1// SPDX-License-Identifier: GPL-2.0+ 2/* 3 * NILFS recovery logic 4 * 5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation. 6 * 7 * Written by Ryusuke Konishi. 8 */ 9 10#include <linux/buffer_head.h> 11#include <linux/blkdev.h> 12#include <linux/swap.h> 13#include <linux/slab.h> 14#include <linux/crc32.h> 15#include "nilfs.h" 16#include "segment.h" 17#include "sufile.h" 18#include "page.h" 19#include "segbuf.h" 20 21/* 22 * Segment check result 23 */ 24enum { 25 NILFS_SEG_VALID, 26 NILFS_SEG_NO_SUPER_ROOT, 27 NILFS_SEG_FAIL_IO, 28 NILFS_SEG_FAIL_MAGIC, 29 NILFS_SEG_FAIL_SEQ, 30 NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT, 31 NILFS_SEG_FAIL_CHECKSUM_FULL, 32 NILFS_SEG_FAIL_CONSISTENCY, 33}; 34 35/* work structure for recovery */ 36struct nilfs_recovery_block { 37 ino_t ino; /* 38 * Inode number of the file that this block 39 * belongs to 40 */ 41 sector_t blocknr; /* block number */ 42 __u64 vblocknr; /* virtual block number */ 43 unsigned long blkoff; /* File offset of the data block (per block) */ 44 struct list_head list; 45}; 46 47 48static int nilfs_warn_segment_error(struct super_block *sb, int err) 49{ 50 const char *msg = NULL; 51 52 switch (err) { 53 case NILFS_SEG_FAIL_IO: 54 nilfs_err(sb, "I/O error reading segment"); 55 return -EIO; 56 case NILFS_SEG_FAIL_MAGIC: 57 msg = "Magic number mismatch"; 58 break; 59 case NILFS_SEG_FAIL_SEQ: 60 msg = "Sequence number mismatch"; 61 break; 62 case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT: 63 msg = "Checksum error in super root"; 64 break; 65 case NILFS_SEG_FAIL_CHECKSUM_FULL: 66 msg = "Checksum error in segment payload"; 67 break; 68 case NILFS_SEG_FAIL_CONSISTENCY: 69 msg = "Inconsistency found"; 70 break; 71 case NILFS_SEG_NO_SUPER_ROOT: 72 msg = "No super root in the last segment"; 73 break; 74 default: 75 nilfs_err(sb, "unrecognized segment error %d", err); 76 return -EINVAL; 77 } 78 nilfs_warn(sb, "invalid segment: %s", msg); 79 return -EINVAL; 80} 81 82/** 83 * nilfs_compute_checksum - compute checksum of blocks continuously 84 * @nilfs: nilfs object 85 * @bhs: buffer head of start block 86 * @sum: place to store result 87 * @offset: offset bytes in the first block 88 * @check_bytes: number of bytes to be checked 89 * @start: DBN of start block 90 * @nblock: number of blocks to be checked 91 */ 92static int nilfs_compute_checksum(struct the_nilfs *nilfs, 93 struct buffer_head *bhs, u32 *sum, 94 unsigned long offset, u64 check_bytes, 95 sector_t start, unsigned long nblock) 96{ 97 unsigned int blocksize = nilfs->ns_blocksize; 98 unsigned long size; 99 u32 crc; 100 101 BUG_ON(offset >= blocksize); 102 check_bytes -= offset; 103 size = min_t(u64, check_bytes, blocksize - offset); 104 crc = crc32_le(nilfs->ns_crc_seed, 105 (unsigned char *)bhs->b_data + offset, size); 106 if (--nblock > 0) { 107 do { 108 struct buffer_head *bh; 109 110 bh = __bread(nilfs->ns_bdev, ++start, blocksize); 111 if (!bh) 112 return -EIO; 113 check_bytes -= size; 114 size = min_t(u64, check_bytes, blocksize); 115 crc = crc32_le(crc, bh->b_data, size); 116 brelse(bh); 117 } while (--nblock > 