_timeout_secs; 1149 if (hang_check) 1150 while (!wait_for_completion_io_timeout(&done, 1151 hang_check * (HZ/2))) 1152 ; 1153 else 1154 wait_for_completion_io(&done); 1155 1156 return blk_status_to_errno(bio->bi_status); 1157 } 1158 EXPORT_SYMBOL(submit_bio_wait); 1159 1160 /** 1161 * bio_advance - increment/complete a bio by some number of bytes 1162 * @bio: bio to advance 1163 * @bytes: number of bytes to complete 1164 * 1165 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1166 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1167 * be updated on the last bvec as well. 1168 * 1169 * @bio will then represent the remaining, uncompleted portion of the io. 1170 */ 1171 void bio_advance(struct bio *bio, unsigned bytes) 1172 { 1173 if (bio_integrity(bio)) 1174 bio_integrity_advance(bio, bytes); 1175 1176 bio_crypt_advance(bio, bytes); 1177 bio_advance_iter(bio, &bio->bi_iter, bytes); 1178 } 1179 EXPORT_SYMBOL(bio_advance); 1180 1181 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1182 struct bio *src, struct bvec_iter *src_iter) 1183 { 1184 struct bio_vec src_bv, dst_bv; 1185 void *src_p, *dst_p; 1186 unsigned bytes; 1187 1188 while (src_iter->bi_size && dst_iter->bi_size) { 1189 src_bv = bio_iter_iovec(src, *src_iter); 1190 dst_bv = bio_iter_iovec(dst, *dst_iter); 1191 1192 bytes = min(src_bv.bv_len, dst_bv.bv_len); 1193 1194 src_p = kmap_atomic(src_bv.bv_page); 1195 dst_p = kmap_atomic(dst_bv.bv_page); 1196 1197 memcpy(dst_p + dst_bv.bv_offset, 1198 src_p + src_bv.bv_offset, 1199 bytes); 1200 1201 kunmap_atomic(dst_p); 1202 kunmap_atomic(src_p); 1203 1204 flush_dcache_page(dst_bv.bv_page); 1205 1206 bio_advance_iter_single(src, src_iter, bytes); 1207 bio_advance_iter_single(dst, dst_iter, bytes); 1208 } 1209 } 1210 EXPORT_SYMBOL(bio_copy_data_iter); 1211 1212 /** 1213 * bio_copy_data - copy contents of data buffers from one bio to another 1214 * @src: source bio 1215 * @dst: destination bio 1216 * 1217 * Stops when it reaches the end of either @src or @dst - that is, copies 1218 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1219 */ 1220 void bio_copy_data(struct bio *dst, struct bio *src) 1221 { 1222 struct bvec_iter src_iter = src->bi_iter; 1223 struct bvec_iter dst_iter = dst->bi_iter; 1224 1225 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1226 } 1227 EXPORT_SYMBOL(bio_copy_data); 1228 1229 /** 1230 * bio_list_copy_data - copy contents of data buffers from one chain of bios to 1231 * another 1232 * @src: source bio list 1233 * @dst: destination bio list 1234 * 1235 * Stops when it reaches the end of either the @src list or @dst list - that is, 1236 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 1237 * bios). 1238 */ 1239 void bio_list_copy_data(struct bio *dst, struct bio *src) 1240 { 1241 struct bvec_iter src_iter = src->bi_iter; 1242 struct bvec_iter dst_iter = dst->bi_iter; 1243 1244 while (1) { 1245 if (!src_iter.bi_size) { 1246 src = src->bi_next; 1247 if (!src) 1248 break; 1249 1250 src_iter = src->bi_iter; 1251 } 1252 1253 if (!dst_iter.bi_size) { 1254 dst = dst->bi_next; 1255 if (!dst) 1256 break; 1257 1258 dst_iter = dst->bi_iter; 1259 } 1260 1261 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1262 } 1263 } 1264 EXPORT_SYMBOL(bio_list_copy_data); 1265 1266 void bio_free_pages(struct bio *bio) 1267 { 1268 struct bio_vec *bvec; 1269 struct bvec_iter_all iter_all; 1270 1271 bio_for_each_segment_all(bvec, bio, iter_all) 1272 __free_page(bvec->bv_page); 1273 } 1274 EXPORT_SYMBOL(bio_free_pages); 1275 1276 /* 1277 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1278 * for performing direct-IO in BIOs. 1279 * 1280 * The problem is that we cannot run set_page_dirty() from interrupt context 1281 * because the required locks are not interrupt-safe. So what we can do is to 1282 * mark the pages dirty _before_ performing IO. And in interrupt context, 1283 * check that the pages are still dirty. If so, fine. If not, redirty them 1284 * in process context. 