1 /* 2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License version 2 as 6 * published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 11 * GNU General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public Licens 14 * along with this program; if not, write to the Free Software 15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- 16 * 17 */ 18 #include <linux/mm.h> 19 #include <linux/swap.h> 20 #include <linux/bio.h> 21 #include <linux/blkdev.h> 22 #include <linux/uio.h> 23 #include <linux/iocontext.h> 24 #include <linux/slab.h> 25 #include <linux/init.h> 26 #include <linux/kernel.h> 27 #include <linux/export.h> 28 #include <linux/mempool.h> 29 #include <linux/workqueue.h> 30 #include <linux/cgroup.h> 31 32 #include <trace/events/block.h> 33 #include "blk.h" 34 35 /* 36 * Test patch to inline a certain number of bi_io_vec's inside the bio 37 * itself, to shrink a bio data allocation from two mempool calls to one 38 */ 39 #define BIO_INLINE_VECS 4 40 41 /* 42 * if you change this list, also change bvec_alloc or things will 43 * break badly! cannot be bigger than what you can fit into an 44 * unsigned short 45 */ 46 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 47 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 48 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 49 }; 50 #undef BV 51 52 /* 53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by 54 * IO code that does not need private memory pools. 55 */ 56 struct bio_set fs_bio_set; 57 EXPORT_SYMBOL(fs_bio_set); 58 59 /* 60 * Our slab pool management 61 */ 62 struct bio_slab { 63 struct kmem_cache *slab; 64 unsigned int slab_ref; 65 unsigned int slab_size; 66 char name[8]; 67 }; 68 static DEFINE_MUTEX(bio_slab_lock); 69 static struct bio_slab *bio_slabs; 70 static unsigned int bio_slab_nr, bio_slab_max; 71 72 static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 73 { 74 unsigned int sz = sizeof(struct bio) + extra_size; 75 struct kmem_cache *slab = NULL; 76 struct bio_slab *bslab, *new_bio_slabs; 77 unsigned int new_bio_slab_max; 78 unsigned int i, entry = -1; 79 80 mutex_lock(&bio_slab_lock); 81 82 i = 0; 83 while (i < bio_slab_nr) { 84 bslab = &bio_slabs[i]; 85 86 if (!bslab->slab && entry == -1) 87 entry = i; 88 else if (bslab->slab_size == sz) { 89 slab = bslab->slab; 90 bslab->slab_ref++; 91 break; 92 } 93 i++; 94 } 95 96 if (slab) 97 goto out_unlock; 98 99 if (bio_slab_nr == bio_slab_max && entry == -1) { 100 new_bio_slab_max = bio_slab_max << 1; 101 new_bio_slabs = krealloc(bio_slabs, 102 new_bio_slab_max * sizeof(struct bio_slab), 103 GFP_KERNEL); 104 if (!new_bio_slabs) 105 goto out_unlock; 106 bio_slab_max = new_bio_slab_max; 107 bio_slabs = new_bio_slabs; 108 } 109 if (entry == -1) 110 entry = bio_slab_nr++; 111 112 bslab = &bio_slabs[entry]; 113 114 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 115 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 116 SLAB_HWCACHE_ALIGN, NULL); 117 if (!slab) 118 goto out_unlock; 119 120 bslab->slab = slab; 121 bslab->slab_ref = 1; 122 bslab->slab_size = sz; 123 out_unlock: 124 mutex_unlock(&bio_slab_lock); 125 return slab; 126 } 127 128 static void bio_put_slab(struct bio_set *bs) 129 { 130 struct bio_slab *bslab = NULL; 131 unsigned int i; 132 133 mutex_lock(&bio_slab_lock); 134 135 for (i = 0; i < bio_slab_nr; i++) { 136 if (bs->bio_slab == bio_slabs[i].slab) { 137 bslab = &bio_slabs[i]; 138 break; 139 } 140 } 141 142 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 143 goto out; 144 145 WARN_ON(!bslab->slab_ref); 146 147 if (--bslab->slab_ref) 148 goto out; 149 150 kmem_cache_destroy(bslab->slab); 151 bslab->slab = NULL; 152 153 out: 154 mutex_unlock(&bio_slab_lock); 155 } 156 157 unsigned int bvec_nr_vecs(unsigned short idx) 158 { 159 return bvec_slabs[idx].nr_vecs; 160 } 161 162 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 163 { 164 if (!idx) 165 return; 166 idx--; 167 168 BIO_BUG_ON(idx >= BVEC_POOL_NR); 169 170 if (idx == BVEC_POOL_MAX) { 171 mempool_free(bv, pool); 172 } else { 173 struct biovec_slab *bvs = bvec_slabs + idx; 174 175 kmem_cache_free(bvs->slab, bv); 176 } 177 } 178 179 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 180 mempool_t *pool) 181 { 182 struct bio_vec *bvl; 183 184 /* 185 * see comment near bvec_array define! 186 */ 187 switch (nr) { 188 case 1: 189 *idx = 0; 190 break; 191 case 2 ... 4: 192 *idx = 1; 193 break; 194 case 5 ... 16: 195 *idx = 2; 196 break; 197 case 17 ... 64: 198 *idx = 3; 199 break; 200 case 65 ... 128: 201 *idx = 4; 202 break; 203 case 129 ... BIO_MAX_PAGES: 204 *idx = 5; 205 break; 206 default: 207 return NULL; 208 } 209 210 /* 211 * idx now points to the pool we want to allocate from. only the 212 * 1-vec entry pool is mempool backed. 213 */ 214 if (*idx == BVEC_POOL_MAX) { 215 fallback: 216 bvl = mempool_alloc(pool, gfp_mask); 217 } else { 218 struct biovec_slab *bvs = bvec_slabs + *idx; 219 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 220 221 /* 222 * Make this allocation restricted and don't dump info on 223 * allocation failures, since we'll fallback to the mempool 224 * in case of failure. 225 */ 226 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 227 228 /* 229 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 230 * is set, retry with the 1-entry mempool 231 */ 232 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 233 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 234 *idx = BVEC_POOL_MAX; 235 goto fallback; 236 } 237 } 238 239 (*idx)++; 240 return bvl; 241 } 242 243 void bio_uninit(struct bio *bio) 244 { 245 bio_disassociate_task(bio); 246 } 247 EXPORT_SYMBOL(bio_uninit); 248 249 static void bio_free(struct bio *bio) 250 { 251 struct bio_set *bs = bio->bi_pool; 252 void *p; 253 254 bio_uninit(bio); 255 256 if (bs) { 257 bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 258 259 /* 260 * If we have front padding, adjust the bio pointer before freeing 261 */ 262 p = bio; 263 p -= bs->front_pad; 264 265 mempool_free(p, &bs->bio_pool); 266 } else { 267 /* Bio was allocated by bio_kmalloc() */ 268 kfree(bio); 269 } 270 } 271 272 /* 273 * Users of this function have their own bio allocation. Subsequently, 274 * they must remember to pair any call to bio_init() with bio_uninit() 275 * when IO has completed, or when the bio is released. 276 */ 277 void bio_init(struct bio *bio, struct bio_vec *table, 278 unsigned short max_vecs) 279 { 280 memset(bio, 0, sizeof(*bio)); 281 atomic_set(&bio->__bi_remaining, 1); 282 atomic_set(&bio->__bi_cnt, 1); 283 284 bio->bi_io_vec = table; 285 bio->bi_max_vecs = max_vecs; 286 } 287 EXPORT_SYMBOL(bio_init); 288 289 /** 290 * bio_reset - reinitialize a bio 291 * @bio: bio to reset 292 * 293 * Description: 294 * After calling bio_reset(), @bio will be in the same state as a freshly 295 * allocated bio returned bio bio_alloc_bioset() - the only fields that are 296 * preserved are the ones that are initialized by bio_alloc_bioset(). See 297 * comment in struct bio. 298 */ 299 void bio_reset(struct bio *bio) 300 { 301 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 302 303 bio_uninit(bio); 304 305 memset(bio, 0, BIO_RESET_BYTES); 306 bio->bi_flags = flags; 307 atomic_set(&bio->__bi_remaining, 1); 308 } 309 EXPORT_SYMBOL(bio_reset); 310 311 static struct bio *__bio_chain_endio(struct bio *bio) 312 { 313 struct bio *parent = bio->bi_private; 314 315 if (!parent->bi_status) 316 parent->bi_status = bio->bi_status; 317 bio_put(bio); 318 return parent; 319 } 320 321 static void bio_chain_endio(struct bio *bio) 322 { 323 bio_endio(__bio_chain_endio(bio)); 324 } 325 326 /** 327 * bio_chain - chain bio completions 328 * @bio: the target bio 329 * @parent: the @bio's parent bio 330 * 331 * The caller won't have a bi_end_io called when @bio completes - instead, 332 * @parent's bi_end_io won't be called until both @parent and @bio have 333 * completed; the chained bio will also be freed when it completes. 