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 21 #include <trace/events/block.h> 22 #include "blk.h" 23 #include "blk-rq-qos.h" 24 25 /* 26 * Test patch to inline a certain number of bi_io_vec's inside the bio 27 * itself, to shrink a bio data allocation from two mempool calls to one 28 */ 29 #define BIO_INLINE_VECS 4 30 31 /* 32 * if you change this list, also change bvec_alloc or things will 33 * break badly! cannot be bigger than what you can fit into an 34 * unsigned short 35 */ 36 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 37 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 38 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 39 }; 40 #undef BV 41 42 /* 43 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 44 * IO code that does not need private memory pools. 45 */ 46 struct bio_set fs_bio_set; 47 EXPORT_SYMBOL(fs_bio_set); 48 49 /* 50 * Our slab pool management 51 */ 52 struct bio_slab { 53 struct kmem_cache *slab; 54 unsigned int slab_ref; 55 unsigned int slab_size; 56 char name[8]; 57 }; 58 static DEFINE_MUTEX(bio_slab_lock); 59 static struct bio_slab *bio_slabs; 60 static unsigned int bio_slab_nr, bio_slab_max; 61 62 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 63 { 64 unsigned int sz = sizeof(struct bio) + extra_size; 65 struct kmem_cache *slab = NULL; 66 struct bio_slab *bslab, *new_bio_slabs; 67 unsigned int new_bio_slab_max; 68 unsigned int i, entry = -1; 69 70 mutex_lock(&bio_slab_lock); 71 72 i = 0; 73 while (i < bio_slab_nr) { 74 bslab = &bio_slabs[i]; 75 76 if (!bslab->slab && entry == -1) 77 entry = i; 78 else if (bslab->slab_size == sz) { 79 slab = bslab->slab; 80 bslab->slab_ref++; 81 break; 82 } 83 i++; 84 } 85 86 if (slab) 87 goto out_unlock; 88 89 if (bio_slab_nr == bio_slab_max && entry == -1) { 90 new_bio_slab_max = bio_slab_max << 1; 91 new_bio_slabs = krealloc(bio_slabs, 92 new_bio_slab_max * sizeof(struct bio_slab), 93 GFP_KERNEL); 94 if (!new_bio_slabs) 95 goto out_unlock; 96 bio_slab_max = new_bio_slab_max; 97 bio_slabs = new_bio_slabs; 98 } 99 if (entry == -1) 100 entry = bio_slab_nr++; 101 102 bslab = &bio_slabs[entry]; 103 104 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 105 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 106 SLAB_HWCACHE_ALIGN, NULL); 107 if (!slab) 108 goto out_unlock; 109 110 bslab->slab = slab; 111 bslab->slab_ref = 1; 112 bslab->slab_size = sz; 113 out_unlock: 114 mutex_unlock(&bio_slab_lock); 115 return slab; 116 } 117 118 static void bio_put_slab(struct bio_set *bs) 119 { 120 struct bio_slab *bslab = NULL; 121 unsigned int i; 122 123 mutex_lock(&bio_slab_lock); 124 125 for (i = 0; i < bio_slab_nr; i++) { 126 if (bs->bio_slab == bio_slabs[i].slab) { 127 bslab = &bio_slabs[i]; 128 break; 129 } 130 } 131 132 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 133 goto out; 134 135 WARN_ON(!bslab->slab_ref); 136 137 if (--bslab->slab_ref) 138 goto out; 139 140 kmem_cache_destroy(bslab->slab); 141 bslab->slab = NULL; 142 143 out: 144 mutex_unlock(&bio_slab_lock); 145 } 146 147 unsigned int bvec_nr_vecs(unsigned short idx) 148 { 149 return bvec_slabs[--idx].nr_vecs; 150 } 151 152 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 153 { 154 if (!idx) 155 return; 156 idx--; 157 158 BIO_BUG_ON(idx >= BVEC_POOL_NR); 159 160 if (idx == BVEC_POOL_MAX) { 161 mempool_free(bv, pool); 162 } else { 163 struct biovec_slab *bvs = bvec_slabs + idx; 164 165 kmem_cache_free(bvs->slab, bv); 166 } 167 } 168 169 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 170 mempool_t *pool) 171 { 172 struct bio_vec *bvl; 173 174 /* 175 * see comment near bvec_array define! 176 */ 177 switch (nr) { 178 case 1: 179 *idx = 0; 180 break; 181 case 2 ... 4: 182 *idx = 1; 183 break; 184 case 5 ... 16: 185 *idx = 2; 186 break; 187 case 17 ... 64: 188 *idx = 3; 189 break; 190 case 65 ... 128: 191 *idx = 4; 192 break; 193 case 129 ... BIO_MAX_PAGES: 194 *idx = 5; 195 break; 196 default: 197 return NULL; 198 } 199 200 /* 201 * idx now points to the pool we want to allocate from. only the 202 * 1-vec entry pool is mempool backed. 203 */ 204 if (*idx == BVEC_POOL_MAX) { 205 fallback: 206 bvl = mempool_alloc(pool, gfp_mask); 207 } else { 208 struct biovec_slab *bvs = bvec_slabs + *idx; 209 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 210 211 /* 212 * Make this allocation restricted and don't dump info on 213 * allocation failures, since we'll fallback to the mempool 214 * in case of failure. 215 */ 216 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 217 218 /* 219 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 220 * is set, retry with the 1-entry mempool 221 */ 222 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 223 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 224 *idx = BVEC_POOL_MAX; 225 goto fallback; 226 } 227 } 228 229 (*idx)++; 230 return bvl; 231 } 232 233 void bio_uninit(struct bio *bio) 234 { 235 bio_disassociate_blkg(bio); 236 237 if (bio_integrity(bio)) 238 bio_integrity_free(bio); 239 } 240 EXPORT_SYMBOL(bio_uninit); 241 242 static void bio_free(struct bio *bio) 243 { 244 struct bio_set *bs = bio->bi_pool; 245 void *p; 246 247 bio_uninit(bio); 248 249 if (bs) { 250 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 251 252 /* 253 * If we have front padding, adjust the bio pointer before freeing 254 */ 255 p = bio; 256 p -= bs->front_pad; 257 258 mempool_free(p, &bs->bio_pool); 259 } else { 260 /* Bio was allocated by bio_kmalloc() */ 261 kfree(bio); 262 } 263 } 264 265 /* 266 * Users of this function have their own bio allocation. Subsequently, 267 * they must remember to pair any call to bio_init() with bio_uninit() 268 * when IO has completed, or when the bio is released. 269 */ 270 void bio_init(struct bio *bio, struct bio_vec *table, 271 unsigned short max_vecs) 272 { 273 memset(bio, 0, sizeof(*bio)); 274 atomic_set(&bio->__bi_remaining, 1); 275 atomic_set(&bio->__bi_cnt, 1); 276 277 bio->bi_io_vec = table; 278 bio->bi_max_vecs = max_vecs; 279 } 280 EXPORT_SYMBOL(bio_init); 281 282 /** 283 * bio_reset - reinitialize a bio 284 * @bio: bio to reset 285 * 286 * Description: 287 * After calling bio_reset(), @bio will be in the same state as a freshly 288 * allocated bio returned bio bio_alloc_bioset() - the only fields that are 289 * preserved are the ones that are initialized by bio_alloc_bioset(). See 290 * comment in struct bio. 291 */ 292 void bio_reset(struct bio *bio) 293 { 294 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 295 296 bio_uninit(bio); 297 298 memset(bio, 0, BIO_RESET_BYTES); 299 bio->bi_flags = flags; 300 atomic_set(&bio->__bi_remaining, 1); 301 } 302 EXPORT_SYMBOL(bio_reset); 303 304 static struct bio *__bio_chain_endio(struct bio *bio) 305 { 306 struct bio *parent = bio->bi_private; 307 308 if (!parent->bi_status) 309 parent->bi_status = bio->bi_status; 310 bio_put(bio); 311 return parent; 312 } 313 314 static void bio_chain_endio(struct bio *bio) 315 { 316 bio_endio(__bio_chain_endio(bio)); 317 } 318 319 /** 320 * bio_chain - chain bio completions 321 * @bio: the target bio 322 * @parent: the @bio's parent bio 323 * 324 * The caller won't have a bi_end_io called when @bio completes - instead, 325 * @parent's bi_end_io won't be called until both @parent and @bio have 326 * completed; the chained bio will also be freed when it completes. 