1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 4 */ 5 #include <linux/mm.h> 6 #include <linux/swap.h> 7 #include <linux/bio.h> 8 #include <linux/blkdev.h> 9 #include <linux/uio.h> 10 #include <linux/iocontext.h> 11 #include <linux/slab.h> 12 #include <linux/init.h> 13 #include <linux/kernel.h> 14 #include <linux/export.h> 15 #include <linux/mempool.h> 16 #include <linux/workqueue.h> 17 #include <linux/cgroup.h> 18 #include <linux/blk-cgroup.h> 19 #include <linux/highmem.h> 20 #include <linux/sched/sysctl.h> 21 #include <linux/blk-crypto.h> 22 #include <linux/xarray.h> 23 24 #include <trace/events/block.h> 25 #include "blk.h" 26 #include "blk-rq-qos.h" 27 28 /* 29 * Test patch to inline a certain number of bi_io_vec's inside the bio 30 * itself, to shrink a bio data allocation from two mempool calls to one 31 */ 32 #define BIO_INLINE_VECS 4 33 34 /* 35 * if you change this list, also change bvec_alloc or things will 36 * break badly! cannot be bigger than what you can fit into an 37 * unsigned short 38 */ 39 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 40 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 41 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 42 }; 43 #undef BV 44 45 /* 46 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 47 * IO code that does not need private memory pools. 48 */ 49 struct bio_set fs_bio_set; 50 EXPORT_SYMBOL(fs_bio_set); 51 52 /* 53 * Our slab pool management 54 */ 55 struct bio_slab { 56 struct kmem_cache *slab; 57 unsigned int slab_ref; 58 unsigned int slab_size; 59 char name[8]; 60 }; 61 static DEFINE_MUTEX(bio_slab_lock); 62 static DEFINE_XARRAY(bio_slabs); 63 64 static struct bio_slab *create_bio_slab(unsigned int size) 65 { 66 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); 67 68 if (!bslab) 69 return NULL; 70 71 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); 72 bslab->slab = kmem_cache_create(bslab->name, size, 73 ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL); 74 if (!bslab->slab) 75 goto fail_alloc_slab; 76 77 bslab->slab_ref = 1; 78 bslab->slab_size = size; 79 80 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) 81 return bslab; 82 83 kmem_cache_destroy(bslab->slab); 84 85 fail_alloc_slab: 86 kfree(bslab); 87 return NULL; 88 } 89 90 static inline unsigned int bs_bio_slab_size(struct bio_set *bs) 91 { 92 return bs->front_pad + sizeof(struct bio) + bs->back_pad; 93 } 94 95 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) 96 { 97 unsigned int size = bs_bio_slab_size(bs); 98 struct bio_slab *bslab; 99 100 mutex_lock(&bio_slab_lock); 101 bslab = xa_load(&bio_slabs, size); 102 if (bslab) 103 bslab->slab_ref++; 104 else 105 bslab = create_bio_slab(size); 106 mutex_unlock(&bio_slab_lock); 107 108 if (bslab) 109 return bslab->slab; 110 return NULL; 111 } 112 113 static void bio_put_slab(struct bio_set *bs) 114 { 115 struct bio_slab *bslab = NULL; 116 unsigned int slab_size = bs_bio_slab_size(bs); 117 118 mutex_lock(&bio_slab_lock); 119 120 bslab = xa_load(&bio_slabs, slab_size); 121 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 122 goto out; 123 124 WARN_ON_ONCE(bslab->slab != bs->bio_slab); 125 126 WARN_ON(!bslab->slab_ref); 127 128 if (--bslab->slab_ref) 129 goto out; 130 131 xa_erase(&bio_slabs, slab_size); 132 133 kmem_cache_destroy(bslab->slab); 134 kfree(bslab); 135 136 out: 137 mutex_unlock(&bio_slab_lock); 138 } 139 140 unsigned int bvec_nr_vecs(unsigned short idx) 141 { 142 return bvec_slabs[--idx].nr_vecs; 143 } 144 145 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 146 { 147 if (!idx) 148 return; 149 idx--; 150 151 BIO_BUG_ON(idx >= BVEC_POOL_NR); 152 153 if (idx == BVEC_POOL_MAX) { 154 mempool_free(bv, pool); 155 } else { 156 struct biovec_slab *bvs = bvec_slabs + idx; 157 158 kmem_cache_free(bvs->slab, bv); 159 } 160 } 161 162 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 163 mempool_t *pool) 164 { 165 struct bio_vec *bvl; 166 167 /* 168 * see comment near bvec_array define! 169 */ 170 switch (nr) { 171 case 1: 172 *idx = 0; 173 break; 174 case 2 ... 4: 175 *idx = 1; 176 break; 177 case 5 ... 16: 178 *idx = 2; 179 break; 180 case 17 ... 64: 181 *idx = 3; 182 break; 183 case 65 ... 128: 184 *idx = 4; 185 break; 186 case 129 ... BIO_MAX_PAGES: 187 *idx = 5; 188 break; 189 default: 190 return NULL; 191 } 192 193 /* 194 * idx now points to the pool we want to allocate from. only the 195 * 1-vec entry pool is mempool backed. 196 */ 197 if (*idx == BVEC_POOL_MAX) { 198 fallback: 199 bvl = mempool_alloc(pool, gfp_mask); 200 } else { 201 struct biovec_slab *bvs = bvec_slabs + *idx; 202 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 203 204 /* 205 * Make this allocation restricted and don't dump info on 206 * allocation failures, since we'll fallback to the mempool 207 * in case of failure. 208 */ 209 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 210 211 /* 212 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 213 * is set, retry with the 1-entry mempool 214 */ 215 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 216 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 217 *idx = BVEC_POOL_MAX; 218 goto fallback; 219 } 220 } 221 222 (*idx)++; 223 return bvl; 224 } 225 226 void bio_uninit(struct bio *bio) 227 { 228 #ifdef CONFIG_BLK_CGROUP 229 if (bio->bi_blkg) { 230 blkg_put(bio->bi_blkg); 231 bio->bi_blkg = NULL; 232 } 233 #endif 234 if (bio_integrity(bio)) 235 bio_integrity_free(bio); 236 237 bio_crypt_free_ctx(bio); 238 } 239 EXPORT_SYMBOL(bio_uninit); 240 241 static void bio_free(struct bio *bio) 242 { 243 struct bio_set *bs = bio->bi_pool; 244 void *p; 245 246 bio_uninit(bio); 247 248 if (bs) { 249 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 250 251 /* 252 * If we have front padding, adjust the bio pointer before freeing 253 */ 254 p = bio; 255 p -= bs->front_pad; 256 257 mempool_free(p, &bs->bio_pool); 258 } else { 259 /* Bio was allocated by bio_kmalloc() */ 260 kfree(bio); 261 } 262 } 263 264 /* 265 * Users of this function have their own bio allocation. Subsequently, 266 * they must remember to pair any call to bio_init() with bio_uninit() 267 * when IO has completed, or when the bio is released. 