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 * bio_put - release a reference to a bio 593 * @bio: bio to release reference to 594 * 595 * Description: 596 * Put a reference to a &struct bio, either one you have gotten with 597 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 598 **/ 599 void bio_put(struct bio *bio) 600 { 601 if (!bio_flagged(bio, BIO_REFFED)) 602 bio_free(bio); 603 else { 604 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 605 606 /* 607 * last put frees it 608 */ 609 if (atomic_dec_and_test(&bio->__bi_cnt)) 610 bio_free(bio); 611 } 612 } 613 EXPORT_SYMBOL(bio_put); 614 615 /** 616 * __bio_clone_fast - clone a bio that shares the original bio's biovec 617 * @bio: destination bio 618 * @bio_src: bio to clone 619 * 620 * Clone a &bio. Caller will own the returned bio, but not 621 * the actual data it points to. Reference count of returned 622 * bio will be one. 623 * 624 * Caller must ensure that @bio_src is not freed before @bio. 625 */ 626 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 627 { 628 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 629 630 /* 631 * most users will be overriding ->bi_disk with a new target, 632 * so we don't set nor calculate new physical/hw segment counts here 633 */ 634 bio->bi_disk = bio_src->bi_disk; 635 bio->bi_partno = bio_src->bi_partno; 636 bio_set_flag(bio, BIO_CLONED); 637 if (bio_flagged(bio_src, BIO_THROTTLED)) 638 bio_set_flag(bio, BIO_THROTTLED); 639 bio->bi_opf = bio_src->bi_opf; 640 bio->bi_ioprio = bio_src->bi_ioprio; 641 bio->bi_write_hint = bio_src->bi_write_hint; 642 bio->bi_iter = bio_src->bi_iter; 643 bio->bi_io_vec = bio_src->bi_io_vec; 644 645 bio_clone_blkg_association(bio, bio_src); 646 blkcg_bio_issue_init(bio); 647 } 648 EXPORT_SYMBOL(__bio_clone_fast); 649 650 /** 651 * bio_clone_fast - clone a bio that shares the original bio's biovec 652 * @bio: bio to clone 653 * @gfp_mask: allocation priority 654 * @bs: bio_set to allocate from 655 * 656 * Like __bio_clone_fast, only also allocates the returned bio 657 */ 658 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 659 { 660 struct bio *b; 661 662 b = bio_alloc_bioset(gfp_mask, 0, bs); 663 if (!b) 664 return NULL; 665 666 __bio_clone_fast(b, bio); 667 668 if (bio_integrity(bio)) { 669 int ret; 670 671 ret = bio_integrity_clone(b, bio, gfp_mask); 672 673 if (ret < 0) { 674 bio_put(b); 675 return NULL; 676 } 677 } 678 679 return b; 680 } 681 EXPORT_SYMBOL(bio_clone_fast); 682 683 const char *bio_devname(struct bio *bio, char *buf) 684 { 685 return disk_name(bio->bi_disk, bio->bi_partno, buf); 686 } 687 EXPORT_SYMBOL(bio_devname); 688 689 static inline bool page_is_mergeable(const struct bio_vec *bv, 690 struct page *page, unsigned int len, unsigned int off, 691 bool *same_page) 692 { 693 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 694 bv->bv_offset + bv->bv_len - 1; 695 phys_addr_t page_addr = page_to_phys(page); 696 697 if (vec_end_addr + 1 != page_addr + off) 698 return false; 699 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 700 return false; 701 702 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 703 if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) 704 return false; 705 return true; 706 } 707 708 static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio, 709 struct page *page, unsigned len, unsigned offset, 710 bool *same_page) 711 { 712 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 713 unsigned long mask = queue_segment_boundary(q); 714 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 715 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 716 717 if ((addr1 | mask) != (addr2 | mask)) 718 return false; 719 if (bv->bv_len + len > queue_max_segment_size(q)) 720 return false; 721 return __bio_try_merge_page(bio, page, len, offset, same_page); 722 } 723 724 /** 725 * __bio_add_pc_page - attempt to add page to passthrough bio 726 * @q: the target queue 727 * @bio: destination bio 728 * @page: page to add 729 * @len: vec entry length 730 * @offset: vec entry offset 731 * @same_page: return if the merge happen inside the same page 732 * 733 * Attempt to add a page to the bio_vec maplist. This can fail for a 734 * number of reasons, such as the bio being full or target block device 735 * limitations. The target block device must allow bio's up to PAGE_SIZE, 736 * so it is always possible to add a single page to an empty bio. 737 * 738 * This should only be used by passthrough bios. 739 */ 740 static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, 741 struct page *page, unsigned int len, unsigned int offset, 742 bool *same_page) 743 { 744 struct bio_vec *bvec; 745 746 /* 747 * cloned bio must not modify vec list 748 */ 749 if (unlikely(bio_flagged(bio, BIO_CLONED))) 750 return 0; 751 752 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 753 return 0; 754 755 if (bio->bi_vcnt > 0) { 756 if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page)) 757 return len; 758 759 /* 760 * If the queue doesn't support SG gaps and adding this segment 761 * would create a gap, disallow it. 762 */ 763 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 764 if (bvec_gap_to_prev(q, bvec, offset)) 765 return 0; 766 } 767 768 if (bio_full(bio, len)) 769 return 0; 770 771 if (bio->bi_vcnt >= queue_max_segments(q)) 772 return 0; 773 774 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 775 bvec->bv_page = page; 776 bvec->bv_len = len; 777 bvec->bv_offset = offset; 778 bio->bi_vcnt++; 779 bio->bi_iter.bi_size += len; 780 return len; 781 } 782 783 int bio_add_pc_page(struct request_queue *q, struct bio *bio, 784 struct page *page, unsigned int len, unsigned int offset) 785 { 786 bool same_page = false; 787 return __bio_add_pc_page(q, bio, page, len, offset, &same_page); 788 } 789 EXPORT_SYMBOL(bio_add_pc_page); 790 791 /** 792 * __bio_try_merge_page - try appending data to an existing bvec. 793 * @bio: destination bio 794 * @page: start page to add 795 * @len: length of the data to add 796 * @off: offset of the data relative to @page 797 * @same_page: return if the segment has been merged inside the same page 798 * 799 * Try to add the data at @page + @off to the last bvec of @bio. This is a 800 * a useful optimisation for file systems with a block size smaller than the 801 * page size. 802 * 803 * Warn if (@len, @off) crosses pages in case that @same_page is true. 804 * 805 * Return %true on success or %false on failure. 