327 * 328 * The caller must not set bi_private or bi_end_io in @bio. 329 */ 330 void bio_chain(struct bio *bio, struct bio *parent) 331 { 332 BUG_ON(bio->bi_private || bio->bi_end_io); 333 334 bio->bi_private = parent; 335 bio->bi_end_io = bio_chain_endio; 336 bio_inc_remaining(parent); 337 } 338 EXPORT_SYMBOL(bio_chain); 339 340 static void bio_alloc_rescue(struct work_struct *work) 341 { 342 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 343 struct bio *bio; 344 345 while (1) { 346 spin_lock(&bs->rescue_lock); 347 bio = bio_list_pop(&bs->rescue_list); 348 spin_unlock(&bs->rescue_lock); 349 350 if (!bio) 351 break; 352 353 generic_make_request(bio); 354 } 355 } 356 357 static void punt_bios_to_rescuer(struct bio_set *bs) 358 { 359 struct bio_list punt, nopunt; 360 struct bio *bio; 361 362 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 363 return; 364 /* 365 * In order to guarantee forward progress we must punt only bios that 366 * were allocated from this bio_set; otherwise, if there was a bio on 367 * there for a stacking driver higher up in the stack, processing it 368 * could require allocating bios from this bio_set, and doing that from 369 * our own rescuer would be bad. 370 * 371 * Since bio lists are singly linked, pop them all instead of trying to 372 * remove from the middle of the list: 373 */ 374 375 bio_list_init(&punt); 376 bio_list_init(&nopunt); 377 378 while ((bio = bio_list_pop(¤t->bio_list[0]))) 379 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 380 current->bio_list[0] = nopunt; 381 382 bio_list_init(&nopunt); 383 while ((bio = bio_list_pop(¤t->bio_list[1]))) 384 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 385 current->bio_list[1] = nopunt; 386 387 spin_lock(&bs->rescue_lock); 388 bio_list_merge(&bs->rescue_list, &punt); 389 spin_unlock(&bs->rescue_lock); 390 391 queue_work(bs->rescue_workqueue, &bs->rescue_work); 392 } 393 394 /** 395 * bio_alloc_bioset - allocate a bio for I/O 396 * @gfp_mask: the GFP_* mask given to the slab allocator 397 * @nr_iovecs: number of iovecs to pre-allocate 398 * @bs: the bio_set to allocate from. 399 * 400 * Description: 401 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 402 * backed by the @bs's mempool. 403 * 404 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 405 * always be able to allocate a bio. This is due to the mempool guarantees. 406 * To make this work, callers must never allocate more than 1 bio at a time 407 * from this pool. Callers that need to allocate more than 1 bio must always 408 * submit the previously allocated bio for IO before attempting to allocate 409 * a new one. Failure to do so can cause deadlocks under memory pressure. 410 * 411 * Note that when running under generic_make_request() (i.e. any block 412 * driver), bios are not submitted until after you return - see the code in 413 * generic_make_request() that converts recursion into iteration, to prevent 414 * stack overflows. 415 * 416 * This would normally mean allocating multiple bios under 417 * generic_make_request() would be susceptible to deadlocks, but we have 418 * deadlock avoidance code that resubmits any blocked bios from a rescuer 419 * thread. 420 * 421 * However, we do not guarantee forward progress for allocations from other 422 * mempools. Doing multiple allocations from the same mempool under 423 * generic_make_request() should be avoided - instead, use bio_set's front_pad 424 * for per bio allocations. 425 * 426 * RETURNS: 427 * Pointer to new bio on success, NULL on failure. 428 */ 429 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 430 struct bio_set *bs) 431 { 432 gfp_t saved_gfp = gfp_mask; 433 unsigned front_pad; 434 unsigned inline_vecs; 435 struct bio_vec *bvl = NULL; 436 struct bio *bio; 437 void *p; 438 439 if (!bs) { 440 if (nr_iovecs > UIO_MAXIOV) 441 return NULL; 442 443 p = kmalloc(sizeof(struct bio) + 444 nr_iovecs * sizeof(struct bio_vec), 445 gfp_mask); 446 front_pad = 0; 447 inline_vecs = nr_iovecs; 448 } else { 449 /* should not use nobvec bioset for nr_iovecs > 0 */ 450 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 451 nr_iovecs > 0)) 452 return NULL; 453 /* 454 * generic_make_request() converts recursion to iteration; this 455 * means if we're running beneath it, any bios we allocate and 456 * submit will not be submitted (and thus freed) until after we 457 * return. 458 * 459 * This exposes us to a potential deadlock if we allocate 460 * multiple bios from the same bio_set() while running 461 * underneath generic_make_request(). If we were to allocate 462 * multiple bios (say a stacking block driver that was splitting 463 * bios), we would deadlock if we exhausted the mempool's 464 * reserve. 465 * 466 * We solve this, and guarantee forward progress, with a rescuer 467 * workqueue per bio_set. If we go to allocate and there are 468 * bios on current->bio_list, we first try the allocation 469 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 470 * bios we would be blocking to the rescuer workqueue before 471 * we retry with the original gfp_flags. 472 */ 473 474 if (current->bio_list && 475 (!bio_list_empty(¤t->bio_list[0]) || 476 !bio_list_empty(¤t->bio_list[1])) && 477 bs->rescue_workqueue) 478 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 479 480 p = mempool_alloc(&bs->bio_pool, gfp_mask); 481 if (!p && gfp_mask != saved_gfp) { 482 punt_bios_to_rescuer(bs); 483 gfp_mask = saved_gfp; 484 p = mempool_alloc(&bs->bio_pool, gfp_mask); 485 } 486 487 front_pad = bs->front_pad; 488 inline_vecs = BIO_INLINE_VECS; 489 } 490 491 if (unlikely(!p)) 492 return NULL; 493 494 bio = p + front_pad; 495 bio_init(bio, NULL, 0); 496 497 if (nr_iovecs > inline_vecs) { 498 unsigned long idx = 0; 499 500 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 501 if (!bvl && gfp_mask != saved_gfp) { 502 punt_bios_to_rescuer(bs); 503 gfp_mask = saved_gfp; 504 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 505 } 506 507 if (unlikely(!bvl)) 508 goto err_free; 509 510 bio->bi_flags |= idx << BVEC_POOL_OFFSET; 511 } else if (nr_iovecs) { 512 bvl = bio->bi_inline_vecs; 513 } 514 515 bio->bi_pool = bs; 516 bio->bi_max_vecs = nr_iovecs; 517 bio->bi_io_vec = bvl; 518 return bio; 519 520 err_free: 521 mempool_free(p, &bs->bio_pool); 522 return NULL; 523 } 524 EXPORT_SYMBOL(bio_alloc_bioset); 525 526 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 527 { 528 unsigned long flags; 529 struct bio_vec bv; 530 struct bvec_iter iter; 531 532 __bio_for_each_segment(bv, bio, iter, start) { 533 char *data = bvec_kmap_irq(&bv, &flags); 534 memset(data, 0, bv.bv_len); 535 flush_dcache_page(bv.bv_page); 536 bvec_kunmap_irq(data, &flags); 537 } 538 } 539 EXPORT_SYMBOL(zero_fill_bio_iter); 540 541 void bio_truncate(struct bio *bio, unsigned new_size) 542 { 543 struct bio_vec bv; 544 struct bvec_iter iter; 545 unsigned int done = 0; 546 bool truncated = false; 547 548 if (new_size >= bio->bi_iter.bi_size) 549 return; 550 551 if (bio_data_dir(bio) != READ) 552 goto exit; 553 554 bio_for_each_segment(bv, bio, iter) { 555 if (done + bv.bv_len > new_size) { 556 unsigned offset; 557 558 if (!truncated) 559 offset = new_size - done; 560 else 561 offset = 0; 562 zero_user(bv.bv_page, offset, bv.bv_len - offset); 563 truncated = true; 564 } 565 done += bv.bv_len; 566 } 567 568 exit: 569 /* 570 * Don't touch bvec table here and make it really immutable, since 571 * fs bio user has to retrieve all pages via bio_for_each_segment_all 572 * in its .end_bio() callback. 573 * 574 * It is enough to truncate bio by updating .bi_size since we can make 575 * correct bvec with the updated .bi_size for drivers. 