268 */ 269 void bio_init(struct bio *bio, struct bio_vec *table, 270 unsigned short max_vecs) 271 { 272 memset(bio, 0, sizeof(*bio)); 273 atomic_set(&bio->__bi_remaining, 1); 274 atomic_set(&bio->__bi_cnt, 1); 275 276 bio->bi_io_vec = table; 277 bio->bi_max_vecs = max_vecs; 278 } 279 EXPORT_SYMBOL(bio_init); 280 281 /** 282 * bio_reset - reinitialize a bio 283 * @bio: bio to reset 284 * 285 * Description: 286 * After calling bio_reset(), @bio will be in the same state as a freshly 287 * allocated bio returned bio bio_alloc_bioset() - the only fields that are 288 * preserved are the ones that are initialized by bio_alloc_bioset(). See 289 * comment in struct bio. 290 */ 291 void bio_reset(struct bio *bio) 292 { 293 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 294 295 bio_uninit(bio); 296 297 memset(bio, 0, BIO_RESET_BYTES); 298 bio->bi_flags = flags; 299 atomic_set(&bio->__bi_remaining, 1); 300 } 301 EXPORT_SYMBOL(bio_reset); 302 303 static struct bio *__bio_chain_endio(struct bio *bio) 304 { 305 struct bio *parent = bio->bi_private; 306 307 if (!parent->bi_status) 308 parent->bi_status = bio->bi_status; 309 bio_put(bio); 310 return parent; 311 } 312 313 static void bio_chain_endio(struct bio *bio) 314 { 315 bio_endio(__bio_chain_endio(bio)); 316 } 317 318 /** 319 * bio_chain - chain bio completions 320 * @bio: the target bio 321 * @parent: the parent bio of @bio 322 * 323 * The caller won't have a bi_end_io called when @bio completes - instead, 324 * @parent's bi_end_io won't be called until both @parent and @bio have 325 * completed; the chained bio will also be freed when it completes. 326 * 327 * The caller must not set bi_private or bi_end_io in @bio. 328 */ 329 void bio_chain(struct bio *bio, struct bio *parent) 330 { 331 BUG_ON(bio->bi_private || bio->bi_end_io); 332 333 bio->bi_private = parent; 334 bio->bi_end_io = bio_chain_endio; 335 bio_inc_remaining(parent); 336 } 337 EXPORT_SYMBOL(bio_chain); 338 339 static void bio_alloc_rescue(struct work_struct *work) 340 { 341 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 342 struct bio *bio; 343 344 while (1) { 345 spin_lock(&bs->rescue_lock); 346 bio = bio_list_pop(&bs->rescue_list); 347 spin_unlock(&bs->rescue_lock); 348 349 if (!bio) 350 break; 351 352 submit_bio_noacct(bio); 353 } 354 } 355 356 static void punt_bios_to_rescuer(struct bio_set *bs) 357 { 358 struct bio_list punt, nopunt; 359 struct bio *bio; 360 361 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 362 return; 363 /* 364 * In order to guarantee forward progress we must punt only bios that 365 * were allocated from this bio_set; otherwise, if there was a bio on 366 * there for a stacking driver higher up in the stack, processing it 367 * could require allocating bios from this bio_set, and doing that from 368 * our own rescuer would be bad. 369 * 370 * Since bio lists are singly linked, pop them all instead of trying to 371 * remove from the middle of the list: 372 */ 373 374 bio_list_init(&punt); 375 bio_list_init(&nopunt); 376 377 while ((bio = bio_list_pop(¤t->bio_list[0]))) 378 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 379 current->bio_list[0] = nopunt; 380 381 bio_list_init(&nopunt); 382 while ((bio = bio_list_pop(¤t->bio_list[1]))) 383 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 384 current->bio_list[1] = nopunt; 385 386 spin_lock(&bs->rescue_lock); 387 bio_list_merge(&bs->rescue_list, &punt); 388 spin_unlock(&bs->rescue_lock); 389 390 queue_work(bs->rescue_workqueue, &bs->rescue_work); 391 } 392 393 /** 394 * bio_alloc_bioset - allocate a bio for I/O 395 * @gfp_mask: the GFP_* mask given to the slab allocator 396 * @nr_iovecs: number of iovecs to pre-allocate 397 * @bs: the bio_set to allocate from. 398 * 399 * Description: 400 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 401 * backed by the @bs's mempool. 402 * 403 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 404 * always be able to allocate a bio. This is due to the mempool guarantees. 405 * To make this work, callers must never allocate more than 1 bio at a time 406 * from this pool. Callers that need to allocate more than 1 bio must always 407 * submit the previously allocated bio for IO before attempting to allocate 408 * a new one. Failure to do so can cause deadlocks under memory pressure. 409 * 410 * Note that when running under submit_bio_noacct() (i.e. any block 411 * driver), bios are not submitted until after you return - see the code in 412 * submit_bio_noacct() that converts recursion into iteration, to prevent 413 * stack overflows. 414 * 415 * This would normally mean allocating multiple bios under 416 * submit_bio_noacct() would be susceptible to deadlocks, but we have 417 * deadlock avoidance code that resubmits any blocked bios from a rescuer 418 * thread. 419 * 420 * However, we do not guarantee forward progress for allocations from other 421 * mempools. Doing multiple allocations from the same mempool under 422 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 423 * for per bio allocations. 424 * 425 * RETURNS: 426 * Pointer to new bio on success, NULL on failure. 427 */ 428 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 429 struct bio_set *bs) 430 { 431 gfp_t saved_gfp = gfp_mask; 432 unsigned front_pad; 433 unsigned inline_vecs; 434 struct bio_vec *bvl = NULL; 435 struct bio *bio; 436 void *p; 437 438 if (!bs) { 439 if (nr_iovecs > UIO_MAXIOV) 440 return NULL; 441 442 p = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 443 front_pad = 0; 444 inline_vecs = nr_iovecs; 445 } else { 446 /* should not use nobvec bioset for nr_iovecs > 0 */ 447 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 448 nr_iovecs > 0)) 449 return NULL; 450 /* 451 * submit_bio_noacct() converts recursion to iteration; this 452 * means if we're running beneath it, any bios we allocate and 453 * submit will not be submitted (and thus freed) until after we 454 * return. 