806 */ 807 bool __bio_try_merge_page(struct bio *bio, struct page *page, 808 unsigned int len, unsigned int off, bool *same_page) 809 { 810 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 811 return false; 812 813 if (bio->bi_vcnt > 0) { 814 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 815 816 if (page_is_mergeable(bv, page, len, off, same_page)) { 817 if (bio->bi_iter.bi_size > UINT_MAX - len) 818 return false; 819 bv->bv_len += len; 820 bio->bi_iter.bi_size += len; 821 return true; 822 } 823 } 824 return false; 825 } 826 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 827 828 /** 829 * __bio_add_page - add page(s) to a bio in a new segment 830 * @bio: destination bio 831 * @page: start page to add 832 * @len: length of the data to add, may cross pages 833 * @off: offset of the data relative to @page, may cross pages 834 * 835 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 836 * that @bio has space for another bvec. 837 */ 838 void __bio_add_page(struct bio *bio, struct page *page, 839 unsigned int len, unsigned int off) 840 { 841 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 842 843 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 844 WARN_ON_ONCE(bio_full(bio, len)); 845 846 bv->bv_page = page; 847 bv->bv_offset = off; 848 bv->bv_len = len; 849 850 bio->bi_iter.bi_size += len; 851 bio->bi_vcnt++; 852 853 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 854 bio_set_flag(bio, BIO_WORKINGSET); 855 } 856 EXPORT_SYMBOL_GPL(__bio_add_page); 857 858 /** 859 * bio_add_page - attempt to add page(s) to bio 860 * @bio: destination bio 861 * @page: start page to add 862 * @len: vec entry length, may cross pages 863 * @offset: vec entry offset relative to @page, may cross pages 864 * 865 * Attempt to add page(s) to the bio_vec maplist. This will only fail 866 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 867 */ 868 int bio_add_page(struct bio *bio, struct page *page, 869 unsigned int len, unsigned int offset) 870 { 871 bool same_page = false; 872 873 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 874 if (bio_full(bio, len)) 875 return 0; 876 __bio_add_page(bio, page, len, offset); 877 } 878 return len; 879 } 880 EXPORT_SYMBOL(bio_add_page); 881 882 void bio_release_pages(struct bio *bio, bool mark_dirty) 883 { 884 struct bvec_iter_all iter_all; 885 struct bio_vec *bvec; 886 887 if (bio_flagged(bio, BIO_NO_PAGE_REF)) 888 return; 889 890 bio_for_each_segment_all(bvec, bio, iter_all) { 891 if (mark_dirty && !PageCompound(bvec->bv_page)) 892 set_page_dirty_lock(bvec->bv_page); 893 put_page(bvec->bv_page); 894 } 895 } 896 897 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 898 { 899 const struct bio_vec *bv = iter->bvec; 900 unsigned int len; 901 size_t size; 902 903 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 904 return -EINVAL; 905 906 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 907 size = bio_add_page(bio, bv->bv_page, len, 908 bv->bv_offset + iter->iov_offset); 909 if (unlikely(size != len)) 910 return -EINVAL; 911 iov_iter_advance(iter, size); 912 return 0; 913 } 914 915 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 916 917 /** 918 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 919 * @bio: bio to add pages to 920 * @iter: iov iterator describing the region to be mapped 921 * 922 * Pins pages from *iter and appends them to @bio's bvec array. The 923 * pages will have to be released using put_page() when done. 924 * For multi-segment *iter, this function only adds pages from the 925 * the next non-empty segment of the iov iterator. 926 */ 927 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 928 { 929 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 930 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 931 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 932 struct page **pages = (struct page **)bv; 933 bool same_page = false; 934 ssize_t size, left; 935 unsigned len, i; 936 size_t offset; 937 938 /* 939 * Move page array up in the allocated memory for the bio vecs as far as 940 * possible so that we can start filling biovecs from the beginning 941 * without overwriting the temporary page array. 942 */ 943 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 944 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 945 946 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 947 if (unlikely(size <= 0)) 948 return size ? size : -EFAULT; 949 950 for (left = size, i = 0; left > 0; left -= len, i++) { 951 struct page *page = pages[i]; 952 953 len = min_t(size_t, PAGE_SIZE - offset, left); 954 955 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 956 if (same_page) 957 put_page(page); 958 } else { 959 if (WARN_ON_ONCE(bio_full(bio, len))) 960 return -EINVAL; 961 __bio_add_page(bio, page, len, offset); 962 } 963 offset = 0; 964 } 965 966 iov_iter_advance(iter, size); 967 return 0; 968 } 969 970 /** 971 * bio_iov_iter_get_pages - add user or kernel pages to a bio 972 * @bio: bio to add pages to 973 * @iter: iov iterator describing the region to be added 974 * 975 * This takes either an iterator pointing to user memory, or one pointing to 976 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 977 * map them into the kernel. On IO completion, the caller should put those 978 * pages. If we're adding kernel pages, and the caller told us it's safe to 979 * do so, we just have to add the pages to the bio directly. We don't grab an 980 * extra reference to those pages (the user should already have that), and we 981 * don't put the page on IO completion. The caller needs to check if the bio is 982 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 983 * released. 