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