0); 118 } 119 *sum = crc; 120 return 0; 121} 122 123/** 124 * nilfs_read_super_root_block - read super root block 125 * @nilfs: nilfs object 126 * @sr_block: disk block number of the super root block 127 * @pbh: address of a buffer_head pointer to return super root buffer 128 * @check: CRC check flag 129 */ 130int nilfs_read_super_root_block(struct the_nilfs *nilfs, sector_t sr_block, 131 struct buffer_head **pbh, int check) 132{ 133 struct buffer_head *bh_sr; 134 struct nilfs_super_root *sr; 135 u32 crc; 136 int ret; 137 138 *pbh = NULL; 139 bh_sr = __bread(nilfs->ns_bdev, sr_block, nilfs->ns_blocksize); 140 if (unlikely(!bh_sr)) { 141 ret = NILFS_SEG_FAIL_IO; 142 goto failed; 143 } 144 145 sr = (struct nilfs_super_root *)bh_sr->b_data; 146 if (check) { 147 unsigned int bytes = le16_to_cpu(sr->sr_bytes); 148 149 if (bytes == 0 || bytes > nilfs->ns_blocksize) { 150 ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT; 151 goto failed_bh; 152 } 153 if (nilfs_compute_checksum( 154 nilfs, bh_sr, &crc, sizeof(sr->sr_sum), bytes, 155 sr_block, 1)) { 156 ret = NILFS_SEG_FAIL_IO; 157 goto failed_bh; 158 } 159 if (crc != le32_to_cpu(sr->sr_sum)) { 160 ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT; 161 goto failed_bh; 162 } 163 } 164 *pbh = bh_sr; 165 return 0; 166 167 failed_bh: 168 brelse(bh_sr); 169 170 failed: 171 return nilfs_warn_segment_error(nilfs->ns_sb, ret); 172} 173 174/** 175 * nilfs_read_log_header - read summary header of the specified log 176 * @nilfs: nilfs object 177 * @start_blocknr: start block number of the log 178 * @sum: pointer to return segment summary structure 179 */ 180static struct buffer_head * 181nilfs_read_log_header(struct the_nilfs *nilfs, sector_t start_blocknr, 182 struct nilfs_segment_summary **sum) 183{ 184 struct buffer_head *bh_sum; 185 186 bh_sum = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize); 187 if (bh_sum) 188 *sum = (struct nilfs_segment_summary *)bh_sum->b_data; 189 return bh_sum; 190} 191 192/** 193 * nilfs_validate_log - verify consistency of log 194 * @nilfs: nilfs object 195 * @seg_seq: sequence number of segment 196 * @bh_sum: buffer head of summary block 197 * @sum: segment summary struct 198 */ 199static int nilfs_validate_log(struct the_nilfs *nilfs, u64 seg_seq, 200 struct buffer_head *bh_sum, 201 struct nilfs_segment_summary *sum) 202{ 203 unsigned long nblock; 204 u32 crc; 205 int ret; 206 207 ret = NILFS_SEG_FAIL_MAGIC; 208 if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC) 209 goto out; 210 211 ret = NILFS_SEG_FAIL_SEQ; 212 if (le64_to_cpu(sum->ss_seq) != seg_seq) 213 goto out; 214 215 nblock = le32_to_cpu(sum->ss_nblocks); 216 ret = NILFS_SEG_FAIL_CONSISTENCY; 217 if (unlikely(nblock == 0 || nblock > nilfs->ns_blocks_per_segment)) 218 /* This limits the number of blocks read in the CRC check */ 219 goto out; 220 221 ret = NILFS_SEG_FAIL_IO; 222 if (nilfs_compute_checksum(nilfs, bh_sum, &crc, sizeof(sum->ss_datasum), 223 ((u64)nblock << nilfs->ns_blocksize_bits), 224 bh_sum->b_blocknr, nblock)) 225 goto out; 226 227 ret = NILFS_SEG_FAIL_CHECKSUM_FULL; 228 if (crc != le32_to_cpu(sum->ss_datasum)) 229 goto out; 230 ret = 0; 231out: 232 return ret; 233} 234 235/** 236 * nilfs_read_summary_info - read an item on summary blocks of a log 237 * @nilfs: nilfs object 238 * @pbh: the current buffer head on summary blocks [in, out] 239 * @offset: the current byte offset on summary blocks [in, out] 240 * @bytes: byte size of the item to be read 241 */ 242static void *nilfs_read_summary_info(struct the_nilfs *nilfs, 243 struct buffer_head **pbh, 244 unsigned int *offset, unsigned int bytes) 245{ 246 void *ptr; 247 sector_t blocknr; 248 249 BUG_ON((*pbh)->b_size < *offset); 250 if (bytes > (*pbh)->b_size - *offset) { 251 blocknr = (*pbh)->b_blocknr; 252 brelse(*pbh); 253 *pbh = __bread(nilfs->ns_bdev, blocknr + 1, 254 nilfs->ns_blocksize); 255 if (unlikely(!*pbh)) 256 return NULL; 257 *offset = 0; 258 } 259 ptr = (*pbh)->b_data + *offset; 260 *offset += bytes; 261 return ptr; 262} 263 264/** 265 * nilfs_skip_summary_info - skip items on summary blocks of a log 266 * @nilfs: nilfs object 267 * @pbh: the current buffer head on summary blocks [in, out] 268 * @offset: the current byte offset on summary blocks [in, out] 269 * @bytes: byte size of the item to be skipped 270 * @count: number of items to be skipped 271 */ 272static void nilfs_skip_summary_info(struct the_nilfs *nilfs, 273 struct buffer_head **pbh, 274 unsigned int *offset, unsigned int bytes, 275 unsigned long count) 276{ 277 unsigned int rest_item_in_current_block 278 = ((*pbh)->b_size - *offset) / bytes; 279 280 if (count <= rest_item_in_current_block) { 281 *offset += bytes * count; 282 } else { 283 sector_t blocknr = (*pbh)->b_blocknr; 284 unsigned int nitem_per_block = (*pbh)->b_size / bytes; 285 unsigned int bcnt; 286 287 count -= rest_item_in_current_block; 288 bcnt = DIV_ROUND_UP(count, nitem_per_block); 289 *offset = bytes * (count - (bcnt - 1) * nitem_per_block); 290 291 brelse(*pbh); 292 *pbh = __bread(nilfs->ns_bdev, blocknr + bcnt, 293 nilfs->ns_blocksize); 294 } 295} 296 297/** 298 * nilfs_scan_dsync_log - get block information of a log written for data sync 299 * @nilfs: nilfs object 300 * @start_blocknr: start block number of the log 301 * @sum: log summary information 302 * @head: list head to add nilfs_recovery_block struct 303 */ 304static int nilfs_scan_dsync_log(struct the_nilfs *nilfs, sector_t start_blocknr, 305 struct nilfs_segment_summary *sum, 306 struct list_head *head) 307{ 308 struct buffer_head *bh; 309 unsigned int offset; 310 u32 nfinfo, sumbytes; 311 sector_t blocknr; 312 ino_t ino; 313 int err = -EIO; 314 315 nfinfo = le32_to_cpu(sum->ss_nfinfo); 316 if (!nfinfo) 317 return 0; 318 319 sumbytes = le32_to_cpu(sum->ss_sumbytes); 320 blocknr = start_blocknr + DIV_ROUND_UP(sumbytes, nilfs->ns_blocksize); 321 bh = __bread(nilfs->ns_bdev, start_blocknr, nilfs->ns_blocksize); 322 if (unlikely(!bh)) 323 goto out; 324 325 offset = le16_to_cpu(sum->ss_bytes); 326 for (;;) { 327 unsigned long nblocks, ndatablk, nnodeblk; 328 struct nilfs_finfo *finfo; 329 330 finfo = nilfs_read_summary_info(nilfs, &bh, &offset, 331 sizeof(*finfo)); 332 if (unlikely(!finfo)) 