334 * 335 * The caller must not set bi_private or bi_end_io in @bio. 336 */ 337 void bio_chain(struct bio *bio, struct bio *parent) 338 { 339 BUG_ON(bio->bi_private || bio->bi_end_io); 340 341 bio->bi_private = parent; 342 bio->bi_end_io = bio_chain_endio; 343 bio_inc_remaining(parent); 344 } 345 EXPORT_SYMBOL(bio_chain); 346 347 static void bio_alloc_rescue(struct work_struct *work) 348 { 349 struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 350 struct bio *bio; 351 352 while (1) { 353 spin_lock(&bs->rescue_lock); 354 bio = bio_list_pop(&bs->rescue_list); 355 spin_unlock(&bs->rescue_lock); 356 357 if (!bio) 358 break; 359 360 generic_make_request(bio); 361 } 362 } 363 364 static void punt_bios_to_rescuer(struct bio_set *bs) 365 { 366 struct bio_list punt, nopunt; 367 struct bio *bio; 368 369 if (WARN_ON_ONCE(!bs->rescue_workqueue)) 370 return; 371 /* 372 * In order to guarantee forward progress we must punt only bios that 373 * were allocated from this bio_set; otherwise, if there was a bio on 374 * there for a stacking driver higher up in the stack, processing it 375 * could require allocating bios from this bio_set, and doing that from 376 * our own rescuer would be bad. 377 * 378 * Since bio lists are singly linked, pop them all instead of trying to 379 * remove from the middle of the list: 380 */ 381 382 bio_list_init(&punt); 383 bio_list_init(&nopunt); 384 385 while ((bio = bio_list_pop(¤t->bio_list[0]))) 386 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 387 current->bio_list[0] = nopunt; 388 389 bio_list_init(&nopunt); 390 while ((bio = bio_list_pop(¤t->bio_list[1]))) 391 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 392 current->bio_list[1] = nopunt; 393 394 spin_lock(&bs->rescue_lock); 395 bio_list_merge(&bs->rescue_list, &punt); 396 spin_unlock(&bs->rescue_lock); 397 398 queue_work(bs->rescue_workqueue, &bs->rescue_work); 399 } 400 401 /** 402 * bio_alloc_bioset - allocate a bio for I/O 403 * @gfp_mask: the GFP_* mask given to the slab allocator 404 * @nr_iovecs: number of iovecs to pre-allocate 405 * @bs: the bio_set to allocate from. 406 * 407 * Description: 408 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 409 * backed by the @bs's mempool. 410 * 411 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 412 * always be able to allocate a bio. This is due to the mempool guarantees. 413 * To make this work, callers must never allocate more than 1 bio at a time 414 * from this pool. Callers that need to allocate more than 1 bio must always 415 * submit the previously allocated bio for IO before attempting to allocate 416 * a new one. Failure to do so can cause deadlocks under memory pressure. 417 * 418 * Note that when running under generic_make_request() (i.e. any block 419 * driver), bios are not submitted until after you return - see the code in 420 * generic_make_request() that converts recursion into iteration, to prevent 421 * stack overflows. 422 * 423 * This would normally mean allocating multiple bios under 424 * generic_make_request() would be susceptible to deadlocks, but we have 425 * deadlock avoidance code that resubmits any blocked bios from a rescuer 426 * thread. 427 * 428 * However, we do not guarantee forward progress for allocations from other 429 * mempools. Doing multiple allocations from the same mempool under 430 * generic_make_request() should be avoided - instead, use bio_set's front_pad 431 * for per bio allocations. 432 * 433 * RETURNS: 434 * Pointer to new bio on success, NULL on failure. 435 */ 436 struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 437 struct bio_set *bs) 438 { 439 gfp_t saved_gfp = gfp_mask; 440 unsigned front_pad; 441 unsigned inline_vecs; 442 struct bio_vec *bvl = NULL; 443 struct bio *bio; 444 void *p; 445 446 if (!bs) { 447 if (nr_iovecs > UIO_MAXIOV) 448 return NULL; 449 450 p = kmalloc(sizeof(struct bio) + 451 nr_iovecs * sizeof(struct bio_vec), 452 gfp_mask); 453 front_pad = 0; 454 inline_vecs = nr_iovecs; 455 } else { 456 /* should not use nobvec bioset for nr_iovecs > 0 */ 457 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 458 nr_iovecs > 0)) 459 return NULL; 460 /* 461 * generic_make_request() converts recursion to iteration; this 462 * means if we're running beneath it, any bios we allocate and 463 * submit will not be submitted (and thus freed) until after we 464 * return. 465 * 466 * This exposes us to a potential deadlock if we allocate 467 * multiple bios from the same bio_set() while running 468 * underneath generic_make_request(). If we were to allocate 469 * multiple bios (say a stacking block driver that was splitting 470 * bios), we would deadlock if we exhausted the mempool's 471 * reserve. 472 * 473 * We solve this, and guarantee forward progress, with a rescuer 474 * workqueue per bio_set. If we go to allocate and there are 475 * bios on current->bio_list, we first try the allocation 476 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 477 * bios we would be blocking to the rescuer workqueue before 478 * we retry with the original gfp_flags. 479 */ 480 481 if (current->bio_list && 482 (!bio_list_empty(¤t->bio_list[0]) || 483 !bio_list_empty(¤t->bio_list[1])) && 484 bs->rescue_workqueue) 485 gfp_mask &= ~__GFP_DIRECT_RECLAIM; 486 487 p = mempool_alloc(&bs->bio_pool, gfp_mask); 488 if (!p && gfp_mask != saved_gfp) { 489 punt_bios_to_rescuer(bs); 490 gfp_mask = saved_gfp; 491 p = mempool_alloc(&bs->bio_pool, gfp_mask); 492 } 493 494 front_pad = bs->front_pad; 495 inline_vecs = BIO_INLINE_VECS; 496 } 497 498 if (unlikely(!p)) 499 return NULL; 500 501 bio = p + front_pad; 502 bio_init(bio, NULL, 0); 503 504 if (nr_iovecs > inline_vecs) { 505 unsigned long idx = 0; 506 507 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 508 if (!bvl && gfp_mask != saved_gfp) { 509 punt_bios_to_rescuer(bs); 510 gfp_mask = saved_gfp; 511 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 512 } 513 514 if (unlikely(!bvl)) 515 goto err_free; 516 517 bio->bi_flags |= idx << BVEC_POOL_OFFSET; 518 } else if (nr_iovecs) { 519 bvl = bio->bi_inline_vecs; 520 } 521 522 bio->bi_pool = bs; 523 bio->bi_max_vecs = nr_iovecs; 524 bio->bi_io_vec = bvl; 525 return bio; 526 527 err_free: 528 mempool_free(p, &bs->bio_pool); 529 return NULL; 530 } 531 EXPORT_SYMBOL(bio_alloc_bioset); 532 533 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 534 { 535 unsigned long flags; 536 struct bio_vec bv; 537 struct bvec_iter iter; 538 539 __bio_for_each_segment(bv, bio, iter, start) { 540 char *data = bvec_kmap_irq(&bv, &flags); 541 memset(data, 0, bv.bv_len); 542 flush_dcache_page(bv.bv_page); 543 bvec_kunmap_irq(data, &flags); 544 } 545 } 546 EXPORT_SYMBOL(zero_fill_bio_iter); 