576 */ 577 bio->bi_iter.bi_size = new_size; 578 } 579 580 /** 581 * bio_put - release a reference to a bio 582 * @bio: bio to release reference to 583 * 584 * Description: 585 * Put a reference to a &struct bio, either one you have gotten with 586 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 587 **/ 588 void bio_put(struct bio *bio) 589 { 590 if (!bio_flagged(bio, BIO_REFFED)) 591 bio_free(bio); 592 else { 593 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 594 595 /* 596 * last put frees it 597 */ 598 if (atomic_dec_and_test(&bio->__bi_cnt)) 599 bio_free(bio); 600 } 601 } 602 EXPORT_SYMBOL(bio_put); 603 604 /** 605 * __bio_clone_fast - clone a bio that shares the original bio's biovec 606 * @bio: destination bio 607 * @bio_src: bio to clone 608 * 609 * Clone a &bio. Caller will own the returned bio, but not 610 * the actual data it points to. Reference count of returned 611 * bio will be one. 612 * 613 * Caller must ensure that @bio_src is not freed before @bio. 614 */ 615 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 616 { 617 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 618 619 /* 620 * most users will be overriding ->bi_disk with a new target, 621 * so we don't set nor calculate new physical/hw segment counts here 622 */ 623 bio->bi_disk = bio_src->bi_disk; 624 bio->bi_partno = bio_src->bi_partno; 625 bio_set_flag(bio, BIO_CLONED); 626 if (bio_flagged(bio_src, BIO_THROTTLED)) 627 bio_set_flag(bio, BIO_THROTTLED); 628 bio->bi_opf = bio_src->bi_opf; 629 bio->bi_ioprio = bio_src->bi_ioprio; 630 bio->bi_write_hint = bio_src->bi_write_hint; 631 bio->bi_iter = bio_src->bi_iter; 632 bio->bi_io_vec = bio_src->bi_io_vec; 633 634 bio_clone_blkg_association(bio, bio_src); 635 blkcg_bio_issue_init(bio); 636 } 637 EXPORT_SYMBOL(__bio_clone_fast); 638 639 /** 640 * bio_clone_fast - clone a bio that shares the original bio's biovec 641 * @bio: bio to clone 642 * @gfp_mask: allocation priority 643 * @bs: bio_set to allocate from 644 * 645 * Like __bio_clone_fast, only also allocates the returned bio 646 */ 647 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 648 { 649 struct bio *b; 650 651 b = bio_alloc_bioset(gfp_mask, 0, bs); 652 if (!b) 653 return NULL; 654 655 __bio_clone_fast(b, bio); 656 657 if (bio_integrity(bio)) { 658 int ret; 659 660 ret = bio_integrity_clone(b, bio, gfp_mask); 661 662 if (ret < 0) { 663 bio_put(b); 664 return NULL; 665 } 666 } 667 668 return b; 669 } 670 EXPORT_SYMBOL(bio_clone_fast); 671 672 static inline bool page_is_mergeable(const struct bio_vec *bv, 673 struct page *page, unsigned int len, unsigned int off, 674 bool *same_page) 675 { 676 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 677 bv->bv_offset + bv->bv_len - 1; 678 phys_addr_t page_addr = page_to_phys(page); 679 680 if (vec_end_addr + 1 != page_addr + off) 681 return false; 682 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 683 return false; 684 685 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 686 if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) 687 return false; 688 return true; 689 } 690 691 static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio, 692 struct page *page, unsigned len, unsigned offset, 693 bool *same_page) 694 { 695 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 696 unsigned long mask = queue_segment_boundary(q); 697 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 698 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 699 700 if ((addr1 | mask) != (addr2 | mask)) 701 return false; 702 if (bv->bv_len + len > queue_max_segment_size(q)) 703 return false; 704 return __bio_try_merge_page(bio, page, len, offset, same_page); 705 } 706 707 /** 708 * __bio_add_pc_page - attempt to add page to passthrough bio 709 * @q: the target queue 710 * @bio: destination bio 711 * @page: page to add 712 * @len: vec entry length 713 * @offset: vec entry offset 714 * @same_page: return if the merge happen inside the same page 715 * 716 * Attempt to add a page to the bio_vec maplist. This can fail for a 717 * number of reasons, such as the bio being full or target block device 718 * limitations. The target block device must allow bio's up to PAGE_SIZE, 719 * so it is always possible to add a single page to an empty bio. 720 * 721 * This should only be used by passthrough bios. 722 */ 723 static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, 724 struct page *page, unsigned int len, unsigned int offset, 725 bool *same_page) 726 { 727 struct bio_vec *bvec; 728 729 /* 730 * cloned bio must not modify vec list 731 */ 732 if (unlikely(bio_flagged(bio, BIO_CLONED))) 733 return 0; 734 735 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 736 return 0; 737 738 if (bio->bi_vcnt > 0) { 739 if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page)) 740 return len; 741 742 /* 743 * If the queue doesn't support SG gaps and adding this segment 744 * would create a gap, disallow it. 745 */ 746 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 747 if (bvec_gap_to_prev(q, bvec, offset)) 748 return 0; 749 } 750 751 if (bio_full(bio, len)) 752 return 0; 753 754 if (bio->bi_vcnt >= queue_max_segments(q)) 755 return 0; 756 757 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 758 bvec->bv_page = page; 759 bvec->bv_len = len; 760 bvec->bv_offset = offset; 761 bio->bi_vcnt++; 762 bio->bi_iter.bi_size += len; 763 return len; 764 } 765 766 int bio_add_pc_page(struct request_queue *q, struct bio *bio, 767 struct page *page, unsigned int len, unsigned int offset) 768 { 769 bool same_page = false; 770 return __bio_add_pc_page(q, bio, page, len, offset, &same_page); 771 } 772 EXPORT_SYMBOL(bio_add_pc_page); 773 774 /** 775 * __bio_try_merge_page - try appending data to an existing bvec. 776 * @bio: destination bio 777 * @page: start page to add 778 * @len: length of the data to add 779 * @off: offset of the data relative to @page 780 * @same_page: return if the segment has been merged inside the same page 781 * 782 * Try to add the data at @page + @off to the last bvec of @bio. This is a 783 * a useful optimisation for file systems with a block size smaller than the 784 * page size. 785 * 786 * Warn if (@len, @off) crosses pages in case that @same_page is true. 787 * 788 * Return %true on success or %false on failure. 789 */ 790 bool __bio_try_merge_page(struct bio *bio, struct page *page, 791 unsigned int len, unsigned int off, bool *same_page) 792 { 793 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 794 return false; 795 796 if (bio->bi_vcnt > 0) { 797 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 798 799 if (page_is_mergeable(bv, page, len, off, same_page)) { 800 if (bio->bi_iter.bi_size > UINT_MAX - len) 801 return false; 802 bv->bv_len += len; 803 bio->bi_iter.bi_size += len; 804 return true; 805 } 806 } 807 return false; 808 } 809 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 810 811 /** 812 * __bio_add_page - add page(s) to a bio in a new segment 813 * @bio: destination bio 814 * @page: start page to add 815 * @len: length of the data to add, may cross pages 816 * @off: offset of the data relative to @page, may cross pages 817 * 818 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 819 * that @bio has space for another bvec. 820 */ 821 void __bio_add_page(struct bio *bio, struct page *page, 822 unsigned int len, unsigned int off) 823 { 824 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 825 826 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 827 WARN_ON_ONCE(bio_full(bio, len)); 828 829 bv->bv_page = page; 830 bv->bv_offset = off; 831 bv->bv_len = len; 832 833 bio->bi_iter.bi_size += len; 834 bio->bi_vcnt++; 835 836 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 837 bio_set_flag(bio, BIO_WORKINGSET); 838 } 839 EXPORT_SYMBOL_GPL(__bio_add_page); 840 841 /** 842 * bio_add_page - attempt to add page(s) to bio 843 * @bio: destination bio 844 * @page: start page to add 845 * @len: vec entry length, may cross pages 846 * @offset: vec entry offset relative to @page, may cross pages 847 * 848 * Attempt to add page(s) to the