333 goto out; 334 335 ino = le64_to_cpu(finfo->fi_ino); 336 nblocks = le32_to_cpu(finfo->fi_nblocks); 337 ndatablk = le32_to_cpu(finfo->fi_ndatablk); 338 nnodeblk = nblocks - ndatablk; 339 340 while (ndatablk-- > 0) { 341 struct nilfs_recovery_block *rb; 342 struct nilfs_binfo_v *binfo; 343 344 binfo = nilfs_read_summary_info(nilfs, &bh, &offset, 345 sizeof(*binfo)); 346 if (unlikely(!binfo)) 347 goto out; 348 349 rb = kmalloc(sizeof(*rb), GFP_NOFS); 350 if (unlikely(!rb)) { 351 err = -ENOMEM; 352 goto out; 353 } 354 rb->ino = ino; 355 rb->blocknr = blocknr++; 356 rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr); 357 rb->blkoff = le64_to_cpu(binfo->bi_blkoff); 358 /* INIT_LIST_HEAD(&rb->list); */ 359 list_add_tail(&rb->list, head); 360 } 361 if (--nfinfo == 0) 362 break; 363 blocknr += nnodeblk; /* always 0 for data sync logs */ 364 nilfs_skip_summary_info(nilfs, &bh, &offset, sizeof(__le64), 365 nnodeblk); 366 if (unlikely(!bh)) 367 goto out; 368 } 369 err = 0; 370 out: 371 brelse(bh); /* brelse(NULL) is just ignored */ 372 return err; 373} 374 375static void dispose_recovery_list(struct list_head *head) 376{ 377 while (!list_empty(head)) { 378 struct nilfs_recovery_block *rb; 379 380 rb = list_first_entry(head, struct nilfs_recovery_block, list); 381 list_del(&rb->list); 382 kfree(rb); 383 } 384} 385 386struct nilfs_segment_entry { 387 struct list_head list; 388 __u64 segnum; 389}; 390 391static int nilfs_segment_list_add(struct list_head *head, __u64 segnum) 392{ 393 struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS); 394 395 if (unlikely(!ent)) 396 return -ENOMEM; 397 398 ent->segnum = segnum; 399 INIT_LIST_HEAD(&ent->list); 400 list_add_tail(&ent->list, head); 401 return 0; 402} 403 404void nilfs_dispose_segment_list(struct list_head *head) 405{ 406 while (!list_empty(head)) { 407 struct nilfs_segment_entry *ent; 408 409 ent = list_first_entry(head, struct nilfs_segment_entry, list); 410 list_del(&ent->list); 411 kfree(ent); 412 } 413} 414 415static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs, 416 struct super_block *sb, 417 struct nilfs_recovery_info *ri) 418{ 419 struct list_head *head = &ri->ri_used_segments; 420 struct nilfs_segment_entry *ent, *n; 421 struct inode *sufile = nilfs->ns_sufile; 422 __u64 segnum[4]; 423 int err; 424 int i; 425 426 segnum[0] = nilfs->ns_segnum; 427 segnum[1] = nilfs->ns_nextnum; 428 segnum[2] = ri->ri_segnum; 429 segnum[3] = ri->ri_nextnum; 430 431 /* 432 * Releasing the next segment of the latest super root. 433 * The next segment is invalidated by this recovery. 434 */ 435 err = nilfs_sufile_free(sufile, segnum[1]); 436 if (unlikely(err)) 437 goto failed; 438 439 for (i = 1; i < 4; i++) { 440 err = nilfs_segment_list_add(head, segnum[i]); 441 if (unlikely(err)) 442 goto failed; 443 } 444 445 /* 446 * Collecting segments written after the latest super root. 447 * These are marked dirty to avoid being reallocated in the next write. 