455 * 456 * This exposes us to a potential deadlock if we allocate 457 * multiple bios from the same bio_set() while running 458 * underneath submit_bio_noacct(). If we were to allocate 459 * multiple bios (say a stacking block driver that was splitting 460 * bios), we would deadlock if we exhausted the mempool's 461 * reserve. 462 * 463 * We solve this, and guarantee forward progress, with a rescuer 464 * workqueue per bio_set. If we go to allocate and there are 465 * bios on current->bio_list, we first try the allocation 466 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 467 * bios we would be blocking to the rescuer workqueue before 468 * we retry with the original gfp_flags. 469 */ 470 471 if (current->bio_list && 472 (!bio_list_empty(¤t->bio_list[0]) || 473 !bio_list_empty(¤t->bio_list[1])) && 474 bs->rescue_workqueue) 475 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 476 477 p = mempool_alloc(&bs->bio_pool, gfp_mask); 478 if (!p && gfp_mask != saved_gfp) { 479 punt_bios_to_rescuer(bs); 480 gfp_mask = saved_gfp; 481 p = mempool_alloc(&bs->bio_pool, gfp_mask); 482 } 483 484 front_pad = bs->front_pad; 485 inline_vecs = BIO_INLINE_VECS; 486 } 487 488 if (unlikely(!p)) 489 return NULL; 490 491 bio = p + front_pad; 492 bio_init(bio, NULL, 0); 493 494 if (nr_iovecs > inline_vecs) { 495 unsigned long idx = 0; 496 497 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 498 if (!bvl && gfp_mask != saved_gfp) { 499 punt_bios_to_rescuer(bs); 500 gfp_mask = saved_gfp; 501 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 502 } 503 504 if (unlikely(!bvl)) 505 goto err_free; 506 507 bio->bi_flags |= idx << BVEC_POOL_OFFSET; 508 bio->bi_max_vecs = bvec_nr_vecs(idx); 509 } else if (nr_iovecs) { 510 bvl = bio->bi_inline_vecs; 511 bio->bi_max_vecs = inline_vecs; 512 } 513 514 bio->bi_pool = bs; 515 bio->bi_io_vec = bvl; 516 return bio; 517 518 err_free: 519 mempool_free(p, &bs->bio_pool); 520 return NULL; 521 } 522 EXPORT_SYMBOL(bio_alloc_bioset); 523 524 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 525 { 526 unsigned long flags; 527 struct bio_vec bv; 528 struct bvec_iter iter; 529 530 __bio_for_each_segment(bv, bio, iter, start) { 531 char *data = bvec_kmap_irq(&bv, &flags); 532 memset(data, 0, bv.bv_len); 533 flush_dcache_page(bv.bv_page); 534 bvec_kunmap_irq(data, &flags); 535 } 536 } 537 EXPORT_SYMBOL(zero_fill_bio_iter); 538 539 /** 540 * bio_truncate - truncate the bio to small size of @new_size 541 * @bio: the bio to be truncated 542 * @new_size: new size for truncating the bio 543 * 544 * Description: 545 * Truncate the bio to new size of @new_size. If bio_op(bio) is 546 * REQ_OP_READ, zero the truncated part. This function should only 547 * be used for handling corner cases, such as bio eod. 548 */ 549 void bio_truncate(struct bio *bio, unsigned new_size) 550 { 551 struct bio_vec bv; 552 struct bvec_iter iter; 553 unsigned int done = 0; 554 bool truncated = false; 555 556 if (new_size >= bio->bi_iter.bi_size) 557 return; 558 559 if (bio_op(bio) != REQ_OP_READ) 560 goto exit; 561 562 bio_for_each_segment(bv, bio, iter) { 563 if (done + bv.bv_len > new_size) { 564 unsigned offset; 565 566 if (!truncated) 567 offset = new_size - done; 568 else 569 offset = 0; 570 zero_user(bv.bv_page, offset, bv.bv_len - offset); 571 truncated = true; 572 } 573 done += bv.bv_len; 574 } 575 576 exit: 577 /* 578 * Don't touch bvec table here and make it really immutable, since 579 * fs bio user has to retrieve all pages via bio_for_each_segment_all 580 * in its .end_bio() callback. 581 * 582 * It is enough to truncate bio by updating .bi_size since we can make 583 * correct bvec with the updated .bi_size for drivers. 584 */ 585 bio->bi_iter.bi_size = new_size; 586 } 587 588 /** 589 * guard_bio_eod - truncate a BIO to fit the block device 590 * @bio: bio to truncate 591 * 592 * This allows us to do IO even on the odd last sectors of a device, even if the 593 * block size is some multiple of the physical sector size. 594 * 595 * We'll just truncate the bio to the size of the device, and clear the end of 596 * the buffer head manually. Truly out-of-range accesses will turn into actual 597 * I/O errors, this only handles the "we need to be able to do I/O at the final 598 * sector" case. 599 */ 600 void guard_bio_eod(struct bio *bio) 601 { 602 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 603 604 if (!maxsector) 605 return; 606 607 /* 608 * If the *whole* IO is past the end of the device, 609 * let it through, and the IO layer will turn it into 610 * an EIO. 611 */ 612 if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 613 return; 614 615 maxsector -= bio->bi_iter.bi_sector; 616 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 617 return; 618 619 bio_truncate(bio, maxsector << 9); 620 } 621 622 /** 623 * bio_put - release a reference to a bio 624 * @bio: bio to release reference to 625 * 626 * Description: 627 * Put a reference to a &struct bio, either one you have gotten with 628 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 629 **/ 630 void bio_put(struct bio *bio) 631 { 632 if (!bio_flagged(bio, BIO_REFFED)) 633 bio_free(bio); 634 else { 635 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 636 637 /* 638 * last put frees it 639 */ 640 if (atomic_dec_and_test(&bio->__bi_cnt)) 641 bio_free(bio); 642 } 643 } 644 EXPORT_SYMBOL(bio_put); 645 646 /** 647 * __bio_clone_fast - clone a bio that shares the original bio's biovec 648 * @bio: destination bio 649 * @bio_src: bio to clone 650 * 651 * Clone a &bio. Caller will own the returned bio, but not 652 * the actual data it points to. Reference count of returned 653 * bio will be one. 654 * 655 * Caller must ensure that @bio_src is not freed before @bio. 656 */ 657 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 658 { 659 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 660 661 /* 662 * most users will be overriding ->bi_bdev with a new target, 663 * so we don't set nor calculate new physical/hw segment counts here 664 */ 665 bio->bi_bdev = bio_src->bi_bdev; 666 