1 /* 2 * Copyright (C) 2011 STRATO. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as 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 GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 #include <linux/sched.h> 20 #include <linux/pagemap.h> 21 #include <linux/writeback.h> 22 #include <linux/blkdev.h> 23 #include <linux/rbtree.h> 24 #include <linux/slab.h> 25 #include <linux/workqueue.h> 26 #include "ctree.h" 27 #include "volumes.h" 28 #include "disk-io.h" 29 #include "ordered-data.h" 30 31 /* 32 * This is only the first step towards a full-features scrub. It reads all 33 * extent and super block and verifies the checksums. In case a bad checksum 34 * is found or the extent cannot be read, good data will be written back if 35 * any can be found. 36 * 37 * Future enhancements: 38 * - To enhance the performance, better read-ahead strategies for the 39 * extent-tree can be employed. 40 * - In case an unrepairable extent is encountered, track which files are 41 * affected and report them 42 * - In case of a read error on files with nodatasum, map the file and read 43 * the extent to trigger a writeback of the good copy 44 * - track and record media errors, throw out bad devices 45 * - add a mode to also read unallocated space 46 * - make the prefetch cancellable 47 */ 48 49 struct scrub_bio; 50 struct scrub_page; 51 struct scrub_dev; 52 static void scrub_bio_end_io(struct bio *bio, int err); 53 static void scrub_checksum(struct btrfs_work *work); 54 static int scrub_checksum_data(struct scrub_dev *sdev, 55 struct scrub_page *spag, void *buffer); 56 static int scrub_checksum_tree_block(struct scrub_dev *sdev, 57 struct scrub_page *spag, u64 logical, 58 void *buffer); 59 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer); 60 static int scrub_fixup_check(struct scrub_bio *sbio, int ix); 61 static void scrub_fixup_end_io(struct bio *bio, int err); 62 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector, 63 struct page *page); 64 static void scrub_fixup(struct scrub_bio *sbio, int ix); 65 66 #define SCRUB_PAGES_PER_BIO 16 /* 64k per bio */ 67 #define SCRUB_BIOS_PER_DEV 16 /* 1 MB per device in flight */ 68 69 struct scrub_page { 70 u64 flags; /* extent flags */ 71 u64 generation; 72 u64 mirror_num; 73 int have_csum; 74 u8 csum[BTRFS_CSUM_SIZE]; 75 }; 76 77 struct scrub_bio { 78 int index; 79 struct scrub_dev *sdev; 80 struct bio *bio; 81 int err; 82 u64 logical; 83 u64 physical; 84 struct scrub_page spag[SCRUB_PAGES_PER_BIO]; 85 u64 count; 86 int next_free; 87 struct btrfs_work work; 88 }; 89 90 struct scrub_dev { 91 struct scrub_bio *bios[SCRUB_BIOS_PER_DEV]; 92 struct btrfs_device *dev; 93 int first_free; 94 int curr; 95 atomic_t in_flight; 96 spinlock_t list_lock; 97 wait_queue_head_t list_wait; 98 u16 csum_size; 99 struct list_head csum_list; 100 atomic_t cancel_req; 101 int readonly; 102 /* 103 * statistics 104 */ 105 struct btrfs_scrub_progress stat; 106 spinlock_t stat_lock; 107 }; 108 109 static void scrub_free_csums(struct scrub_dev *sdev) 110 { 111 while (!list_empty(&sdev->csum_list)) { 112 struct btrfs_ordered_sum *sum; 113 sum = list_first_entry(&sdev->csum_list, 114 struct btrfs_ordered_sum, list); 115 list_del(&sum->list); 116 kfree(sum); 117 } 118 } 119 120 static void scrub_free_bio(struct bio *bio) 121 { 122 int i; 123 struct page *last_page = NULL; 124 125 if (!bio) 126 return; 127 128 for (i = 0; i < bio->bi_vcnt; ++i) { 129 if (bio->bi_io_vec[i].bv_page == last_page) 130 continue; 131 last_page = bio->bi_io_vec[i].bv_page; 132 __free_page(last_page); 133 } 134 bio_put(bio); 135 } 136 137 static noinline_for_stack void scrub_free_dev(struct scrub_dev *sdev) 138 { 139 int i; 140 141 if (!sdev) 142 return; 143 144 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) { 145 struct scrub_bio *sbio = sdev->bios[i]; 146 147 if (!sbio) 148 break; 149 150 scrub_free_bio(sbio->bio); 151 kfree(sbio); 152 } 153 154 scrub_free_csums(sdev); 155 kfree(sdev); 156 } 157 158 static noinline_for_stack 159 struct scrub_dev *scrub_setup_dev(struct btrfs_device *dev) 160 { 161 struct scrub_dev *sdev; 162 int i; 163 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; 164 165 sdev = kzalloc(sizeof(*sdev), GFP_NOFS); 166 if (!sdev) 167 goto nomem; 168 sdev->dev = dev; 169 for (i = 0; i < SCRUB_BIOS_PER_DEV; ++i) { 170 struct scrub_bio *sbio; 171 172 sbio = kzalloc(sizeof(*sbio), GFP_NOFS); 173 if (!sbio) 174 goto nomem; 175 sdev->bios[i] = sbio; 176 177 sbio->index = i; 178 sbio->sdev = sdev; 179 sbio->count = 0; 180 sbio->work.func = scrub_checksum; 181 182 if (i != SCRUB_BIOS_PER_DEV-1) 