bio_vec maplist. This will only fail 849 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 850 */ 851 int bio_add_page(struct bio *bio, struct page *page, 852 unsigned int len, unsigned int offset) 853 { 854 bool same_page = false; 855 856 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 857 if (bio_full(bio, len)) 858 return 0; 859 __bio_add_page(bio, page, len, offset); 860 } 861 return len; 862 } 863 EXPORT_SYMBOL(bio_add_page); 864 865 void bio_release_pages(struct bio *bio, bool mark_dirty) 866 { 867 struct bvec_iter_all iter_all; 868 struct bio_vec *bvec; 869 870 if (bio_flagged(bio, BIO_NO_PAGE_REF)) 871 return; 872 873 bio_for_each_segment_all(bvec, bio, iter_all) { 874 if (mark_dirty && !PageCompound(bvec->bv_page)) 875 set_page_dirty_lock(bvec->bv_page); 876 put_page(bvec->bv_page); 877 } 878 } 879 880 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 881 { 882 const struct bio_vec *bv = iter->bvec; 883 unsigned int len; 884 size_t size; 885 886 if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 887 return -EINVAL; 888 889 len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 890 size = bio_add_page(bio, bv->bv_page, len, 891 bv->bv_offset + iter->iov_offset); 892 if (unlikely(size != len)) 893 return -EINVAL; 894 iov_iter_advance(iter, size); 895 return 0; 896 } 897 898 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 899 900 /** 901 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 902 * @bio: bio to add pages to 903 * @iter: iov iterator describing the region to be mapped 904 * 905 * Pins pages from *iter and appends them to @bio's bvec array. The 906 * pages will have to be released using put_page() when done. 907 * For multi-segment *iter, this function only adds pages from the 908 * the next non-empty segment of the iov iterator. 909 */ 910 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 911 { 912 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 913 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 914 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 915 struct page **pages = (struct page **)bv; 916 bool same_page = false; 917 ssize_t size, left; 918 unsigned len, i; 919 size_t offset; 920 921 /* 922 * Move page array up in the allocated memory for the bio vecs as far as 923 * possible so that we can start filling biovecs from the beginning 924 * without overwriting the temporary page array. 925 */ 926 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 927 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 928 929 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 930 if (unlikely(size <= 0)) 931 return size ? size : -EFAULT; 932 933 for (left = size, i = 0; left > 0; left -= len, i++) { 934 struct page *page = pages[i]; 935 936 len = min_t(size_t, PAGE_SIZE - offset, left); 937 938 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 939 if (same_page) 940 put_page(page); 941 } else { 942 if (WARN_ON_ONCE(bio_full(bio, len))) 943 return -EINVAL; 944 __bio_add_page(bio, page, len, offset); 945 } 946 offset = 0; 947 } 948 949 iov_iter_advance(iter, size); 950 return 0; 951 } 952 953 /** 954 * bio_iov_iter_get_pages - add user or kernel pages to a bio 955 * @bio: bio to add pages to 956 * @iter: iov iterator describing the region to be added 957 * 958 * This takes either an iterator pointing to user memory, or one pointing to 959 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 960 * map them into the kernel. On IO completion, the caller should put those 961 * pages. If we're adding kernel pages, and the caller told us it's safe to 962 * do so, we just have to add the pages to the bio directly. We don't grab an 963 * extra reference to those pages (the user should already have that), and we 964 * don't put the page on IO completion. The caller needs to check if the bio is 965 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 966 * released. 967 * 968 * The function tries, but does not guarantee, to pin as many pages as 969 * fit into the bio, or are requested in *iter, whatever is smaller. If 970 * MM encounters an error pinning the requested pages, it stops. Error 971 * is returned only if 0 pages could be pinned. 972 */ 973 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 974 { 975 const bool is_bvec = iov_iter_is_bvec(iter); 976 int ret; 977 978 if (WARN_ON_ONCE(bio->bi_vcnt)) 979 return -EINVAL; 980 981 do { 982 if (is_bvec) 983 ret = __bio_iov_bvec_add_pages(bio, iter); 984 else 985 ret = __bio_iov_iter_get_pages(bio, iter); 986 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 987 988 if (is_bvec) 989 bio_set_flag(bio, BIO_NO_PAGE_REF); 990 return bio->bi_vcnt ? 0 : ret; 991 } 992 993 static void submit_bio_wait_endio(struct bio *bio) 994 { 995 complete(bio->bi_private); 996 } 997 998 /** 999 * submit_bio_wait - submit a bio, and wait until it completes 1000 * @bio: The &struct bio which describes the I/O 1001 * 1002 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1003 * bio_endio() on failure. 1004 * 1005 * WARNING: Unlike to how submit_bio() is usually used, this function does not 1006 * result in bio reference to be consumed. The caller must drop the reference 1007 * on his own. 1008 */ 1009 int submit_bio_wait(struct bio *bio) 1010 { 1011 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 1012 1013 bio->bi_private = &done; 1014 bio->bi_end_io = submit_bio_wait_endio; 1015 bio->bi_opf |= REQ_SYNC; 1016 submit_bio(bio); 1017 wait_for_completion_io(&done); 1018 1019 return blk_status_to_errno(bio->bi_status); 1020 } 1021 EXPORT_SYMBOL(submit_bio_wait); 1022 1023 /** 1024 * bio_advance - increment/complete a bio by some number of bytes 1025 * @bio: bio to advance 1026 * @bytes: number of bytes to complete 1027 * 1028 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1029 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1030 * be updated on the last bvec as well. 1031 * 1032 * @bio will then represent the remaining, uncompleted portion of the io. 1033 */ 1034 void bio_advance(struct bio *bio, unsigned bytes) 1035 { 1036 if (bio_integrity(bio)) 1037 bio_integrity_advance(bio, bytes); 1038 1039 bio_advance_iter(bio, &bio->bi_iter, bytes); 1040 } 1041 EXPORT_SYMBOL(bio_advance); 1042 1043 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1044 struct bio *src, struct bvec_iter *src_iter) 1045 { 1046 struct bio_vec src_bv, dst_bv; 1047 void *src_p, *dst_p; 1048 unsigned bytes; 1049 1050 while (src_iter->bi_size && dst_iter->bi_size) { 1051 src_bv = bio_iter_iovec(src, *src_iter); 1052 dst_bv = bio_iter_iovec(dst, *dst_iter); 1053 1054 bytes = min(src_bv.bv_len, dst_bv.bv_len); 1055 1056 src_p = kmap_atomic(src_bv.bv_page); 1057 dst_p = kmap_atomic(dst_bv.bv_page); 1058 1059 memcpy(dst_p + dst_bv.bv_offset, 1060 src_p + src_bv.bv_offset, 1061 bytes); 1062 1063 kunmap_atomic(dst_p); 1064 kunmap_atomic(src_p); 1065 1066 flush_dcache_page(dst_bv.bv_page); 1067 1068 bio_advance_iter(src, src_iter, bytes); 1069 bio_advance_iter(dst, dst_iter, bytes); 1070 } 1071 } 1072 EXPORT_SYMBOL(bio_copy_data_iter); 1073 1074 /** 1075 * bio_copy_data - copy contents of data buffers from one bio to another 1076 * @src: source bio 1077 * @dst: destination bio 1078 * 1079 * Stops when it reaches the end of either @src or @dst - that is, copies 1080 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1081 */ 1082 void bio_copy_data(struct bio *dst, struct bio *src) 1083 { 1084 struct bvec_iter src_iter = src->bi_iter; 1085 struct bvec_iter dst_iter = dst->bi_iter; 1086 1087 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1088 } 1089 EXPORT_SYMBOL(bio_copy_data); 1090 1091 /** 1092 * bio_list_copy_data - copy contents of data buffers from one chain of bios to 1093 * another 1094 * @src: source bio list 1095 * @dst: destination bio list 1096 * 1097 * Stops when it reaches the end of either the @src list or @dst list - that is, 1098 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 1099 * bios). 