547 548 /** 549 * bio_put - release a reference to a bio 550 * @bio: bio to release reference to 551 * 552 * Description: 553 * Put a reference to a &struct bio, either one you have gotten with 554 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 555 **/ 556 void bio_put(struct bio *bio) 557 { 558 if (!bio_flagged(bio, BIO_REFFED)) 559 bio_free(bio); 560 else { 561 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 562 563 /* 564 * last put frees it 565 */ 566 if (atomic_dec_and_test(&bio->__bi_cnt)) 567 bio_free(bio); 568 } 569 } 570 EXPORT_SYMBOL(bio_put); 571 572 inline int bio_phys_segments(struct request_queue *q, struct bio *bio) 573 { 574 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 575 blk_recount_segments(q, bio); 576 577 return bio->bi_phys_segments; 578 } 579 EXPORT_SYMBOL(bio_phys_segments); 580 581 /** 582 * __bio_clone_fast - clone a bio that shares the original bio's biovec 583 * @bio: destination bio 584 * @bio_src: bio to clone 585 * 586 * Clone a &bio. Caller will own the returned bio, but not 587 * the actual data it points to. Reference count of returned 588 * bio will be one. 589 * 590 * Caller must ensure that @bio_src is not freed before @bio. 591 */ 592 void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 593 { 594 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 595 596 /* 597 * most users will be overriding ->bi_disk with a new target, 598 * so we don't set nor calculate new physical/hw segment counts here 599 */ 600 bio->bi_disk = bio_src->bi_disk; 601 bio->bi_partno = bio_src->bi_partno; 602 bio_set_flag(bio, BIO_CLONED); 603 if (bio_flagged(bio_src, BIO_THROTTLED)) 604 bio_set_flag(bio, BIO_THROTTLED); 605 bio->bi_opf = bio_src->bi_opf; 606 bio->bi_write_hint = bio_src->bi_write_hint; 607 bio->bi_iter = bio_src->bi_iter; 608 bio->bi_io_vec = bio_src->bi_io_vec; 609 610 bio_clone_blkcg_association(bio, bio_src); 611 } 612 EXPORT_SYMBOL(__bio_clone_fast); 613 614 /** 615 * bio_clone_fast - clone a bio that shares the original bio's biovec 616 * @bio: bio to clone 617 * @gfp_mask: allocation priority 618 * @bs: bio_set to allocate from 619 * 620 * Like __bio_clone_fast, only also allocates the returned bio 621 */ 622 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 623 { 624 struct bio *b; 625 626 b = bio_alloc_bioset(gfp_mask, 0, bs); 627 if (!b) 628 return NULL; 629 630 __bio_clone_fast(b, bio); 631 632 if (bio_integrity(bio)) { 633 int ret; 634 635 ret = bio_integrity_clone(b, bio, gfp_mask); 636 637 if (ret < 0) { 638 bio_put(b); 639 return NULL; 640 } 641 } 642 643 return b; 644 } 645 EXPORT_SYMBOL(bio_clone_fast); 646 647 /** 648 * bio_clone_bioset - clone a bio 649 * @bio_src: bio to clone 650 * @gfp_mask: allocation priority 651 * @bs: bio_set to allocate from 652 * 653 * Clone bio. Caller will own the returned bio, but not the actual data it 654 * points to. Reference count of returned bio will be one. 655 */ 656 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, 657 struct bio_set *bs) 658 { 659 struct bvec_iter iter; 660 struct bio_vec bv; 661 struct bio *bio; 662 663 /* 664 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from 665 * bio_src->bi_io_vec to bio->bi_io_vec. 666 * 667 * We can't do that anymore, because: 668 * 669 * - The point of cloning the biovec is to produce a bio with a biovec 670 * the caller can modify: bi_idx and bi_bvec_done should be 0. 671 * 672 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if 673 * we tried to clone the whole thing bio_alloc_bioset() would fail. 674 * But the clone should succeed as long as the number of biovecs we 675 * actually need to allocate is fewer than BIO_MAX_PAGES. 676 * 677 * - Lastly, bi_vcnt should not be looked at or relied upon by code 678 * that does not own the bio - reason being drivers don't use it for 679 * iterating over the biovec anymore, so expecting it to be kept up 680 * to date (i.e. for clones that share the parent biovec) is just 681 * asking for trouble and would force extra work on 682 * __bio_clone_fast() anyways. 683 */ 684 685 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs); 686 if (!bio) 687 return NULL; 688 bio->bi_disk = bio_src->bi_disk; 689 bio->bi_opf = bio_src->bi_opf; 690 bio->bi_write_hint = bio_src->bi_write_hint; 691 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; 692 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; 693 694 switch (bio_op(bio)) { 695 case REQ_OP_DISCARD: 696 case REQ_OP_SECURE_ERASE: 697 case REQ_OP_WRITE_ZEROES: 698 break; 699 case REQ_OP_WRITE_SAME: 700 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; 701 break; 702 default: 703 bio_for_each_segment(bv, bio_src, iter) 704 bio->bi_io_vec[bio->bi_vcnt++] = bv; 705 break; 706 } 707 708 if (bio_integrity(bio_src)) { 709 int ret; 710 711 ret = bio_integrity_clone(bio, bio_src, gfp_mask); 712 if (ret < 0) { 713 bio_put(bio); 714 return NULL; 715 } 716 } 717 718 bio_clone_blkcg_association(bio, bio_src); 719 720 return bio; 721 } 722 EXPORT_SYMBOL(bio_clone_bioset); 723 724 /** 725 * bio_add_pc_page - attempt to add page to 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 * 732 * Attempt to add a page to the bio_vec maplist. This can fail for a 733 * number of reasons, such as the bio being full or target block device 734 * limitations. The target block device must allow bio's up to PAGE_SIZE, 735 * so it is always possible to add a single page to an empty bio. 736 * 737 * This should only be used by REQ_PC bios. 738 */ 739 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page 740 *page, unsigned int len, unsigned int offset) 741 { 742 int retried_segments = 0; 743 struct bio_vec *bvec; 744 745 /* 746 * cloned bio must not modify vec list 747 */ 748 if (unlikely(bio_flagged(bio, BIO_CLONED))) 749 return 0; 750 751 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 752 return 0; 753 754 /* 755 * For filesystems with a blocksize smaller than the pagesize 756 * we will often be called with the same page as last time and 757 * a consecutive offset. Optimize this special case. 758 */ 759 if (bio->bi_vcnt > 0) { 760 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; 761 762 if (page == prev->bv_page && 763 offset == prev->bv_offset + prev->bv_len) { 764 prev->bv_len += len; 765 bio->bi_iter.bi_size += len; 766 goto done; 767 } 768 769 /* 770 * If the queue doesn't support SG gaps and adding this 771 * offset would create a gap, disallow it. 772 */ 773 if (bvec_gap_to_prev(q, prev, offset)) 774 return 0; 775 } 776 777 if (bio_full(bio)) 778 return 0; 779 780 /* 781 * setup the new entry, we might clear it again later if we 782 * cannot add the page 783 */ 784 bvec = &bio->bi_io_vec[bio->bi_vcnt]; 785 bvec->bv_page = page; 786 bvec->bv_len = len; 787 bvec->bv_offset = offset; 788 bio->bi_vcnt++; 789 bio->bi_phys_segments++; 790 bio->bi_iter.bi_size += len; 791 792 /* 793 * Perform a recount if the number of segments is greater 794 * than queue_max_segments(q). 