448 */ 449 list_for_each_entry_safe(ent, n, head, list) { 450 if (ent->segnum != segnum[0]) { 451 err = nilfs_sufile_scrap(sufile, ent->segnum); 452 if (unlikely(err)) 453 goto failed; 454 } 455 list_del(&ent->list); 456 kfree(ent); 457 } 458 459 /* Allocate new segments for recovery */ 460 err = nilfs_sufile_alloc(sufile, &segnum[0]); 461 if (unlikely(err)) 462 goto failed; 463 464 nilfs->ns_pseg_offset = 0; 465 nilfs->ns_seg_seq = ri->ri_seq + 2; 466 nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0]; 467 468 failed: 469 /* No need to recover sufile because it will be destroyed on error */ 470 return err; 471} 472 473static int nilfs_recovery_copy_block(struct the_nilfs *nilfs, 474 struct nilfs_recovery_block *rb, 475 loff_t pos, struct page *page) 476{ 477 struct buffer_head *bh_org; 478 size_t from = pos & ~PAGE_MASK; 479 void *kaddr; 480 481 bh_org = __bread(nilfs->ns_bdev, rb->blocknr, nilfs->ns_blocksize); 482 if (unlikely(!bh_org)) 483 return -EIO; 484 485 kaddr = kmap_local_page(page); 486 memcpy(kaddr + from, bh_org->b_data, bh_org->b_size); 487 kunmap_local(kaddr); 488 brelse(bh_org); 489 return 0; 490} 491 492static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs, 493 struct super_block *sb, 494 struct nilfs_root *root, 495 struct list_head *head, 496 unsigned long *nr_salvaged_blocks) 497{ 498 struct inode *inode; 499 struct nilfs_recovery_block *rb, *n; 500 unsigned int blocksize = nilfs->ns_blocksize; 501 struct page *page; 502 loff_t pos; 503 int err = 0, err2 = 0; 504 505 list_for_each_entry_safe(rb, n, head, list) { 506 inode = nilfs_iget(sb, root, rb->ino); 507 if (IS_ERR(inode)) { 508 err = PTR_ERR(inode); 509 inode = NULL; 510 goto failed_inode; 511 } 512 513 pos = rb->blkoff << inode->i_blkbits; 514 err = block_write_begin(inode->i_mapping, pos, blocksize, 515 &page, nilfs_get_block); 516 if (unlikely(err)) { 517 loff_t isize = inode->i_size; 518 519 if (pos + blocksize > isize) 520 nilfs_write_failed(inode->i_mapping, 521 pos + blocksize); 522 goto failed_inode; 523 } 524 525 err = nilfs_recovery_copy_block(nilfs, rb, pos, page); 526 if (unlikely(err)) 527 goto failed_page; 528 529 err = nilfs_set_file_dirty(inode, 1); 530 if (unlikely(err)) 531 goto failed_page; 532 533 block_write_end(NULL, inode->i_mapping, pos, blocksize, 534 blocksize, page, NULL); 535 536 unlock_page(page); 537 put_page(page); 538 539 (*nr_salvaged_blocks)++; 540 goto next; 541 542 failed_page: 543 unlock_page(page); 544 put_page(page); 545 546 failed_inode: 547 nilfs_warn(sb, 548 "error %d recovering data block (ino=%lu, block-offset=%llu)", 549 err, (unsigned long)rb->ino, 550 (unsigned long long)rb->blkoff); 551 if (!err2) 552 err2 = err; 553 next: 554 iput(inode); /* iput(NULL) is just ignored */ 555 list_del_init(&rb->list); 556 kfree(rb); 557 } 558 return err2; 559} 560 561/** 562 * nilfs_do_roll_forward - salvage logical segments newer than the latest 563 * checkpoint 564 * @nilfs: nilfs object 565 * @sb: super block instance 566 * @ri: pointer to a nilfs_recovery_info 567 */ 568static int nilfs_do_roll_forward(struct the_nilfs *nilfs, 569 struct super_block *sb, 570 struct nilfs_root *root, 571 struct nilfs_recovery_info *ri) 572{ 573 struct buffer_head *bh_sum = NULL; 574 struct