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 22 #include <trace/events/block.h> 23 #include "blk.h" 24 #include "blk-rq-qos.h" 25 26 /* 27 * Test patch to inline a certain number of bi_io_vec's inside the bio 28 * itself, to shrink a bio data allocation from two mempool calls to one 29 */ 30 #define BIO_INLINE_VECS 4 31 32 /* 33 * if you change this list, also change bvec_alloc or things will 34 * break badly! cannot be bigger than what you can fit into an 35 * unsigned short 36 */ 37 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 38 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 39 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 40 }; 41 #undef BV 42 43 /* 44 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 45 * IO code that does not need private memory pools. 46 */ 47 struct bio_set fs_bio_set; 48 EXPORT_SYMBOL(fs_bio_set); 49 50 /* 51 * Our slab pool management 52 */ 53 struct bio_slab { 54 struct kmem_cache *slab; 55 unsigned int slab_ref; 56 unsigned int slab_size; 57 char name[8]; 58 }; 59 static DEFINE_MUTEX(bio_slab_lock); 60 static struct bio_slab *bio_slabs; 61 static unsigned int bio_slab_nr, bio_slab_max; 62 63 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 64 { 65 unsigned int sz = sizeof(struct bio) + extra_size; 66 struct kmem_cache *slab = NULL; 67 struct bio_slab *bslab, *new_bio_slabs; 68 unsigned int new_bio_slab_max; 69 unsigned int i, entry = -1; 70 71 mutex_lock(&bio_slab_lock); 72 73 i = 0; 74 while (i < bio_slab_nr) { 75 bslab = &bio_slabs[i]; 76 77 if (!bslab->slab && entry == -1) 78 entry = i; 79 else if (bslab->slab_size == sz) { 80 slab = bslab->slab; 81 bslab->slab_ref++; 82 break; 83 } 84 i++; 85 } 86 87 if (slab) 88 goto out_unlock; 89 90 if (bio_slab_nr == bio_slab_max && entry == -1) { 91 new_bio_slab_max = bio_slab_max << 1; 92 new_bio_slabs = krealloc(bio_slabs, 93 new_bio_slab_max * sizeof(struct bio_slab), 94 GFP_KERNEL); 95 if (!new_bio_slabs) 96 goto out_unlock; 97 bio_slab_max = new_bio_slab_max; 98 bio_slabs = new_bio_slabs; 99 } 100 if (entry == -1) 101 entry = bio_slab_nr++; 102 103 bslab = &bio_slabs[entry]; 104 105 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 106 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 107 SLAB_HWCACHE_ALIGN, NULL); 108 if (!slab) 109 goto out_unlock; 110 111 bslab->slab = slab; 112 bslab->slab_ref = 1; 113 bslab->slab_size = sz; 114 out_unlock: 115 mutex_unlock(&bio_slab_lock); 116 return slab; 117 } 118 119 static void bio_put_slab(struct bio_set *bs) 120 { 121 struct bio_slab *bslab = NULL; 122 unsigned int i; 123 124 mutex_lock(&bio_slab_lock); 125 126 for (i = 0; i < bio_slab_nr; i++) { 127 if (bs->bio_slab == bio_slabs[i].slab) { 128 bslab = &bio_slabs[i]; 129 break; 130 } 131 } 132 133 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 134 goto out; 135 136 WARN_ON(!bslab->slab_ref); 137 138 if (--bslab->slab_ref) 139 goto out; 140 141 kmem_cache_destroy(bslab->slab); 142 bslab->slab = NULL; 143 144 out: 145 mutex_unlock(&bio_slab_lock); 146 } 147 148 unsigned int bvec_nr_vecs(unsigned short idx) 149 { 150 return bvec_slabs[--idx].nr_vecs; 151 } 152 153 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 154 { 155 if (!idx) 156 return; 157 idx--; 158 159 BIO_BUG_ON(idx >= BVEC_POOL_NR); 160 161 if (idx == BVEC_POOL_MAX) { 162 mempool_free(bv, pool); 163 } else { 164 struct biovec_slab *bvs = bvec_slabs + idx; 165 166 kmem_cache_free(bvs->slab, bv); 167 } 168 } 169 170 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 171 mempool_t *pool) 172 { 173 struct bio_vec *bvl; 174 175 /* 176 * see comment near bvec_array define! 177 */ 178 switch (nr) { 179 case 1: 180 *idx = 0; 181 break; 182 case 2 ... 4: 183 *idx = 1; 184 break; 185 case 5 ... 16: 186 *idx = 2; 187 break; 188 case 17 ... 64: 189 *idx = 3; 190 break; 191 case 65 ... 128: 192 *idx = 4; 193 break; 194 case 129 ... BIO_MAX_PAGES: 195 *idx = 5; 196 break; 197 default: 198 return NULL; 199 } 200 201 /* 202 * idx now points to the pool we want to allocate from. only the 203 * 1-vec entry pool is mempool backed. 204 */ 205 if (*idx == BVEC_POOL_MAX) { 206 fallback: 207 bvl = mempool_alloc(pool, gfp_mask); 208 } else { 209 struct biovec_slab *bvs = bvec_slabs + *idx; 210 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 211 212 /* 213 * Make this allocation restricted and don't dump info on 214 * allocation failures, since we'll fallback to the mempool 215 * in case of failure. 216 */ 217 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 218 219 /* 220 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 221 * is set, retry with the 1-entry mempool 222 */ 223 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 224 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 225 *idx = BVEC_POOL_MAX; 226 goto fallback; 227 } 228 } 229 230 (*idx)++; 231 return bvl; 232 } 233 234 void bio_uninit(struct bio *bio) 235 { 236 bio_disassociate_blkg(bio); 237 238 if (bio_integrity(bio)) 239 bio_integrity_free(bio); 240 } 241 EXPORT_SYMBOL(bio_uninit); 242 243 static void bio_free(struct bio *bio) 244 { 245 struct bio_set *bs = bio->bi_pool; 246 void *p; 247 248 bio_uninit(bio); 249 250 if (bs) { 251 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 252 253 /* 254 * If we have front padding, adjust the bio pointer before freeing 255 */ 256 p = bio; 257 p -= bs->front_pad; 258 259 mempool_free(p, &bs->bio_pool); 260 } else { 261 /* Bio was allocated by bio_kmalloc() */ 262 kfree(bio); 263 } 264 } 265 266 /* 267 * Users of this function have their own bio allocation. Subsequently, 268 * they must remember to pair any call to bio_init() with bio_uninit() 269 * when IO has completed, or when the bio is released. 270 */ 271 void bio_init(struct bio *bio, struct bio_vec *table, 272 unsigned short max_vecs) 273 { 274 memset(bio, 0, sizeof(*bio)); 275 atomic_set(&bio->__bi_remaining, 1); 276 atomic_set(&bio->__bi_cnt, 1); 277 278 bio->bi_io_vec = table; 279 bio->bi_max_vecs = max_vecs; 280 } 281 EXPORT_SYMBOL(bio_init); 282 283 /** 284 * bio_reset - reinitialize a bio 285 * @bio: bio to reset 286 * 287 * Description: 288 * After calling bio_reset(), @bio will be in the same state as a freshly 289 * allocated bio returned bio bio_alloc_bioset() - the only fields that are 290 * preserved are the ones that are initialized by bio_alloc_bioset(). See 291 * comment in struct bio. 292 */ 293 void bio_reset(struct bio *bio) 294 { 295 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 296 297 bio_uninit(bio); 298 299 memset(bio, 0, BIO_RESET_BYTES); 300 bio->bi_flags = flags; 301 atomic_set(&bio->__bi_remaining, 1); 302 } 303 EXPORT_SYMBOL(bio_reset); 304 305 static struct bio *__bio_chain_endio(struct bio *bio) 306 { 307 struct bio *parent = bio->bi_private; 308 309 if (!parent->bi_status) 310 parent->bi_status = bio->bi_status; 311 bio_put(bio); 312 return parent; 313 } 314 315 static void bio_chain_endio(struct bio *bio) 316 { 317 bio_endio(__bio_chain_endio(bio)); 318 } 319 320 /** 321 * bio_chain - chain bio completions 322 * @bio: the target bio 323 * @parent: the @bio's parent bio 324 * 325 * The caller won't have a bi_end_io called when @bio completes - instead, 326 * @parent's bi_end_io won't be called until both @parent and @bio have 327 * completed; the chained bio will also be freed when it completes. 