984 * 985 * The function tries, but does not guarantee, to pin as many pages as 986 * fit into the bio, or are requested in *iter, whatever is smaller. If 987 * MM encounters an error pinning the requested pages, it stops. Error 988 * is returned only if 0 pages could be pinned. 989 */ 990 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 991 { 992 const bool is_bvec = iov_iter_is_bvec(iter); 993 int ret; 994 995 if (WARN_ON_ONCE(bio->bi_vcnt)) 996 return -EINVAL; 997 998 do { 999 if (is_bvec) 1000 ret = __bio_iov_bvec_add_pages(bio, iter); 1001 else 1002 ret = __bio_iov_iter_get_pages(bio, iter); 1003 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1004 1005 if (is_bvec) 1006 bio_set_flag(bio, BIO_NO_PAGE_REF); 1007 return bio->bi_vcnt ? 0 : ret; 1008 } 1009 1010 static void submit_bio_wait_endio(struct bio *bio) 1011 { 1012 complete(bio->bi_private); 1013 } 1014 1015 /** 1016 * submit_bio_wait - submit a bio, and wait until it completes 1017 * @bio: The &struct bio which describes the I/O 1018 * 1019 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1020 * bio_endio() on failure. 1021 * 1022 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1023 * result in bio reference to be consumed. The caller must drop the reference 1024 * on his own. 1025 */ 1026 int submit_bio_wait(struct bio *bio) 1027 { 1028 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 1029 unsigned long hang_check; 1030 1031 bio->bi_private = &done; 1032 bio->bi_end_io = submit_bio_wait_endio; 1033 bio->bi_opf |= REQ_SYNC; 1034 submit_bio(bio); 1035 1036 /* Prevent hang_check timer from firing at us during very long I/O */ 1037 hang_check = sysctl_hung_task_timeout_secs; 1038 if (hang_check) 1039 while (!wait_for_completion_io_timeout(&done, 1040 hang_check * (HZ/2))) 1041 ; 1042 else 1043 wait_for_completion_io(&done); 1044 1045 return blk_status_to_errno(bio->bi_status); 1046 } 1047 EXPORT_SYMBOL(submit_bio_wait); 1048 1049 /** 1050 * bio_advance - increment/complete a bio by some number of bytes 1051 * @bio: bio to advance 1052 * @bytes: number of bytes to complete 1053 * 1054 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1055 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1056 * be updated on the last bvec as well. 1057 * 1058 * @bio will then represent the remaining, uncompleted portion of the io. 1059 */ 1060 void bio_advance(struct bio *bio, unsigned bytes) 1061 { 1062 if (bio_integrity(bio)) 1063 bio_integrity_advance(bio, bytes); 1064 1065 bio_advance_iter(bio, &bio->bi_iter, bytes); 1066 } 1067 EXPORT_SYMBOL(bio_advance); 1068 1069 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1070 struct bio *src, struct bvec_iter *src_iter) 1071 { 1072 struct bio_vec src_bv, dst_bv; 1073 void *src_p, *dst_p; 1074 unsigned bytes; 1075 1076 while (src_iter->bi_size && dst_iter->bi_size) { 1077 src_bv = bio_iter_iovec(src, *src_iter); 1078 dst_bv = bio_iter_iovec(dst, *dst_iter); 1079 1080 bytes = min(src_bv.bv_len, dst_bv.bv_len); 1081 1082 src_p = kmap_atomic(src_bv.bv_page); 1083 dst_p = kmap_atomic(dst_bv.bv_page); 1084 1085 memcpy(dst_p + dst_bv.bv_offset, 1086 src_p + src_bv.bv_offset, 1087 bytes); 1088 1089 kunmap_atomic(dst_p); 1090 kunmap_atomic(src_p); 1091 1092 flush_dcache_page(dst_bv.bv_page); 1093 1094 bio_advance_iter(src, src_iter, bytes); 1095 bio_advance_iter(dst, dst_iter, bytes); 1096 } 1097 } 1098 EXPORT_SYMBOL(bio_copy_data_iter); 1099 1100 /** 1101 * bio_copy_data - copy contents of data buffers from one bio to another 1102 * @src: source bio 1103 * @dst: destination bio 1104 * 1105 * Stops when it reaches the end of either @src or @dst - that is, copies 1106 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1107 */ 1108 void bio_copy_data(struct bio *dst, struct bio *src) 1109 { 1110 struct bvec_iter src_iter = src->bi_iter; 1111 struct bvec_iter dst_iter = dst->bi_iter; 1112 1113 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1114 } 1115 EXPORT_SYMBOL(bio_copy_data); 1116 1117 /** 1118 * bio_list_copy_data - copy contents of data buffers from one chain of bios to 1119 * another 1120 * @src: source bio list 1121 * @dst: destination bio list 1122 * 1123 * Stops when it reaches the end of either the @src list or @dst list - that is, 1124 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 1125 * bios). 1126 */ 1127 void bio_list_copy_data(struct bio *dst, struct bio *src) 1128 { 1129 struct bvec_iter src_iter = src->bi_iter; 1130 struct bvec_iter dst_iter = dst->bi_iter; 1131 1132 while (1) { 1133 if (!src_iter.bi_size) { 1134 src = src->bi_next; 1135 if (!src) 1136 break; 1137 1138 src_iter = src->bi_iter; 1139 } 1140 1141 if (!dst_iter.bi_size) { 1142 dst = dst->bi_next; 1143 if (!dst) 1144 break; 1145 1146 dst_iter = dst->bi_iter; 1147 } 1148 1149 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1150 } 1151 } 1152 EXPORT_SYMBOL(bio_list_copy_data); 1153 1154 struct bio_map_data { 1155 int is_our_pages; 1156 struct iov_iter iter; 1157 struct iovec iov[]; 1158 }; 1159 1160 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1161 gfp_t gfp_mask) 1162 { 1163 struct bio_map_data *bmd; 1164 if (data->nr_segs > UIO_MAXIOV) 1165 return NULL; 1166 1167 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); 1168 if (!bmd) 1169 return NULL; 1170 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 1171 bmd->iter = *data; 1172 bmd->iter.iov = bmd->iov; 1173 return bmd; 1174 } 1175 1176 /** 1177 * bio_copy_from_iter - copy all pages from iov_iter to bio 1178 * @bio: The &struct bio which describes the I/O as destination 1179 * @iter: iov_iter as source 1180 * 1181 * Copy all pages from iov_iter to bio. 1182 * Returns 0 on success, or error on failure. 