1285 * 1286 * We special-case compound pages here: normally this means reads into hugetlb 1287 * pages. The logic in here doesn't really work right for compound pages 1288 * because the VM does not uniformly chase down the head page in all cases. 1289 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1290 * handle them at all. So we skip compound pages here at an early stage. 1291 * 1292 * Note that this code is very hard to test under normal circumstances because 1293 * direct-io pins the pages with get_user_pages(). This makes 1294 * is_page_cache_freeable return false, and the VM will not clean the pages. 1295 * But other code (eg, flusher threads) could clean the pages if they are mapped 1296 * pagecache. 1297 * 1298 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1299 * deferred bio dirtying paths. 1300 */ 1301 1302 /* 1303 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1304 */ 1305 void bio_set_pages_dirty(struct bio *bio) 1306 { 1307 struct bio_vec *bvec; 1308 struct bvec_iter_all iter_all; 1309 1310 bio_for_each_segment_all(bvec, bio, iter_all) { 1311 if (!PageCompound(bvec->bv_page)) 1312 set_page_dirty_lock(bvec->bv_page); 1313 } 1314 } 1315 1316 /* 1317 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1318 * If they are, then fine. If, however, some pages are clean then they must 1319 * have been written out during the direct-IO read. So we take another ref on 1320 * the BIO and re-dirty the pages in process context. 1321 * 1322 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1323 * here on. It will run one put_page() against each page and will run one 1324 * bio_put() against the BIO. 1325 */ 1326 1327 static void bio_dirty_fn(struct work_struct *work); 1328 1329 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1330 static DEFINE_SPINLOCK(bio_dirty_lock); 1331 static struct bio *bio_dirty_list; 1332 1333 /* 1334 * This runs in process context 1335 */ 1336 static void bio_dirty_fn(struct work_struct *work) 1337 { 1338 struct bio *bio, *next; 1339 1340 spin_lock_irq(&bio_dirty_lock); 1341 next = bio_dirty_list; 1342 bio_dirty_list = NULL; 1343 spin_unlock_irq(&bio_dirty_lock); 1344 1345 while ((bio = next) != NULL) { 1346 next = bio->bi_private; 1347 1348 bio_release_pages(bio, true); 1349 bio_put(bio); 1350 } 1351 } 1352 1353 void bio_check_pages_dirty(struct bio *bio) 1354 { 1355 struct bio_vec *bvec; 1356 unsigned long flags; 1357 struct bvec_iter_all iter_all; 1358 1359 bio_for_each_segment_all(bvec, bio, iter_all) { 1360 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1361 goto defer; 1362 } 1363 1364 bio_release_pages(bio, false); 1365 bio_put(bio); 1366 return; 1367 defer: 1368 spin_lock_irqsave(&bio_dirty_lock, flags); 1369 bio->bi_private = bio_dirty_list; 1370 bio_dirty_list = bio; 1371 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1372 schedule_work(&bio_dirty_work); 1373 } 1374 1375 static inline bool bio_remaining_done(struct bio *bio) 1376 { 1377 /* 1378 * If we're not chaining, then ->__bi_remaining is always 1 and 1379 * we always end io on the first invocation. 1380 */ 1381 if (!bio_flagged(bio, BIO_CHAIN)) 1382 return true; 1383 1384 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1385 1386 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1387 bio_clear_flag(bio, BIO_CHAIN); 1388 return true; 1389 } 1390 1391 return false; 1392 } 1393 1394 /** 1395 * bio_endio - end I/O on a bio 1396 * @bio: bio 1397 * 1398 * Description: 1399 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1400 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1401 * bio unless they own it and thus know that it has an end_io function. 