183 sdev->bios[i]->next_free = i + 1; 184 else 185 sdev->bios[i]->next_free = -1; 186 } 187 sdev->first_free = 0; 188 sdev->curr = -1; 189 atomic_set(&sdev->in_flight, 0); 190 atomic_set(&sdev->cancel_req, 0); 191 sdev->csum_size = btrfs_super_csum_size(&fs_info->super_copy); 192 INIT_LIST_HEAD(&sdev->csum_list); 193 194 spin_lock_init(&sdev->list_lock); 195 spin_lock_init(&sdev->stat_lock); 196 init_waitqueue_head(&sdev->list_wait); 197 return sdev; 198 199 nomem: 200 scrub_free_dev(sdev); 201 return ERR_PTR(-ENOMEM); 202 } 203 204 /* 205 * scrub_recheck_error gets called when either verification of the page 206 * failed or the bio failed to read, e.g. with EIO. In the latter case, 207 * recheck_error gets called for every page in the bio, even though only 208 * one may be bad 209 */ 210 static void scrub_recheck_error(struct scrub_bio *sbio, int ix) 211 { 212 if (sbio->err) { 213 if (scrub_fixup_io(READ, sbio->sdev->dev->bdev, 214 (sbio->physical + ix * PAGE_SIZE) >> 9, 215 sbio->bio->bi_io_vec[ix].bv_page) == 0) { 216 if (scrub_fixup_check(sbio, ix) == 0) 217 return; 218 } 219 } 220 221 scrub_fixup(sbio, ix); 222 } 223 224 static int scrub_fixup_check(struct scrub_bio *sbio, int ix) 225 { 226 int ret = 1; 227 struct page *page; 228 void *buffer; 229 u64 flags = sbio->spag[ix].flags; 230 231 page = sbio->bio->bi_io_vec[ix].bv_page; 232 buffer = kmap_atomic(page, KM_USER0); 233 if (flags & BTRFS_EXTENT_FLAG_DATA) { 234 ret = scrub_checksum_data(sbio->sdev, 235 sbio->spag + ix, buffer); 236 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 237 ret = scrub_checksum_tree_block(sbio->sdev, 238 sbio->spag + ix, 239 sbio->logical + ix * PAGE_SIZE, 240 buffer); 241 } else { 242 WARN_ON(1); 243 } 244 kunmap_atomic(buffer, KM_USER0); 245 246 return ret; 247 } 248 249 static void scrub_fixup_end_io(struct bio *bio, int err) 250 { 251 complete((struct completion *)bio->bi_private); 252 } 253 254 static void scrub_fixup(struct scrub_bio *sbio, int ix) 255 { 256 struct scrub_dev *sdev = sbio->sdev; 257 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info; 258 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; 259 struct btrfs_multi_bio *multi = NULL; 260 u64 logical = sbio->logical + ix * PAGE_SIZE; 261 u64 length; 262 int i; 263 int ret; 264 DECLARE_COMPLETION_ONSTACK(complete); 265 266 if ((sbio->spag[ix].flags & BTRFS_EXTENT_FLAG_DATA) && 267 (sbio->spag[ix].have_csum == 0)) { 268 /* 269 * nodatasum, don't try to fix anything 270 * FIXME: we can do better, open the inode and trigger a 271 * writeback 272 */ 273 goto uncorrectable; 274 } 275 276 length = PAGE_SIZE; 277 ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length, 278 &multi, 0); 279 if (ret || !multi || length < PAGE_SIZE) { 280 printk(KERN_ERR 281 "scrub_fixup: btrfs_map_block failed us for %llu\n", 282 (unsigned long long)logical); 283 WARN_ON(1); 284 return; 285 } 286 287 if (multi->num_stripes == 1) 288 /* there aren't any replicas */ 289 goto uncorrectable; 290 291 /* 292 * first find a good copy 293 */ 294 for (i = 0; i < multi->num_stripes; ++i) { 295 if (i == sbio->spag[ix].mirror_num) 296 continue; 297 298 if (scrub_fixup_io(READ, multi->stripes[i].dev->bdev, 299 multi->stripes[i].physical >> 9, 300 sbio->bio->bi_io_vec[ix].bv_page)) { 301 /* I/O-error, this is not a good copy */ 302 continue; 303 } 304 305 if (scrub_fixup_check(sbio, ix) == 0) 306 break; 307 } 308 if (i == multi->num_stripes) 309 goto uncorrectable; 310 311 if (!sdev->readonly) { 312 /* 313 * bi_io_vec[ix].bv_page now contains good data, write it back 314 */ 315 if (scrub_fixup_io(WRITE, sdev->dev->bdev, 316 (sbio->physical + ix * PAGE_SIZE) >> 9, 317 sbio->bio->bi_io_vec[ix].bv_page)) { 318 /* I/O-error, writeback failed, give up */ 319 goto uncorrectable; 320 } 321 } 322 323 kfree(multi); 324 spin_lock(&sdev->stat_lock); 325 ++sdev->stat.corrected_errors; 326 spin_unlock(&sdev->stat_lock); 327 328 if (printk_ratelimit()) 329 printk(KERN_ERR "btrfs: fixed up at %llu\n", 330 (unsigned long long)logical); 331 return; 332 333 uncorrectable: 334 kfree(multi); 335 spin_lock(&sdev->stat_lock); 336 ++sdev->stat.uncorrectable_errors; 337 spin_unlock(&sdev->stat_lock); 338 339 if (printk_ratelimit()) 340 printk(KERN_ERR "btrfs: unable to fixup at %llu\n", 341 (unsigned long long)logical); 342 } 343 344 static int scrub_fixup_io(int rw, struct block_device *bdev, sector_t sector, 345 struct page *page) 346 { 347 struct bio *bio = NULL; 348 int ret; 349 DECLARE_COMPLETION_ONSTACK(complete); 350 351 bio = bio_alloc(GFP_NOFS, 1); 352 bio->bi_bdev = bdev; 353 bio->bi_sector = sector; 354 bio_add_page(bio, page, PAGE_SIZE, 0); 355 bio->bi_end_io = scrub_fixup_end_io; 356 bio->bi_private = &complete; 357 submit_bio(rw, bio); 358 359 /* this will also unplug the queue */ 360 wait_for_completion(&complete); 361 362 ret = !test_bit(BIO_UPTODATE, &bio->bi_flags); 363 bio_put(bio); 364 return ret; 365 } 366 367 static void scrub_bio_end_io(struct bio *bio, int err) 368 { 369 struct scrub_bio *sbio = bio->bi_private; 370 struct scrub_dev *sdev = sbio->sdev; 371 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info; 372 373 sbio->err = err; 374 sbio->bio = bio; 375 376 btrfs_queue_worker(&fs_info->scrub_workers, &sbio->work); 377 } 378 379 static void scrub_checksum(struct btrfs_work *work) 380 { 381 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); 382 struct scrub_dev *sdev = sbio->sdev; 383 struct page *page; 384 void *buffer; 385 int i; 386 u64 flags; 387 u64 logical; 388 int ret; 389 390 if (sbio->err) { 391 for (i = 0; i < sbio->count; ++i) 392 scrub_recheck_error(sbio, i); 393 394 sbio->bio->bi_flags &= ~(BIO_POOL_MASK - 1); 395 sbio->bio->bi_flags |= 1 << BIO_UPTODATE; 396 sbio->bio->bi_phys_segments = 0; 397 sbio->bio->bi_idx = 0; 398 399 for (i = 0; i < sbio->count; i++) { 400 struct bio_vec *bi; 401 bi = &sbio->bio->bi_io_vec[i]; 402 bi->bv_offset = 0; 403 bi->bv_len = PAGE_SIZE; 404 } 405 406 spin_lock(&sdev->stat_lock); 407 ++sdev->stat.read_errors; 408 spin_unlock(&sdev->stat_lock); 409 goto out; 410 } 411 for (i = 0; i < sbio->count; ++i) { 412 page = sbio->bio->bi_io_vec[i].bv_page; 413 buffer = kmap_atomic(page, KM_USER0); 414 flags = sbio->spag[i].flags; 415 logical = sbio->logical + i * PAGE_SIZE; 416 ret = 0; 417 if (flags & BTRFS_EXTENT_FLAG_DATA) { 418 ret = scrub_checksum_data(sdev, sbio->spag + i, buffer); 419 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 420 ret = scrub_checksum_tree_block(sdev, sbio->spag + i, 421 logical, buffer); 422 } else if (flags & BTRFS_EXTENT_FLAG_SUPER) { 423 BUG_ON(i); 424 (void)scrub_checksum_super(sbio, buffer); 425 } else { 426 WARN_ON(1); 427 } 428 kunmap_atomic(buffer, KM_USER0); 429 if (ret) 430 scrub_recheck_error(sbio, i); 431 } 432 433 out: 434 scrub_free_bio(sbio->bio); 435 sbio->bio = NULL; 436 spin_lock(&sdev->list_lock); 437 sbio->next_free = sdev->first_free; 438 sdev->first_free = sbio->index; 439 spin_unlock(&sdev->list_lock); 440 atomic_dec(&sdev->in_flight); 441 wake_up(&sdev->list_wait); 442 } 443 444 static int scrub_checksum_data(struct scrub_dev *sdev, 445 struct scrub_page *spag, void *buffer) 446 { 447 u8 csum[BTRFS_CSUM_SIZE]; 448 u32 crc = ~(u32)0; 449 int fail = 0; 450 struct btrfs_root *root = sdev->dev->dev_root; 451 452 if (!spag->have_csum) 453 return 0; 454 455 crc = btrfs_csum_data(root, buffer, crc, PAGE_SIZE); 456 btrfs_csum_final(crc, csum); 457 if (memcmp(csum, spag->csum, sdev->csum_size)) 458 fail = 1; 459 460 spin_lock(&sdev->stat_lock); 461 ++sdev->stat.data_extents_scrubbed; 462 sdev->stat.data_bytes_scrubbed += PAGE_SIZE; 463 if (fail) 464 ++sdev->stat.csum_errors; 465 spin_unlock(&sdev->stat_lock); 466 467 return fail; 468 } 469 470 static int scrub_checksum_tree_block(struct scrub_dev *sdev, 471 struct scrub_page *spag, u64 logical, 472 void *buffer) 473 { 474 struct btrfs_header *h; 475 struct btrfs_root *root = sdev->dev->dev_root; 476 struct btrfs_fs_info *fs_info = root->fs_info; 477 u8 csum[BTRFS_CSUM_SIZE]; 478 u32 crc = ~(u32)0; 479 int fail = 0; 480 int crc_fail = 0; 481 482 /* 483 * we don't use the getter functions here, as we 484 * a) don't have an extent buffer and 485 * b) the page is already kmapped 486 */ 487 h = (struct btrfs_header *)buffer; 488 489 if (logical != le64_to_cpu(h->bytenr)) 490 ++fail; 491 492 if (spag->generation != le64_to_cpu(h->generation)) 493 ++fail; 494 495 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) 496 ++fail; 497 498 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, 499 BTRFS_UUID_SIZE)) 500 ++fail; 501 502 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc, 503 PAGE_SIZE - BTRFS_CSUM_SIZE); 504 btrfs_csum_final(crc, csum); 505 if (memcmp(csum, h->csum, sdev->csum_size)) 506 ++crc_fail; 507 508 spin_lock(&sdev->stat_lock); 509 ++sdev->stat.tree_extents_scrubbed; 510 sdev->stat.tree_bytes_scrubbed += PAGE_SIZE; 511 if (crc_fail) 512 ++sdev->stat.csum_errors; 513 if (fail) 514 ++sdev->stat.verify_errors; 515 spin_unlock(&sdev->stat_lock); 516 517 return fail || crc_fail; 518 } 519 520 static int scrub_checksum_super(struct scrub_bio *sbio, void *buffer) 521 { 522 struct btrfs_super_block *s; 523 u64 logical; 