328 * 329 * The caller must not set bi_private or bi_end_io in @bio. 330 */ 331 void bio_chain(struct bio *bio, struct bio *parent) 332 { 333 BUG_ON(bio->bi_private || bio->bi_end_io); 334 335 bio->bi_private = parent; 336 bio->bi_end_io = bio_chain_endio; 337 bio_inc_remaining(parent); 338 } 339 EXPORT_SYMBOL(bio_chain); 340 341 static void bio_alloc_rescue(struct work_struct *work) 342 { 343 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 344 struct bio *bio; 345 346 while (1) { 347 spin_lock(&bs->rescue_lock); 348 bio = bio_list_pop(&bs->rescue_list); 349 spin_unlock(&bs->rescue_lock); 350 351 if (!bio) 352 break; 353 354 generic_make_request(bio); 355 } 356 } 357 358 static void punt_bios_to_rescuer(struct bio_set *bs) 359 { 360 struct bio_list punt, nopunt; 361 struct bio *bio; 362 363 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 364 return; 365 /* 366 * In order to guarantee forward progress we must punt only bios that 367 * were allocated from this bio_set; otherwise, if there was a bio on 368 * there for a stacking driver higher up in the stack, processing it 369 * could require allocating bios from this bio_set, and doing that from 370 * our own rescuer would be bad. 371 * 372 * Since bio lists are singly linked, pop them all instead of trying to 373 * remove from the middle of the list: 374 */ 375 376 bio_list_init(&punt); 377 bio_list_init(&nopunt); 378 379 while ((bio = bio_list_pop(¤t->bio_list[0]))) 380 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 381 current->bio_list[0] = nopunt; 382 383 bio_list_init(&nopunt); 384 while ((bio = bio_list_pop(¤t->bio_list[1]))) 385 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 386 current->bio_list[1] = nopunt; 387 388 spin_lock(&bs->rescue_lock); 389 bio_list_merge(&bs->rescue_list, &punt); 390 spin_unlock(&bs->rescue_lock); 391 392 queue_work(bs->rescue_workqueue, &bs->rescue_work); 393 } 394 395 /** 396 * bio_alloc_bioset - allocate a bio for I/O 397 * @gfp_mask: the GFP_* mask given to the slab allocator 398 * @nr_iovecs: number of iovecs to pre-allocate 399 * @bs: the bio_set to allocate from. 400 * 401 * Description: 402 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 403 * backed by the @bs's mempool. 404 * 405 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 406 * always be able to allocate a bio. This is due to the mempool guarantees. 407 * To make this work, callers must never allocate more than 1 bio at a time 408 * from this pool. Callers that need to allocate more than 1 bio must always 409 * submit the previously allocated bio for IO before attempting to allocate 410 * a new one. Failure to do so can cause deadlocks under memory pressure. 411 * 412 * Note that when running under generic_make_request() (i.e. any block 413 * driver), bios are not submitted until after you return - see the code in 414 * generic_make_request() that converts recursion into iteration, to prevent 415 * stack overflows. 416 * 417 * This would normally mean allocating multiple bios under 418 * generic_make_request() would be susceptible to deadlocks, but we have 419 * deadlock avoidance code that resubmits any blocked bios from a rescuer 420 * thread. 421 * 422 * However, we do not guarantee forward progress for allocations from other 423 * mempools. Doing multiple allocations from the same mempool under 424 * generic_make_request() should be avoided - instead, use bio_set's front_pad 425 * for per bio allocations. 426 * 427 * RETURNS: 428 * Pointer to new bio on success, NULL on failure. 429 */ 430 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 431 struct bio_set *bs) 432 { 433 gfp_t saved_gfp = gfp_mask; 434 unsigned front_pad; 435 unsigned inline_vecs; 436 struct bio_vec *bvl = NULL; 437 struct bio *bio; 438 void *p; 439 440 if (!bs) { 441 if (nr_iovecs > UIO_MAXIOV) 442 return NULL; 443 444 p = kmalloc(sizeof(struct bio) + 445 nr_iovecs * sizeof(struct bio_vec), 446 gfp_mask); 447 front_pad = 0; 448 inline_vecs = nr_iovecs; 449 } else { 450 /* should not use nobvec bioset for nr_iovecs > 0 */ 451 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 452 nr_iovecs > 0)) 453 return NULL; 454 /* 455 * generic_make_request() converts recursion to iteration; this 456 * means if we're running beneath it, any bios we allocate and 457 * submit will not be submitted (and thus freed) until after we 458 * return. 459 * 460 * This exposes us to a potential deadlock if we allocate 461 * multiple bios from the same bio_set() while running 462 * underneath generic_make_request(). If we were to allocate 463 * multiple bios (say a stacking block driver that was splitting 464 * bios), we would deadlock if we exhausted the mempool's 465 * reserve. 466 * 467 * We solve this, and guarantee forward progress, with a rescuer 468 * workqueue per bio_set. If we go to allocate and there are 469 * bios on current->bio_list, we first try the allocation 470 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 471 * bios we would be blocking to the rescuer workqueue before 472 * we retry with the original gfp_flags. 473 */ 474 475 if (current->bio_list && 476 (!bio_list_empty(¤t->bio_list[0]) || 477 !bio_list_empty(¤t->bio_list[1])) && 478 bs->rescue_workqueue) 479 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 480 481 p = mempool_alloc(&bs->bio_pool, gfp_mask); 482 if (!p && gfp_mask != saved_gfp) { 483 punt_bios_to_rescuer(bs); 484 gfp_mask = saved_gfp; 485 p = mempool_alloc(&bs->bio_pool, gfp_mask); 486 } 487 488 front_pad = bs->front_pad; 489 inline_vecs = BIO_INLINE_VECS; 490 } 491 492 if (unlikely(!p)) 493 return NULL; 494 495 bio = p + front_pad; 496 bio_init(bio, NULL, 0); 497 498 if (nr_iovecs > inline_vecs) { 499 unsigned long idx = 0; 500 501 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 502 if (!bvl && gfp_mask != saved_gfp) { 503 punt_bios_to_rescuer(bs); 504 gfp_mask = saved_gfp; 505 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 506 } 507 508 if (unlikely(!bvl)) 509 goto err_free; 510 511 bio->bi_flags |= idx << BVEC_POOL_OFFSET; 512 } else if (nr_iovecs) { 513 bvl = bio->bi_inline_vecs; 514 } 515 516 bio->bi_pool = bs; 517 bio->bi_max_vecs = nr_iovecs; 518 bio->bi_io_vec = bvl; 519 return bio; 520 521 err_free: 522 mempool_free(p, &bs->bio_pool); 523 return NULL; 524 } 525 EXPORT_SYMBOL(bio_alloc_bioset); 526 527 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 528 { 529 unsigned long flags; 530 struct bio_vec bv; 531 struct bvec_iter iter; 532 533 __bio_for_each_segment(bv, bio, iter, start) { 534 char *data = bvec_kmap_irq(&bv, &flags); 535 memset(data, 0, bv.bv_len); 536 flush_dcache_page(bv.bv_page); 537 bvec_kunmap_irq(data, &flags); 538 } 539 } 540 EXPORT_SYMBOL(zero_fill_bio_iter); 541 542 /** 543 * bio_truncate - truncate the bio to small size of @new_size 544 * @bio: the bio to be truncated 545 * @new_size: new size for truncating the bio 546 * 547 * Description: 548 * Truncate the bio to new size of @new_size. If bio_op(bio) is 549 * REQ_OP_READ, zero the truncated part. This function should only 550 * be used for handling corner cases, such as bio eod. 551 */ 552 void bio_truncate(struct bio *bio, unsigned new_size) 553 { 554 struct bio_vec bv; 555 struct bvec_iter iter; 556 unsigned int done = 0; 557 bool truncated = false; 558 559 if (new_size >= bio->bi_iter.bi_size) 560 return; 561 562 if (bio_op(bio) != REQ_OP_READ) 563 goto exit; 564 565 bio_for_each_segment(bv, bio, iter) { 566 if (done + bv.bv_len > new_size) { 567 unsigned offset; 568 569 if (!truncated) 570 offset = new_size - done; 571 else 572 offset = 0; 573 zero_user(bv.bv_page, offset, bv.bv_len - offset); 574 truncated = true; 575 } 576 done += bv.bv_len; 577 } 578 579 exit: 580 /* 581 * Don't touch bvec table here and make it really immutable, since 582 * fs bio user has to retrieve all pages via bio_for_each_segment_all 583 * in its .end_bio() callback. 584 * 585 * It is enough to truncate bio by updating .bi_size since we can make 586 * correct bvec with the updated .bi_size for drivers. 