795 */ 796 797 while (bio->bi_phys_segments > queue_max_segments(q)) { 798 799 if (retried_segments) 800 goto failed; 801 802 retried_segments = 1; 803 blk_recount_segments(q, bio); 804 } 805 806 /* If we may be able to merge these biovecs, force a recount */ 807 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) 808 bio_clear_flag(bio, BIO_SEG_VALID); 809 810 done: 811 return len; 812 813 failed: 814 bvec->bv_page = NULL; 815 bvec->bv_len = 0; 816 bvec->bv_offset = 0; 817 bio->bi_vcnt--; 818 bio->bi_iter.bi_size -= len; 819 blk_recount_segments(q, bio); 820 return 0; 821 } 822 EXPORT_SYMBOL(bio_add_pc_page); 823 824 /** 825 * __bio_try_merge_page - try appending data to an existing bvec. 826 * @bio: destination bio 827 * @page: page to add 828 * @len: length of the data to add 829 * @off: offset of the data in @page 830 * 831 * Try to add the data at @page + @off to the last bvec of @bio. This is a 832 * a useful optimisation for file systems with a block size smaller than the 833 * page size. 834 * 835 * Return %true on success or %false on failure. 836 */ 837 bool __bio_try_merge_page(struct bio *bio, struct page *page, 838 unsigned int len, unsigned int off) 839 { 840 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 841 return false; 842 843 if (bio->bi_vcnt > 0) { 844 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 845 846 if (page == bv->bv_page && off == bv->bv_offset + bv->bv_len) { 847 bv->bv_len += len; 848 bio->bi_iter.bi_size += len; 849 return true; 850 } 851 } 852 return false; 853 } 854 EXPORT_SYMBOL_GPL(__bio_try_merge_page); 855 856 /** 857 * __bio_add_page - add page to a bio in a new segment 858 * @bio: destination bio 859 * @page: page to add 860 * @len: length of the data to add 861 * @off: offset of the data in @page 862 * 863 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 864 * that @bio has space for another bvec. 865 */ 866 void __bio_add_page(struct bio *bio, struct page *page, 867 unsigned int len, unsigned int off) 868 { 869 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 870 871 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 872 WARN_ON_ONCE(bio_full(bio)); 873 874 bv->bv_page = page; 875 bv->bv_offset = off; 876 bv->bv_len = len; 877 878 bio->bi_iter.bi_size += len; 879 bio->bi_vcnt++; 880 } 881 EXPORT_SYMBOL_GPL(__bio_add_page); 882 883 /** 884 * bio_add_page - attempt to add page to bio 885 * @bio: destination bio 886 * @page: page to add 887 * @len: vec entry length 888 * @offset: vec entry offset 889 * 890 * Attempt to add a page to the bio_vec maplist. This will only fail 891 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 892 */ 893 int bio_add_page(struct bio *bio, struct page *page, 894 unsigned int len, unsigned int offset) 895 { 896 if (!__bio_try_merge_page(bio, page, len, offset)) { 897 if (bio_full(bio)) 898 return 0; 899 __bio_add_page(bio, page, len, offset); 900 } 901 return len; 902 } 903 EXPORT_SYMBOL(bio_add_page); 904 905 /** 906 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 907 * @bio: bio to add pages to 908 * @iter: iov iterator describing the region to be mapped 909 * 910 * Pins pages from *iter and appends them to @bio's bvec array. The 911 * pages will have to be released using put_page() when done. 912 * For multi-segment *iter, this function only adds pages from the 913 * the next non-empty segment of the iov iterator. 914 */ 915 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 916 { 917 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt, idx; 918 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 919 struct page **pages = (struct page **)bv; 920 size_t offset; 921 ssize_t size; 922 923 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 924 if (unlikely(size <= 0)) 925 return size ? size : -EFAULT; 926 idx = nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE; 927 928 /* 929 * Deep magic below: We need to walk the pinned pages backwards 930 * because we are abusing the space allocated for the bio_vecs 931 * for the page array. Because the bio_vecs are larger than the 932 * page pointers by definition this will always work. But it also 933 * means we can't use bio_add_page, so any changes to it's semantics 934 * need to be reflected here as well. 935 */ 936 bio->bi_iter.bi_size += size; 937 bio->bi_vcnt += nr_pages; 938 939 while (idx--) { 940 bv[idx].bv_page = pages[idx]; 941 bv[idx].bv_len = PAGE_SIZE; 942 bv[idx].bv_offset = 0; 943 } 944 945 bv[0].bv_offset += offset; 946 bv[0].bv_len -= offset; 947 bv[nr_pages - 1].bv_len -= nr_pages * PAGE_SIZE - offset - size; 948 949 iov_iter_advance(iter, size); 950 return 0; 951 } 952 953 /** 954 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 955 * @bio: bio to add pages to 956 * @iter: iov iterator describing the region to be mapped 957 * 958 * Pins pages from *iter and appends them to @bio's bvec array. The 959 * pages will have to be released using put_page() when done. 960 * The function tries, but does not guarantee, to pin as many pages as 961 * fit into the bio, or are requested in *iter, whatever is smaller. 962 * If MM encounters an error pinning the requested pages, it stops. 963 * Error is returned only if 0 pages could be pinned. 964 */ 965 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 966 { 967 unsigned short orig_vcnt = bio->bi_vcnt; 968 969 do { 970 int ret = __bio_iov_iter_get_pages(bio, iter); 971 972 if (unlikely(ret)) 973 return bio->bi_vcnt > orig_vcnt ? 0 : ret; 974 975 } while (iov_iter_count(iter) && !bio_full(bio)); 976 977 return 0; 978 } 979 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 980 981 static void submit_bio_wait_endio(struct bio *bio) 982 { 983 complete(bio->bi_private); 984 } 985 986 /** 987 * submit_bio_wait - submit a bio, and wait until it completes 988 * @bio: The &struct bio which describes the I/O 989 * 990 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 991 * bio_endio() on failure. 992 * 993 * WARNING: Unlike to how submit_bio() is usually used, this function does not 994 * result in bio reference to be consumed. The caller must drop the reference 995 * on his own. 996 */ 997 int submit_bio_wait(struct bio *bio) 998 { 999 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 1000 1001 bio->bi_private = &done; 1002 bio->bi_end_io = submit_bio_wait_endio; 1003 bio->bi_opf |= REQ_SYNC; 1004 submit_bio(bio); 1005 wait_for_completion_io(&done); 1006 1007 return blk_status_to_errno(bio->bi_status); 1008 } 1009 EXPORT_SYMBOL(submit_bio_wait); 1010 1011 /** 1012 * bio_advance - increment/complete a bio by some number of bytes 1013 * @bio: bio to advance 1014 * @bytes: number of bytes to complete 1015 * 1016 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1017 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1018 * be updated on the last bvec as well. 1019 * 1020 * @bio will then represent the remaining, uncompleted portion of the io. 1021 */ 1022 void bio_advance(struct bio *bio, unsigned bytes) 1023 { 1024 if (bio_integrity(bio)) 1025 bio_integrity_advance(bio, bytes); 1026 1027 bio_advance_iter(bio, &bio->bi_iter, bytes); 1028 } 1029 EXPORT_SYMBOL(bio_advance); 1030 1031 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1032 struct bio *src, struct bvec_iter *src_iter) 1033 { 1034 struct bio_vec src_bv, dst_bv; 1035 void *src_p, *dst_p; 1036 unsigned bytes; 1037 1038 while (src_iter->bi_size && dst_iter->bi_size) { 1039 src_bv = bio_iter_iovec(src, *src_iter); 1040 dst_bv = bio_iter_iovec(dst, *dst_iter); 1041 1042 bytes = min(src_bv.bv_len, dst_bv.bv_len); 1043 1044 src_p = kmap_atomic(src_bv.bv_page); 1045 dst_p = kmap_atomic(dst_bv.bv_page); 1046 1047 memcpy(dst_p + dst_bv.bv_offset, 1048 src_p + src_bv.bv_offset, 1049 bytes); 1050 1051 kunmap_atomic(dst_p); 1052 kunmap_atomic(src_p); 1053 1054 flush_dcache_page(dst_bv.bv_page); 1055 1056 bio_advance_iter(src, src_iter, bytes); 1057 bio_advance_iter(dst, dst_iter, bytes); 1058 } 1059 } 1060 EXPORT_SYMBOL(bio_copy_data_iter); 1061 1062 /** 1063 * bio_copy_data - copy contents of data buffers from one bio to another 1064 * @src: source bio 1065 * @dst: destination bio 1066 * 1067 * Stops when it reaches the end of either @src or @dst - that is, copies 1068 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 1069 */ 1070 void bio_copy_data(struct bio *dst, struct bio *src) 1071 { 1072 struct bvec_iter src_iter = src->bi_iter; 1073 struct bvec_iter dst_iter = dst->bi_iter; 1074 1075 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1076 } 1077 EXPORT_SYMBOL(bio_copy_data); 1078 1079 /** 1080 * bio_list_copy_data - copy contents of data buffers from one chain of bios to 1081 * another 1082 * @src: source bio list 1083 * @dst: destination bio list 1084 * 1085 * Stops when it reaches the end of either the @src list or @dst list - that is, 1086 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 1087 * bios). 1088 */ 1089 void bio_list_copy_data(struct bio *dst, struct bio *src) 1090 { 1091 struct bvec_iter src_iter = src->bi_iter; 1092 struct bvec_iter dst_iter = dst->bi_iter; 1093 1094 while (1) { 1095 if (!src_iter.bi_size) { 1096 src = src->bi_next; 1097 if (!src) 1098 break; 1099 1100 src_iter = src->bi_iter; 1101 } 1102 1103 if (!dst_iter.bi_size) { 1104 dst = dst->bi_next; 1105 if (!dst) 1106 break; 1107 1108 dst_iter = dst->bi_iter; 1109 } 1110 1111 bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1112 } 1113 } 1114 EXPORT_SYMBOL(bio_list_copy_data); 1115 1116 struct bio_map_data { 1117 int is_our_pages; 1118 struct iov_iter iter; 1119 struct iovec iov[]; 1120 }; 1121 1122 static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1123 gfp_t gfp_mask) 1124 { 1125 struct bio_map_data *bmd; 1126 if (data->nr_segs > UIO_MAXIOV) 1127 return NULL; 1128 1129 bmd = kmalloc(sizeof(struct bio_map_data) + 1130 sizeof(struct iovec) * data->nr_segs, gfp_mask); 1131 if (!bmd) 1132 return NULL; 1133 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 1134 bmd->iter = *data; 1135 bmd->iter.iov = bmd->iov; 1136 return bmd; 1137 } 1138 1139 /** 1140 * bio_copy_from_iter - copy all pages from iov_iter to bio 1141 * @bio: The &struct bio which describes the I/O as destination 1142 * @iter: iov_iter as source 1143 * 1144 * Copy all pages from iov_iter to bio. 1145 * Returns 0 on success, or error on failure. 1146 */ 1147 static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1148 { 1149 int i; 1150 struct bio_vec *bvec; 1151 1152 bio_for_each_segment_all(bvec, bio, i) { 1153 ssize_t ret; 1154 1155 ret = copy_page_from_iter(bvec->bv_page, 1156 bvec->bv_offset, 1157 bvec->bv_len, 1158 iter); 1159 1160 if (!iov_iter_count(iter)) 1161 break; 1162 1163 if (ret < bvec->bv_len) 1164 return -EFAULT; 1165 } 1166 1167 return 0; 1168 } 1169 1170 /** 1171 * bio_copy_to_iter - copy all pages from bio to iov_iter 1172 * @bio: The &struct bio which describes the I/O as source 1173 * @iter: iov_iter as destination 1174 * 1175 * Copy all pages from bio to iov_iter. 1176 * Returns 0 on success, or error on failure. 1177 */ 1178 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 1179 { 1180 int i; 1181 struct bio_vec *bvec; 1182 1183 bio_for_each_segment_all(bvec, bio, i) { 1184 ssize_t ret; 1185 1186 ret = copy_page_to_iter(bvec->bv_page, 1187 bvec->bv_offset, 1188 bvec->bv_len, 1189 &iter); 1190 1191 if (!iov_iter_count(&iter)) 1192 break; 1193 1194 if (ret < bvec->bv_len) 1195 return -EFAULT; 1196 } 1197 1198 return 0; 1199 } 1200 1201 void bio_free_pages(struct bio *bio) 1202 { 1203 struct bio_vec *bvec; 1204 int i; 1205 1206 bio_for_each_segment_all(bvec, bio, i) 1207 __free_page(bvec->bv_page); 1208 } 1209 EXPORT_SYMBOL(bio_free_pages); 1210 1211 /** 1212 * bio_uncopy_user - finish previously mapped bio 1213 * @bio: bio being terminated 1214 * 1215 * Free pages allocated from bio_copy_user_iov() and write back data 1216 * to user space in case of a read. 1217 */ 1218 int bio_uncopy_user(struct bio *bio) 1219 { 1220 struct bio_map_data *bmd = bio->bi_private; 1221 int ret = 0; 1222 1223 if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1224 /* 1225 * if we're in a workqueue, the request is orphaned, so 1226 * don't copy into a random user address space, just free 1227 * and return -EINTR so user space doesn't expect any data. 1228 */ 1229 if (!current->mm) 1230 ret = -EINTR; 1231 else if (bio_data_dir(bio) == READ) 1232 ret = bio_copy_to_iter(bio, bmd->iter); 1233 if (bmd->is_our_pages) 1234 bio_free_pages(bio); 1235 } 1236 kfree(bmd); 1237 bio_put(bio); 1238 return ret; 1239 } 1240 1241 /** 1242 * bio_copy_user_iov - copy user data to bio 1243 * @q: destination block queue 1244 * @map_data: pointer to the rq_map_data holding pages (if necessary) 1245 * @iter: iovec iterator 1246 * @gfp_mask: memory allocation flags 1247 * 1248 * Prepares and returns a bio for indirect user io, bouncing data 1249 * to/from kernel pages as necessary. Must be paired with 1250 * call bio_uncopy_user() on io completion. 1251 */ 1252 struct bio *bio_copy_user_iov(struct request_queue *q, 1253 struct rq_map_data *map_data, 1254 struct iov_iter *iter, 1255 gfp_t gfp_mask) 1256 { 1257 struct bio_map_data *bmd; 1258 struct page *page; 1259 struct bio *bio; 1260 int i = 0, ret; 1261 int nr_pages; 1262 unsigned int len = iter->count; 1263 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1264 1265 bmd = bio_alloc_map_data(iter, gfp_mask); 1266 if (!bmd) 1267 return ERR_PTR(-ENOMEM); 1268 1269 /* 1270 * We need to do a deep copy of the iov_iter including the iovecs. 1271 * The caller provided iov might point to an on-stack or otherwise 1272 * shortlived one. 1273 */ 1274 bmd->is_our_pages = map_data ? 0 : 1; 1275 1276 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1277 if (nr_pages > BIO_MAX_PAGES) 1278 nr_pages = BIO_MAX_PAGES; 1279 1280 ret = -ENOMEM; 1281 bio = bio_kmalloc(gfp_mask, nr_pages); 1282 if (!bio) 1283 goto out_bmd; 1284 1285 ret = 0; 1286 1287 if (map_data) { 1288 nr_pages = 1 << map_data->page_order; 1289 i = map_data->offset / PAGE_SIZE; 1290 } 1291 while (len) { 1292 unsigned int bytes = PAGE_SIZE; 1293 1294 bytes -= offset; 1295 1296 if (bytes > len) 1297 bytes = len; 1298 1299 if (map_data) { 1300 if (i == map_data->nr_entries * nr_pages) { 1301 ret = -ENOMEM; 1302 break; 1303 } 1304 1305 page = map_data->pages[i / nr_pages]; 1306 page += (i % nr_pages); 1307 1308 i++; 1309 } else { 1310 page = alloc_page(q->bounce_gfp | gfp_mask); 1311 if (!page) { 1312 ret = -ENOMEM; 1313 break; 1314 } 1315 } 1316 1317 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) 1318 break; 1319 1320 len -= bytes; 1321 offset = 0; 1322 } 1323 1324 if (ret) 1325 goto cleanup; 1326 1327 if (map_data) 1328 map_data->offset += bio->bi_iter.bi_size; 1329 1330 /* 1331 * success 1332 */ 1333 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) || 1334 (map_data && map_data->from_user)) { 1335 ret = bio_copy_from_iter(bio, iter); 1336 if (ret) 1337 goto cleanup; 1338 } else { 1339 iov_iter_advance(iter, bio->bi_iter.bi_size); 1340 } 1341 1342 bio->bi_private = bmd; 1343 if (map_data && map_data->null_mapped) 1344 bio_set_flag(bio, BIO_NULL_MAPPED); 1345 return bio; 1346 cleanup: 1347 if (!map_data) 1348 bio_free_pages(bio); 1349 bio_put(bio); 1350 out_bmd: 1351 kfree(bmd); 1352 return ERR_PTR(ret); 1353 } 1354 1355 /** 1356 * bio_map_user_iov - map user iovec into bio 1357 * @q: the struct request_queue for the bio 1358 * @iter: iovec iterator 1359 * @gfp_mask: memory allocation flags 1360 * 1361 * Map the user space address into a bio suitable for io to a block 1362 * device. Returns an error pointer in case of error. 