bio_set_flag(bio, BIO_CLONED); 667 if (bio_flagged(bio_src, BIO_THROTTLED)) 668 bio_set_flag(bio, BIO_THROTTLED); 669 bio->bi_opf = bio_src->bi_opf; 670 bio->bi_ioprio = bio_src->bi_ioprio; 671 bio->bi_write_hint = bio_src->bi_write_hint; 672 bio->bi_iter = bio_src->bi_iter; 673 bio->bi_io_vec = bio_src->bi_io_vec; 674 675 bio_clone_blkg_association(bio, bio_src); 676 blkcg_bio_issue_init(bio); 677 } 678 EXPORT_SYMBOL(__bio_clone_fast); 679 680 /** 681 * bio_clone_fast - clone a bio that shares the original bio's biovec 682 * @bio: bio to clone 683 * @gfp_mask: allocation priority 684 * @bs: bio_set to allocate from 685 * 686 * Like __bio_clone_fast, only also allocates the returned bio 687 */ 688 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 689 { 690 struct bio *b; 691 692 b = bio_alloc_bioset(gfp_mask, 0, bs); 693 if (!b) 694 return NULL; 695 696 __bio_clone_fast(b, bio); 697 698 if (bio_crypt_clone(b, bio, gfp_mask) < 0) 699 goto err_put; 700 701 if (bio_integrity(bio) && 702 bio_integrity_clone(b, bio, gfp_mask) < 0) 703 goto err_put; 704 705 return b; 706 707 err_put: 708 bio_put(b); 709 return NULL; 710 } 711 EXPORT_SYMBOL(bio_clone_fast); 712 713 const char *bio_devname(struct bio *bio, char *buf) 714 { 715 return bdevname(bio->bi_bdev, buf); 716 } 717 EXPORT_SYMBOL(bio_devname); 718 719 static inline bool page_is_mergeable(const struct bio_vec *bv, 720 struct page *page, unsigned int len, unsigned int off, 721 bool *same_page) 722 { 723 size_t bv_end = bv->bv_offset + bv->bv_len; 724 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 725 phys_addr_t page_addr = page_to_phys(page); 726 727 if (vec_end_addr + 1 != page_addr + off) 728 return false; 729 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 730 return false; 731 732 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 733 if (*same_page) 734 return true; 735 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 736 } 737 738 /* 739 * Try to merge a page into a segment, while obeying the hardware segment 740 * size limit. This is not for normal read/write bios, but for passthrough 741 * or Zone Append operations that we can't split. 742 */ 743 static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 744 struct page *page, unsigned len, 745 unsigned offset, bool *same_page) 746 { 747 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 748 unsigned long mask = queue_segment_boundary(q); 749 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 750 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 751 752 if ((addr1 | mask) != (addr2 | mask)) 753 return false; 754 if (bv->bv_len + len > queue_max_segment_size(q)) 755 return false; 756 return __bio_try_merge_page(bio, page, len, offset, same_page); 757 } 758 759 /** 760 * bio_add_hw_page - attempt to add a page to a bio with hw constraints 761 * @q: the target queue 762 * @bio: destination bio 763 * @page: page to add 764 * @len: vec entry length 765 * @offset: vec entry offset 766 * @max_sectors: maximum number of sectors that can be added 767 * @same_page: return if the segment has been merged inside the same page 768 * 769 * Add a page to a bio while respecting the hardware max_sectors, max_segment 770 * and gap limitations. 771 */ 772 int bio_add_hw_page(struct request_queue *q, struct bio *bio, 773 struct page *page, unsigned int len, unsigned int offset, 774 unsigned int max_sectors, bool *same_page) 775 { 776 struct bio_vec *bvec; 777 778 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 779 return 0; 780 781 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 782 return 0; 783 784 if (bio->bi_vcnt > 0) { 785 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 786 return len; 787 788 /* 789 * If the queue doesn't support SG gaps and adding this segment 790 * would create a gap, disallow it. 791 */ 792 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 793 if (bvec_gap_to_prev(q, bvec, offset)) 794 return 0; 795 } 796 797 if (bio_full(bio, len)) 798 return 0; 799 800 if (bio->bi_vcnt >= queue_max_segments(q)) 801 return 0; 802 803 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 804 bvec->bv_page = page; 805 bvec->bv_len = len; 806 bvec->bv_offset = offset; 807 bio->bi_vcnt++; 808 bio->bi_iter.bi_size += len; 809 return len; 810 } 811 812 /** 813 * bio_add_pc_page - attempt to add page to passthrough bio 814 * @q: the target queue 815 * @bio: destination bio 816 * @page: page to add 817 * @len: vec entry length 818 * @offset: vec entry offset 819 * 820 * Attempt to add a page to the bio_vec maplist. This can fail for a 821 * number of reasons, such as the bio being full or target block device 822 * limitations. The target block device must allow bio's up to PAGE_SIZE, 823 * so it is always possible to add a single page to an empty bio. 824 * 825 * This should only be used by passthrough bios. 826 */ 827 int bio_add_pc_page(struct request_queue *q, struct bio *bio, 828 struct page *page, unsigned int len, unsigned int offset) 829 { 830 bool same_page = false; 831 return bio_add_hw_page(q, bio, page, len, offset, 832 queue_max_hw_sectors(q), &same_page); 833 } 834 EXPORT_SYMBOL(bio_add_pc_page); 835 836 /** 837 * __bio_try_merge_page - try appending data to an existing bvec. 838 * @bio: destination bio 839 * @page: start page to add 840 * @len: length of the data to add 841 * @off: offset of the data relative to @page 842 * @same_page: return if the segment has been merged inside the same page 843 * 844 * Try to add the data at @page + @off to the last bvec of @bio. This is a 845 * useful optimisation for file systems with a block size smaller than the 846 * page size. 847 * 848 * Warn if (@len, @off) crosses pages in case that @same_page is true. 849 * 850 * Return %true on success or %false on failure. 