1183 */ 1184 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1185 { 1186 struct bio_vec *bvec; 1187 struct bvec_iter_all iter_all; 1188 1189 bio_for_each_segment_all(bvec, bio, iter_all) { 1190 ssize_t ret; 1191 1192 ret = copy_page_from_iter(bvec->bv_page, 1193 bvec->bv_offset, 1194 bvec->bv_len, 1195 iter); 1196 1197 if (!iov_iter_count(iter)) 1198 break; 1199 1200 if (ret < bvec->bv_len) 1201 return -EFAULT; 1202 } 1203 1204 return 0; 1205 } 1206 1207 /** 1208 * bio_copy_to_iter - copy all pages from bio to iov_iter 1209 * @bio: The &struct bio which describes the I/O as source 1210 * @iter: iov_iter as destination 1211 * 1212 * Copy all pages from bio to iov_iter. 1213 * Returns 0 on success, or error on failure. 1214 */ 1215 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 1216 { 1217 struct bio_vec *bvec; 1218 struct bvec_iter_all iter_all; 1219 1220 bio_for_each_segment_all(bvec, bio, iter_all) { 1221 ssize_t ret; 1222 1223 ret = copy_page_to_iter(bvec->bv_page, 1224 bvec->bv_offset, 1225 bvec->bv_len, 1226 &iter); 1227 1228 if (!iov_iter_count(&iter)) 1229 break; 1230 1231 if (ret < bvec->bv_len) 1232 return -EFAULT; 1233 } 1234 1235 return 0; 1236 } 1237 1238 void bio_free_pages(struct bio *bio) 1239 { 1240 struct bio_vec *bvec; 1241 struct bvec_iter_all iter_all; 1242 1243 bio_for_each_segment_all(bvec, bio, iter_all) 1244 __free_page(bvec->bv_page); 1245 } 1246 EXPORT_SYMBOL(bio_free_pages); 1247 1248 /** 1249 * bio_uncopy_user - finish previously mapped bio 1250 * @bio: bio being terminated 1251 * 1252 * Free pages allocated from bio_copy_user_iov() and write back data 1253 * to user space in case of a read. 1254 */ 1255 int bio_uncopy_user(struct bio *bio) 1256 { 1257 struct bio_map_data *bmd = bio->bi_private; 1258 int ret = 0; 1259 1260 if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1261 /* 1262 * if we're in a workqueue, the request is orphaned, so 1263 * don't copy into a random user address space, just free 1264 * and return -EINTR so user space doesn't expect any data. 1265 */ 1266 if (!current->mm) 1267 ret = -EINTR; 1268 else if (bio_data_dir(bio) == READ) 1269 ret = bio_copy_to_iter(bio, bmd->iter); 1270 if (bmd->is_our_pages) 1271 bio_free_pages(bio); 1272 } 1273 kfree(bmd); 1274 bio_put(bio); 1275 return ret; 1276 } 1277 1278 /** 1279 * bio_copy_user_iov - copy user data to bio 1280 * @q: destination block queue 1281 * @map_data: pointer to the rq_map_data holding pages (if necessary) 1282 * @iter: iovec iterator 1283 * @gfp_mask: memory allocation flags 1284 * 1285 * Prepares and returns a bio for indirect user io, bouncing data 1286 * to/from kernel pages as necessary. Must be paired with 1287 * call bio_uncopy_user() on io completion. 1288 */ 1289 struct bio *bio_copy_user_iov(struct request_queue *q, 1290 struct rq_map_data *map_data, 1291 struct iov_iter *iter, 1292 gfp_t gfp_mask) 1293 { 1294 struct bio_map_data *bmd; 1295 struct page *page; 1296 struct bio *bio; 1297 int i = 0, ret; 1298 int nr_pages; 1299 unsigned int len = iter->count; 1300 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1301 1302 bmd = bio_alloc_map_data(iter, gfp_mask); 1303 if (!bmd) 1304 return ERR_PTR(-ENOMEM); 1305 1306 /* 1307 * We need to do a deep copy of the iov_iter including the iovecs. 1308 * The caller provided iov might point to an on-stack or otherwise 1309 * shortlived one. 1310 */ 1311 bmd->is_our_pages = map_data ? 0 : 1; 1312 1313 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1314 if (nr_pages > BIO_MAX_PAGES) 1315 nr_pages = BIO_MAX_PAGES; 1316 1317 ret = -ENOMEM; 1318 bio = bio_kmalloc(gfp_mask, nr_pages); 1319 if (!bio) 1320 goto out_bmd; 1321 1322 ret = 0; 1323 1324 if (map_data) { 1325 nr_pages = 1 << map_data->page_order; 1326 i = map_data->offset / PAGE_SIZE; 1327 } 1328 while (len) { 1329 unsigned int bytes = PAGE_SIZE; 1330 1331 bytes -= offset; 1332 1333 if (bytes > len) 1334 bytes = len; 1335 1336 if (map_data) { 1337 if (i == map_data->nr_entries * nr_pages) { 1338 ret = -ENOMEM; 1339 break; 1340 } 1341 1342 page = map_data->pages[i / nr_pages]; 1343 page += (i % nr_pages); 1344 1345 i++; 1346 } else { 1347 page = alloc_page(q->bounce_gfp | gfp_mask); 1348 if (!page) { 1349 ret = -ENOMEM; 1350 break; 1351 } 1352 } 1353 1354 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { 1355 if (!map_data) 1356 __free_page(page); 1357 break; 1358 } 1359 1360 len -= bytes; 1361 offset = 0; 1362 } 1363 1364 if (ret) 1365 goto cleanup; 1366 1367 if (map_data) 1368 map_data->offset += bio->bi_iter.bi_size; 1369 1370 /* 1371 * success 1372 */ 1373 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1374 (map_data && map_data->from_user)) { 1375 ret = bio_copy_from_iter(bio, iter); 1376 if (ret) 1377 goto cleanup; 1378 } else { 1379 if (bmd->is_our_pages) 1380 zero_fill_bio(bio); 1381 iov_iter_advance(iter, bio->bi_iter.bi_size); 1382 } 1383 1384 bio->bi_private = bmd; 1385 if (map_data && map_data->null_mapped) 1386 bio_set_flag(bio, BIO_NULL_MAPPED); 1387 return bio; 1388 cleanup: 1389 if (!map_data) 1390 bio_free_pages(bio); 1391 bio_put(bio); 1392 out_bmd: 1393 kfree(bmd); 1394 return ERR_PTR(ret); 1395 } 1396 1397 /** 1398 * bio_map_user_iov - map user iovec into bio 1399 * @q: the struct request_queue for the bio 1400 * @iter: iovec iterator 1401 * @gfp_mask: memory allocation flags 1402 * 1403 * Map the user space address into a bio suitable for io to a block 1404 * device. Returns an error pointer in case of error. 