524 struct scrub_dev *sdev = sbio->sdev; 525 struct btrfs_root *root = sdev->dev->dev_root; 526 struct btrfs_fs_info *fs_info = root->fs_info; 527 u8 csum[BTRFS_CSUM_SIZE]; 528 u32 crc = ~(u32)0; 529 int fail = 0; 530 531 s = (struct btrfs_super_block *)buffer; 532 logical = sbio->logical; 533 534 if (logical != le64_to_cpu(s->bytenr)) 535 ++fail; 536 537 if (sbio->spag[0].generation != le64_to_cpu(s->generation)) 538 ++fail; 539 540 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) 541 ++fail; 542 543 crc = btrfs_csum_data(root, buffer + BTRFS_CSUM_SIZE, crc, 544 PAGE_SIZE - BTRFS_CSUM_SIZE); 545 btrfs_csum_final(crc, csum); 546 if (memcmp(csum, s->csum, sbio->sdev->csum_size)) 547 ++fail; 548 549 if (fail) { 550 /* 551 * if we find an error in a super block, we just report it. 552 * They will get written with the next transaction commit 553 * anyway 554 */ 555 spin_lock(&sdev->stat_lock); 556 ++sdev->stat.super_errors; 557 spin_unlock(&sdev->stat_lock); 558 } 559 560 return fail; 561 } 562 563 static int scrub_submit(struct scrub_dev *sdev) 564 { 565 struct scrub_bio *sbio; 566 struct bio *bio; 567 int i; 568 569 if (sdev->curr == -1) 570 return 0; 571 572 sbio = sdev->bios[sdev->curr]; 573 574 bio = bio_alloc(GFP_NOFS, sbio->count); 575 if (!bio) 576 goto nomem; 577 578 bio->bi_private = sbio; 579 bio->bi_end_io = scrub_bio_end_io; 580 bio->bi_bdev = sdev->dev->bdev; 581 bio->bi_sector = sbio->physical >> 9; 582 583 for (i = 0; i < sbio->count; ++i) { 584 struct page *page; 585 int ret; 586 587 page = alloc_page(GFP_NOFS); 588 if (!page) 589 goto nomem; 590 591 ret = bio_add_page(bio, page, PAGE_SIZE, 0); 592 if (!ret) { 593 __free_page(page); 594 goto nomem; 595 } 596 } 597 598 sbio->err = 0; 599 sdev->curr = -1; 600 atomic_inc(&sdev->in_flight); 601 602 submit_bio(READ, bio); 603 604 return 0; 605 606 nomem: 607 scrub_free_bio(bio); 608 609 return -ENOMEM; 610 } 611 612 static int scrub_page(struct scrub_dev *sdev, u64 logical, u64 len, 613 u64 physical, u64 flags, u64 gen, u64 mirror_num, 614 u8 *csum, int force) 615 { 616 struct scrub_bio *sbio; 617 618 again: 619 /* 620 * grab a fresh bio or wait for one to become available 621 */ 622 while (sdev->curr == -1) { 623 spin_lock(&sdev->list_lock); 624 sdev->curr = sdev->first_free; 625 if (sdev->curr != -1) { 626 sdev->first_free = sdev->bios[sdev->curr]->next_free; 627 sdev->bios[sdev->curr]->next_free = -1; 628 sdev->bios[sdev->curr]->count = 0; 629 spin_unlock(&sdev->list_lock); 630 } else { 631 spin_unlock(&sdev->list_lock); 632 wait_event(sdev->list_wait, sdev->first_free != -1); 633 } 634 } 635 sbio = sdev->bios[sdev->curr]; 636 if (sbio->count == 0) { 637 sbio->physical = physical; 638 sbio->logical = logical; 639 } else if (sbio->physical + sbio->count * PAGE_SIZE != physical || 640 sbio->logical + sbio->count * PAGE_SIZE != logical) { 641 int ret; 642 643 ret = scrub_submit(sdev); 644 if (ret) 645 return ret; 646 goto again; 647 } 648 sbio->spag[sbio->count].flags = flags; 649 sbio->spag[sbio->count].generation = gen; 650 sbio->spag[sbio->count].have_csum = 0; 651 sbio->spag[sbio->count].mirror_num = mirror_num; 652 if (csum) { 653 sbio->spag[sbio->count].have_csum = 1; 654 memcpy(sbio->spag[sbio->count].csum, csum, sdev->csum_size); 655 } 656 ++sbio->count; 657 if (sbio->count == SCRUB_PAGES_PER_BIO || force) { 658 int ret; 659 660 ret = scrub_submit(sdev); 661 if (ret) 662 return ret; 663 } 664 665 return 0; 666 } 667 668 static int scrub_find_csum(struct scrub_dev *sdev, u64 logical, u64 len, 669 u8 *csum) 670 { 671 struct btrfs_ordered_sum *sum = NULL; 672 int ret = 0; 673 unsigned long i; 674 unsigned long num_sectors; 675 u32 sectorsize = sdev->dev->dev_root->sectorsize; 676 677 while (!list_empty(&sdev->csum_list)) { 678 sum = list_first_entry(&sdev->csum_list, 679 struct btrfs_ordered_sum, list); 680 if (sum->bytenr > logical) 681 return 0; 682 if (sum->bytenr + sum->len > logical) 683 break; 684 685 ++sdev->stat.csum_discards; 686 list_del(&sum->list); 687 kfree(sum); 688 sum = NULL; 689 } 690 if (!sum) 691 return 0; 692 693 num_sectors = sum->len / sectorsize; 694 for (i = 0; i < num_sectors; ++i) { 695 if (sum->sums[i].bytenr == logical) { 696 memcpy(csum, &sum->sums[i].sum, sdev->csum_size); 697 ret = 1; 698 break; 699 } 700 } 701 if (ret && i == num_sectors - 1) { 702 list_del(&sum->list); 703 kfree(sum); 704 } 705 return ret; 706 } 707 708 /* scrub extent tries to collect up to 64 kB for each bio */ 709 static int scrub_extent(struct scrub_dev *sdev, u64 logical, u64 len, 710 u64 physical, u64 flags, u64 gen, u64 