851 */ 852 bool __bio_try_merge_page(struct bio *bio, struct page *page, 853 unsigned int len, unsigned int off, bool *same_page) 854 { 855 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 856 return false; 857 858 if (bio->bi_vcnt > 0) { 859 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 860 861 if (page_is_mergeable(bv, page, len, off, same_page)) { 862 if (bio->bi_iter.bi_size > UINT_MAX - len) { 863 *same_page = false; 864 return false; 865 } 866 bv->bv_len += len; 867 bio->bi_iter.bi_size += len; 868 return true; 869 } 870 } 871 return false; 872 } 873 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 874 875 /** 876 * __bio_add_page - add page(s) to a bio in a new segment 877 * @bio: destination bio 878 * @page: start page to add 879 * @len: length of the data to add, may cross pages 880 * @off: offset of the data relative to @page, may cross pages 881 * 882 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 883 * that @bio has space for another bvec. 884 */ 885 void __bio_add_page(struct bio *bio, struct page *page, 886 unsigned int len, unsigned int off) 887 { 888 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 889 890 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 891 WARN_ON_ONCE(bio_full(bio, len)); 892 893 bv->bv_page = page; 894 bv->bv_offset = off; 895 bv->bv_len = len; 896 897 bio->bi_iter.bi_size += len; 898 bio->bi_vcnt++; 899 900 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 901 bio_set_flag(bio, BIO_WORKINGSET); 902 } 903 EXPORT_SYMBOL_GPL(__bio_add_page); 904 905 /** 906 * bio_add_page - attempt to add page(s) to bio 907 * @bio: destination bio 908 * @page: start page to add 909 * @len: vec entry length, may cross pages 910 * @offset: vec entry offset relative to @page, may cross pages 911 * 912 * Attempt to add page(s) to the bio_vec maplist. This will only fail 913 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 914 */ 915 int bio_add_page(struct bio *bio, struct page *page, 916 unsigned int len, unsigned int offset) 917 { 918 bool same_page = false; 919 920 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 921 if (bio_full(bio, len)) 922 return 0; 923 __bio_add_page(bio, page, len, offset); 924 } 925 return len; 926 } 927 EXPORT_SYMBOL(bio_add_page); 928 929 void bio_release_pages(struct bio *bio, bool mark_dirty) 930 { 931 struct bvec_iter_all iter_all; 932 struct bio_vec *bvec; 933 934 if (bio_flagged(bio, BIO_NO_PAGE_REF)) 935 return; 936 937 bio_for_each_segment_all(bvec, bio, iter_all) { 938 if (mark_dirty && !PageCompound(bvec->bv_page)) 939 set_page_dirty_lock(bvec->bv_page); 940 put_page(bvec->bv_page); 941 } 942 } 943 EXPORT_SYMBOL_GPL(bio_release_pages); 944 945 static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 946 { 947 WARN_ON_ONCE(BVEC_POOL_IDX(bio) != 0); 948 949 bio->bi_vcnt = iter->nr_segs; 950 bio->bi_max_vecs = iter->nr_segs; 951 bio->bi_io_vec = (struct bio_vec *)iter->bvec; 952 bio->bi_iter.bi_bvec_done = iter->iov_offset; 953 bio->bi_iter.bi_size = iter->count; 954 955 iov_iter_advance(iter, iter->count); 956 return 0; 957 } 958 959 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 960 961 /** 962 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 963 * @bio: bio to add pages to 964 * @iter: iov iterator describing the region to be mapped 965 * 966 * Pins pages from *iter and appends them to @bio's bvec array. The 967 * pages will have to be released using put_page() when done. 968 * For multi-segment *iter, this function only adds pages from the 969 * next non-empty segment of the iov iterator. 970 */ 971 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 972 { 973 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 974 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 975 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 976 struct page **pages = (struct page **)bv; 977 bool same_page = false; 978 ssize_t size, left; 979 unsigned len, i; 980 size_t offset; 981 982 /* 983 * Move page array up in the allocated memory for the bio vecs as far as 984 * possible so that we can start filling biovecs from the beginning 985 * without overwriting the temporary page array. 986 */ 987 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 988 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 989 990 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 991 if (unlikely(size <= 0)) 992 return size ? size : -EFAULT; 993 994 for (left = size, i = 0; left > 0; left -= len, i++) { 995 struct page *page = pages[i]; 996 997 len = min_t(size_t, PAGE_SIZE - offset, left); 998 999 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1000 if (same_page) 1001 put_page(page); 1002 } else { 1003 if (WARN_ON_ONCE(bio_full(bio, len))) 1004 return -EINVAL; 1005 __bio_add_page(bio, page, len, offset); 1006 } 1007 offset = 0; 1008 } 1009 1010 iov_iter_advance(iter, size); 1011 return 0; 1012 } 1013 1014 static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 1015 { 1016 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1017 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1018 struct request_queue *q = bio->bi_bdev->bd_disk->queue; 1019 unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 1020 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 1021 struct page **pages = (struct page **)bv; 1022 ssize_t size, left; 1023 unsigned len, i; 1024 size_t offset; 1025 int ret = 0; 1026 1027 if (WARN_ON_ONCE(!max_append_sectors)) 1028 return 0; 1029 1030 /* 1031 * Move page array up in the allocated memory for the bio vecs as far as 1032 * possible so that we can start filling biovecs from the beginning 1033 * without overwriting the temporary page array. 