1405 */ 1406 struct bio *bio_map_user_iov(struct request_queue *q, 1407 struct iov_iter *iter, 1408 gfp_t gfp_mask) 1409 { 1410 int j; 1411 struct bio *bio; 1412 int ret; 1413 1414 if (!iov_iter_count(iter)) 1415 return ERR_PTR(-EINVAL); 1416 1417 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1418 if (!bio) 1419 return ERR_PTR(-ENOMEM); 1420 1421 while (iov_iter_count(iter)) { 1422 struct page **pages; 1423 ssize_t bytes; 1424 size_t offs, added = 0; 1425 int npages; 1426 1427 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1428 if (unlikely(bytes <= 0)) { 1429 ret = bytes ? bytes : -EFAULT; 1430 goto out_unmap; 1431 } 1432 1433 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1434 1435 if (unlikely(offs & queue_dma_alignment(q))) { 1436 ret = -EINVAL; 1437 j = 0; 1438 } else { 1439 for (j = 0; j < npages; j++) { 1440 struct page *page = pages[j]; 1441 unsigned int n = PAGE_SIZE - offs; 1442 bool same_page = false; 1443 1444 if (n > bytes) 1445 n = bytes; 1446 1447 if (!__bio_add_pc_page(q, bio, page, n, offs, 1448 &same_page)) { 1449 if (same_page) 1450 put_page(page); 1451 break; 1452 } 1453 1454 added += n; 1455 bytes -= n; 1456 offs = 0; 1457 } 1458 iov_iter_advance(iter, added); 1459 } 1460 /* 1461 * release the pages we didn't map into the bio, if any 1462 */ 1463 while (j < npages) 1464 put_page(pages[j++]); 1465 kvfree(pages); 1466 /* couldn't stuff something into bio? */ 1467 if (bytes) 1468 break; 1469 } 1470 1471 bio_set_flag(bio, BIO_USER_MAPPED); 1472 1473 /* 1474 * subtle -- if bio_map_user_iov() ended up bouncing a bio, 1475 * it would normally disappear when its bi_end_io is run. 1476 * however, we need it for the unmap, so grab an extra 1477 * reference to it 1478 */ 1479 bio_get(bio); 1480 return bio; 1481 1482 out_unmap: 1483 bio_release_pages(bio, false); 1484 bio_put(bio); 1485 return ERR_PTR(ret); 1486 } 1487 1488 /** 1489 * bio_unmap_user - unmap a bio 1490 * @bio: the bio being unmapped 1491 * 1492 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 1493 * process context. 1494 * 1495 * bio_unmap_user() may sleep. 1496 */ 1497 void bio_unmap_user(struct bio *bio) 1498 { 1499 bio_release_pages(bio, bio_data_dir(bio) == READ); 1500 bio_put(bio); 1501 bio_put(bio); 1502 } 1503 1504 static void bio_invalidate_vmalloc_pages(struct bio *bio) 1505 { 1506 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE 1507 if (bio->bi_private && !op_is_write(bio_op(bio))) { 1508 unsigned long i, len = 0; 1509 1510 for (i = 0; i < bio->bi_vcnt; i++) 1511 len += bio->bi_io_vec[i].bv_len; 1512 invalidate_kernel_vmap_range(bio->bi_private, len); 1513 } 1514 #endif 1515 } 1516 1517 static void bio_map_kern_endio(struct bio *bio) 1518 { 1519 bio_invalidate_vmalloc_pages(bio); 1520 bio_put(bio); 1521 } 1522 1523 /** 1524 * bio_map_kern - map kernel address into bio 1525 * @q: the struct request_queue for the bio 1526 * @data: pointer to buffer to map 1527 * @len: length in bytes 1528 * @gfp_mask: allocation flags for bio allocation 1529 * 1530 * Map the kernel address into a bio suitable for io to a block 1531 * device. Returns an error pointer in case of error. 1532 */ 1533 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 1534 gfp_t gfp_mask) 1535 { 1536 unsigned long kaddr = (unsigned long)data; 1537 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1538 unsigned long start = kaddr >> PAGE_SHIFT; 1539 const int nr_pages = end - start; 1540 bool is_vmalloc = is_vmalloc_addr(data); 1541 struct page *page; 1542 int offset, i; 1543 struct bio *bio; 1544 1545 bio = bio_kmalloc(gfp_mask, nr_pages); 1546 if (!bio) 1547 return ERR_PTR(-ENOMEM); 1548 1549 if (is_vmalloc) { 1550 flush_kernel_vmap_range(data, len); 1551 bio->bi_private = data; 1552 } 1553 1554 offset = offset_in_page(kaddr); 1555 for (i = 0; i < nr_pages; i++) { 1556 unsigned int bytes = PAGE_SIZE - offset; 1557 1558 if (len <= 0) 1559 break; 1560 1561 if (bytes > len) 1562 bytes = len; 1563 1564 if (!is_vmalloc) 1565 page = virt_to_page(data); 1566 else 1567 page = vmalloc_to_page(data); 1568 if (bio_add_pc_page(q, bio, page, bytes, 1569 offset) < bytes) { 1570 /* we don't support partial mappings */ 1571 bio_put(bio); 1572 return ERR_PTR(-EINVAL); 1573 } 1574 1575 data += bytes; 1576 len -= bytes; 1577 offset = 0; 1578 } 1579 1580 bio->bi_end_io = bio_map_kern_endio; 1581 return bio; 1582 } 1583 1584 static void bio_copy_kern_endio(struct bio *bio) 1585 { 1586 bio_free_pages(bio); 1587 bio_put(bio); 1588 } 1589 1590 static void bio_copy_kern_endio_read(struct bio *bio) 1591 { 1592 char *p = bio->bi_private; 1593 struct bio_vec *bvec; 1594 struct bvec_iter_all iter_all; 1595 1596 bio_for_each_segment_all(bvec, bio, iter_all) { 1597 memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1598 p += bvec->bv_len; 1599 } 1600 1601 bio_copy_kern_endio(bio); 1602 } 1603 1604 /** 1605 * bio_copy_kern - copy kernel address into bio 1606 * @q: the struct request_queue for the bio 1607 * @data: pointer to buffer to copy 1608 * @len: length in bytes 1609 * @gfp_mask: allocation flags for bio and page allocation 1610 * @reading: data direction is READ 1611 * 1612 * copy the kernel address into a bio suitable for io to a block 1613 * device. Returns an error pointer in case of error. 