mirror_num) 711 { 712 int ret; 713 u8 csum[BTRFS_CSUM_SIZE]; 714 715 while (len) { 716 u64 l = min_t(u64, len, PAGE_SIZE); 717 int have_csum = 0; 718 719 if (flags & BTRFS_EXTENT_FLAG_DATA) { 720 /* push csums to sbio */ 721 have_csum = scrub_find_csum(sdev, logical, l, csum); 722 if (have_csum == 0) 723 ++sdev->stat.no_csum; 724 } 725 ret = scrub_page(sdev, logical, l, physical, flags, gen, 726 mirror_num, have_csum ? csum : NULL, 0); 727 if (ret) 728 return ret; 729 len -= l; 730 logical += l; 731 physical += l; 732 } 733 return 0; 734 } 735 736 static noinline_for_stack int scrub_stripe(struct scrub_dev *sdev, 737 struct map_lookup *map, int num, u64 base, u64 length) 738 { 739 struct btrfs_path *path; 740 struct btrfs_fs_info *fs_info = sdev->dev->dev_root->fs_info; 741 struct btrfs_root *root = fs_info->extent_root; 742 struct btrfs_root *csum_root = fs_info->csum_root; 743 struct btrfs_extent_item *extent; 744 struct blk_plug plug; 745 u64 flags; 746 int ret; 747 int slot; 748 int i; 749 u64 nstripes; 750 int start_stripe; 751 struct extent_buffer *l; 752 struct btrfs_key key; 753 u64 physical; 754 u64 logical; 755 u64 generation; 756 u64 mirror_num; 757 758 u64 increment = map->stripe_len; 759 u64 offset; 760 761 nstripes = length; 762 offset = 0; 763 do_div(nstripes, map->stripe_len); 764 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 765 offset = map->stripe_len * num; 766 increment = map->stripe_len * map->num_stripes; 767 mirror_num = 0; 768 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 769 int factor = map->num_stripes / map->sub_stripes; 770 offset = map->stripe_len * (num / map->sub_stripes); 771 increment = map->stripe_len * factor; 772 mirror_num = num % map->sub_stripes; 773 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { 774 increment = map->stripe_len; 775 mirror_num = num % map->num_stripes; 776 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 777 increment = map->stripe_len; 778 mirror_num = num % map->num_stripes; 779 } else { 780 increment = map->stripe_len; 781 mirror_num = 0; 782 } 783 784 path = btrfs_alloc_path(); 785 if (!path) 786 return -ENOMEM; 787 788 path->reada = 2; 789 path->search_commit_root = 1; 790 path->skip_locking = 1; 791 792 /* 793 * find all extents for each stripe and just read them to get 794 * them into the page cache 795 * FIXME: we can do better. build a more intelligent prefetching 796 */ 797 logical = base + offset; 798 physical = map->stripes[num].physical; 799 ret = 0; 800 for (i = 0; i < nstripes; ++i) { 801 key.objectid = logical; 802 key.type = BTRFS_EXTENT_ITEM_KEY; 803 key.offset = (u64)0; 804 805 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 806 if (ret < 0) 807 goto out; 808 809 l = path->nodes[0]; 810 slot = path->slots[0]; 811 btrfs_item_key_to_cpu(l, &key, slot); 812 if (key.objectid != logical) { 813 ret = btrfs_previous_item(root, path, 0, 814 BTRFS_EXTENT_ITEM_KEY); 815 if (ret < 0) 816 goto out; 817 } 818 819 while (1) { 820 l = path->nodes[0]; 821 slot = path->slots[0]; 822 if (slot >= btrfs_header_nritems(l)) { 823 ret = btrfs_next_leaf(root, path); 824 if (ret == 0) 825 continue; 826 if (ret < 0) 827 goto out; 828 829 break; 830 } 831 btrfs_item_key_to_cpu(l, &key, slot); 832 833 if (key.objectid >= logical + map->stripe_len) 834 break; 835 836 path->slots[0]++; 837 } 838 btrfs_release_path(path); 839 logical += increment; 840 physical += map->stripe_len; 841 cond_resched(); 842 } 843 844 /* 845 * collect all data csums for the stripe to avoid seeking during 846 * the scrub. This might currently (crc32) end up to be about 1MB 847 */ 848 start_stripe = 0; 849 blk_start_plug(&plug); 850 again: 851 logical = base + offset + start_stripe * increment; 852 for (i = start_stripe; i < nstripes; ++i) { 853 ret = btrfs_lookup_csums_range(csum_root, logical, 854 logical + map->stripe_len - 1, 855 &sdev->csum_list, 1); 856 if (ret) 857 goto out; 858 859 logical += increment; 860 cond_resched(); 861 } 862 /* 863 * now find all extents for each stripe and scrub them 864 */ 865 logical = base + offset + start_stripe * increment; 866 physical = map->stripes[num].physical + start_stripe * map->stripe_len; 867 ret = 0; 868 for (i = start_stripe; i < nstripes; ++i) { 869 /* 870 * canceled? 