1034 */ 1035 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1036 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 1037 1038 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 1039 if (unlikely(size <= 0)) 1040 return size ? size : -EFAULT; 1041 1042 for (left = size, i = 0; left > 0; left -= len, i++) { 1043 struct page *page = pages[i]; 1044 bool same_page = false; 1045 1046 len = min_t(size_t, PAGE_SIZE - offset, left); 1047 if (bio_add_hw_page(q, bio, page, len, offset, 1048 max_append_sectors, &same_page) != len) { 1049 ret = -EINVAL; 1050 break; 1051 } 1052 if (same_page) 1053 put_page(page); 1054 offset = 0; 1055 } 1056 1057 iov_iter_advance(iter, size - left); 1058 return ret; 1059 } 1060 1061 /** 1062 * bio_iov_iter_get_pages - add user or kernel pages to a bio 1063 * @bio: bio to add pages to 1064 * @iter: iov iterator describing the region to be added 1065 * 1066 * This takes either an iterator pointing to user memory, or one pointing to 1067 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 1068 * map them into the kernel. On IO completion, the caller should put those 1069 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1070 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1071 * to ensure the bvecs and pages stay referenced until the submitted I/O is 1072 * completed by a call to ->ki_complete() or returns with an error other than 1073 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1074 * on IO completion. If it isn't, then pages should be released. 1075 * 1076 * The function tries, but does not guarantee, to pin as many pages as 1077 * fit into the bio, or are requested in @iter, whatever is smaller. If 1078 * MM encounters an error pinning the requested pages, it stops. Error 1079 * is returned only if 0 pages could be pinned. 1080 * 1081 * It's intended for direct IO, so doesn't do PSI tracking, the caller is 1082 * responsible for setting BIO_WORKINGSET if necessary. 1083 */ 1084 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1085 { 1086 int ret = 0; 1087 1088 if (iov_iter_is_bvec(iter)) { 1089 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 1090 return -EINVAL; 1091 bio_iov_bvec_set(bio, iter); 1092 bio_set_flag(bio, BIO_NO_PAGE_REF); 1093 return 0; 1094 } else { 1095 do { 1096 if (bio_op(bio) == REQ_OP_ZONE_APPEND) 1097 ret = __bio_iov_append_get_pages(bio, iter); 1098 else 1099 ret = __bio_iov_iter_get_pages(bio, iter); 1100 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1101 } 1102 1103 /* don't account direct I/O as memory stall */ 1104 bio_clear_flag(bio, BIO_WORKINGSET); 1105 return bio->bi_vcnt ? 0 : ret; 1106 } 1107 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 1108 1109 static void submit_bio_wait_endio(struct bio *bio) 1110 { 1111 complete(bio->bi_private); 1112 } 1113 1114 /** 1115 * submit_bio_wait - submit a bio, and wait until it completes 1116 * @bio: The &struct bio which describes the I/O 1117 * 1118 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1119 * bio_endio() on failure. 1120 * 1121 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1122 * result in bio reference to be consumed. The caller must drop the reference 1123 * on his own. 1124 */ 1125 int submit_bio_wait(struct bio *bio) 1126 { 1127 DECLARE_COMPLETION_ONSTACK_MAP(done, 1128 bio->bi_bdev->bd_disk->lockdep_map); 1129 unsigned long hang_check; 1130 1131 bio->bi_private = &done; 1132 bio->bi_end_io = submit_bio_wait_endio; 1133 bio->bi_opf |= REQ_SYNC; 1134 submit_bio(bio); 1135 1136 /* Prevent hang_check timer from firing at us during very long I/O */ 1137 hang_check = sysctl_hung_task_timeout_secs; 1138 if (hang_check) 1139 while (!wait_for_completion_io_timeout(&done, 1140 hang_check * (HZ/2))) 1141 ; 1142 else 1143 wait_for_completion_io(&done); 1144 1145 return blk_status_to_errno(bio->bi_status); 1146 } 1147 EXPORT_SYMBOL(submit_bio_wait); 1148 1149 /** 1150 * bio_advance - increment/complete a bio by some number of bytes 1151 * @bio: bio to advance 1152 * @bytes: number of bytes to complete 1153 * 1154 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1155 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1156 * be updated on the last bvec as well. 1157 * 1158 * @bio will then represent the remaining, uncompleted portion of the io. 1159 */ 1160 void bio_advance(struct bio *bio, unsigned bytes) 1161 { 1162 if (bio_integrity(bio)) 1163 bio_integrity_advance(bio, bytes); 1164 1165 bio_crypt_advance(bio, bytes); 1166 bio_advance_iter(bio, &bio->bi_iter, bytes); 1167 } 1168 EXPORT_SYMBOL(bio_advance); 1169 1170 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1171 struct bio *src, struct bvec_iter *src_iter) 1172 { 1173 struct bio_vec src_bv, dst_bv; 1174 void *src_p, *dst_p; 1175 unsigned bytes; 1176 1177 while (src_iter->bi_size && dst_iter->bi_size) { 1178 src_bv = bio_iter_iovec(src, *src_iter); 1179 dst_bv = bio_iter_iovec(dst, *dst_iter); 1180 1181 bytes = min(src_bv.bv_len, dst_bv.bv_len); 1182 1183 src_p = kmap_atomic(src_bv.bv_page); 1184 dst_p = kmap_atomic(dst_bv.bv_page); 1185 1186 memcpy(dst_p + dst_bv.bv_offset, 1187 src_p + src_bv.bv_offset, 1188 bytes); 1189 1190 kunmap_atomic(dst_p); 1191 kunmap_atomic(src_p); 1192 1193 flush_dcache_page(dst_bv.bv_page); 1194 1195 bio_advance_iter_single(src, src_iter, bytes); 1196 bio_advance_iter_single(dst, dst_iter, bytes); 1197 } 1198 } 1199 EXPORT_SYMBOL(bio_copy_data_iter); 1200 1201 /** 1202 * bio_copy_data - copy contents of data buffers from one bio to another 1203 * @src: source bio 1204 * @dst: destination bio 1205 * 1206 * Stops when it reaches the end of either @src or @dst - that is, copies 1207 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1208 */ 1209 void bio_copy_data(struct bio *dst, struct bio *src) 1210 { 1211 struct bvec_iter src_iter = src->bi_iter; 1212 struct bvec_iter dst_iter = dst->bi_iter; 1213 1214 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1215 } 1216 EXPORT_SYMBOL(bio_copy_data); 1217 1218 /** 1219 * bio_list_copy_data - copy contents of data buffers from one chain of bios to 1220 * another 1221 * @src: source bio list 1222 * @dst: destination bio list 1223 * 1224 * Stops when it reaches the end of either the @src list or @dst list - that is, 1225 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 1226 * bios). 