1614 */ 1615 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1616 gfp_t gfp_mask, int reading) 1617 { 1618 unsigned long kaddr = (unsigned long)data; 1619 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1620 unsigned long start = kaddr >> PAGE_SHIFT; 1621 struct bio *bio; 1622 void *p = data; 1623 int nr_pages = 0; 1624 1625 /* 1626 * Overflow, abort 1627 */ 1628 if (end < start) 1629 return ERR_PTR(-EINVAL); 1630 1631 nr_pages = end - start; 1632 bio = bio_kmalloc(gfp_mask, nr_pages); 1633 if (!bio) 1634 return ERR_PTR(-ENOMEM); 1635 1636 while (len) { 1637 struct page *page; 1638 unsigned int bytes = PAGE_SIZE; 1639 1640 if (bytes > len) 1641 bytes = len; 1642 1643 page = alloc_page(q->bounce_gfp | gfp_mask); 1644 if (!page) 1645 goto cleanup; 1646 1647 if (!reading) 1648 memcpy(page_address(page), p, bytes); 1649 1650 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 1651 break; 1652 1653 len -= bytes; 1654 p += bytes; 1655 } 1656 1657 if (reading) { 1658 bio->bi_end_io = bio_copy_kern_endio_read; 1659 bio->bi_private = data; 1660 } else { 1661 bio->bi_end_io = bio_copy_kern_endio; 1662 } 1663 1664 return bio; 1665 1666 cleanup: 1667 bio_free_pages(bio); 1668 bio_put(bio); 1669 return ERR_PTR(-ENOMEM); 1670 } 1671 1672 /* 1673 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1674 * for performing direct-IO in BIOs. 1675 * 1676 * The problem is that we cannot run set_page_dirty() from interrupt context 1677 * because the required locks are not interrupt-safe. So what we can do is to 1678 * mark the pages dirty _before_ performing IO. And in interrupt context, 1679 * check that the pages are still dirty. If so, fine. If not, redirty them 1680 * in process context. 1681 * 1682 * We special-case compound pages here: normally this means reads into hugetlb 1683 * pages. The logic in here doesn't really work right for compound pages 1684 * because the VM does not uniformly chase down the head page in all cases. 1685 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1686 * handle them at all. So we skip compound pages here at an early stage. 1687 * 1688 * Note that this code is very hard to test under normal circumstances because 1689 * direct-io pins the pages with get_user_pages(). This makes 1690 * is_page_cache_freeable return false, and the VM will not clean the pages. 1691 * But other code (eg, flusher threads) could clean the pages if they are mapped 1692 * pagecache. 1693 * 1694 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1695 * deferred bio dirtying paths. 1696 */ 1697 1698 /* 1699 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1700 */ 1701 void bio_set_pages_dirty(struct bio *bio) 1702 { 1703 struct bio_vec *bvec; 1704 struct bvec_iter_all iter_all; 1705 1706 bio_for_each_segment_all(bvec, bio, iter_all) { 1707 if (!PageCompound(bvec->bv_page)) 1708 set_page_dirty_lock(bvec->bv_page); 1709 } 1710 } 1711 1712 /* 1713 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1714 * If they are, then fine. If, however, some pages are clean then they must 1715 * have been written out during the direct-IO read. So we take another ref on 1716 * the BIO and re-dirty the pages in process context. 1717 * 1718 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1719 * here on. It will run one put_page() against each page and will run one 1720 * bio_put() against the BIO. 1721 */ 1722 1723 static void bio_dirty_fn(struct work_struct *work); 1724 1725 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1726 static DEFINE_SPINLOCK(bio_dirty_lock); 1727 static struct bio *bio_dirty_list; 1728 1729 /* 1730 * This runs in process context 1731 */ 1732 static void bio_dirty_fn(struct work_struct *work) 1733 { 1734 struct bio *bio, *next; 1735 1736 spin_lock_irq(&bio_dirty_lock); 1737 next = bio_dirty_list; 1738 bio_dirty_list = NULL; 1739 spin_unlock_irq(&bio_dirty_lock); 1740 1741 while ((bio = next) != NULL) { 1742 next = bio->bi_private; 1743 1744 bio_release_pages(bio, true); 1745 bio_put(bio); 1746 } 1747 } 1748 1749 void bio_check_pages_dirty(struct bio *bio) 1750 { 1751 struct bio_vec *bvec; 1752 unsigned long flags; 1753 struct bvec_iter_all iter_all; 1754 1755 bio_for_each_segment_all(bvec, bio, iter_all) { 1756 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1757 goto defer; 1758 } 1759 1760 bio_release_pages(bio, false); 1761 bio_put(bio); 1762 return; 1763 defer: 1764 spin_lock_irqsave(&bio_dirty_lock, flags); 1765 bio->bi_private = bio_dirty_list; 1766 bio_dirty_list = bio; 1767 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1768 schedule_work(&bio_dirty_work); 1769 } 1770 1771 void update_io_ticks(struct hd_struct *part, unsigned long now) 1772 { 1773 unsigned long stamp; 1774 again: 1775 stamp = READ_ONCE(part->stamp); 1776 if (unlikely(stamp != now)) { 1777 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 1778 __part_stat_add(part, io_ticks, 1); 1779 } 1780 } 1781 if (part->partno) { 1782 part = &part_to_disk(part)->part0; 1783 goto again; 1784 } 1785 } 1786 1787 void generic_start_io_acct(struct request_queue *q, int op, 1788 unsigned long sectors, struct hd_struct *part) 1789 { 1790 const int sgrp = op_stat_group(op); 1791 1792 part_stat_lock(); 1793 1794 update_io_ticks(part, jiffies); 1795 part_stat_inc(part, ios[sgrp]); 1796 part_stat_add(part, sectors[sgrp], sectors); 1797 part_inc_in_flight(q, part, op_is_write(op)); 1798 1799 part_stat_unlock(); 1800 } 1801 EXPORT_SYMBOL(generic_start_io_acct); 1802 1803 void generic_end_io_acct(struct request_queue *q, int req_op, 1804 struct hd_struct *part, unsigned long start_time) 1805 { 1806 unsigned long now = jiffies; 1807 unsigned long duration = now - start_time; 1808 const int sgrp = op_stat_group(req_op); 1809 1810 part_stat_lock(); 1811 1812 update_io_ticks(part, now); 1813 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 1814 part_stat_add(part, time_in_queue, duration); 1815 part_dec_in_flight(q, part, op_is_write(req_op)); 1816 1817 part_stat_unlock(); 1818 } 1819 EXPORT_SYMBOL(generic_end_io_acct); 1820 1821 static inline bool bio_remaining_done(struct bio *bio) 1822 { 1823 /* 1824 * If we're not chaining, then ->__bi_remaining is always 1 and 1825 * we always end io on the first invocation. 1826 */ 1827 if (!bio_flagged(bio, BIO_CHAIN)) 1828 return true; 1829 1830 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1831 1832 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1833 bio_clear_flag(bio, BIO_CHAIN); 1834 return true; 1835 } 1836 1837 return false; 1838 } 1839 1840 /** 1841 * bio_endio - end I/O on a bio 1842 * @bio: bio 1843 * 1844 * Description: 1845 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1846 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1847 * bio unless they own it and thus know that it has an end_io function. 