1100 */ 1101 void bio_list_copy_data(struct bio *dst, struct bio *src) 1102 { 1103 struct bvec_iter src_iter = src->bi_iter; 1104 struct bvec_iter dst_iter = dst->bi_iter; 1105 1106 while (1) { 1107 if (!src_iter.bi_size) { 1108 src = src->bi_next; 1109 if (!src) 1110 break; 1111 1112 src_iter = src->bi_iter; 1113 } 1114 1115 if (!dst_iter.bi_size) { 1116 dst = dst->bi_next; 1117 if (!dst) 1118 break; 1119 1120 dst_iter = dst->bi_iter; 1121 } 1122 1123 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1124 } 1125 } 1126 EXPORT_SYMBOL(bio_list_copy_data); 1127 1128 struct bio_map_data { 1129 int is_our_pages; 1130 struct iov_iter iter; 1131 struct iovec iov[]; 1132 }; 1133 1134 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1135 gfp_t gfp_mask) 1136 { 1137 struct bio_map_data *bmd; 1138 if (data->nr_segs > UIO_MAXIOV) 1139 return NULL; 1140 1141 bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); 1142 if (!bmd) 1143 return NULL; 1144 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 1145 bmd->iter = *data; 1146 bmd->iter.iov = bmd->iov; 1147 return bmd; 1148 } 1149 1150 /** 1151 * bio_copy_from_iter - copy all pages from iov_iter to bio 1152 * @bio: The &struct bio which describes the I/O as destination 1153 * @iter: iov_iter as source 1154 * 1155 * Copy all pages from iov_iter to bio. 1156 * Returns 0 on success, or error on failure. 1157 */ 1158 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1159 { 1160 struct bio_vec *bvec; 1161 struct bvec_iter_all iter_all; 1162 1163 bio_for_each_segment_all(bvec, bio, iter_all) { 1164 ssize_t ret; 1165 1166 ret = copy_page_from_iter(bvec->bv_page, 1167 bvec->bv_offset, 1168 bvec->bv_len, 1169 iter); 1170 1171 if (!iov_iter_count(iter)) 1172 break; 1173 1174 if (ret < bvec->bv_len) 1175 return -EFAULT; 1176 } 1177 1178 return 0; 1179 } 1180 1181 /** 1182 * bio_copy_to_iter - copy all pages from bio to iov_iter 1183 * @bio: The &struct bio which describes the I/O as source 1184 * @iter: iov_iter as destination 1185 * 1186 * Copy all pages from bio to iov_iter. 1187 * Returns 0 on success, or error on failure. 1188 */ 1189 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 1190 { 1191 struct bio_vec *bvec; 1192 struct bvec_iter_all iter_all; 1193 1194 bio_for_each_segment_all(bvec, bio, iter_all) { 1195 ssize_t ret; 1196 1197 ret = copy_page_to_iter(bvec->bv_page, 1198 bvec->bv_offset, 1199 bvec->bv_len, 1200 &iter); 1201 1202 if (!iov_iter_count(&iter)) 1203 break; 1204 1205 if (ret < bvec->bv_len) 1206 return -EFAULT; 1207 } 1208 1209 return 0; 1210 } 1211 1212 void bio_free_pages(struct bio *bio) 1213 { 1214 struct bio_vec *bvec; 1215 struct bvec_iter_all iter_all; 1216 1217 bio_for_each_segment_all(bvec, bio, iter_all) 1218 __free_page(bvec->bv_page); 1219 } 1220 EXPORT_SYMBOL(bio_free_pages); 1221 1222 /** 1223 * bio_uncopy_user - finish previously mapped bio 1224 * @bio: bio being terminated 1225 * 1226 * Free pages allocated from bio_copy_user_iov() and write back data 1227 * to user space in case of a read. 1228 */ 1229 int bio_uncopy_user(struct bio *bio) 1230 { 1231 struct bio_map_data *bmd = bio->bi_private; 1232 int ret = 0; 1233 1234 if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1235 /* 1236 * if we're in a workqueue, the request is orphaned, so 1237 * don't copy into a random user address space, just free 1238 * and return -EINTR so user space doesn't expect any data. 1239 */ 1240 if (!current->mm) 1241 ret = -EINTR; 1242 else if (bio_data_dir(bio) == READ) 1243 ret = bio_copy_to_iter(bio, bmd->iter); 1244 if (bmd->is_our_pages) 1245 bio_free_pages(bio); 1246 } 1247 kfree(bmd); 1248 bio_put(bio); 1249 return ret; 1250 } 1251 1252 /** 1253 * bio_copy_user_iov - copy user data to bio 1254 * @q: destination block queue 1255 * @map_data: pointer to the rq_map_data holding pages (if necessary) 1256 * @iter: iovec iterator 1257 * @gfp_mask: memory allocation flags 1258 * 1259 * Prepares and returns a bio for indirect user io, bouncing data 1260 * to/from kernel pages as necessary. Must be paired with 1261 * call bio_uncopy_user() on io completion. 1262 */ 1263 struct bio *bio_copy_user_iov(struct request_queue *q, 1264 struct rq_map_data *map_data, 1265 struct iov_iter *iter, 1266 gfp_t gfp_mask) 1267 { 1268 struct bio_map_data *bmd; 1269 struct page *page; 1270 struct bio *bio; 1271 int i = 0, ret; 1272 int nr_pages; 1273 unsigned int len = iter->count; 1274 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1275 1276 bmd = bio_alloc_map_data(iter, gfp_mask); 1277 if (!bmd) 1278 return ERR_PTR(-ENOMEM); 1279 1280 /* 1281 * We need to do a deep copy of the iov_iter including the iovecs. 1282 * The caller provided iov might point to an on-stack or otherwise 1283 * shortlived one. 1284 */ 1285 bmd->is_our_pages = map_data ? 0 : 1; 1286 1287 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1288 if (nr_pages > BIO_MAX_PAGES) 1289 nr_pages = BIO_MAX_PAGES; 1290 1291 ret = -ENOMEM; 1292 bio = bio_kmalloc(gfp_mask, nr_pages); 1293 if (!bio) 1294 goto out_bmd; 1295 1296 ret = 0; 1297 1298 if (map_data) { 1299 nr_pages = 1 << map_data->page_order; 1300 i = map_data->offset / PAGE_SIZE; 1301 } 1302 while (len) { 1303 unsigned int bytes = PAGE_SIZE; 1304 1305 bytes -= offset; 1306 1307 if (bytes > len) 1308 bytes = len; 1309 1310 if (map_data) { 1311 if (i == map_data->nr_entries * nr_pages) { 1312 ret = -ENOMEM; 1313 break; 1314 } 1315 1316 page = map_data->pages[i / nr_pages]; 1317 page += (i % nr_pages); 1318 1319 i++; 1320 } else { 1321 page = alloc_page(q->bounce_gfp | gfp_mask); 1322 if (!page) { 1323 ret = -ENOMEM; 1324 break; 1325 } 1326 } 1327 1328 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { 1329 if (!map_data) 1330 __free_page(page); 1331 break; 1332 } 1333 1334 len -= bytes; 1335 offset = 0; 1336 } 1337 1338 if (ret) 1339 goto cleanup; 1340 1341 if (map_data) 1342 map_data->offset += bio->bi_iter.bi_size; 1343 1344 /* 1345 * success 1346 */ 1347 if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1348 (map_data && map_data->from_user)) { 1349 ret = bio_copy_from_iter(bio, iter); 1350 if (ret) 1351 goto cleanup; 1352 } else { 1353 if (bmd->is_our_pages) 1354 zero_fill_bio(bio); 1355 iov_iter_advance(iter, bio->bi_iter.bi_size); 1356 } 1357 1358 bio->bi_private = bmd; 1359 if (map_data && map_data->null_mapped) 1360 bio_set_flag(bio, BIO_NULL_MAPPED); 1361 return bio; 1362 cleanup: 1363 if (!map_data) 1364 bio_free_pages(bio); 1365 bio_put(bio); 1366 out_bmd: 1367 kfree(bmd); 1368 return ERR_PTR(ret); 1369 } 1370 1371 /** 1372 * bio_map_user_iov - map user iovec into bio 1373 * @q: the struct request_queue for the bio 1374 * @iter: iovec iterator 1375 * @gfp_mask: memory allocation flags 1376 * 1377 * Map the user space address into a bio suitable for io to a block 1378 * device. Returns an error pointer in case of error. 1379 */ 1380 struct bio *bio_map_user_iov(struct request_queue *q, 1381 struct iov_iter *iter, 1382 gfp_t gfp_mask) 1383 { 1384 int j; 1385 struct bio *bio; 1386 int ret; 1387 1388 if (!iov_iter_count(iter)) 1389 return ERR_PTR(-EINVAL); 1390 1391 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1392 if (!bio) 1393 return ERR_PTR(-ENOMEM); 1394 1395 while (iov_iter_count(iter)) { 1396 struct page **pages; 1397 ssize_t bytes; 1398 size_t offs, added = 0; 1399 int npages; 1400 1401 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1402 if (unlikely(bytes <= 0)) { 1403 ret = bytes ? bytes : -EFAULT; 1404 goto out_unmap; 1405 } 1406 1407 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1408 1409 if (unlikely(offs & queue_dma_alignment(q))) { 1410 ret = -EINVAL; 1411 j = 0; 1412 } else { 1413 for (j = 0; j < npages; j++) { 1414 struct page *page = pages[j]; 1415 unsigned int n = PAGE_SIZE - offs; 1416 bool same_page = false; 1417 1418 if (n > bytes) 1419 n = bytes; 1420 1421 if (!