1363 */ 1364 struct bio *bio_map_user_iov(struct request_queue *q, 1365 struct iov_iter *iter, 1366 gfp_t gfp_mask) 1367 { 1368 int j; 1369 struct bio *bio; 1370 int ret; 1371 struct bio_vec *bvec; 1372 1373 if (!iov_iter_count(iter)) 1374 return ERR_PTR(-EINVAL); 1375 1376 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1377 if (!bio) 1378 return ERR_PTR(-ENOMEM); 1379 1380 while (iov_iter_count(iter)) { 1381 struct page **pages; 1382 ssize_t bytes; 1383 size_t offs, added = 0; 1384 int npages; 1385 1386 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1387 if (unlikely(bytes <= 0)) { 1388 ret = bytes ? bytes : -EFAULT; 1389 goto out_unmap; 1390 } 1391 1392 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1393 1394 if (unlikely(offs & queue_dma_alignment(q))) { 1395 ret = -EINVAL; 1396 j = 0; 1397 } else { 1398 for (j = 0; j < npages; j++) { 1399 struct page *page = pages[j]; 1400 unsigned int n = PAGE_SIZE - offs; 1401 unsigned short prev_bi_vcnt = bio->bi_vcnt; 1402 1403 if (n > bytes) 1404 n = bytes; 1405 1406 if (!bio_add_pc_page(q, bio, page, n, offs)) 1407 break; 1408 1409 /* 1410 * check if vector was merged with previous 1411 * drop page reference if needed 1412 */ 1413 if (bio->bi_vcnt == prev_bi_vcnt) 1414 put_page(page); 1415 1416 added += n; 1417 bytes -= n; 1418 offs = 0; 1419 } 1420 iov_iter_advance(iter, added); 1421 } 1422 /* 1423 * release the pages we didn't map into the bio, if any 1424 */ 1425 while (j < npages) 1426 put_page(pages[j++]); 1427 kvfree(pages); 1428 /* couldn't stuff something into bio? */ 1429 if (bytes) 1430 break; 1431 } 1432 1433 bio_set_flag(bio, BIO_USER_MAPPED); 1434 1435 /* 1436 * subtle -- if bio_map_user_iov() ended up bouncing a bio, 1437 * it would normally disappear when its bi_end_io is run. 1438 * however, we need it for the unmap, so grab an extra 1439 * reference to it 1440 */ 1441 bio_get(bio); 1442 return bio; 1443 1444 out_unmap: 1445 bio_for_each_segment_all(bvec, bio, j) { 1446 put_page(bvec->bv_page); 1447 } 1448 bio_put(bio); 1449 return ERR_PTR(ret); 1450 } 1451 1452 static void __bio_unmap_user(struct bio *bio) 1453 { 1454 struct bio_vec *bvec; 1455 int i; 1456 1457 /* 1458 * make sure we dirty pages we wrote to 1459 */ 1460 bio_for_each_segment_all(bvec, bio, i) { 1461 if (bio_data_dir(bio) == READ) 1462 set_page_dirty_lock(bvec->bv_page); 1463 1464 put_page(bvec->bv_page); 1465 } 1466 1467 bio_put(bio); 1468 } 1469 1470 /** 1471 * bio_unmap_user - unmap a bio 1472 * @bio: the bio being unmapped 1473 * 1474 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 1475 * process context. 1476 * 1477 * bio_unmap_user() may sleep. 1478 */ 1479 void bio_unmap_user(struct bio *bio) 1480 { 1481 __bio_unmap_user(bio); 1482 bio_put(bio); 1483 } 1484 1485 static void bio_map_kern_endio(struct bio *bio) 1486 { 1487 bio_put(bio); 1488 } 1489 1490 /** 1491 * bio_map_kern - map kernel address into bio 1492 * @q: the struct request_queue for the bio 1493 * @data: pointer to buffer to map 1494 * @len: length in bytes 1495 * @gfp_mask: allocation flags for bio allocation 1496 * 1497 * Map the kernel address into a bio suitable for io to a block 1498 * device. Returns an error pointer in case of error. 1499 */ 1500 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 1501 gfp_t gfp_mask) 1502 { 1503 unsigned long kaddr = (unsigned long)data; 1504 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1505 unsigned long start = kaddr >> PAGE_SHIFT; 1506 const int nr_pages = end - start; 1507 int offset, i; 1508 struct bio *bio; 1509 1510 bio = bio_kmalloc(gfp_mask, nr_pages); 1511 if (!bio) 1512 return ERR_PTR(-ENOMEM); 1513 1514 offset = offset_in_page(kaddr); 1515 for (i = 0; i < nr_pages; i++) { 1516 unsigned int bytes = PAGE_SIZE - offset; 1517 1518 if (len <= 0) 1519 break; 1520 1521 if (bytes > len) 1522 bytes = len; 1523 1524 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 1525 offset) < bytes) { 1526 /* we don't support partial mappings */ 1527 bio_put(bio); 1528 return ERR_PTR(-EINVAL); 1529 } 1530 1531 data += bytes; 1532 len -= bytes; 1533 offset = 0; 1534 } 1535 1536 bio->bi_end_io = bio_map_kern_endio; 1537 return bio; 1538 } 1539 EXPORT_SYMBOL(bio_map_kern); 1540 1541 static void bio_copy_kern_endio(struct bio *bio) 1542 { 1543 bio_free_pages(bio); 1544 bio_put(bio); 1545 } 1546 1547 static void bio_copy_kern_endio_read(struct bio *bio) 1548 { 1549 char *p = bio->bi_private; 1550 struct bio_vec *bvec; 1551 int i; 1552 1553 bio_for_each_segment_all(bvec, bio, i) { 1554 memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1555 p += bvec->bv_len; 1556 } 1557 1558 bio_copy_kern_endio(bio); 1559 } 1560 1561 /** 1562 * bio_copy_kern - copy kernel address into bio 1563 * @q: the struct request_queue for the bio 1564 * @data: pointer to buffer to copy 1565 * @len: length in bytes 1566 * @gfp_mask: allocation flags for bio and page allocation 1567 * @reading: data direction is READ 1568 * 1569 * copy the kernel address into a bio suitable for io to a block 1570 * device. Returns an error pointer in case of error. 1571 */ 1572 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1573 gfp_t gfp_mask, int reading) 1574 { 1575 unsigned long kaddr = (unsigned long)data; 1576 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1577 unsigned long start = kaddr >> PAGE_SHIFT; 1578 struct bio *bio; 1579 void *p = data; 1580 int nr_pages = 0; 1581 1582 /* 1583 * Overflow, abort 1584 */ 1585 if (end < start) 1586 return ERR_PTR(-EINVAL); 1587 1588 nr_pages = end - start; 1589 bio = bio_kmalloc(gfp_mask, nr_pages); 1590 if (!bio) 1591 return ERR_PTR(-ENOMEM); 1592 1593 while (len) { 1594 struct page *page; 1595 unsigned int bytes = PAGE_SIZE; 1596 1597 if (bytes > len) 1598 bytes = len; 1599 1600 page = alloc_page(q->bounce_gfp | gfp_mask); 1601 if (!page) 1602 goto cleanup; 1603 1604 if (!reading) 1605 memcpy(page_address(page), p, bytes); 1606 1607 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 1608 break; 1609 1610 len -= bytes; 1611 p += bytes; 1612 } 1613 1614 if (reading) { 1615 bio->bi_end_io = bio_copy_kern_endio_read; 1616 bio->bi_private = data; 1617 } else { 1618 bio->bi_end_io = bio_copy_kern_endio; 1619 } 1620 1621 return bio; 1622 1623 cleanup: 1624 bio_free_pages(bio); 1625 bio_put(bio); 1626 return ERR_PTR(-ENOMEM); 1627 } 1628 1629 /* 1630 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1631 * for performing direct-IO in BIOs. 1632 * 1633 * The problem is that we cannot run set_page_dirty() from interrupt context 1634 * because the required locks are not interrupt-safe. So what we can do is to 1635 * mark the pages dirty _before_ performing IO. And in interrupt context, 1636 * check that the pages are still dirty. If so, fine. If not, redirty them 1637 * in process context. 