1402 * 1403 * bio_endio() can be called several times on a bio that has been chained 1404 * using bio_chain(). The ->bi_end_io() function will only be called the 1405 * last time. At this point the BLK_TA_COMPLETE tracing event will be 1406 * generated if BIO_TRACE_COMPLETION is set. 1407 **/ 1408 void bio_endio(struct bio *bio) 1409 { 1410 again: 1411 if (!bio_remaining_done(bio)) 1412 return; 1413 if (!bio_integrity_endio(bio)) 1414 return; 1415 1416 if (bio->bi_bdev) 1417 rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio); 1418 1419 /* 1420 * Need to have a real endio function for chained bios, otherwise 1421 * various corner cases will break (like stacking block devices that 1422 * save/restore bi_end_io) - however, we want to avoid unbounded 1423 * recursion and blowing the stack. Tail call optimization would 1424 * handle this, but compiling with frame pointers also disables 1425 * gcc's sibling call optimization. 1426 */ 1427 if (bio->bi_end_io == bio_chain_endio) { 1428 bio = __bio_chain_endio(bio); 1429 goto again; 1430 } 1431 1432 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1433 trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio); 1434 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1435 } 1436 1437 blk_throtl_bio_endio(bio); 1438 /* release cgroup info */ 1439 bio_uninit(bio); 1440 if (bio->bi_end_io) 1441 bio->bi_end_io(bio); 1442 } 1443 EXPORT_SYMBOL(bio_endio); 1444 1445 /** 1446 * bio_split - split a bio 1447 * @bio: bio to split 1448 * @sectors: number of sectors to split from the front of @bio 1449 * @gfp: gfp mask 1450 * @bs: bio set to allocate from 1451 * 1452 * Allocates and returns a new bio which represents @sectors from the start of 1453 * @bio, and updates @bio to represent the remaining sectors. 1454 * 1455 * Unless this is a discard request the newly allocated bio will point 1456 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1457 * neither @bio nor @bs are freed before the split bio. 1458 */ 1459 struct bio *bio_split(struct bio *bio, int sectors, 1460 gfp_t gfp, struct bio_set *bs) 1461 { 1462 struct bio *split; 1463 1464 BUG_ON(sectors <= 0); 1465 BUG_ON(sectors >= bio_sectors(bio)); 1466 1467 /* Zone append commands cannot be split */ 1468 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 1469 return NULL; 1470 1471 split = bio_clone_fast(bio, gfp, bs); 1472 if (!split) 1473 return NULL; 1474 1475 split->bi_iter.bi_size = sectors << 9; 1476 1477 if (bio_integrity(split)) 1478 bio_integrity_trim(split); 1479 1480 bio_advance(bio, split->bi_iter.bi_size); 1481 1482 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1483 bio_set_flag(split, BIO_TRACE_COMPLETION); 1484 1485 return split; 1486 } 1487 EXPORT_SYMBOL(bio_split); 1488 1489 /** 1490 * bio_trim - trim a bio 1491 * @bio: bio to trim 1492 * @offset: number of sectors to trim from the front of @bio 1493 * @size: size we want to trim @bio to, in sectors 1494 */ 1495 void bio_trim(struct bio *bio, int offset, int size) 1496 { 1497 /* 'bio' is a cloned bio which we need to trim to match 1498 * the given offset and size. 1499 */ 1500 1501 size <<= 9; 1502 if (offset == 0 && size == bio->bi_iter.bi_size) 1503 return; 1504 1505 bio_advance(bio, offset << 9); 1506 bio->bi_iter.bi_size = size; 1507 1508 if (bio_integrity(bio)) 1509 bio_integrity_trim(bio); 1510 1511 } 1512 EXPORT_SYMBOL_GPL(bio_trim); 1513 1514 /* 1515 * create memory pools for biovec's in a bio_set. 1516 * use the global biovec slabs created for general use. 1517 */ 1518 int biovec_init_pool(mempool_t *pool, int pool_entries) 1519 { 1520 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1521 1522 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1523 } 1524 1525 /* 1526 * bioset_exit - exit a bioset initialized with bioset_init() 1527 * 1528 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1529 * kzalloc()). 