871 */ 872 if (atomic_read(&fs_info->scrub_cancel_req) || 873 atomic_read(&sdev->cancel_req)) { 874 ret = -ECANCELED; 875 goto out; 876 } 877 /* 878 * check to see if we have to pause 879 */ 880 if (atomic_read(&fs_info->scrub_pause_req)) { 881 /* push queued extents */ 882 scrub_submit(sdev); 883 wait_event(sdev->list_wait, 884 atomic_read(&sdev->in_flight) == 0); 885 atomic_inc(&fs_info->scrubs_paused); 886 wake_up(&fs_info->scrub_pause_wait); 887 mutex_lock(&fs_info->scrub_lock); 888 while (atomic_read(&fs_info->scrub_pause_req)) { 889 mutex_unlock(&fs_info->scrub_lock); 890 wait_event(fs_info->scrub_pause_wait, 891 atomic_read(&fs_info->scrub_pause_req) == 0); 892 mutex_lock(&fs_info->scrub_lock); 893 } 894 atomic_dec(&fs_info->scrubs_paused); 895 mutex_unlock(&fs_info->scrub_lock); 896 wake_up(&fs_info->scrub_pause_wait); 897 scrub_free_csums(sdev); 898 start_stripe = i; 899 goto again; 900 } 901 902 key.objectid = logical; 903 key.type = BTRFS_EXTENT_ITEM_KEY; 904 key.offset = (u64)0; 905 906 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 907 if (ret < 0) 908 goto out; 909 910 l = path->nodes[0]; 911 slot = path->slots[0]; 912 btrfs_item_key_to_cpu(l, &key, slot); 913 if (key.objectid != logical) { 914 ret = btrfs_previous_item(root, path, 0, 915 BTRFS_EXTENT_ITEM_KEY); 916 if (ret < 0) 917 goto out; 918 } 919 920 while (1) { 921 l = path->nodes[0]; 922 slot = path->slots[0]; 923 if (slot >= btrfs_header_nritems(l)) { 924 ret = btrfs_next_leaf(root, path); 925 if (ret == 0) 926 continue; 927 if (ret < 0) 928 goto out; 929 930 break; 931 } 932 btrfs_item_key_to_cpu(l, &key, slot); 933 934 if (key.objectid + key.offset <= logical) 935 goto next; 936 937 if (key.objectid >= logical + map->stripe_len) 938 break; 939 940 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) 941 goto next; 942 943 extent = btrfs_item_ptr(l, slot, 944 struct btrfs_extent_item); 945 flags = btrfs_extent_flags(l, extent); 946 generation = btrfs_extent_generation(l, extent); 947 948 if (key.objectid < logical && 949 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { 950 printk(KERN_ERR 951 "btrfs scrub: tree block %llu spanning " 952 "stripes, ignored. logical=%llu\n", 953 (unsigned long long)key.objectid, 954 (unsigned long long)logical); 955 goto next; 956 } 957 958 /* 959 * trim extent to this stripe 960 */ 961 if (key.objectid < logical) { 962 key.offset -= logical - key.objectid; 963 key.objectid = logical; 964 } 965 if (key.objectid + key.offset > 966 logical + map->stripe_len) { 967 key.offset = logical + map->stripe_len - 968 key.objectid; 969 } 970 971 ret = scrub_extent(sdev, key.objectid, key.offset, 972 key.objectid - logical + physical, 973 flags, generation, mirror_num); 974 if (ret) 975 goto out; 976 977 next: 978 path->slots[0]++; 979 } 980 btrfs_release_path(path); 981 logical += increment; 982 physical += map->stripe_len; 983 spin_lock(&sdev->stat_lock); 984 sdev->stat.last_physical = physical; 985 spin_unlock(&sdev->stat_lock); 986 } 987 /* push queued extents */ 988 scrub_submit(sdev); 989 990 out: 991 blk_finish_plug(&plug); 992 btrfs_free_path(path); 993 return ret < 0 ? ret : 0; 994 } 995 996 static noinline_for_stack int scrub_chunk(struct scrub_dev *sdev, 997 u64 chunk_tree, u64 chunk_objectid, u64 chunk_offset, u64 length) 998 { 999 struct btrfs_mapping_tree *map_tree = 1000 &sdev->dev->dev_root->fs_info->mapping_tree; 1001 struct map_lookup *map; 1002 struct extent_map *em; 1003 int i; 1004 int ret = -EINVAL; 1005 1006 read_lock(&map_tree->map_tree.lock); 1007 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); 1008 read_unlock(&map_tree->map_tree.lock); 1009 1010 if (!em) 1011 return -EINVAL; 1012 1013 map = (struct map_lookup *)em->bdev; 1014 if (em->start != chunk_offset) 1015 goto out; 1016 1017 if (em->len < length) 1018 goto out; 1019 1020 for (i = 0; i < map->num_stripes; ++i) { 1021 if (map->stripes[i].dev == sdev->dev) { 1022 ret = scrub_stripe(sdev, map, i, chunk_offset, length); 1023 if (ret) 1024 goto out; 1025 } 1026 } 1027 out: 1028 free_extent_map(em); 1029 1030 return ret; 1031 } 1032 1033 static noinline_for_stack 1034 int scrub_enumerate_chunks(struct scrub_dev *sdev, u64 start, u64 end) 1035 { 1036 struct btrfs_dev_extent *dev_extent = NULL; 1037 struct btrfs_path *path; 1038 struct btrfs_root *root = sdev->dev->dev_root; 1039 struct btrfs_fs_info *fs_info = root->fs_info; 1040 u64 length; 1041 u64 chunk_tree; 1042 u64 chunk_objectid; 1043 u64 chunk_offset; 1044 int ret; 1045 int slot; 1046 struct extent_buffer *l; 1047 struct btrfs_key key; 1048 struct btrfs_key found_key; 1049 struct btrfs_block_group_cache *cache; 1050 1051 path = btrfs_alloc_path(); 1052 if (!path) 1053 return -ENOMEM; 1054 1055 path->reada = 2; 1056 path->search_commit_root = 1; 1057 path->skip_locking = 1; 1058 1059 key.objectid = sdev->dev->devid; 1060 key.offset = 0ull; 1061 key.type = BTRFS_DEV_EXTENT_KEY; 1062 1063 1064 