587 */ 588 bio->bi_iter.bi_size = new_size; 589 } 590 591 /** 592 * guard_bio_eod - truncate a BIO to fit the block device 593 * @bio: bio to truncate 594 * 595 * This allows us to do IO even on the odd last sectors of a device, even if the 596 * block size is some multiple of the physical sector size. 597 * 598 * We'll just truncate the bio to the size of the device, and clear the end of 599 * the buffer head manually. Truly out-of-range accesses will turn into actual 600 * I/O errors, this only handles the "we need to be able to do I/O at the final 601 * sector" case. 602 */ 603 void guard_bio_eod(struct bio *bio) 604 { 605 sector_t maxsector; 606 struct hd_struct *part; 607 608 rcu_read_lock(); 609 part = __disk_get_part(bio->bi_disk, bio->bi_partno); 610 if (part) 611 maxsector = part_nr_sects_read(part); 612 else 613 maxsector = get_capacity(bio->bi_disk); 614 rcu_read_unlock(); 615 616 if (!maxsector) 617 return; 618 619 /* 620 * If the *whole* IO is past the end of the device, 621 * let it through, and the IO layer will turn it into 622 * an EIO. 623 */ 624 if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 625 return; 626 627 maxsector -= bio->bi_iter.bi_sector; 628 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 629 return; 630 631 bio_truncate(bio, maxsector << 9); 632 } 633 634 /** 635 * bio_put - release a reference to a bio 636 * @bio: bio to release reference to 637 * 638 * Description: 639 * Put a reference to a &struct bio, either one you have gotten with 640 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 641 **/ 642 void bio_put(struct bio *bio) 643 { 644 if (!bio_flagged(bio, BIO_REFFED)) 645 bio_free(bio); 646 else { 647 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 648 649 /* 650 * last put frees it 651 */ 652 if (atomic_dec_and_test(&bio->__bi_cnt)) 653 bio_free(bio); 654 } 655 } 656 EXPORT_SYMBOL(bio_put); 657 658 /** 659 * __bio_clone_fast - clone a bio that shares the original bio's biovec 660 * @bio: destination bio 661 * @bio_src: bio to clone 662 * 663 * Clone a &bio. Caller will own the returned bio, but not 664 * the actual data it points to. Reference count of returned 665 * bio will be one. 666 * 667 * Caller must ensure that @bio_src is not freed before @bio. 668 */ 669 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 670 { 671 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 672 673 /* 674 * most users will be overriding ->bi_disk with a new target, 675 * so we don't set nor calculate new physical/hw segment counts here 676 */ 677 bio->bi_disk = bio_src->bi_disk; 678 bio->bi_partno = bio_src->bi_partno; 679 bio_set_flag(bio, BIO_CLONED); 680 if (bio_flagged(bio_src, BIO_THROTTLED)) 681 bio_set_flag(bio, BIO_THROTTLED); 682 bio->bi_opf = bio_src->bi_opf; 683 bio->bi_ioprio = bio_src->bi_ioprio; 684 bio->bi_write_hint = bio_src->bi_write_hint; 685 bio->bi_iter = bio_src->bi_iter; 686 bio->bi_io_vec = bio_src->bi_io_vec; 687 688 bio_clone_blkg_association(bio, bio_src); 689 blkcg_bio_issue_init(bio); 690 } 691 EXPORT_SYMBOL(__bio_clone_fast); 692 693 /** 694 * bio_clone_fast - clone a bio that shares the original bio's biovec 695 * @bio: bio to clone 696 * @gfp_mask: allocation priority 697 * @bs: bio_set to allocate from 698 * 699 * Like __bio_clone_fast, only also allocates the returned bio 700 */ 701 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 702 { 703 struct bio *b; 704 705 b = bio_alloc_bioset(gfp_mask, 0, bs); 706 if (!b) 707 return NULL; 708 709 __bio_clone_fast(b, bio); 710 711 if (bio_integrity(bio)) { 712 int ret; 713 714 ret = bio_integrity_clone(b, bio, gfp_mask); 715 716 if (ret < 0) { 717 bio_put(b); 718 return NULL; 719 } 720 } 721 722 return b; 723 } 724 EXPORT_SYMBOL(bio_clone_fast); 725 726 const char *bio_devname(struct bio *bio, char *buf) 727 { 728 return disk_name(bio->bi_disk, bio->bi_partno, buf); 729 } 730 EXPORT_SYMBOL(bio_devname); 731 732 static inline bool page_is_mergeable(const struct bio_vec *bv, 733 struct page *page, unsigned int len, unsigned int off, 734 bool *same_page) 735 { 736 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 737 bv->bv_offset + bv->bv_len - 1; 738 phys_addr_t page_addr = page_to_phys(page); 739 740 if (vec_end_addr + 1 != page_addr + off) 741 return false; 742 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 743 return false; 744 745 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 746 if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) 747 return false; 748 return true; 749 } 750 751 static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio, 752 struct page *page, unsigned len, unsigned offset, 753 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_pc_page - attempt to add page to passthrough bio 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 * @same_page: return if the merge happen inside the same page 775 * 776 * Attempt to add a page to the bio_vec maplist. This can fail for a 777 * number of reasons, such as the bio being full or target block device 778 * limitations. The target block device must allow bio's up to PAGE_SIZE, 779 * so it is always possible to add a single page to an empty bio. 780 * 781 * This should only be used by passthrough bios. 782 */ 783 static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, 784 struct page *page, unsigned int len, unsigned int offset, 785 bool *same_page) 786 { 787 struct bio_vec *bvec; 788 789 /* 790 * cloned bio must not modify vec list 791 */ 792 if (unlikely(bio_flagged(bio, BIO_CLONED))) 793 return 0; 794 795 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 796 return 0; 797 798 if (bio->bi_vcnt > 0) { 799 if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page)) 800 return len; 801 802 /* 803 * If the queue doesn't support SG gaps and adding this segment 804 * would create a gap, disallow it. 805 */ 806 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 807 if (bvec_gap_to_prev(q, bvec, offset)) 808 return 0; 809 } 810 811 if (bio_full(bio, len)) 812 return 0; 813 814 if (bio->bi_vcnt >= queue_max_segments(q)) 815 return 0; 816 817 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 818 bvec->bv_page = page; 819 bvec->bv_len = len; 820 bvec->bv_offset = offset; 821 bio->bi_vcnt++; 822 bio->bi_iter.bi_size += len; 823 return len; 824 } 825 826 int bio_add_pc_page(struct request_queue *q, struct bio *bio, 827 struct page *page, unsigned int len, unsigned int offset) 828 { 829 bool same_page = false; 830 return __bio_add_pc_page(q, bio, page, len, offset, &same_page); 831 } 832 EXPORT_SYMBOL(bio_add_pc_page); 833 834 /** 835 * __bio_try_merge_page - try appending data to an existing bvec. 836 * @bio: destination bio 837 * @page: start page to add 838 * @len: length of the data to add 839 * @off: offset of the data relative to @page 840 * @same_page: return if the segment has been merged inside the same page 841 * 842 * Try to add the data at @page + @off to the last bvec of @bio. This is a 843 * a useful optimisation for file systems with a block size smaller than the 844 * page size. 845 * 846 * Warn if (@len, @off) crosses pages in case that @same_page is true. 847 * 848 * Return %true on success or %false on failure. 