1530 */ 1531 void bioset_exit(struct bio_set *bs) 1532 { 1533 if (bs->rescue_workqueue) 1534 destroy_workqueue(bs->rescue_workqueue); 1535 bs->rescue_workqueue = NULL; 1536 1537 mempool_exit(&bs->bio_pool); 1538 mempool_exit(&bs->bvec_pool); 1539 1540 bioset_integrity_free(bs); 1541 if (bs->bio_slab) 1542 bio_put_slab(bs); 1543 bs->bio_slab = NULL; 1544 } 1545 EXPORT_SYMBOL(bioset_exit); 1546 1547 /** 1548 * bioset_init - Initialize a bio_set 1549 * @bs: pool to initialize 1550 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1551 * @front_pad: Number of bytes to allocate in front of the returned bio 1552 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1553 * and %BIOSET_NEED_RESCUER 1554 * 1555 * Description: 1556 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1557 * to ask for a number of bytes to be allocated in front of the bio. 1558 * Front pad allocation is useful for embedding the bio inside 1559 * another structure, to avoid allocating extra data to go with the bio. 1560 * Note that the bio must be embedded at the END of that structure always, 1561 * or things will break badly. 1562 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1563 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1564 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1565 * dispatch queued requests when the mempool runs out of space. 1566 * 1567 */ 1568 int bioset_init(struct bio_set *bs, 1569 unsigned int pool_size, 1570 unsigned int front_pad, 1571 int flags) 1572 { 1573 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1574 1575 bs->front_pad = front_pad; 1576 1577 spin_lock_init(&bs->rescue_lock); 1578 bio_list_init(&bs->rescue_list); 1579 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1580 1581 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1582 if (!bs->bio_slab) 1583 return -ENOMEM; 1584 1585 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1586 goto bad; 1587 1588 if ((flags & BIOSET_NEED_BVECS) && 1589 biovec_init_pool(&bs->bvec_pool, pool_size)) 1590 goto bad; 1591 1592 if (!(flags & BIOSET_NEED_RESCUER)) 1593 return 0; 1594 1595 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1596 if (!bs->rescue_workqueue) 1597 goto bad; 1598 1599 return 0; 1600 bad: 1601 bioset_exit(bs); 1602 return -ENOMEM; 1603 } 1604 EXPORT_SYMBOL(bioset_init); 1605 1606 /* 1607 * Initialize and setup a new bio_set, based on the settings from 1608 * another bio_set. 1609 */ 1610 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 1611 { 1612 int flags; 1613 1614 flags = 0; 1615 if (src->bvec_pool.min_nr) 1616 flags |= BIOSET_NEED_BVECS; 1617 if (src->rescue_workqueue) 1618 flags |= BIOSET_NEED_RESCUER; 1619 1620 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 1621 } 1622 EXPORT_SYMBOL(bioset_init_from_src); 1623 1624 static void __init biovec_init_slabs(void) 1625 { 1626 int i; 1627 1628 for (i = 0; i < BVEC_POOL_NR; i++) { 1629 int size; 1630 struct biovec_slab *bvs = bvec_slabs + i; 1631 1632 if (bvs->nr_vecs <= BIO_INLINE_VECS) { 1633 bvs->slab = NULL; 1634 continue; 1635 } 1636 1637 size = bvs->nr_vecs * sizeof(struct bio_vec); 1638 bvs->slab = kmem_cache_create(bvs->name, size, 0, 1639 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 1640 } 1641 } 1642 1643 static int __init init_bio(void) 1644 { 1645 bio_slab_max = 2; 1646 bio_slab_nr = 0; 1647 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 1648 GFP_KERNEL); 1649 1650 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 1651 1652 if (!bio_slabs) 1653 panic("bio: can't allocate bios\n"); 1654 1655 bio_integrity_init(); 1656 biovec_init_slabs(); 1657 1658 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1659 panic("bio: can't allocate bios\n"); 1660 1661 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1662 panic("bio: can't create integrity pool\n"); 1663 1664 return 0; 1665 } 1666 subsys_initcall(init_bio); 1667