1227 */ 1228 void bio_list_copy_data(struct bio *dst, struct bio *src) 1229 { 1230 struct bvec_iter src_iter = src->bi_iter; 1231 struct bvec_iter dst_iter = dst->bi_iter; 1232 1233 while (1) { 1234 if (!src_iter.bi_size) { 1235 src = src->bi_next; 1236 if (!src) 1237 break; 1238 1239 src_iter = src->bi_iter; 1240 } 1241 1242 if (!dst_iter.bi_size) { 1243 dst = dst->bi_next; 1244 if (!dst) 1245 break; 1246 1247 dst_iter = dst->bi_iter; 1248 } 1249 1250 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1251 } 1252 } 1253 EXPORT_SYMBOL(bio_list_copy_data); 1254 1255 void bio_free_pages(struct bio *bio) 1256 { 1257 struct bio_vec *bvec; 1258 struct bvec_iter_all iter_all; 1259 1260 bio_for_each_segment_all(bvec, bio, iter_all) 1261 __free_page(bvec->bv_page); 1262 } 1263 EXPORT_SYMBOL(bio_free_pages); 1264 1265 /* 1266 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1267 * for performing direct-IO in BIOs. 1268 * 1269 * The problem is that we cannot run set_page_dirty() from interrupt context 1270 * because the required locks are not interrupt-safe. So what we can do is to 1271 * mark the pages dirty _before_ performing IO. And in interrupt context, 1272 * check that the pages are still dirty. If so, fine. If not, redirty them 1273 * in process context. 1274 * 1275 * We special-case compound pages here: normally this means reads into hugetlb 1276 * pages. The logic in here doesn't really work right for compound pages 1277 * because the VM does not uniformly chase down the head page in all cases. 1278 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1279 * handle them at all. So we skip compound pages here at an early stage. 1280 * 1281 * Note that this code is very hard to test under normal circumstances because 1282 * direct-io pins the pages with get_user_pages(). This makes 1283 * is_page_cache_freeable return false, and the VM will not clean the pages. 1284 * But other code (eg, flusher threads) could clean the pages if they are mapped 1285 * pagecache. 1286 * 1287 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1288 * deferred bio dirtying paths. 1289 */ 1290 1291 /* 1292 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1293 */ 1294 void bio_set_pages_dirty(struct bio *bio) 1295 { 1296 struct bio_vec *bvec; 1297 struct bvec_iter_all iter_all; 1298 1299 bio_for_each_segment_all(bvec, bio, iter_all) { 1300 if (!PageCompound(bvec->bv_page)) 1301 set_page_dirty_lock(bvec->bv_page); 1302 } 1303 } 1304 1305 /* 1306 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1307 * If they are, then fine. If, however, some pages are clean then they must 1308 * have been written out during the direct-IO read. So we take another ref on 1309 * the BIO and re-dirty the pages in process context. 1310 * 1311 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1312 * here on. It will run one put_page() against each page and will run one 1313 * bio_put() against the BIO. 1314 */ 1315 1316 static void bio_dirty_fn(struct work_struct *work); 1317 1318 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1319 static DEFINE_SPINLOCK(bio_dirty_lock); 1320 static struct bio *bio_dirty_list; 1321 1322 /* 1323 * This runs in process context 1324 */ 1325 static void bio_dirty_fn(struct work_struct *work) 1326 { 1327 struct bio *bio, *next; 1328 1329 spin_lock_irq(&bio_dirty_lock); 1330 next = bio_dirty_list; 1331 bio_dirty_list = NULL; 1332 spin_unlock_irq(&bio_dirty_lock); 1333 1334 while ((bio = next) != NULL) { 1335 next = bio->bi_private; 1336 1337 bio_release_pages(bio, true); 1338 bio_put(bio); 1339 } 1340 } 1341 1342 void bio_check_pages_dirty(struct bio *bio) 1343 { 1344 struct bio_vec *bvec; 1345 unsigned long flags; 1346 struct bvec_iter_all iter_all; 1347 1348 bio_for_each_segment_all(bvec, bio, iter_all) { 1349 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1350 goto defer; 1351 } 1352 1353 bio_release_pages(bio, false); 1354 bio_put(bio); 1355 return; 1356 defer: 1357 spin_lock_irqsave(&bio_dirty_lock, flags); 1358 bio->bi_private = bio_dirty_list; 1359 bio_dirty_list = bio; 1360 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1361 schedule_work(&bio_dirty_work); 1362 } 1363 1364 static inline bool bio_remaining_done(struct bio *bio) 1365 { 1366 /* 1367 * If we're not chaining, then ->__bi_remaining is always 1 and 1368 * we always end io on the first invocation. 1369 */ 1370 if (!bio_flagged(bio, BIO_CHAIN)) 1371 return true; 1372 1373 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1374 1375 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1376 bio_clear_flag(bio, BIO_CHAIN); 1377 return true; 1378 } 1379 1380 return false; 1381 } 1382 1383 /** 1384 * bio_endio - end I/O on a bio 1385 * @bio: bio 1386 * 1387 * Description: 1388 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1389 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1390 * bio unless they own it and thus know that it has an end_io function. 1391 * 1392 * bio_endio() can be called several times on a bio that has been chained 1393 * using bio_chain(). The ->bi_end_io() function will only be called the 1394 * last time. At this point the BLK_TA_COMPLETE tracing event will be 1395 * generated if BIO_TRACE_COMPLETION is set. 1396 **/ 1397 void bio_endio(struct bio *bio) 1398 { 1399 again: 1400 if (!bio_remaining_done(bio)) 1401 return; 1402 if (!bio_integrity_endio(bio)) 1403 return; 1404 1405 if (bio->bi_bdev) 1406 rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio); 1407 1408 /* 1409 * Need to have a real endio function for chained bios, otherwise 1410 * various corner cases will break (like stacking block devices that 1411 * save/restore bi_end_io) - however, we want to avoid unbounded 1412 * recursion and blowing the stack. Tail call optimization would 1413 * handle this, but compiling with frame pointers also disables 1414 * gcc's sibling call optimization. 1415 */ 1416 if (bio->bi_end_io == bio_chain_endio) { 1417 bio = __bio_chain_endio(bio); 1418 goto again; 1419 } 1420 1421 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1422 trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio); 1423 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1424 } 1425 1426 blk_throtl_bio_endio(bio); 1427 /* release cgroup info */ 1428 bio_uninit(bio); 1429 if (bio->bi_end_io) 1430 bio->bi_end_io(bio); 1431 } 1432 EXPORT_SYMBOL(bio_endio); 1433 1434 /** 1435 * bio_split - split a bio 1436 * @bio: bio to split 1437 * @sectors: number of sectors to split from the front of @bio 1438 * @gfp: gfp mask 1439 * @bs: bio set to allocate from 1440 * 1441 * Allocates and returns a new bio which represents @sectors from the start of 1442 * @bio, and updates @bio to represent the remaining sectors. 