1638 * 1639 * We special-case compound pages here: normally this means reads into hugetlb 1640 * pages. The logic in here doesn't really work right for compound pages 1641 * because the VM does not uniformly chase down the head page in all cases. 1642 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1643 * handle them at all. So we skip compound pages here at an early stage. 1644 * 1645 * Note that this code is very hard to test under normal circumstances because 1646 * direct-io pins the pages with get_user_pages(). This makes 1647 * is_page_cache_freeable return false, and the VM will not clean the pages. 1648 * But other code (eg, flusher threads) could clean the pages if they are mapped 1649 * pagecache. 1650 * 1651 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1652 * deferred bio dirtying paths. 1653 */ 1654 1655 /* 1656 * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1657 */ 1658 void bio_set_pages_dirty(struct bio *bio) 1659 { 1660 struct bio_vec *bvec; 1661 int i; 1662 1663 bio_for_each_segment_all(bvec, bio, i) { 1664 struct page *page = bvec->bv_page; 1665 1666 if (page && !PageCompound(page)) 1667 set_page_dirty_lock(page); 1668 } 1669 } 1670 EXPORT_SYMBOL_GPL(bio_set_pages_dirty); 1671 1672 static void bio_release_pages(struct bio *bio) 1673 { 1674 struct bio_vec *bvec; 1675 int i; 1676 1677 bio_for_each_segment_all(bvec, bio, i) { 1678 struct page *page = bvec->bv_page; 1679 1680 if (page) 1681 put_page(page); 1682 } 1683 } 1684 1685 /* 1686 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1687 * If they are, then fine. If, however, some pages are clean then they must 1688 * have been written out during the direct-IO read. So we take another ref on 1689 * the BIO and the offending pages and re-dirty the pages in process context. 1690 * 1691 * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1692 * here on. It will run one put_page() against each page and will run one 1693 * bio_put() against the BIO. 1694 */ 1695 1696 static void bio_dirty_fn(struct work_struct *work); 1697 1698 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1699 static DEFINE_SPINLOCK(bio_dirty_lock); 1700 static struct bio *bio_dirty_list; 1701 1702 /* 1703 * This runs in process context 1704 */ 1705 static void bio_dirty_fn(struct work_struct *work) 1706 { 1707 unsigned long flags; 1708 struct bio *bio; 1709 1710 spin_lock_irqsave(&bio_dirty_lock, flags); 1711 bio = bio_dirty_list; 1712 bio_dirty_list = NULL; 1713 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1714 1715 while (bio) { 1716 struct bio *next = bio->bi_private; 1717 1718 bio_set_pages_dirty(bio); 1719 bio_release_pages(bio); 1720 bio_put(bio); 1721 bio = next; 1722 } 1723 } 1724 1725 void bio_check_pages_dirty(struct bio *bio) 1726 { 1727 struct bio_vec *bvec; 1728 int nr_clean_pages = 0; 1729 int i; 1730 1731 bio_for_each_segment_all(bvec, bio, i) { 1732 struct page *page = bvec->bv_page; 1733 1734 if (PageDirty(page) || PageCompound(page)) { 1735 put_page(page); 1736 bvec->bv_page = NULL; 1737 } else { 1738 nr_clean_pages++; 1739 } 1740 } 1741 1742 if (nr_clean_pages) { 1743 unsigned long flags; 1744 1745 spin_lock_irqsave(&bio_dirty_lock, flags); 1746 bio->bi_private = bio_dirty_list; 1747 bio_dirty_list = bio; 1748 spin_unlock_irqrestore(&bio_dirty_lock, flags); 1749 schedule_work(&bio_dirty_work); 1750 } else { 1751 bio_put(bio); 1752 } 1753 } 1754 EXPORT_SYMBOL_GPL(bio_check_pages_dirty); 1755 1756 void generic_start_io_acct(struct request_queue *q, int rw, 1757 unsigned long sectors, struct hd_struct *part) 1758 { 1759 int cpu = part_stat_lock(); 1760 1761 part_round_stats(q, cpu, part); 1762 part_stat_inc(cpu, part, ios[rw]); 1763 part_stat_add(cpu, part, sectors[rw], sectors); 1764 part_inc_in_flight(q, part, rw); 1765 1766 part_stat_unlock(); 1767 } 1768 EXPORT_SYMBOL(generic_start_io_acct); 1769 1770 void generic_end_io_acct(struct request_queue *q, int rw, 1771 struct hd_struct *part, unsigned long start_time) 1772 { 1773 unsigned long duration = jiffies - start_time; 1774 int cpu = part_stat_lock(); 1775 1776 part_stat_add(cpu, part, ticks[rw], duration); 1777 part_round_stats(q, cpu, part); 1778 part_dec_in_flight(q, part, rw); 1779 1780 part_stat_unlock(); 1781 } 1782 EXPORT_SYMBOL(generic_end_io_acct); 1783 1784 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1785 void bio_flush_dcache_pages(struct bio *bi) 1786 { 1787 struct bio_vec bvec; 1788 struct bvec_iter iter; 1789 1790 bio_for_each_segment(bvec, bi, iter) 1791 flush_dcache_page(bvec.bv_page); 1792 } 1793 EXPORT_SYMBOL(bio_flush_dcache_pages); 1794 #endif 1795 1796 static inline bool bio_remaining_done(struct bio *bio) 1797 { 1798 /* 1799 * If we're not chaining, then ->__bi_remaining is always 1 and 1800 * we always end io on the first invocation. 1801 */ 1802 if (!bio_flagged(bio, BIO_CHAIN)) 1803 return true; 1804 1805 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1806 1807 if (atomic_dec_and_test(&bio->__bi_remaining)) { 1808 bio_clear_flag(bio, BIO_CHAIN); 1809 return true; 1810 } 1811 1812 return false; 1813 } 1814 1815 /** 1816 * bio_endio - end I/O on a bio 1817 * @bio: bio 1818 * 1819 * Description: 1820 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 1821 * way to end I/O on a bio. No one should call bi_end_io() directly on a 1822 * bio unless they own it and thus know that it has an end_io function. 1823 * 1824 * bio_endio() can be called several times on a bio that has been chained 1825 * using bio_chain(). The ->bi_end_io() function will only be called the 1826 * last time. At this point the BLK_TA_COMPLETE tracing event will be 1827 * generated if BIO_TRACE_COMPLETION is set. 1828 **/ 1829 void bio_endio(struct bio *bio) 1830 { 1831 again: 1832 if (!bio_remaining_done(bio)) 1833 return; 1834 if (!bio_integrity_endio(bio)) 1835 return; 1836 1837 /* 1838 * Need to have a real endio function for chained bios, otherwise 1839 * various corner cases will break (like stacking block devices that 1840 * save/restore bi_end_io) - however, we want to avoid unbounded 1841 * recursion and blowing the stack. Tail call optimization would 1842 * handle this, but compiling with frame pointers also disables 1843 * gcc's sibling call optimization. 1844 */ 1845 if (bio->bi_end_io == bio_chain_endio) { 1846 bio = __bio_chain_endio(bio); 1847 goto again; 1848 } 1849 1850 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1851 trace_block_bio_complete(bio->bi_disk->queue, bio, 1852 blk_status_to_errno(bio->bi_status)); 1853 bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1854 } 1855 1856 blk_throtl_bio_endio(bio); 1857 /* release cgroup info */ 1858 bio_uninit(bio); 1859 if (bio->bi_end_io) 1860 bio->bi_end_io(bio); 1861 } 1862 EXPORT_SYMBOL(bio_endio); 1863 1864 /** 1865 * bio_split - split a bio 1866 * @bio: bio to split 1867 * @sectors: number of sectors to split from the front of @bio 1868 * @gfp: gfp mask 1869 * @bs: bio set to allocate from 1870 * 1871 * Allocates and returns a new bio which represents @sectors from the start of 1872 * @bio, and updates @bio to represent the remaining sectors. 1873 * 1874 * Unless this is a discard request the newly allocated bio will point 1875 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that 1876 * @bio is not freed before the split. 