1848 * 1849 * bio_endio() can be called several times on a bio that has been chained 1850 * using bio_chain(). The ->bi_end_io() function will only be called the 1851 * last time. At this point the BLK_TA_COMPLETE tracing event will be 1852 * generated if BIO_TRACE_COMPLETION is set. 1853 **/ 1854 void bio_endio(struct bio *bio) 1855 { 1856 again: 1857 if (!bio_remaining_done(bio)) 1858 return; 1859 if (!bio_integrity_endio(bio)) 1860 return; 1861 1862 if (bio->bi_disk) 1863 rq_qos_done_bio(bio->bi_disk->queue, bio); 1864 1865 /* 1866 * Need to have a real endio function for chained bios, otherwise 1867 * various corner cases will break (like stacking block devices that 1868 * save/restore bi_end_io) - however, we want to avoid unbounded 1869 * recursion and blowing the stack. Tail call optimization would 1870 * handle this, but compiling with frame pointers also disables 1871 * gcc's sibling call optimization. 1872 */ 1873 if (bio->bi_end_io == bio_chain_endio) { 1874 bio = __bio_chain_endio(bio); 1875 goto again; 1876 } 1877 1878 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1879 trace_block_bio_complete(bio->bi_disk->queue, bio, 1880 blk_status_to_errno(bio->bi_status)); 1881 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1882 } 1883 1884 blk_throtl_bio_endio(bio); 1885 /* release cgroup info */ 1886 bio_uninit(bio); 1887 if (bio->bi_end_io) 1888 bio->bi_end_io(bio); 1889 } 1890 EXPORT_SYMBOL(bio_endio); 1891 1892 /** 1893 * bio_split - split a bio 1894 * @bio: bio to split 1895 * @sectors: number of sectors to split from the front of @bio 1896 * @gfp: gfp mask 1897 * @bs: bio set to allocate from 1898 * 1899 * Allocates and returns a new bio which represents @sectors from the start of 1900 * @bio, and updates @bio to represent the remaining sectors. 1901 * 1902 * Unless this is a discard request the newly allocated bio will point 1903 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1904 * neither @bio nor @bs are freed before the split bio. 1905 */ 1906 struct bio *bio_split(struct bio *bio, int sectors, 1907 gfp_t gfp, struct bio_set *bs) 1908 { 1909 struct bio *split; 1910 1911 BUG_ON(sectors <= 0); 1912 BUG_ON(sectors >= bio_sectors(bio)); 1913 1914 split = bio_clone_fast(bio, gfp, bs); 1915 if (!split) 1916 return NULL; 1917 1918 split->bi_iter.bi_size = sectors << 9; 1919 1920 if (bio_integrity(split)) 1921 bio_integrity_trim(split); 1922 1923 bio_advance(bio, split->bi_iter.bi_size); 1924 1925 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1926 bio_set_flag(split, BIO_TRACE_COMPLETION); 1927 1928 return split; 1929 } 1930 EXPORT_SYMBOL(bio_split); 1931 1932 /** 1933 * bio_trim - trim a bio 1934 * @bio: bio to trim 1935 * @offset: number of sectors to trim from the front of @bio 1936 * @size: size we want to trim @bio to, in sectors 1937 */ 1938 void bio_trim(struct bio *bio, int offset, int size) 1939 { 1940 /* 'bio' is a cloned bio which we need to trim to match 1941 * the given offset and size. 1942 */ 1943 1944 size <<= 9; 1945 if (offset == 0 && size == bio->bi_iter.bi_size) 1946 return; 1947 1948 bio_advance(bio, offset << 9); 1949 bio->bi_iter.bi_size = size; 1950 1951 if (bio_integrity(bio)) 1952 bio_integrity_trim(bio); 1953 1954 } 1955 EXPORT_SYMBOL_GPL(bio_trim); 1956 1957 /* 1958 * create memory pools for biovec's in a bio_set. 1959 * use the global biovec slabs created for general use. 1960 */ 1961 int biovec_init_pool(mempool_t *pool, int pool_entries) 1962 { 1963 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1964 1965 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1966 } 1967 1968 /* 1969 * bioset_exit - exit a bioset initialized with bioset_init() 1970 * 1971 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1972 * kzalloc()). 1973 */ 1974 void bioset_exit(struct bio_set *bs) 1975 { 1976 if (bs->rescue_workqueue) 1977 destroy_workqueue(bs->rescue_workqueue); 1978 bs->rescue_workqueue = NULL; 1979 1980 mempool_exit(&bs->bio_pool); 1981 mempool_exit(&bs->bvec_pool); 1982 1983 bioset_integrity_free(bs); 1984 if (bs->bio_slab) 1985 bio_put_slab(bs); 1986 bs->bio_slab = NULL; 1987 } 1988 EXPORT_SYMBOL(bioset_exit); 1989 1990 /** 1991 * bioset_init - Initialize a bio_set 1992 * @bs: pool to initialize 1993 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1994 * @front_pad: Number of bytes to allocate in front of the returned bio 1995 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1996 * and %BIOSET_NEED_RESCUER 1997 * 1998 * Description: 1999 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 2000 * to ask for a number of bytes to be allocated in front of the bio. 2001 * Front pad allocation is useful for embedding the bio inside 2002 * another structure, to avoid allocating extra data to go with the bio. 2003 * Note that the bio must be embedded at the END of that structure always, 2004 * or things will break badly. 2005 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 2006 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 2007 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 2008 * dispatch queued requests when the mempool runs out of space. 2009 * 2010 */ 2011 int bioset_init(struct bio_set *bs, 2012 unsigned int pool_size, 2013 unsigned int front_pad, 2014 int flags) 2015 { 2016 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 2017 2018 bs->front_pad = front_pad; 2019 2020 spin_lock_init(&bs->rescue_lock); 2021 bio_list_init(&bs->rescue_list); 2022 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 2023 2024 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 2025 if (!bs->bio_slab) 2026 return -ENOMEM; 2027 2028 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2029 goto bad; 2030 2031 if ((flags & BIOSET_NEED_BVECS) && 2032 biovec_init_pool(&bs->bvec_pool, pool_size)) 2033 goto bad; 2034 2035 if (!