1443 * 1444 * Unless this is a discard request the newly allocated bio will point 1445 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1446 * neither @bio nor @bs are freed before the split bio. 1447 */ 1448 struct bio *bio_split(struct bio *bio, int sectors, 1449 gfp_t gfp, struct bio_set *bs) 1450 { 1451 struct bio *split; 1452 1453 BUG_ON(sectors <= 0); 1454 BUG_ON(sectors >= bio_sectors(bio)); 1455 1456 /* Zone append commands cannot be split */ 1457 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 1458 return NULL; 1459 1460 split = bio_clone_fast(bio, gfp, bs); 1461 if (!split) 1462 return NULL; 1463 1464 split->bi_iter.bi_size = sectors << 9; 1465 1466 if (bio_integrity(split)) 1467 bio_integrity_trim(split); 1468 1469 bio_advance(bio, split->bi_iter.bi_size); 1470 1471 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1472 bio_set_flag(split, BIO_TRACE_COMPLETION); 1473 1474 return split; 1475 } 1476 EXPORT_SYMBOL(bio_split); 1477 1478 /** 1479 * bio_trim - trim a bio 1480 * @bio: bio to trim 1481 * @offset: number of sectors to trim from the front of @bio 1482 * @size: size we want to trim @bio to, in sectors 1483 */ 1484 void bio_trim(struct bio *bio, int offset, int size) 1485 { 1486 /* 'bio' is a cloned bio which we need to trim to match 1487 * the given offset and size. 1488 */ 1489 1490 size <<= 9; 1491 if (offset == 0 && size == bio->bi_iter.bi_size) 1492 return; 1493 1494 bio_advance(bio, offset << 9); 1495 bio->bi_iter.bi_size = size; 1496 1497 if (bio_integrity(bio)) 1498 bio_integrity_trim(bio); 1499 1500 } 1501 EXPORT_SYMBOL_GPL(bio_trim); 1502 1503 /* 1504 * create memory pools for biovec's in a bio_set. 1505 * use the global biovec slabs created for general use. 1506 */ 1507 int biovec_init_pool(mempool_t *pool, int pool_entries) 1508 { 1509 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1510 1511 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1512 } 1513 1514 /* 1515 * bioset_exit - exit a bioset initialized with bioset_init() 1516 * 1517 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1518 * kzalloc()). 1519 */ 1520 void bioset_exit(struct bio_set *bs) 1521 { 1522 if (bs->rescue_workqueue) 1523 destroy_workqueue(bs->rescue_workqueue); 1524 bs->rescue_workqueue = NULL; 1525 1526 mempool_exit(&bs->bio_pool); 1527 mempool_exit(&bs->bvec_pool); 1528 1529 bioset_integrity_free(bs); 1530 if (bs->bio_slab) 1531 bio_put_slab(bs); 1532 bs->bio_slab = NULL; 1533 } 1534 EXPORT_SYMBOL(bioset_exit); 1535 1536 /** 1537 * bioset_init - Initialize a bio_set 1538 * @bs: pool to initialize 1539 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1540 * @front_pad: Number of bytes to allocate in front of the returned bio 1541 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1542 * and %BIOSET_NEED_RESCUER 1543 * 1544 * Description: 1545 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1546 * to ask for a number of bytes to be allocated in front of the bio. 1547 * Front pad allocation is useful for embedding the bio inside 1548 * another structure, to avoid allocating extra data to go with the bio. 1549 * Note that the bio must be embedded at the END of that structure always, 1550 * or things will break badly. 1551 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1552 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1553 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1554 * dispatch queued requests when the mempool runs out of space. 1555 * 1556 */ 1557 int bioset_init(struct bio_set *bs, 1558 unsigned int pool_size, 1559 unsigned int front_pad, 1560 int flags) 1561 { 1562 bs->front_pad = front_pad; 1563 if (flags & BIOSET_NEED_BVECS) 1564 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1565 else 1566 bs->back_pad = 0; 1567 1568 spin_lock_init(&bs->rescue_lock); 1569 bio_list_init(&bs->rescue_list); 1570 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1571 1572 bs->bio_slab = bio_find_or_create_slab(bs); 1573 if (!bs->bio_slab) 1574 return -ENOMEM; 1575 1576 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1577 goto bad; 1578 1579 if ((flags & BIOSET_NEED_BVECS) && 1580 biovec_init_pool(&bs->bvec_pool, pool_size)) 1581 goto bad; 1582 1583 if (!(flags & BIOSET_NEED_RESCUER)) 1584 return 0; 1585 1586 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1587 if (!bs->rescue_workqueue) 1588 goto bad; 1589 1590 return 0; 1591 bad: 1592 bioset_exit(bs); 1593 return -ENOMEM; 1594 } 1595 EXPORT_SYMBOL(bioset_init); 1596 1597 /* 1598 * Initialize and setup a new bio_set, based on the settings from 1599 * another bio_set. 1600 */ 1601 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 1602 { 1603 int flags; 1604 1605 flags = 0; 1606 if (src->bvec_pool.min_nr) 1607 flags |= BIOSET_NEED_BVECS; 1608 if (src->rescue_workqueue) 1609 flags |= BIOSET_NEED_RESCUER; 1610 1611 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 1612 } 1613 EXPORT_SYMBOL(bioset_init_from_src); 1614 1615 static void __init biovec_init_slabs(void) 1616 { 1617 int i; 1618 1619 for (i = 0; i < BVEC_POOL_NR; i++) { 1620 int size; 1621 struct biovec_slab *bvs = bvec_slabs + i; 1622 1623 if (bvs->nr_vecs <= BIO_INLINE_VECS) { 1624 bvs->slab = NULL; 1625 continue; 1626 } 1627 1628 size = bvs->nr_vecs * sizeof(struct bio_vec); 1629 bvs->slab = kmem_cache_create(bvs->name, size, 0, 1630 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 1631 } 1632 } 1633 1634 static int __init init_bio(void) 1635 { 1636 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 1637 1638 bio_integrity_init(); 1639 biovec_init_slabs(); 1640 1641 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1642 panic("bio: can't allocate bios\n"); 1643 1644 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1645 panic("bio: can't create integrity pool\n"); 1646 1647 return 0; 1648 } 1649 subsys_initcall(init_bio); 1650