__bio_add_pc_page(q, bio, page, n, offs, 1422 &same_page)) { 1423 if (same_page) 1424 put_page(page); 1425 break; 1426 } 1427 1428 added += n; 1429 bytes -= n; 1430 offs = 0; 1431 } 1432 iov_iter_advance(iter, added); 1433 } 1434 /* 1435 * release the pages we didn't map into the bio, if any 1436 */ 1437 while (j < npages) 1438 put_page(pages[j++]); 1439 kvfree(pages); 1440 /* couldn't stuff something into bio? */ 1441 if (bytes) 1442 break; 1443 } 1444 1445 bio_set_flag(bio, BIO_USER_MAPPED); 1446 1447 /* 1448 * subtle -- if bio_map_user_iov() ended up bouncing a bio, 1449 * it would normally disappear when its bi_end_io is run. 1450 * however, we need it for the unmap, so grab an extra 1451 * reference to it 1452 */ 1453 bio_get(bio); 1454 return bio; 1455 1456 out_unmap: 1457 bio_release_pages(bio, false); 1458 bio_put(bio); 1459 return ERR_PTR(ret); 1460 } 1461 1462 /** 1463 * bio_unmap_user - unmap a bio 1464 * @bio: the bio being unmapped 1465 * 1466 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 1467 * process context. 1468 * 1469 * bio_unmap_user() may sleep. 1470 */ 1471 void bio_unmap_user(struct bio *bio) 1472 { 1473 bio_release_pages(bio, bio_data_dir(bio) == READ); 1474 bio_put(bio); 1475 bio_put(bio); 1476 } 1477 1478 static void bio_invalidate_vmalloc_pages(struct bio *bio) 1479 { 1480 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE 1481 if (bio->bi_private && !op_is_write(bio_op(bio))) { 1482 unsigned long i, len = 0; 1483 1484 for (i = 0; i < bio->bi_vcnt; i++) 1485 len += bio->bi_io_vec[i].bv_len; 1486 invalidate_kernel_vmap_range(bio->bi_private, len); 1487 } 1488 #endif 1489 } 1490 1491 static void bio_map_kern_endio(struct bio *bio) 1492 { 1493 bio_invalidate_vmalloc_pages(bio); 1494 bio_put(bio); 1495 } 1496 1497 /** 1498 * bio_map_kern - map kernel address into bio 1499 * @q: the struct request_queue for the bio 1500 * @data: pointer to buffer to map 1501 * @len: length in bytes 1502 * @gfp_mask: allocation flags for bio allocation 1503 * 1504 * Map the kernel address into a bio suitable for io to a block 1505 * device. Returns an error pointer in case of error. 1506 */ 1507 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 1508 gfp_t gfp_mask) 1509 { 1510 unsigned long kaddr = (unsigned long)data; 1511 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1512 unsigned long start = kaddr >> PAGE_SHIFT; 1513 const int nr_pages = end - start; 1514 bool is_vmalloc = is_vmalloc_addr(data); 1515 struct page *page; 1516 int offset, i; 1517 struct bio *bio; 1518 1519 bio = bio_kmalloc(gfp_mask, nr_pages); 1520 if (!bio) 1521 return ERR_PTR(-ENOMEM); 1522 1523 if (is_vmalloc) { 1524 flush_kernel_vmap_range(data, len); 1525 bio->bi_private = data; 1526 } 1527 1528 offset = offset_in_page(kaddr); 1529 for (i = 0; i < nr_pages; i++) { 1530 unsigned int bytes = PAGE_SIZE - offset; 1531 1532 if (len <= 0) 1533 break; 1534 1535 if (bytes > len) 1536 bytes = len; 1537 1538 if (!is_vmalloc) 1539 page = virt_to_page(data); 1540 else 1541 page = vmalloc_to_page(data); 1542 if (bio_add_pc_page(q, bio, page, bytes, 1543 offset) < bytes) { 1544 /* we don't support partial mappings */ 1545 bio_put(bio); 1546 return ERR_PTR(-EINVAL); 1547 } 1548 1549 data += bytes; 1550 len -= bytes; 1551 offset = 0; 1552 } 1553 1554 bio->bi_end_io = bio_map_kern_endio; 1555 return bio; 1556 } 1557 1558 static void bio_copy_kern_endio(struct bio *bio) 1559 { 1560 bio_free_pages(bio); 1561 bio_put(bio); 1562 } 1563 1564 static void bio_copy_kern_endio_read(struct bio *bio) 1565 { 1566 char *p = bio->bi_private; 1567 struct bio_vec *bvec; 1568 struct bvec_iter_all iter_all; 1569 1570 bio_for_each_segment_all(bvec, bio, iter_all) { 1571 memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1572 p += bvec->bv_len; 1573 } 1574 1575 bio_copy_kern_endio(bio); 1576 } 1577 1578 /** 1579 * bio_copy_kern - copy kernel address into bio 1580 * @q: the struct request_queue for the bio 1581 * @data: pointer to buffer to copy 1582 * @len: length in bytes 1583 * @gfp_mask: allocation flags for bio and page allocation 1584 * @reading: data direction is READ 1585 * 1586 * copy the kernel address into a bio suitable for io to a block 1587 * device. Returns an error pointer in case of error. 1588 */ 1589 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1590 gfp_t gfp_mask, int reading) 1591 { 1592 unsigned long kaddr = (unsigned long)data; 1593 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1594 unsigned long start = kaddr >> PAGE_SHIFT; 1595 struct bio *bio; 1596 void *p = data; 1597 int nr_pages = 0; 1598 1599 /* 1600 * Overflow, abort 1601 */ 1602 if (end < start) 1603 return ERR_PTR(-EINVAL); 1604 1605 nr_pages = end - start; 1606 bio = bio_kmalloc(gfp_mask, nr_pages); 1607 if (!bio) 1608 return ERR_PTR(-ENOMEM); 1609 1610 while (len) { 1611 struct page *page; 1612 unsigned int bytes = PAGE_SIZE; 1613 1614 if (bytes > len) 1615 bytes = len; 1616 1617 page = alloc_page(q->bounce_gfp | gfp_mask); 1618 if (!page) 1619 goto cleanup; 1620 1621 if (!reading) 1622 memcpy(page_address(page), p, bytes); 1623 1624 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 1625 break; 1626 1627 len -= bytes; 1628 p += bytes; 1629 } 1630 1631 if (reading) { 1632 bio->bi_end_io = bio_copy_kern_endio_read; 1633 bio->bi_private = data; 1634 } else { 1635 bio->bi_end_io = bio_copy_kern_endio; 1636 } 1637 1638 return bio; 1639 1640 cleanup: 1641 bio_free_pages(bio); 1642 bio_put(bio); 1643 return ERR_PTR(-ENOMEM); 1644 } 1645 1646 /* 1647 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1648 * for performing direct-IO in BIOs. 1649 * 1650 * The problem is that we cannot run set_page_dirty() from interrupt context 1651 * because the required locks are not interrupt-safe. So what we can do is to 1652 * mark the pages dirty _before_ performing IO. And in interrupt context, 1653 * check that the pages are still dirty. If so, fine. If not, redirty them 1654 * in process context. 1655 * 1656 * We special-case compound pages here: normally this means reads into hugetlb 1657 * pages. The logic in here doesn't really work right for compound pages 1658 * because the VM does not uniformly chase down the head page in all cases. 1659 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1660 * handle them at all. So we skip compound pages here at an early stage. 1661 * 1662 * Note that this code is very hard to test under normal circumstances because 1663 * direct-io pins the pages with get_user_pages(). This makes 1664 * is_page_cache_freeable return false, and the VM will not clean the pages. 1665 * But other code (eg, flusher threads) could clean the pages if they are mapped 1666 * pagecache. 1667 * 1668 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1669 * deferred bio dirtying paths. 1670 */ 1671 1672 /* 1673 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1674 */ 1675 void bio_set_pages_dirty(struct bio *bio) 1676 { 1677 struct bio_vec *bvec; 1678 struct bvec_iter_all iter_all; 1679 1680 bio_for_each_segment_all(bvec, bio, iter_all) { 1681 if (!PageCompound(bvec->bv_page)) 1682 set_page_dirty_lock(bvec->bv_page); 1683 } 1684 } 1685 1686 /* 1687 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1688 * If they are, then fine. If, however, some pages are clean then they must 1689 * have been written out during the direct-IO read. So we take another ref on 1690 * the BIO and re-dirty the pages in process context. 