849 */ 850 bool __bio_try_merge_page(struct bio *bio, struct page *page, 851 unsigned int len, unsigned int off, bool *same_page) 852 { 853 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 854 return false; 855 856 if (bio->bi_vcnt > 0) { 857 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 858 859 if (page_is_mergeable(bv, page, len, off, same_page)) { 860 if (bio->bi_iter.bi_size > UINT_MAX - len) 861 return false; 862 bv->bv_len += len; 863 bio->bi_iter.bi_size += len; 864 return true; 865 } 866 } 867 return false; 868 } 869 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 870 871 /** 872 * __bio_add_page - add page(s) to a bio in a new segment 873 * @bio: destination bio 874 * @page: start page to add 875 * @len: length of the data to add, may cross pages 876 * @off: offset of the data relative to @page, may cross pages 877 * 878 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 879 * that @bio has space for another bvec. 880 */ 881 void __bio_add_page(struct bio *bio, struct page *page, 882 unsigned int len, unsigned int off) 883 { 884 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 885 886 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 887 WARN_ON_ONCE(bio_full(bio, len)); 888 889 bv->bv_page = page; 890 bv->bv_offset = off; 891 bv->bv_len = len; 892 893 bio->bi_iter.bi_size += len; 894 bio->bi_vcnt++; 895 896 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 897 bio_set_flag(bio, BIO_WORKINGSET); 898 } 899 EXPORT_SYMBOL_GPL(__bio_add_page); 900 901 /** 902 * bio_add_page - attempt to add page(s) to bio 903 * @bio: destination bio 904 * @page: start page to add 905 * @len: vec entry length, may cross pages 906 * @offset: vec entry offset relative to @page, may cross pages 907 * 908 * Attempt to add page(s) to the bio_vec maplist. This will only fail 909 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 910 */ 911 int bio_add_page(struct bio *bio, struct page *page, 912 unsigned int len, unsigned int offset) 913 { 914 bool same_page = false; 915 916 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 917 if (bio_full(bio, len)) 918 return 0; 919 __bio_add_page(bio, page, len, offset); 920 } 921 return len; 922 } 923 EXPORT_SYMBOL(bio_add_page); 924 925 void bio_release_pages(struct bio *bio, bool mark_dirty) 926 { 927 struct bvec_iter_all iter_all; 928 struct bio_vec *bvec; 929 930 if (bio_flagged(bio, BIO_NO_PAGE_REF)) 931 return; 932 933 bio_for_each_segment_all(bvec, bio, iter_all) { 934 if (mark_dirty && !PageCompound(bvec->bv_page)) 935 set_page_dirty_lock(bvec->bv_page); 936 put_page(bvec->bv_page); 937 } 938 } 939 940 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 941 { 942 const struct bio_vec *bv = iter->bvec; 943 unsigned int len; 944 size_t size; 945 946 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 947 return -EINVAL; 948 949 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 950 size = bio_add_page(bio, bv->bv_page, len, 951 bv->bv_offset + iter->iov_offset); 952 if (unlikely(size != len)) 953 return -EINVAL; 954 iov_iter_advance(iter, size); 955 return 0; 956 } 957 958 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 959 960 /** 961 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 962 * @bio: bio to add pages to 963 * @iter: iov iterator describing the region to be mapped 964 * 965 * Pins pages from *iter and appends them to @bio's bvec array. The 966 * pages will have to be released using put_page() when done. 967 * For multi-segment *iter, this function only adds pages from the 968 * the next non-empty segment of the iov iterator. 969 */ 970 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 971 { 972 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 973 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 974 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 975 struct page **pages = (struct page **)bv; 976 bool same_page = false; 977 ssize_t size, left; 978 unsigned len, i; 979 size_t offset; 980 981 /* 982 * Move page array up in the allocated memory for the bio vecs as far as 983 * possible so that we can start filling biovecs from the beginning 984 * without overwriting the temporary page array. 985 */ 986 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 987 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 988 989 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 990 if (unlikely(size <= 0)) 991 return size ? size : -EFAULT; 992 993 for (left = size, i = 0; left > 0; left -= len, i++) { 994 struct page *page = pages[i]; 995 996 len = min_t(size_t, PAGE_SIZE - offset, left); 997 998 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 999 if (same_page) 1000 put_page(page); 1001 } else { 1002 if (WARN_ON_ONCE(bio_full(bio, len))) 1003 return -EINVAL; 1004 __bio_add_page(bio, page, len, offset); 1005 } 1006 offset = 0; 1007 } 1008 1009 iov_iter_advance(iter, size); 1010 return 0; 1011 } 1012 1013 /** 1014 * bio_iov_iter_get_pages - add user or kernel pages to a bio 1015 * @bio: bio to add pages to 1016 * @iter: iov iterator describing the region to be added 1017 * 1018 * This takes either an iterator pointing to user memory, or one pointing to 1019 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 1020 * map them into the kernel. On IO completion, the caller should put those 1021 * pages. If we're adding kernel pages, and the caller told us it's safe to 1022 * do so, we just have to add the pages to the bio directly. We don't grab an 1023 * extra reference to those pages (the user should already have that), and we 1024 * don't put the page on IO completion. The caller needs to check if the bio is 1025 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 1026 * released. 1027 * 1028 * The function tries, but does not guarantee, to pin as many pages as 1029 * fit into the bio, or are requested in *iter, whatever is smaller. If 1030 * MM encounters an error pinning the requested pages, it stops. Error 1031 * is returned only if 0 pages could be pinned. 1032 */ 1033 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 1034 { 1035 const bool is_bvec = iov_iter_is_bvec(iter); 1036 int ret; 1037 1038 if (WARN_ON_ONCE(bio->bi_vcnt)) 1039 return -EINVAL; 1040 1041 do { 1042 if (is_bvec) 1043 ret = __bio_iov_bvec_add_pages(bio, iter); 1044 else 1045 ret = __bio_iov_iter_get_pages(bio, iter); 1046 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1047 1048 if (is_bvec) 1049 bio_set_flag(bio, BIO_NO_PAGE_REF); 1050 return bio->bi_vcnt ? 0 : ret; 1051 } 1052 1053 static void submit_bio_wait_endio(struct bio *bio) 1054 { 1055 complete(bio->bi_private); 1056 } 1057 1058 /** 1059 * submit_bio_wait - submit a bio, and wait until it completes 1060 * @bio: The &struct bio which describes the I/O 1061 * 1062 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1063 * bio_endio() on failure. 1064 * 1065 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1066 * result in bio reference to be consumed. The caller must drop the reference 1067 * on his own. 1068 */ 1069 int submit_bio_wait(struct bio *bio) 1070 { 1071 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 1072 unsigned long hang_check; 1073 1074 bio->bi_private = &done; 1075 bio->bi_end_io = submit_bio_wait_endio; 1076 bio->bi_opf |= REQ_SYNC; 1077 submit_bio(bio); 1078 1079 /* Prevent hang_check timer from firing at us during very long I/O */ 1080 hang_check = sysctl_hung_task_timeout_secs; 1081 if (hang_check) 1082 while (!wait_for_completion_io_timeout(&done, 1083 hang_check * (HZ/2))) 1084 ; 1085 else 1086 wait_for_completion_io(&done); 1087 1088 return blk_status_to_errno(bio->bi_status); 1089 } 1090 EXPORT_SYMBOL(submit_bio_wait); 1091 1092 /** 1093 * bio_advance - increment/complete a bio by some number of bytes 1094 * @bio: bio to advance 1095 * @bytes: number of bytes to complete 1096 * 1097 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1098 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1099 * be updated on the last bvec as well. 1100 * 1101 * @bio will then represent the remaining, uncompleted portion of the io. 1102 */ 1103 void bio_advance(struct bio *bio, unsigned bytes) 1104 { 1105 if (bio_integrity(bio)) 1106 bio_integrity_advance(bio, bytes); 1107 1108 bio_advance_iter(bio, &bio->bi_iter, bytes); 1109 } 1110 EXPORT_SYMBOL(bio_advance); 1111 1112 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1113 struct bio *src, struct bvec_iter *src_iter) 1114 { 1115 struct bio_vec src_bv, dst_bv; 1116 void *src_p, *dst_p; 1117 unsigned bytes; 1118 1119 while (src_iter->bi_size && dst_iter->bi_size) { 1120 src_bv = bio_iter_iovec(src, *src_iter); 1121 dst_bv = bio_iter_iovec(dst, *dst_iter); 1122 1123 bytes = min(src_bv.bv_len, dst_bv.bv_len); 1124 1125 src_p = kmap_atomic(src_bv.bv_page); 1126 dst_p = kmap_atomic(dst_bv.bv_page); 1127 1128 memcpy(dst_p + dst_bv.bv_offset, 1129 src_p + src_bv.bv_offset, 1130 bytes); 1131 1132 kunmap_atomic(dst_p); 1133 kunmap_atomic(src_p); 1134 1135 flush_dcache_page(dst_bv.bv_page); 1136 1137 bio_advance_iter(src, src_iter, bytes); 1138 bio_advance_iter(dst, dst_iter, bytes); 1139 } 1140 } 1141 EXPORT_SYMBOL(bio_copy_data_iter); 1142 1143 /** 1144 * bio_copy_data - copy contents of data buffers from one bio to another 1145 * @src: source bio 1146 * @dst: destination bio 1147 * 1148 * Stops when it reaches the end of either @src or @dst - that is, copies 1149 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1150 */ 1151 void bio_copy_data(struct bio *dst, struct bio *src) 1152 { 1153 struct bvec_iter src_iter = src->bi_iter; 1154 struct bvec_iter dst_iter = dst->bi_iter; 1155 1156 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1157 } 1158 EXPORT_SYMBOL(bio_copy_data); 1159 1160 /** 1161 * bio_list_copy_data - copy contents of data buffers from one chain of bios to 1162 * another 1163 * @src: source bio list 1164 * @dst: destination bio list 1165 * 1166 * Stops when it reaches the end of either the @src list or @dst list - that is, 1167 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 1168 * bios). 