while (1) { 1065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 1066 if (ret < 0) 1067 goto out; 1068 ret = 0; 1069 1070 l = path->nodes[0]; 1071 slot = path->slots[0]; 1072 1073 btrfs_item_key_to_cpu(l, &found_key, slot); 1074 1075 if (found_key.objectid != sdev->dev->devid) 1076 break; 1077 1078 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) 1079 break; 1080 1081 if (found_key.offset >= end) 1082 break; 1083 1084 if (found_key.offset < key.offset) 1085 break; 1086 1087 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1088 length = btrfs_dev_extent_length(l, dev_extent); 1089 1090 if (found_key.offset + length <= start) { 1091 key.offset = found_key.offset + length; 1092 btrfs_release_path(path); 1093 continue; 1094 } 1095 1096 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); 1097 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); 1098 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 1099 1100 /* 1101 * get a reference on the corresponding block group to prevent 1102 * the chunk from going away while we scrub it 1103 */ 1104 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 1105 if (!cache) { 1106 ret = -ENOENT; 1107 goto out; 1108 } 1109 ret = scrub_chunk(sdev, chunk_tree, chunk_objectid, 1110 chunk_offset, length); 1111 btrfs_put_block_group(cache); 1112 if (ret) 1113 break; 1114 1115 key.offset = found_key.offset + length; 1116 btrfs_release_path(path); 1117 } 1118 1119 out: 1120 btrfs_free_path(path); 1121 return ret; 1122 } 1123 1124 static noinline_for_stack int scrub_supers(struct scrub_dev *sdev) 1125 { 1126 int i; 1127 u64 bytenr; 1128 u64 gen; 1129 int ret; 1130 struct btrfs_device *device = sdev->dev; 1131 struct btrfs_root *root = device->dev_root; 1132 1133 gen = root->fs_info->last_trans_committed; 1134 1135 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 1136 bytenr = btrfs_sb_offset(i); 1137 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 1138 break; 1139 1140 ret = scrub_page(sdev, bytenr, PAGE_SIZE, bytenr, 1141 BTRFS_EXTENT_FLAG_SUPER, gen, i, NULL, 1); 1142 if (ret) 1143 return ret; 1144 } 1145 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0); 1146 1147 return 0; 1148 } 1149 1150 /* 1151 * get a reference count on fs_info->scrub_workers. start worker if necessary 1152 */ 1153 static noinline_for_stack int scrub_workers_get(struct btrfs_root *root) 1154 { 1155 struct btrfs_fs_info *fs_info = root->fs_info; 1156 1157 mutex_lock(&fs_info->scrub_lock); 1158 if (fs_info->scrub_workers_refcnt == 0) 1159 btrfs_start_workers(&fs_info->scrub_workers, 1); 1160 ++fs_info->scrub_workers_refcnt; 1161 mutex_unlock(&fs_info->scrub_lock); 1162 1163 return 0; 1164 } 1165 1166 static noinline_for_stack void scrub_workers_put(struct btrfs_root *root) 1167 { 1168 struct btrfs_fs_info *fs_info = root->fs_info; 1169 1170 mutex_lock(&fs_info->scrub_lock); 1171 if (--fs_info->scrub_workers_refcnt == 0) 1172 btrfs_stop_workers(&fs_info->scrub_workers); 1173 WARN_ON(fs_info->scrub_workers_refcnt < 0); 1174 mutex_unlock(&fs_info->scrub_lock); 1175 } 1176 1177 1178 int btrfs_scrub_dev(struct btrfs_root *root, u64 devid, u64 start, u64 end, 1179 struct btrfs_scrub_progress *progress, int readonly) 1180 { 1181 struct scrub_dev *sdev; 1182 struct btrfs_fs_info *fs_info = root->fs_info; 1183 int ret; 1184 struct btrfs_device *dev; 1185 1186 if (btrfs_fs_closing(root->fs_info)) 1187 return -EINVAL; 1188 1189 /* 1190 * check some assumptions 1191 */ 1192 if (root->sectorsize != PAGE_SIZE || 1193 root->sectorsize != root->leafsize || 1194 root->sectorsize != root->nodesize) { 1195 printk(KERN_ERR "btrfs_scrub: size assumptions fail\n"); 1196 return -EINVAL; 1197 } 1198 1199 ret = scrub_workers_get(root); 1200 if (ret) 1201 return ret; 1202 1203 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 1204 dev = btrfs_find_device(root, devid, NULL, NULL); 1205 if (!dev || dev->missing) { 1206 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1207 scrub_workers_put(root); 1208 return -ENODEV; 1209 } 1210 mutex_lock(&fs_info->scrub_lock); 1211 1212 if (!dev->in_fs_metadata) { 1213 mutex_unlock(&fs_info->scrub_lock); 1214 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1215 scrub_workers_put(root); 1216 return -ENODEV; 1217 } 1218 1219 if (dev->scrub_device) { 1220 mutex_unlock(&fs_info->scrub_lock); 1221 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1222 scrub_workers_put(root); 1223 return -EINPROGRESS; 1224 } 1225 sdev = scrub_setup_dev(dev); 1226 if (IS_ERR(sdev)) { 1227 mutex_unlock(&fs_info->scrub_lock); 1228 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1229 