1877 */ 1878 struct bio *bio_split(struct bio *bio, int sectors, 1879 gfp_t gfp, struct bio_set *bs) 1880 { 1881 struct bio *split; 1882 1883 BUG_ON(sectors <= 0); 1884 BUG_ON(sectors >= bio_sectors(bio)); 1885 1886 split = bio_clone_fast(bio, gfp, bs); 1887 if (!split) 1888 return NULL; 1889 1890 split->bi_iter.bi_size = sectors << 9; 1891 1892 if (bio_integrity(split)) 1893 bio_integrity_trim(split); 1894 1895 bio_advance(bio, split->bi_iter.bi_size); 1896 bio->bi_iter.bi_done = 0; 1897 1898 if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 1899 bio_set_flag(split, BIO_TRACE_COMPLETION); 1900 1901 return split; 1902 } 1903 EXPORT_SYMBOL(bio_split); 1904 1905 /** 1906 * bio_trim - trim a bio 1907 * @bio: bio to trim 1908 * @offset: number of sectors to trim from the front of @bio 1909 * @size: size we want to trim @bio to, in sectors 1910 */ 1911 void bio_trim(struct bio *bio, int offset, int size) 1912 { 1913 /* 'bio' is a cloned bio which we need to trim to match 1914 * the given offset and size. 1915 */ 1916 1917 size <<= 9; 1918 if (offset == 0 && size == bio->bi_iter.bi_size) 1919 return; 1920 1921 bio_clear_flag(bio, BIO_SEG_VALID); 1922 1923 bio_advance(bio, offset << 9); 1924 1925 bio->bi_iter.bi_size = size; 1926 1927 if (bio_integrity(bio)) 1928 bio_integrity_trim(bio); 1929 1930 } 1931 EXPORT_SYMBOL_GPL(bio_trim); 1932 1933 /* 1934 * create memory pools for biovec's in a bio_set. 1935 * use the global biovec slabs created for general use. 1936 */ 1937 int biovec_init_pool(mempool_t *pool, int pool_entries) 1938 { 1939 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1940 1941 return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1942 } 1943 1944 /* 1945 * bioset_exit - exit a bioset initialized with bioset_init() 1946 * 1947 * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1948 * kzalloc()). 1949 */ 1950 void bioset_exit(struct bio_set *bs) 1951 { 1952 if (bs->rescue_workqueue) 1953 destroy_workqueue(bs->rescue_workqueue); 1954 bs->rescue_workqueue = NULL; 1955 1956 mempool_exit(&bs->bio_pool); 1957 mempool_exit(&bs->bvec_pool); 1958 1959 bioset_integrity_free(bs); 1960 if (bs->bio_slab) 1961 bio_put_slab(bs); 1962 bs->bio_slab = NULL; 1963 } 1964 EXPORT_SYMBOL(bioset_exit); 1965 1966 /** 1967 * bioset_init - Initialize a bio_set 1968 * @bs: pool to initialize 1969 * @pool_size: Number of bio and bio_vecs to cache in the mempool 1970 * @front_pad: Number of bytes to allocate in front of the returned bio 1971 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1972 * and %BIOSET_NEED_RESCUER 1973 * 1974 * Description: 1975 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1976 * to ask for a number of bytes to be allocated in front of the bio. 1977 * Front pad allocation is useful for embedding the bio inside 1978 * another structure, to avoid allocating extra data to go with the bio. 1979 * Note that the bio must be embedded at the END of that structure always, 1980 * or things will break badly. 1981 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1982 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1983 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1984 * dispatch queued requests when the mempool runs out of space. 1985 * 1986 */ 1987 int bioset_init(struct bio_set *bs, 1988 unsigned int pool_size, 1989 unsigned int front_pad, 1990 int flags) 1991 { 1992 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1993 1994 bs->front_pad = front_pad; 1995 1996 spin_lock_init(&bs->rescue_lock); 1997 bio_list_init(&bs->rescue_list); 1998 INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1999 2000 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 2001 if (!bs->bio_slab) 2002 return -ENOMEM; 2003 2004 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2005 goto bad; 2006 2007 if ((flags & BIOSET_NEED_BVECS) && 2008 biovec_init_pool(&bs->bvec_pool, pool_size)) 2009 goto bad; 2010 2011 if (!(flags & BIOSET_NEED_RESCUER)) 2012 return 0; 2013 2014 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2015 if (!bs->rescue_workqueue) 2016 goto bad; 2017 2018 return 0; 2019 bad: 2020 bioset_exit(bs); 2021 return -ENOMEM; 2022 } 2023 EXPORT_SYMBOL(bioset_init); 2024 2025 /* 2026 * Initialize and setup a new bio_set, based on the settings from 2027 * another bio_set. 2028 */ 2029 int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 2030 { 2031 int flags; 2032 2033 flags = 0; 2034 if (src->bvec_pool.min_nr) 2035 flags |= BIOSET_NEED_BVECS; 2036 if (src->rescue_workqueue) 2037 flags |= BIOSET_NEED_RESCUER; 2038 2039 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 2040 } 2041 EXPORT_SYMBOL(bioset_init_from_src); 2042 2043 #ifdef CONFIG_BLK_CGROUP 2044 2045 /** 2046 * bio_associate_blkcg - associate a bio with the specified blkcg 2047 * @bio: target bio 2048 * @blkcg_css: css of the blkcg to associate 2049 * 2050 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will 2051 * treat @bio as if it were issued by a task which belongs to the blkcg. 2052 * 2053 * This function takes an extra reference of @blkcg_css which will be put 2054 * when @bio is released. The caller must own @bio and is responsible for 2055 * synchronizing calls to this function. 2056 */ 2057 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css) 2058 { 2059 if (unlikely(bio->bi_css)) 2060 return -EBUSY; 2061 css_get(blkcg_css); 2062 bio->bi_css = blkcg_css; 2063 return 0; 2064 } 2065 EXPORT_SYMBOL_GPL(bio_associate_blkcg); 2066 2067 /** 2068 * bio_disassociate_task - undo bio_associate_current() 2069 * @bio: target bio 2070 */ 2071 void bio_disassociate_task(struct bio *bio) 2072 { 2073 if (bio->bi_ioc) { 2074 put_io_context(bio->bi_ioc); 2075 bio->bi_ioc = NULL; 2076 } 2077 if (bio->bi_css) { 2078 css_put(bio->bi_css); 2079 bio->bi_css = NULL; 2080 } 2081 } 2082 2083 /** 2084 * bio_clone_blkcg_association - clone blkcg association from src to dst bio 2085 * @dst: destination bio 2086 * @src: source bio 2087 */ 2088 void bio_clone_blkcg_association(struct bio *dst, struct bio *src) 2089 { 2090 if (src->bi_css) 2091 WARN_ON(bio_associate_blkcg(dst, src->bi_css)); 2092 } 2093 EXPORT_SYMBOL_GPL(bio_clone_blkcg_association); 2094 #endif /* CONFIG_BLK_CGROUP */ 2095 2096 static void __init biovec_init_slabs(void) 2097 { 2098 int i; 2099 2100 for (i = 0; i < BVEC_POOL_NR; i++) { 2101 int size; 2102 struct biovec_slab *bvs = bvec_slabs + i; 2103 2104 if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2105 bvs->slab = NULL; 2106 continue; 2107 } 2108 2109 size = bvs->nr_vecs * sizeof(struct bio_vec); 2110 bvs->slab = kmem_cache_create(bvs->name, size, 0, 2111 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2112 } 2113 } 2114 2115 static int __init init_bio(void) 2116 { 2117 bio_slab_max = 2; 2118 bio_slab_nr = 0; 2119 bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 2120 GFP_KERNEL); 2121 if (!bio_slabs) 2122 panic("bio: can't allocate bios\n"); 2123 2124 bio_integrity_init(); 2125 biovec_init_slabs(); 2126 2127 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2128 panic("bio: can't allocate bios\n"); 2129 2130 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2131 panic("bio: can't create integrity pool\n"); 2132 2133 return 0; 2134 } 2135 subsys_initcall(init_bio); 2136