1169 */ 1170 void bio_list_copy_data(struct bio *dst, struct bio *src) 1171 { 1172 struct bvec_iter src_iter = src->bi_iter; 1173 struct bvec_iter dst_iter = dst->bi_iter; 1174 1175 while (1) { 1176 if (!src_iter.bi_size) { 1177 src = src->bi_next; 1178 if (!src) 1179 break; 1180 1181 src_iter = src->bi_iter; 1182 } 1183 1184 if (!dst_iter.bi_size) { 1185 dst = dst->bi_next; 1186 if (!dst) 1187 break; 1188 1189 dst_iter = dst->bi_iter; 1190 } 1191 1192 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1193 } 1194 } 1195 EXPORT_SYMBOL(bio_list_copy_data); 1196 1197 struct bio_map_data { 1198 int is_our_pages; 1199 struct iov_iter iter; 1200 struct iovec iov[]; 1201 }; 1202 1203 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1204 gfp_t gfp_mask) 1205 { 1206 struct bio_map_data *bmd; 1207 if (data->nr_segs > UIO_MAXIOV) 1208 return NULL; 1209 1210 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); 1211 if (!bmd) 1212 return NULL; 1213 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 1214 bmd->iter = *data; 1215 bmd->iter.iov = bmd->iov; 1216 return bmd; 1217 } 1218 1219 /** 1220 * bio_copy_from_iter - copy all pages from iov_iter to bio 1221 * @bio: The &struct bio which describes the I/O as destination 1222 * @iter: iov_iter as source 1223 * 1224 * Copy all pages from iov_iter to bio. 1225 * Returns 0 on success, or error on failure. 1226 */ 1227 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1228 { 1229 struct bio_vec *bvec; 1230 struct bvec_iter_all iter_all; 1231 1232 bio_for_each_segment_all(bvec, bio, iter_all) { 1233 ssize_t ret; 1234 1235 ret = copy_page_from_iter(bvec->bv_page, 1236 bvec->bv_offset, 1237 bvec->bv_len, 1238 iter); 1239 1240 if (!iov_iter_count(iter)) 1241 break; 1242 1243 if (ret < bvec->bv_len) 1244 return -EFAULT; 1245 } 1246 1247 return 0; 1248 } 1249 1250 /** 1251 * bio_copy_to_iter - copy all pages from bio to iov_iter 1252 * @bio: The &struct bio which describes the I/O as source 1253 * @iter: iov_iter as destination 1254 * 1255 * Copy all pages from bio to iov_iter. 1256 * Returns 0 on success, or error on failure. 1257 */ 1258 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 1259 { 1260 struct bio_vec *bvec; 1261 struct bvec_iter_all iter_all; 1262 1263 bio_for_each_segment_all(bvec, bio, iter_all) { 1264 ssize_t ret; 1265 1266 ret = copy_page_to_iter(bvec->bv_page, 1267 bvec->bv_offset, 1268 bvec->bv_len, 1269 &iter); 1270 1271 if (!iov_iter_count(&iter)) 1272 break; 1273 1274 if (ret < bvec->bv_len) 1275 return -EFAULT; 1276 } 1277 1278 return 0; 1279 } 1280 1281 void bio_free_pages(struct bio *bio) 1282 { 1283 struct bio_vec *bvec; 1284 struct bvec_iter_all iter_all; 1285 1286 bio_for_each_segment_all(bvec, bio, iter_all) 1287 __free_page(bvec->bv_page); 1288 } 1289 EXPORT_SYMBOL(bio_free_pages); 1290 1291 /** 1292 * bio_uncopy_user - finish previously mapped bio 1293 * @bio: bio being terminated 1294 * 1295 * Free pages allocated from bio_copy_user_iov() and write back data 1296 * to user space in case of a read. 1297 */ 1298 int bio_uncopy_user(struct bio *bio) 1299 { 1300 struct bio_map_data *bmd = bio->bi_private; 1301 int ret = 0; 1302 1303 if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1304 /* 1305 * if we're in a workqueue, the request is orphaned, so 1306 * don't copy into a random user address space, just free 1307 * and return -EINTR so user space doesn't expect any data. 1308 */ 1309 if (!current->mm) 1310 ret = -EINTR; 1311 else if (bio_data_dir(bio) == READ) 1312 ret = bio_copy_to_iter(bio, bmd->iter); 1313 if (bmd->is_our_pages) 1314 bio_free_pages(bio); 1315 } 1316 kfree(bmd); 1317 bio_put(bio); 1318 return ret; 1319 } 1320 1321 /** 1322 * bio_copy_user_iov - copy user data to bio 1323 * @q: destination block queue 1324 * @map_data: pointer to the rq_map_data holding pages (if necessary) 1325 * @iter: iovec iterator 1326 * @gfp_mask: memory allocation flags 1327 * 1328 * Prepares and returns a bio for indirect user io, bouncing data 1329 * to/from kernel pages as necessary. Must be paired with 1330 * call bio_uncopy_user() on io completion. 1331 */ 1332 struct bio *bio_copy_user_iov(struct request_queue *q, 1333 struct rq_map_data *map_data, 1334 struct iov_iter *iter, 1335 gfp_t gfp_mask) 1336 { 1337 struct bio_map_data *bmd; 1338 struct page *page; 1339 struct bio *bio; 1340 int i = 0, ret; 1341 int nr_pages; 1342 unsigned int len = iter->count; 1343 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1344 1345 bmd = bio_alloc_map_data(iter, gfp_mask); 1346 if (!bmd) 1347 return ERR_PTR(-ENOMEM); 1348 1349 /* 1350 * We need to do a deep copy of the iov_iter including the iovecs. 1351 * The caller provided iov might point to an on-stack or otherwise 1352 * shortlived one. 1353 */ 1354 bmd->is_our_pages = map_data ? 0 : 1; 1355 1356 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1357 if (nr_pages > BIO_MAX_PAGES) 1358 nr_pages = BIO_MAX_PAGES; 1359 1360 ret = -ENOMEM; 1361 bio = bio_kmalloc(gfp_mask, nr_pages); 1362 if (!bio) 1363 goto out_bmd; 1364 1365 ret = 0; 1366 1367 if (map_data) { 1368 nr_pages = 1 << map_data->page_order; 1369 i = map_data->offset / PAGE_SIZE; 1370 } 1371 while (len) { 1372 unsigned int bytes = PAGE_SIZE; 1373 1374 bytes -= offset; 1375 1376 if (bytes > len) 1377 bytes = len; 1378 1379 if (map_data) { 1380 if (i == map_data->nr_entries * nr_pages) { 1381 ret = -ENOMEM; 1382 break; 1383 } 1384 1385 page = map_data->pages[i / nr_pages]; 1386 page += (i % nr_pages); 1387 1388 i++; 1389 } else { 1390 page = alloc_page(q->bounce_gfp | gfp_mask); 1391 if (!page) { 1392 ret = -ENOMEM; 1393 break; 1394 } 1395 } 1396 1397 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { 1398 if (!map_data) 1399 __free_page(page); 1400 break; 1401 } 1402 1403 len -= bytes; 1404 offset = 0; 1405 } 1406 1407 if (ret) 1408 goto cleanup; 1409 1410 if (map_data) 1411 map_data->offset += bio->bi_iter.bi_size; 1412 1413 /* 1414 * success 1415 */ 1416 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1417 (map_data && map_data->from_user)) { 1418 ret = bio_copy_from_iter(bio, iter); 1419 if (ret) 1420 goto cleanup; 1421 } else { 1422 if (bmd->is_our_pages) 1423 zero_fill_bio(bio); 1424 iov_iter_advance(iter, bio->bi_iter.bi_size); 1425 } 1426 1427 bio->bi_private = bmd; 1428 if (map_data && map_data->null_mapped) 1429 bio_set_flag(bio, BIO_NULL_MAPPED); 1430 return bio; 1431 cleanup: 1432 if (!map_data) 1433 bio_free_pages(bio); 1434 bio_put(bio); 1435 out_bmd: 1436 kfree(bmd); 1437 return ERR_PTR(ret); 1438 } 1439 1440 /** 1441 * bio_map_user_iov - map user iovec into bio 1442 * @q: the struct request_queue for the bio 1443 * @iter: iovec iterator 1444 * @gfp_mask: memory allocation flags 1445 * 1446 * Map the user space address into a bio suitable for io to a block 1447 * device. Returns an error pointer in case of error. 