1691 * 1692 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1693 * here on. It will run one put_page() against each page and will run one 1694 * bio_put() against the BIO. 1695 */ 1696 1697 static void bio_dirty_fn(struct work_struct *work); 1698 1699 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1700 static DEFINE_SPINLOCK(bio_dirty_lock); 1701 static struct bio *bio_dirty_list; 1702 1703 /* 1704 * This runs in process context 1705 */ 1706 static void bio_dirty_fn(struct work_struct *work) 1707 { 1708 struct bio *bio, *next; 1709 1710 spin_lock_irq(&bio_dirty_lock); 1711 next = bio_dirty_list; 1712 bio_dirty_list = NULL; 1713 spin_unlock_irq(&bio_dirty_lock); 1714 1715 while ((bio = next) != NULL) { 1716 next = bio->bi_private; 1717 1718 bio_release_pages(bio, true); 1719 bio_put(bio); 1720 } 1721 } 1722 1723 void bio_check_pages_dirty(struct bio *bio) 1724 { 1725 struct bio_vec *bvec; 1726 unsigned long flags; 1727 struct bvec_iter_all iter_all; 1728 1729 bio_for_each_segment_all(bvec, bio, iter_all) { 1730 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 1731 goto defer; 1732 } 1733 1734 bio_release_pages(bio, false); 1735 bio_put(bio); 1736 return; 1737 defer: 1738 spin_lock_irqsave(&bio_dirty_lock, flags); 1739 bio->bi_private = bio_dirty_list; 1740 bio_dirty_list = bio; 1741 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1742 schedule_work(&bio_dirty_work); 1743 } 1744 1745 void update_io_ticks(struct hd_struct *part, unsigned long now) 1746 { 1747 unsigned long stamp; 1748 again: 1749 stamp = READ_ONCE(part->stamp); 1750 if (unlikely(stamp != now)) { 1751 if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 1752 __part_stat_add(part, io_ticks, 1); 1753 } 1754 } 1755 if (part->partno) { 1756 part = &part_to_disk(part)->part0; 1757 goto again; 1758 } 1759 } 1760 1761 void generic_start_io_acct(struct request_queue *q, int op, 1762 unsigned long sectors, struct hd_struct *part) 1763 { 1764 const int sgrp = op_stat_group(op); 1765 1766 part_stat_lock(); 1767 1768 update_io_ticks(part, jiffies); 1769 part_stat_inc(part, ios[sgrp]); 1770 part_stat_add(part, sectors[sgrp], sectors); 1771 part_inc_in_flight(q, part, op_is_write(op)); 1772 1773 part_stat_unlock(); 1774 } 1775 EXPORT_SYMBOL(generic_start_io_acct); 1776 1777 void generic_end_io_acct(struct request_queue *q, int req_op, 1778 struct hd_struct *part, unsigned long start_time) 1779 { 1780 unsigned long now = jiffies; 1781 unsigned long duration = now - start_time; 1782 const int sgrp = op_stat_group(req_op); 1783 1784 part_stat_lock(); 1785 1786 update_io_ticks(part, now); 1787 part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 1788 part_stat_add(part, time_in_queue, duration); 1789 part_dec_in_flight(q, part, op_is_write(req_op)); 1790 1791 part_stat_unlock(); 1792 } 1793 EXPORT_SYMBOL(generic_end_io_acct); 1794 1795 static inline bool bio_remaining_done(struct bio *bio) 1796 { 1797 /* 1798 * If we're not chaining, then ->__bi_remaining is always 1 and 1799 * we always end io on the first invocation. 1800 */ 1801 if (!bio_flagged(bio, BIO_CHAIN)) 1802 return true; 1803 1804 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1805 1806 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1807 bio_clear_flag(bio, BIO_CHAIN); 1808 return true; 1809 } 1810 1811 return false; 1812 } 1813 1814 /** 1815 * bio_endio - end I/O on a bio 1816 * @bio: bio 1817 * 1818 * Description: 1819 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1820 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1821 * bio unless they own it and thus know that it has an end_io function. 1822 * 1823 * bio_endio() can be called several times on a bio that has been chained 1824 * using bio_chain(). The ->bi_end_io() function will only be called the 1825 * last time. At this point the BLK_TA_COMPLETE tracing event will be 1826 * generated if BIO_TRACE_COMPLETION is set. 1827 **/ 1828 void bio_endio(struct bio *bio) 1829 { 1830 again: 1831 if (!bio_remaining_done(bio)) 1832 return; 1833 if (!bio_integrity_endio(bio)) 1834 return; 1835 1836 if (bio->bi_disk) 1837 rq_qos_done_bio(bio->bi_disk->queue, bio); 1838 1839 /* 1840 * Need to have a real endio function for chained bios, otherwise 1841 * various corner cases will break (like stacking block devices that 1842 * save/restore bi_end_io) - however, we want to avoid unbounded 1843 * recursion and blowing the stack. Tail call optimization would 1844 * handle this, but compiling with frame pointers also disables 1845 * gcc's sibling call optimization. 1846 */ 1847 if (bio->bi_end_io == bio_chain_endio) { 1848 bio = __bio_chain_endio(bio); 1849 goto again; 1850 } 1851 1852 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1853 trace_block_bio_complete(bio->bi_disk->queue, bio, 1854 blk_status_to_errno(bio->bi_status)); 1855 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1856 } 1857 1858 blk_throtl_bio_endio(bio); 1859 /* release cgroup info */ 1860 bio_uninit(bio); 1861 if (bio->bi_end_io) 1862 bio->bi_end_io(bio); 1863 } 1864 EXPORT_SYMBOL(bio_endio); 1865 1866 /** 1867 * bio_split - split a bio 1868 * @bio: bio to split 1869 * @sectors: number of sectors to split from the front of @bio 1870 * @gfp: gfp mask 1871 * @bs: bio set to allocate from 1872 * 1873 * Allocates and returns a new bio which represents @sectors from the start of 1874 * @bio, and updates @bio to represent the remaining sectors. 1875 * 1876 * Unless this is a discard request the newly allocated bio will point 1877 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1878 * neither @bio nor @bs are freed before the split bio. 1879 */ 1880 struct bio *bio_split(struct bio *bio, int sectors, 1881 gfp_t gfp, struct bio_set *bs) 1882 { 1883 struct bio *split; 1884 1885 BUG_ON(sectors <= 0); 1886 BUG_ON(sectors >= bio_sectors(bio)); 1887 1888 split = bio_clone_fast(bio, gfp, bs); 1889 if (!split) 1890 return NULL; 1891 1892 split->bi_iter.bi_size = sectors << 9; 1893 1894 if (bio_integrity(split)) 1895 bio_integrity_trim(split); 1896 1897 bio_advance(bio, split->bi_iter.bi_size); 1898 1899 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1900 bio_set_flag(split, BIO_TRACE_COMPLETION); 1901 1902 return split; 1903 } 1904 EXPORT_SYMBOL(bio_split); 1905 1906 /** 1907 * bio_trim - trim a bio 1908 * @bio: bio to trim 1909 * @offset: number of sectors to trim from the front of @bio 1910 * @size: size we want to trim @bio to, in sectors 1911 */ 1912 void bio_trim(struct bio *bio, int offset, int size) 1913 { 1914 /* 'bio' is a cloned bio which we need to trim to match 1915 * the given offset and size. 1916 */ 1917 1918 size <<= 9; 1919 if (offset == 0 && size == bio->bi_iter.bi_size) 1920 return; 1921 1922 bio_advance(bio, offset << 9); 1923 bio->bi_iter.bi_size = size; 1924 1925 if (bio_integrity(bio)) 1926 bio_integrity_trim(bio); 1927 1928 } 1929 EXPORT_SYMBOL_GPL(bio_trim); 1930 1931 /* 1932 * create memory pools for biovec's in a bio_set. 1933 * use the global biovec slabs created for general use. 1934 */ 1935 int biovec_init_pool(mempool_t *pool, int pool_entries) 1936 { 1937 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1938 1939 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1940 } 1941 1942 /* 1943 * bioset_exit - exit a bioset initialized with bioset_init() 1944 * 1945 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1946 * kzalloc()). 