(flags & BIOSET_NEED_RESCUER)) 2036 return 0; 2037 2038 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2039 if (!bs->rescue_workqueue) 2040 goto bad; 2041 2042 return 0; 2043 bad: 2044 bioset_exit(bs); 2045 return -ENOMEM; 2046 } 2047 EXPORT_SYMBOL(bioset_init); 2048 2049 /* 2050 * Initialize and setup a new bio_set, based on the settings from 2051 * another bio_set. 2052 */ 2053 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 2054 { 2055 int flags; 2056 2057 flags = 0; 2058 if (src->bvec_pool.min_nr) 2059 flags |= BIOSET_NEED_BVECS; 2060 if (src->rescue_workqueue) 2061 flags |= BIOSET_NEED_RESCUER; 2062 2063 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 2064 } 2065 EXPORT_SYMBOL(bioset_init_from_src); 2066 2067 #ifdef CONFIG_BLK_CGROUP 2068 2069 /** 2070 * bio_disassociate_blkg - puts back the blkg reference if associated 2071 * @bio: target bio 2072 * 2073 * Helper to disassociate the blkg from @bio if a blkg is associated. 2074 */ 2075 void bio_disassociate_blkg(struct bio *bio) 2076 { 2077 if (bio->bi_blkg) { 2078 blkg_put(bio->bi_blkg); 2079 bio->bi_blkg = NULL; 2080 } 2081 } 2082 EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2083 2084 /** 2085 * __bio_associate_blkg - associate a bio with the a blkg 2086 * @bio: target bio 2087 * @blkg: the blkg to associate 2088 * 2089 * This tries to associate @bio with the specified @blkg. Association failure 2090 * is handled by walking up the blkg tree. Therefore, the blkg associated can 2091 * be anything between @blkg and the root_blkg. This situation only happens 2092 * when a cgroup is dying and then the remaining bios will spill to the closest 2093 * alive blkg. 2094 * 2095 * A reference will be taken on the @blkg and will be released when @bio is 2096 * freed. 2097 */ 2098 static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2099 { 2100 bio_disassociate_blkg(bio); 2101 2102 bio->bi_blkg = blkg_tryget_closest(blkg); 2103 } 2104 2105 /** 2106 * bio_associate_blkg_from_css - associate a bio with a specified css 2107 * @bio: target bio 2108 * @css: target css 2109 * 2110 * Associate @bio with the blkg found by combining the css's blkg and the 2111 * request_queue of the @bio. This falls back to the queue's root_blkg if 2112 * the association fails with the css. 2113 */ 2114 void bio_associate_blkg_from_css(struct bio *bio, 2115 struct cgroup_subsys_state *css) 2116 { 2117 struct request_queue *q = bio->bi_disk->queue; 2118 struct blkcg_gq *blkg; 2119 2120 rcu_read_lock(); 2121 2122 if (!css || !css->parent) 2123 blkg = q->root_blkg; 2124 else 2125 blkg = blkg_lookup_create(css_to_blkcg(css), q); 2126 2127 __bio_associate_blkg(bio, blkg); 2128 2129 rcu_read_unlock(); 2130 } 2131 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2132 2133 #ifdef CONFIG_MEMCG 2134 /** 2135 * bio_associate_blkg_from_page - associate a bio with the page's blkg 2136 * @bio: target bio 2137 * @page: the page to lookup the blkcg from 2138 * 2139 * Associate @bio with the blkg from @page's owning memcg and the respective 2140 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2141 * root_blkg. 2142 */ 2143 void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 2144 { 2145 struct cgroup_subsys_state *css; 2146 2147 if (!page->mem_cgroup) 2148 return; 2149 2150 rcu_read_lock(); 2151 2152 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2153 bio_associate_blkg_from_css(bio, css); 2154 2155 rcu_read_unlock(); 2156 } 2157 #endif /* CONFIG_MEMCG */ 2158 2159 /** 2160 * bio_associate_blkg - associate a bio with a blkg 2161 * @bio: target bio 2162 * 2163 * Associate @bio with the blkg found from the bio's css and request_queue. 2164 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 2165 * already associated, the css is reused and association redone as the 2166 * request_queue may have changed. 2167 */ 2168 void bio_associate_blkg(struct bio *bio) 2169 { 2170 struct cgroup_subsys_state *css; 2171 2172 rcu_read_lock(); 2173 2174 if (bio->bi_blkg) 2175 css = &bio_blkcg(bio)->css; 2176 else 2177 css = blkcg_css(); 2178 2179 bio_associate_blkg_from_css(bio, css); 2180 2181 rcu_read_unlock(); 2182 } 2183 EXPORT_SYMBOL_GPL(bio_associate_blkg); 2184 2185 /** 2186 * bio_clone_blkg_association - clone blkg association from src to dst bio 2187 * @dst: destination bio 2188 * @src: source bio 2189 */ 2190 void bio_clone_blkg_association(struct bio *dst, struct bio *src) 2191 { 2192 rcu_read_lock(); 2193 2194 if (src->bi_blkg) 2195 __bio_associate_blkg(dst, src->bi_blkg); 2196 2197 rcu_read_unlock(); 2198 } 2199 EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2200 #endif /* CONFIG_BLK_CGROUP */ 2201 2202 static void __init biovec_init_slabs(void) 2203 { 2204 int i; 2205 2206 for (i = 0; i < BVEC_POOL_NR; i++) { 2207 int size; 2208 struct biovec_slab *bvs = bvec_slabs + i; 2209 2210 if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2211 bvs->slab = NULL; 2212 continue; 2213 } 2214 2215 size = bvs->nr_vecs * sizeof(struct bio_vec); 2216 bvs->slab = kmem_cache_create(bvs->name, size, 0, 2217 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2218 } 2219 } 2220 2221 static int __init init_bio(void) 2222 { 2223 bio_slab_max = 2; 2224 bio_slab_nr = 0; 2225 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 2226 GFP_KERNEL); 2227 2228 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 2229 2230 if (!bio_slabs) 2231 panic("bio: can't allocate bios\n"); 2232 2233 bio_integrity_init(); 2234 biovec_init_slabs(); 2235 2236 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2237 panic("bio: can't allocate bios\n"); 2238 2239 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2240 panic("bio: can't create integrity pool\n"); 2241 2242 return 0; 2243 } 2244 subsys_initcall(init_bio); 2245