1448 */ 1449 struct bio *bio_map_user_iov(struct request_queue *q, 1450 struct iov_iter *iter, 1451 gfp_t gfp_mask) 1452 { 1453 int j; 1454 struct bio *bio; 1455 int ret; 1456 1457 if (!iov_iter_count(iter)) 1458 return ERR_PTR(-EINVAL); 1459 1460 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1461 if (!bio) 1462 return ERR_PTR(-ENOMEM); 1463 1464 while (iov_iter_count(iter)) { 1465 struct page **pages; 1466 ssize_t bytes; 1467 size_t offs, added = 0; 1468 int npages; 1469 1470 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1471 if (unlikely(bytes <= 0)) { 1472 ret = bytes ? bytes : -EFAULT; 1473 goto out_unmap; 1474 } 1475 1476 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1477 1478 if (unlikely(offs & queue_dma_alignment(q))) { 1479 ret = -EINVAL; 1480 j = 0; 1481 } else { 1482 for (j = 0; j < npages; j++) { 1483 struct page *page = pages[j]; 1484 unsigned int n = PAGE_SIZE - offs; 1485 bool same_page = false; 1486 1487 if (n > bytes) 1488 n = bytes; 1489 1490 if (!__bio_add_pc_page(q, bio, page, n, offs, 1491 &same_page)) { 1492 if (same_page) 1493 put_page(page); 1494 break; 1495 } 1496 1497 added += n; 1498 bytes -= n; 1499 offs = 0; 1500 } 1501 iov_iter_advance(iter, added); 1502 } 1503 /* 1504 * release the pages we didn't map into the bio, if any 1505 */ 1506 while (j < npages) 1507 put_page(pages[j++]); 1508 kvfree(pages); 1509 /* couldn't stuff something into bio? */ 1510 if (bytes) 1511 break; 1512 } 1513 1514 bio_set_flag(bio, BIO_USER_MAPPED); 1515 1516 /* 1517 * subtle -- if bio_map_user_iov() ended up bouncing a bio, 1518 * it would normally disappear when its bi_end_io is run. 1519 * however, we need it for the unmap, so grab an extra 1520 * reference to it 1521 */ 1522 bio_get(bio); 1523 return bio; 1524 1525 out_unmap: 1526 bio_release_pages(bio, false); 1527 bio_put(bio); 1528 return ERR_PTR(ret); 1529 } 1530 1531 /** 1532 * bio_unmap_user - unmap a bio 1533 * @bio: the bio being unmapped 1534 * 1535 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 1536 * process context. 1537 * 1538 * bio_unmap_user() may sleep. 1539 */ 1540 void bio_unmap_user(struct bio *bio) 1541 { 1542 bio_release_pages(bio, bio_data_dir(bio) == READ); 1543 bio_put(bio); 1544 bio_put(bio); 1545 } 1546 1547 static void bio_invalidate_vmalloc_pages(struct bio *bio) 1548 { 1549 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE 1550 if (bio->bi_private && !op_is_write(bio_op(bio))) { 1551 unsigned long i, len = 0; 1552 1553 for (i = 0; i < bio->bi_vcnt; i++) 1554 len += bio->bi_io_vec[i].bv_len; 1555 invalidate_kernel_vmap_range(bio->bi_private, len); 1556 } 1557 #endif 1558 } 1559 1560 static void bio_map_kern_endio(struct bio *bio) 1561 { 1562 bio_invalidate_vmalloc_pages(bio); 1563 bio_put(bio); 1564 } 1565 1566 /** 1567 * bio_map_kern - map kernel address into bio 1568 * @q: the struct request_queue for the bio 1569 * @data: pointer to buffer to map 1570 * @len: length in bytes 1571 * @gfp_mask: allocation flags for bio allocation 1572 * 1573 * Map the kernel address into a bio suitable for io to a block 1574 * device. Returns an error pointer in case of error. 1575 */ 1576 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 1577 gfp_t gfp_mask) 1578 { 1579 unsigned long kaddr = (unsigned long)data; 1580 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1581 unsigned long start = kaddr >> PAGE_SHIFT; 1582 const int nr_pages = end - start; 1583 bool is_vmalloc = is_vmalloc_addr(data); 1584 struct page *page; 1585 int offset, i; 1586 struct bio *bio; 1587 1588 bio = bio_kmalloc(gfp_mask, nr_pages); 1589 if (!bio) 1590 return ERR_PTR(-ENOMEM); 1591 1592 if (is_vmalloc) { 1593 flush_kernel_vmap_range(data, len); 1594 bio->bi_private = data; 1595 } 1596 1597 offset = offset_in_page(kaddr); 1598 for (i = 0; i < nr_pages; i++) { 1599 unsigned int bytes = PAGE_SIZE - offset; 1600 1601 if (len <= 0) 1602 break; 1603 1604 if (bytes > len) 1605 bytes = len; 1606 1607 if (!is_vmalloc) 1608 page = virt_to_page(data); 1609 else 1610 page = vmalloc_to_page(data); 1611 if (bio_add_pc_page(q, bio, page, bytes, 1612 offset) < bytes) { 1613 /* we don't support partial mappings */ 1614 bio_put(bio); 1615 return ERR_PTR(-EINVAL); 1616 } 1617 1618 data += bytes; 1619 len -= bytes; 1620 offset = 0; 1621 } 1622 1623 bio->bi_end_io = bio_map_kern_endio; 1624 return bio; 1625 } 1626 1627 static void bio_copy_kern_endio(struct bio *bio) 1628 { 1629 bio_free_pages(bio); 1630 bio_put(bio); 1631 } 1632 1633 static void bio_copy_kern_endio_read(struct bio *bio) 1634 { 1635 char *p = bio->bi_private; 1636 struct bio_vec *bvec; 1637 struct bvec_iter_all iter_all; 1638 1639 bio_for_each_segment_all(bvec, bio, iter_all) { 1640 memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1641 p += bvec->bv_len; 1642 } 1643 1644 bio_copy_kern_endio(bio); 1645 } 1646 1647 /** 1648 * bio_copy_kern - copy kernel address into bio 1649 * @q: the struct request_queue for the bio 1650 * @data: pointer to buffer to copy 1651 * @len: length in bytes 1652 * @gfp_mask: allocation flags for bio and page allocation 1653 * @reading: data direction is READ 1654 * 1655 * copy the kernel address into a bio suitable for io to a block 1656 * device. Returns an error pointer in case of error. 1657 */ 1658 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1659 gfp_t gfp_mask, int reading) 1660 { 1661 unsigned long kaddr = (unsigned long)data; 1662 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1663 unsigned long start = kaddr >> PAGE_SHIFT; 1664 struct bio *bio; 1665 void *p = data; 1666 int nr_pages = 0; 1667 1668 /* 1669 * Overflow, abort 1670 */ 1671 if (end < start) 1672 return ERR_PTR(-EINVAL); 1673 1674 nr_pages = end - start; 1675 bio = bio_kmalloc(gfp_mask, nr_pages); 1676 if (!bio) 1677 return ERR_PTR(-ENOMEM); 1678 1679 while (len) { 1680 struct page *page; 1681 unsigned int bytes = PAGE_SIZE; 1682 1683 if (bytes > len) 1684 bytes = len; 1685 1686 page = alloc_page(q->bounce_gfp | gfp_mask); 1687 if (!page) 1688 goto cleanup; 1689 1690 if (!reading) 1691 memcpy(page_address(page), p, bytes); 1692 1693 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 1694 break; 1695 1696 len -= bytes; 1697 p += bytes; 1698 } 1699 1700 if (reading) { 1701 bio->bi_end_io = bio_copy_kern_endio_read; 1702 bio->bi_private = data; 1703 } else { 1704 bio->bi_end_io = bio_copy_kern_endio; 1705 } 1706 1707 return bio; 1708 1709 cleanup: 1710 bio_free_pages(bio); 1711 bio_put(bio); 1712 return ERR_PTR(-ENOMEM); 1713 } 1714 1715 /* 1716 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1717 * for performing direct-IO in BIOs. 1718 * 1719 * The problem is that we cannot run set_page_dirty() from interrupt context 1720 * because the required locks are not interrupt-safe. So what we can do is to 1721 * mark the pages dirty _before_ performing IO. And in interrupt context, 1722 * check that the pages are still dirty. If so, fine. If not, redirty them 1723 * in process context. 1724 * 1725 * We special-case compound pages here: normally this means reads into hugetlb 1726 * pages. The logic in here doesn't really work right for compound pages 1727 * because the VM does not uniformly chase down the head page in all cases. 1728 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1729 * handle them at all. So we skip compound pages here at an early stage. 1730 * 1731 * Note that this code is very hard to test under normal circumstances because 1732 * direct-io pins the pages with get_user_pages(). This makes 1733 * is_page_cache_freeable return false, and the VM will not clean the pages. 1734 * But other code (eg, flusher threads) could clean the pages if they are mapped 1735 * pagecache. 1736 * 1737 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1738 * deferred bio dirtying paths. 