1947 */ 1948 void bioset_exit(struct bio_set *bs) 1949 { 1950 if (bs->rescue_workqueue) 1951 destroy_workqueue(bs->rescue_workqueue); 1952 bs->rescue_workqueue = NULL; 1953 1954 mempool_exit(&bs->bio_pool); 1955 mempool_exit(&bs->bvec_pool); 1956 1957 bioset_integrity_free(bs); 1958 if (bs->bio_slab) 1959 bio_put_slab(bs); 1960 bs->bio_slab = NULL; 1961 } 1962 EXPORT_SYMBOL(bioset_exit); 1963 1964 /** 1965 * bioset_init - Initialize a bio_set 1966 * @bs: pool to initialize 1967 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1968 * @front_pad: Number of bytes to allocate in front of the returned bio 1969 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1970 * and %BIOSET_NEED_RESCUER 1971 * 1972 * Description: 1973 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1974 * to ask for a number of bytes to be allocated in front of the bio. 1975 * Front pad allocation is useful for embedding the bio inside 1976 * another structure, to avoid allocating extra data to go with the bio. 1977 * Note that the bio must be embedded at the END of that structure always, 1978 * or things will break badly. 1979 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1980 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1981 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1982 * dispatch queued requests when the mempool runs out of space. 1983 * 1984 */ 1985 int bioset_init(struct bio_set *bs, 1986 unsigned int pool_size, 1987 unsigned int front_pad, 1988 int flags) 1989 { 1990 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1991 1992 bs->front_pad = front_pad; 1993 1994 spin_lock_init(&bs->rescue_lock); 1995 bio_list_init(&bs->rescue_list); 1996 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1997 1998 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1999 if (!bs->bio_slab) 2000 return -ENOMEM; 2001 2002 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2003 goto bad; 2004 2005 if ((flags & BIOSET_NEED_BVECS) && 2006 biovec_init_pool(&bs->bvec_pool, pool_size)) 2007 goto bad; 2008 2009 if (!(flags & BIOSET_NEED_RESCUER)) 2010 return 0; 2011 2012 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2013 if (!bs->rescue_workqueue) 2014 goto bad; 2015 2016 return 0; 2017 bad: 2018 bioset_exit(bs); 2019 return -ENOMEM; 2020 } 2021 EXPORT_SYMBOL(bioset_init); 2022 2023 /* 2024 * Initialize and setup a new bio_set, based on the settings from 2025 * another bio_set. 2026 */ 2027 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 2028 { 2029 int flags; 2030 2031 flags = 0; 2032 if (src->bvec_pool.min_nr) 2033 flags |= BIOSET_NEED_BVECS; 2034 if (src->rescue_workqueue) 2035 flags |= BIOSET_NEED_RESCUER; 2036 2037 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 2038 } 2039 EXPORT_SYMBOL(bioset_init_from_src); 2040 2041 #ifdef CONFIG_BLK_CGROUP 2042 2043 /** 2044 * bio_disassociate_blkg - puts back the blkg reference if associated 2045 * @bio: target bio 2046 * 2047 * Helper to disassociate the blkg from @bio if a blkg is associated. 2048 */ 2049 void bio_disassociate_blkg(struct bio *bio) 2050 { 2051 if (bio->bi_blkg) { 2052 blkg_put(bio->bi_blkg); 2053 bio->bi_blkg = NULL; 2054 } 2055 } 2056 EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2057 2058 /** 2059 * __bio_associate_blkg - associate a bio with the a blkg 2060 * @bio: target bio 2061 * @blkg: the blkg to associate 2062 * 2063 * This tries to associate @bio with the specified @blkg. Association failure 2064 * is handled by walking up the blkg tree. Therefore, the blkg associated can 2065 * be anything between @blkg and the root_blkg. This situation only happens 2066 * when a cgroup is dying and then the remaining bios will spill to the closest 2067 * alive blkg. 2068 * 2069 * A reference will be taken on the @blkg and will be released when @bio is 2070 * freed. 2071 */ 2072 static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2073 { 2074 bio_disassociate_blkg(bio); 2075 2076 bio->bi_blkg = blkg_tryget_closest(blkg); 2077 } 2078 2079 /** 2080 * bio_associate_blkg_from_css - associate a bio with a specified css 2081 * @bio: target bio 2082 * @css: target css 2083 * 2084 * Associate @bio with the blkg found by combining the css's blkg and the 2085 * request_queue of the @bio. This falls back to the queue's root_blkg if 2086 * the association fails with the css. 2087 */ 2088 void bio_associate_blkg_from_css(struct bio *bio, 2089 struct cgroup_subsys_state *css) 2090 { 2091 struct request_queue *q = bio->bi_disk->queue; 2092 struct blkcg_gq *blkg; 2093 2094 rcu_read_lock(); 2095 2096 if (!css || !css->parent) 2097 blkg = q->root_blkg; 2098 else 2099 blkg = blkg_lookup_create(css_to_blkcg(css), q); 2100 2101 __bio_associate_blkg(bio, blkg); 2102 2103 rcu_read_unlock(); 2104 } 2105 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2106 2107 #ifdef CONFIG_MEMCG 2108 /** 2109 * bio_associate_blkg_from_page - associate a bio with the page's blkg 2110 * @bio: target bio 2111 * @page: the page to lookup the blkcg from 2112 * 2113 * Associate @bio with the blkg from @page's owning memcg and the respective 2114 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2115 * root_blkg. 2116 */ 2117 void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 2118 { 2119 struct cgroup_subsys_state *css; 2120 2121 if (!page->mem_cgroup) 2122 return; 2123 2124 rcu_read_lock(); 2125 2126 css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2127 bio_associate_blkg_from_css(bio, css); 2128 2129 rcu_read_unlock(); 2130 } 2131 #endif /* CONFIG_MEMCG */ 2132 2133 /** 2134 * bio_associate_blkg - associate a bio with a blkg 2135 * @bio: target bio 2136 * 2137 * Associate @bio with the blkg found from the bio's css and request_queue. 2138 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 2139 * already associated, the css is reused and association redone as the 2140 * request_queue may have changed. 2141 */ 2142 void bio_associate_blkg(struct bio *bio) 2143 { 2144 struct cgroup_subsys_state *css; 2145 2146 rcu_read_lock(); 2147 2148 if (bio->bi_blkg) 2149 css = &bio_blkcg(bio)->css; 2150 else 2151 css = blkcg_css(); 2152 2153 bio_associate_blkg_from_css(bio, css); 2154 2155 rcu_read_unlock(); 2156 } 2157 EXPORT_SYMBOL_GPL(bio_associate_blkg); 2158 2159 /** 2160 * bio_clone_blkg_association - clone blkg association from src to dst bio 2161 * @dst: destination bio 2162 * @src: source bio 2163 */ 2164 void bio_clone_blkg_association(struct bio *dst, struct bio *src) 2165 { 2166 rcu_read_lock(); 2167 2168 if (src->bi_blkg) 2169 __bio_associate_blkg(dst, src->bi_blkg); 2170 2171 rcu_read_unlock(); 2172 } 2173 EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2174 #endif /* CONFIG_BLK_CGROUP */ 2175 2176 static void __init biovec_init_slabs(void) 2177 { 2178 int i; 2179 2180 for (i = 0; i < BVEC_POOL_NR; i++) { 2181 int size; 2182 struct biovec_slab *bvs = bvec_slabs + i; 2183 2184 if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2185 bvs->slab = NULL; 2186 continue; 2187 } 2188 2189 size = bvs->nr_vecs * sizeof(struct bio_vec); 2190 bvs->slab = kmem_cache_create(bvs->name, size, 0, 2191 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2192 } 2193 } 2194 2195 static int __init init_bio(void) 2196 { 2197 bio_slab_max = 2; 2198 bio_slab_nr = 0; 2199 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 2200 GFP_KERNEL); 2201 2202 BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 2203 2204 if (!bio_slabs) 2205 panic("bio: can't allocate bios\n"); 2206 2207 bio_integrity_init(); 2208 biovec_init_slabs(); 2209 2210 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2211 panic("bio: can't allocate bios\n"); 2212 2213 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2214 panic("bio: can't create integrity pool\n"); 2215 2216 return 0; 2217 } 2218 subsys_initcall(init_bio); 2219