scrub_workers_put(root); 1230 return PTR_ERR(sdev); 1231 } 1232 sdev->readonly = readonly; 1233 dev->scrub_device = sdev; 1234 1235 atomic_inc(&fs_info->scrubs_running); 1236 mutex_unlock(&fs_info->scrub_lock); 1237 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1238 1239 down_read(&fs_info->scrub_super_lock); 1240 ret = scrub_supers(sdev); 1241 up_read(&fs_info->scrub_super_lock); 1242 1243 if (!ret) 1244 ret = scrub_enumerate_chunks(sdev, start, end); 1245 1246 wait_event(sdev->list_wait, atomic_read(&sdev->in_flight) == 0); 1247 1248 atomic_dec(&fs_info->scrubs_running); 1249 wake_up(&fs_info->scrub_pause_wait); 1250 1251 if (progress) 1252 memcpy(progress, &sdev->stat, sizeof(*progress)); 1253 1254 mutex_lock(&fs_info->scrub_lock); 1255 dev->scrub_device = NULL; 1256 mutex_unlock(&fs_info->scrub_lock); 1257 1258 scrub_free_dev(sdev); 1259 scrub_workers_put(root); 1260 1261 return ret; 1262 } 1263 1264 int btrfs_scrub_pause(struct btrfs_root *root) 1265 { 1266 struct btrfs_fs_info *fs_info = root->fs_info; 1267 1268 mutex_lock(&fs_info->scrub_lock); 1269 atomic_inc(&fs_info->scrub_pause_req); 1270 while (atomic_read(&fs_info->scrubs_paused) != 1271 atomic_read(&fs_info->scrubs_running)) { 1272 mutex_unlock(&fs_info->scrub_lock); 1273 wait_event(fs_info->scrub_pause_wait, 1274 atomic_read(&fs_info->scrubs_paused) == 1275 atomic_read(&fs_info->scrubs_running)); 1276 mutex_lock(&fs_info->scrub_lock); 1277 } 1278 mutex_unlock(&fs_info->scrub_lock); 1279 1280 return 0; 1281 } 1282 1283 int btrfs_scrub_continue(struct btrfs_root *root) 1284 { 1285 struct btrfs_fs_info *fs_info = root->fs_info; 1286 1287 atomic_dec(&fs_info->scrub_pause_req); 1288 wake_up(&fs_info->scrub_pause_wait); 1289 return 0; 1290 } 1291 1292 int btrfs_scrub_pause_super(struct btrfs_root *root) 1293 { 1294 down_write(&root->fs_info->scrub_super_lock); 1295 return 0; 1296 } 1297 1298 int btrfs_scrub_continue_super(struct btrfs_root *root) 1299 { 1300 up_write(&root->fs_info->scrub_super_lock); 1301 return 0; 1302 } 1303 1304 int btrfs_scrub_cancel(struct btrfs_root *root) 1305 { 1306 struct btrfs_fs_info *fs_info = root->fs_info; 1307 1308 mutex_lock(&fs_info->scrub_lock); 1309 if (!atomic_read(&fs_info->scrubs_running)) { 1310 mutex_unlock(&fs_info->scrub_lock); 1311 return -ENOTCONN; 1312 } 1313 1314 atomic_inc(&fs_info->scrub_cancel_req); 1315 while (atomic_read(&fs_info->scrubs_running)) { 1316 mutex_unlock(&fs_info->scrub_lock); 1317 wait_event(fs_info->scrub_pause_wait, 1318 atomic_read(&fs_info->scrubs_running) == 0); 1319 mutex_lock(&fs_info->scrub_lock); 1320 } 1321 atomic_dec(&fs_info->scrub_cancel_req); 1322 mutex_unlock(&fs_info->scrub_lock); 1323 1324 return 0; 1325 } 1326 1327 int btrfs_scrub_cancel_dev(struct btrfs_root *root, struct btrfs_device *dev) 1328 { 1329 struct btrfs_fs_info *fs_info = root->fs_info; 1330 struct scrub_dev *sdev; 1331 1332 mutex_lock(&fs_info->scrub_lock); 1333 sdev = dev->scrub_device; 1334 if (!sdev) { 1335 mutex_unlock(&fs_info->scrub_lock); 1336 return -ENOTCONN; 1337 } 1338 atomic_inc(&sdev->cancel_req); 1339 while (dev->scrub_device) { 1340 mutex_unlock(&fs_info->scrub_lock); 1341 wait_event(fs_info->scrub_pause_wait, 1342 dev->scrub_device == NULL); 1343 mutex_lock(&fs_info->scrub_lock); 1344 } 1345 mutex_unlock(&fs_info->scrub_lock); 1346 1347 return 0; 1348 } 1349 int btrfs_scrub_cancel_devid(struct btrfs_root *root, u64 devid) 1350 { 1351 struct btrfs_fs_info *fs_info = root->fs_info; 1352 struct btrfs_device *dev; 1353 int ret; 1354 1355 /* 1356 * we have to hold the device_list_mutex here so the device 1357 * does not go away in cancel_dev. FIXME: find a better solution 1358 */ 1359 mutex_lock(&fs_info->fs_devices->device_list_mutex); 1360 dev = btrfs_find_device(root, devid, NULL, NULL); 1361 if (!dev) { 1362 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 1363 return -ENODEV; 1364 } 1365 ret = btrfs_scrub_cancel_dev(root, dev); 1366 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 1367 1368 return ret; 1369 } 1370 1371 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, 1372 struct btrfs_scrub_progress *progress) 1373 { 1374 struct btrfs_device *dev; 1375 struct scrub_dev *sdev = NULL; 1376 1377 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 1378 dev = btrfs_find_device(root, devid, NULL, NULL); 1379 if (dev) 1380 sdev = dev->scrub_device; 1381 if (sdev) 1382 memcpy(progress, &sdev->stat, sizeof(*progress)); 1383 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 1384 1385 return dev ? (sdev ? 0 : -ENOTCONN) : -ENODEV; 1386 } 1387