1739 */ 1740 1741 /* 1742 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1743 */ 1744 void bio_set_pages_dirty(struct bio *bio) 1745 { 1746 struct bio_vec *bvec; 1747 struct bvec_iter_all iter_all; 1748 1749 bio_for_each_segment_all(bvec, bio, iter_all) { 1750 if (!PageCompound(bvec->bv_page)) 1751 set_page_dirty_lock(bvec->bv_page); 1752 } 1753 } 1754 1755 /* 1756 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1757 * If they are, then fine. If, however, some pages are clean then they must 1758 * have been written out during the direct-IO read. So we take another ref on 1759 * the BIO and re-dirty the pages in process context. 1760 * 1761 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1762 * here on. It will run one put_page() against each page and will run one 1763 * bio_put() against the BIO. 1764 */ 1765 1766 static void bio_dirty_fn(struct work_struct *work); 1767 1768 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1769 static DEFINE_SPINLOCK(bio_dirty_lock); 1770 static struct bio *bio_dirty_list; 1771 1772 /* 1773 * This runs in process context 1774 */ 1775 static void bio_dirty_fn(struct work_struct *work) 1776 { 1777 struct bio *bio, *next; 1778 1779 spin_lock_irq(&bio_dirty_lock); 1780 next = bio_dirty_list; 1781 bio_dirty_list = NULL; 1782 spin_unlock_irq(&bio_dirty_lock); 1783 1784 while ((bio = next) != NULL) { 1785 next = bio->bi_private; 1786 1787 bio_release_pages(bio, true); 1788 bio_put(bio); 1789 } 1790 } 1791 1792 void bio_check_pages_dirty(struct bio *bio) 1793 { 1794 struct bio_vec *bvec; 1795 unsigned long flags; 1796 struct bvec_iter_all iter_all; 1797 1798 bio_for_each_segment_all(bvec, bio, iter_all) { 1799 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1800 goto defer; 1801 } 1802 1803 bio_release_pages(bio, false); 1804 bio_put(bio); 1805 return; 1806 defer: 1807 spin_lock_irqsave(&bio_dirty_lock, flags); 1808 bio->bi_private = bio_dirty_list; 1809 bio_dirty_list = bio; 1810 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1811 schedule_work(&bio_dirty_work); 1812 } 1813 1814 void update_io_ticks(struct hd_struct *part, unsigned long now, bool end) 1815 { 1816 unsigned long stamp; 1817 again: 1818 stamp = READ_ONCE(part->stamp); 1819 if (unlikely(stamp != now)) { 1820 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 1821 __part_stat_add(part, io_ticks, end ? now - stamp : 1); 1822 } 1823 } 1824 if (part->partno) { 1825 part = &part_to_disk(part)->part0; 1826 goto again; 1827 } 1828 } 1829 1830 void generic_start_io_acct(struct request_queue *q, int op, 1831 unsigned long sectors, struct hd_struct *part) 1832 { 1833 const int sgrp = op_stat_group(op); 1834 1835 part_stat_lock(); 1836 1837 update_io_ticks(part, jiffies, false); 1838 part_stat_inc(part, ios[sgrp]); 1839 part_stat_add(part, sectors[sgrp], sectors); 1840 part_inc_in_flight(q, part, op_is_write(op)); 1841 1842 part_stat_unlock(); 1843 } 1844 EXPORT_SYMBOL(generic_start_io_acct); 1845 1846 void generic_end_io_acct(struct request_queue *q, int req_op, 1847 struct hd_struct *part, unsigned long start_time) 1848 { 1849 unsigned long now = jiffies; 1850 unsigned long duration = now - start_time; 1851 const int sgrp = op_stat_group(req_op); 1852 1853 part_stat_lock(); 1854 1855 update_io_ticks(part, now, true); 1856 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 1857 part_dec_in_flight(q, part, op_is_write(req_op)); 1858 1859 part_stat_unlock(); 1860 } 1861 EXPORT_SYMBOL(generic_end_io_acct); 1862 1863 static inline bool bio_remaining_done(struct bio *bio) 1864 { 1865 /* 1866 * If we're not chaining, then ->__bi_remaining is always 1 and 1867 * we always end io on the first invocation. 1868 */ 1869 if (!bio_flagged(bio, BIO_CHAIN)) 1870 return true; 1871 1872 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1873 1874 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1875 bio_clear_flag(bio, BIO_CHAIN); 1876 return true; 1877 } 1878 1879 return false; 1880 } 1881 1882 /** 1883 * bio_endio - end I/O on a bio 1884 * @bio: bio 1885 * 1886 * Description: 1887 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1888 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1889 * bio unless they own it and thus know that it has an end_io function. 1890 * 1891 * bio_endio() can be called several times on a bio that has been chained 1892 * using bio_chain(). The ->bi_end_io() function will only be called the 1893 * last time. At this point the BLK_TA_COMPLETE tracing event will be 1894 * generated if BIO_TRACE_COMPLETION is set. 1895 **/ 1896 void bio_endio(struct bio *bio) 1897 { 1898 again: 1899 if (!bio_remaining_done(bio)) 1900 return; 1901 if (!bio_integrity_endio(bio)) 1902 return; 1903 1904 if (bio->bi_disk) 1905 rq_qos_done_bio(bio->bi_disk->queue, bio); 1906 1907 /* 1908 * Need to have a real endio function for chained bios, otherwise 1909 * various corner cases will break (like stacking block devices that 1910 * save/restore bi_end_io) - however, we want to avoid unbounded 1911 * recursion and blowing the stack. Tail call optimization would 1912 * handle this, but compiling with frame pointers also disables 1913 * gcc's sibling call optimization. 1914 */ 1915 if (bio->bi_end_io == bio_chain_endio) { 1916 bio = __bio_chain_endio(bio); 1917 goto again; 1918 } 1919 1920 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1921 trace_block_bio_complete(bio->bi_disk->queue, bio, 1922 blk_status_to_errno(bio->bi_status)); 1923 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1924 } 1925 1926 blk_throtl_bio_endio(bio); 1927 /* release cgroup info */ 1928 bio_uninit(bio); 1929 if (bio->bi_end_io) 1930 bio->bi_end_io(bio); 1931 } 1932 EXPORT_SYMBOL(bio_endio); 1933 1934 /** 1935 * bio_split - split a bio 1936 * @bio: bio to split 1937 * @sectors: number of sectors to split from the front of @bio 1938 * @gfp: gfp mask 1939 * @bs: bio set to allocate from 1940 * 1941 * Allocates and returns a new bio which represents @sectors from the start of 1942 * @bio, and updates @bio to represent the remaining sectors. 1943 * 1944 * Unless this is a discard request the newly allocated bio will point 1945 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1946 * neither @bio nor @bs are freed before the split bio. 1947 */ 1948 struct bio *bio_split(struct bio *bio, int sectors, 1949 gfp_t gfp, struct bio_set *bs) 1950 { 1951 struct bio *split; 1952 1953 BUG_ON(sectors <= 0); 1954 BUG_ON(sectors >= bio_sectors(bio)); 1955 1956 split = bio_clone_fast(bio, gfp, bs); 1957 if (!split) 1958 return NULL; 1959 1960 split->bi_iter.bi_size = sectors << 9; 1961 1962 if (bio_integrity(split)) 1963 bio_integrity_trim(split); 1964 1965 bio_advance(bio, split->bi_iter.bi_size); 1966 1967 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1968 bio_set_flag(split, BIO_TRACE_COMPLETION); 1969 1970 return split; 1971 } 1972 EXPORT_SYMBOL(bio_split); 1973 1974 /** 1975 * bio_trim - trim a bio 1976 * @bio: bio to trim 1977 * @offset: number of sectors to trim from the front of @bio 1978 * @size: size we want to trim @bio to, in sectors 1979 */ 1980 void bio_trim(struct bio *bio, int offset, int size) 1981 { 1982 /* 'bio' is a cloned bio which we need to trim to match 1983 * the given offset and size. 1984 */ 1985 1986 size <<= 9; 1987 if (offset == 0 && size == bio->bi_iter.bi_size) 1988 return; 1989 1990 bio_advance(bio, offset << 9); 1991 bio->bi_iter.bi_size = size; 1992 1993 if (bio_integrity(bio)) 1994 bio_integrity_trim(bio); 1995 1996 } 1997 EXPORT_SYMBOL_GPL(bio_trim); 1998 1999 /* 2000 * create memory pools for biovec's in a bio_set. 2001 * use the global biovec slabs created for general use. 2002 */ 2003 int biovec_init_pool(mempool_t *pool, int pool_entries) 2004 { 2005 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 2006 2007 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 2008 } 2009 2010 /* 2011 * bioset_exit - exit a bioset initialized with bioset_init() 2012 * 2013 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 2014 * kzalloc()). 2015 */ 2016 void bioset_exit(struct bio_set *bs) 2017 { 2018 if (bs->rescue_workqueue) 2019 destroy_workqueue(bs->rescue_workqueue); 2020 bs->rescue_workqueue = NULL; 2021 2022 mempool_exit(&bs->bio_pool); 2023 mempool_exit(&bs->bvec_pool); 2024 2025 bioset_integrity_free(bs); 2026 if (bs->bio_slab) 2027 bio_put_slab(bs); 2028 bs->bio_slab = NULL; 2029 } 2030 EXPORT_SYMBOL(bioset_exit); 2031 2032 /** 2033 * bioset_init - Initialize a bio_set 2034 * @bs: pool to initialize 2035 * @pool_size: Number of bio and bio_vecs to cache in the mempool 2036 * @front_pad: Number of bytes to allocate in front of the returned bio 2037 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 2038 * and %BIOSET_NEED_RESCUER 2039 * 2040 * Description: 2041 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 2042 * to ask for a number of bytes to be allocated in front of the bio. 2043 * Front pad allocation is useful for embedding the bio inside 2044 * another structure, to avoid allocating extra data to go with the bio. 2045 * Note that the bio must be embedded at the END of that structure always, 2046 * or things will break badly. 2047 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 2048 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 2049 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 2050 * dispatch queued requests when the mempool runs out of space. 2051 * 2052 */ 2053 int bioset_init(struct bio_set *bs, 2054 unsigned int pool_size, 2055 unsigned int front_pad, 2056 int flags) 2057 { 2058 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 2059 2060 bs->front_pad = front_pad; 2061 2062 spin_lock_init(&bs->rescue_lock); 2063 bio_list_init(&bs->rescue_list); 2064 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 2065 2066 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 2067 if (!bs->bio_slab) 2068 return -ENOMEM; 2069 2070 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2071 goto bad; 2072 2073 if ((flags & BIOSET_NEED_BVECS) && 2074 biovec_init_pool(&bs->bvec_pool, pool_size)) 2075 goto bad; 2076 2077 if (!(flags & BIOSET_NEED_RESCUER)) 2078 return 0; 2079 2080 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2081 if (!bs->rescue_workqueue) 2082 goto bad; 2083 2084 return 0; 2085 bad: 2086 bioset_exit(bs); 2087 return -ENOMEM; 2088 } 2089 EXPORT_SYMBOL(bioset_init); 2090 2091 /* 2092 * Initialize and setup a new bio_set, based on the settings from 2093 * another bio_set. 2094 */ 2095 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 2096 { 2097 int flags; 2098 2099 flags = 0; 2100 if (src->bvec_pool.min_nr) 2101 flags |= BIOSET_NEED_BVECS; 2102 if (src->rescue_workqueue) 2103 flags |= BIOSET_NEED_RESCUER; 2104 2105 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 2106 } 2107 EXPORT_SYMBOL(bioset_init_from_src); 2108 2109 #ifdef CONFIG_BLK_CGROUP 2110 2111 /** 2112 * bio_disassociate_blkg - puts back the blkg reference if associated 2113 * @bio: target bio 2114 * 2115 * Helper to disassociate the blkg from @bio if a blkg is associated. 2116 */ 2117 void bio_disassociate_blkg(struct bio *bio) 2118 { 2119 if (bio->bi_blkg) { 2120 blkg_put(bio->bi_blkg); 2121 bio->bi_blkg = NULL; 2122 } 2123 } 2124 EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2125 2126 /** 2127 * __bio_associate_blkg - associate a bio with the a blkg 2128 * @bio: target bio 2129 * @blkg: the blkg to associate 2130 * 2131 * This tries to associate @bio with the specified @blkg. Association failure 2132 * is handled by walking up the blkg tree. Therefore, the blkg associated can 2133 * be anything between @blkg and the root_blkg. This situation only happens 2134 * when a cgroup is dying and then the remaining bios will spill to the closest 2135 * alive blkg. 2136 * 2137 * A reference will be taken on the @blkg and will be released when @bio is 2138 * freed. 2139 */ 2140 static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2141 { 2142 bio_disassociate_blkg(bio); 2143 2144 bio->bi_blkg = blkg_tryget_closest(blkg); 2145 } 2146 2147 /** 2148 * bio_associate_blkg_from_css - associate a bio with a specified css 2149 * @bio: target bio 2150 * @css: target css 2151 * 2152 * Associate @bio with the blkg found by combining the css's blkg and the 2153 * request_queue of the @bio. This falls back to the queue's root_blkg if 2154 * the association fails with the css. 2155 */ 2156 void bio_associate_blkg_from_css(struct bio *bio, 2157 struct cgroup_subsys_state *css) 2158 { 2159 struct request_queue *q = bio->bi_disk->queue; 2160 struct blkcg_gq *blkg; 2161 2162 rcu_read_lock(); 2163 2164 if (!css || !css->parent) 2165 blkg = q->root_blkg; 2166 else 2167 blkg = blkg_lookup_create(css_to_blkcg(css), q); 2168 2169 __bio_associate_blkg(bio, blkg); 2170 2171 rcu_read_unlock(); 2172 } 2173 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2174 2175 #ifdef CONFIG_MEMCG 2176 /** 2177 * bio_associate_blkg_from_page - associate a bio with the page's blkg 2178 * @bio: target bio 2179 * @page: the page to lookup the blkcg from 2180 * 2181 * Associate @bio with the blkg from @page's owning memcg and the respective 2182 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2183 * root_blkg. 2184 */ 2185 void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 2186 { 2187 struct cgroup_subsys_state *css; 2188 2189 if (!page->mem_cgroup) 2190 return; 2191 2192 rcu_read_lock(); 2193 2194 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2195 bio_associate_blkg_from_css(bio, css); 2196 2197 rcu_read_unlock(); 2198 } 2199 #endif /* CONFIG_MEMCG */ 2200 2201 /** 2202 * bio_associate_blkg - associate a bio with a blkg 2203 * @bio: target bio 2204 * 2205 * Associate @bio with the blkg found from the bio's css and request_queue. 2206 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 2207 * already associated, the css is reused and association redone as the 2208 * request_queue may have changed. 2209 */ 2210 void bio_associate_blkg(struct bio *bio) 2211 { 2212 struct cgroup_subsys_state *css; 2213 2214 rcu_read_lock(); 2215 2216 if (bio->bi_blkg) 2217 css = &bio_blkcg(bio)->css; 2218 else 2219 css = blkcg_css(); 2220 2221 bio_associate_blkg_from_css(bio, css); 2222 2223 rcu_read_unlock(); 2224 } 2225 EXPORT_SYMBOL_GPL(bio_associate_blkg); 2226 2227 /** 2228 * bio_clone_blkg_association - clone blkg association from src to dst bio 2229 * @dst: destination bio 2230 * @src: source bio 2231 */ 2232 void bio_clone_blkg_association(struct bio *dst, struct bio *src) 2233 { 2234 rcu_read_lock(); 2235 2236 if (src->bi_blkg) 2237 __bio_associate_blkg(dst, src->bi_blkg); 2238 2239 rcu_read_unlock(); 2240 } 2241 EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2242 #endif /* CONFIG_BLK_CGROUP */ 2243 2244 static void __init biovec_init_slabs(void) 2245 { 2246 int i; 2247 2248 for (i = 0; i < BVEC_POOL_NR; i++) { 2249 int size; 2250 struct biovec_slab *bvs = bvec_slabs + i; 2251 2252 if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2253 bvs->slab = NULL; 2254 continue; 2255 } 2256 2257 size = bvs->nr_vecs * sizeof(struct bio_vec); 2258 bvs->slab = kmem_cache_create(bvs->name, size, 0, 2259 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2260 } 2261 } 2262 2263 static int __init init_bio(void) 2264 { 2265 bio_slab_max = 2; 2266 bio_slab_nr = 0; 2267 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 2268 GFP_KERNEL); 2269 2270 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 2271 2272 if (!bio_slabs) 2273 panic("bio: can't allocate bios\n"); 2274 2275 bio_integrity_init(); 2276 biovec_init_slabs(); 2277 2278 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2279 panic("bio: can't allocate bios\n"); 2280 2281 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2282 panic("bio: can't create integrity pool\n"); 2283 2284 return 0; 2285 } 2286 subsys_initcall(init_bio); 2287