nilfs_segment_summary *sum = NULL; 575 sector_t pseg_start; 576 sector_t seg_start, seg_end; /* Starting/ending DBN of full segment */ 577 unsigned long nsalvaged_blocks = 0; 578 unsigned int flags; 579 u64 seg_seq; 580 __u64 segnum, nextnum = 0; 581 int empty_seg = 0; 582 int err = 0, ret; 583 LIST_HEAD(dsync_blocks); /* list of data blocks to be recovered */ 584 enum { 585 RF_INIT_ST, 586 RF_DSYNC_ST, /* scanning data-sync segments */ 587 }; 588 int state = RF_INIT_ST; 589 590 pseg_start = ri->ri_lsegs_start; 591 seg_seq = ri->ri_lsegs_start_seq; 592 segnum = nilfs_get_segnum_of_block(nilfs, pseg_start); 593 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); 594 595 while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) { 596 brelse(bh_sum); 597 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum); 598 if (!bh_sum) { 599 err = -EIO; 600 goto failed; 601 } 602 603 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum); 604 if (ret) { 605 if (ret == NILFS_SEG_FAIL_IO) { 606 err = -EIO; 607 goto failed; 608 } 609 goto strayed; 610 } 611 612 flags = le16_to_cpu(sum->ss_flags); 613 if (flags & NILFS_SS_SR) 614 goto confused; 615 616 /* Found a valid partial segment; do recovery actions */ 617 nextnum = nilfs_get_segnum_of_block(nilfs, 618 le64_to_cpu(sum->ss_next)); 619 empty_seg = 0; 620 nilfs->ns_ctime = le64_to_cpu(sum->ss_create); 621 if (!(flags & NILFS_SS_GC)) 622 nilfs->ns_nongc_ctime = nilfs->ns_ctime; 623 624 switch (state) { 625 case RF_INIT_ST: 626 if (!(flags & NILFS_SS_LOGBGN) || 627 !(flags & NILFS_SS_SYNDT)) 628 goto try_next_pseg; 629 state = RF_DSYNC_ST; 630 fallthrough; 631 case RF_DSYNC_ST: 632 if (!(flags & NILFS_SS_SYNDT)) 633 goto confused; 634 635 err = nilfs_scan_dsync_log(nilfs, pseg_start, sum, 636 &dsync_blocks); 637 if (unlikely(err)) 638 goto failed; 639 if (flags & NILFS_SS_LOGEND) { 640 err = nilfs_recover_dsync_blocks( 641 nilfs, sb, root, &dsync_blocks, 642 &nsalvaged_blocks); 643 if (unlikely(err)) 644 goto failed; 645 state = RF_INIT_ST; 646 } 647 break; /* Fall through to try_next_pseg */ 648 } 649 650 try_next_pseg: 651 if (pseg_start == ri->ri_lsegs_end) 652 break; 653 pseg_start += le32_to_cpu(sum->ss_nblocks); 654 if (pseg_start < seg_end) 655 continue; 656 goto feed_segment; 657 658 strayed: 659 if (pseg_start == ri->ri_lsegs_end) 660 break; 661 662 feed_segment: 663 /* Looking to the next full segment */ 664 if (empty_seg++) 665 break; 666 seg_seq++; 667 segnum = nextnum; 668 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); 669 pseg_start = seg_start; 670 } 671 672 if (nsalvaged_blocks) { 673 nilfs_info(sb, "salvaged %lu blocks", nsalvaged_blocks); 674 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE; 675 } 676 out: 677 brelse(bh_sum); 678 dispose_recovery_list(&dsync_blocks); 679 return err; 680 681 confused: 682 err = -EINVAL; 683 failed: 684 nilfs_err(sb, 685 "error %d roll-forwarding partial segment at blocknr = %llu", 686 err, (unsigned long long)pseg_start); 687 goto out; 688} 689 690static void nilfs_finish_roll_forward(struct the_nilfs *nilfs, 691 struct nilfs_recovery_info *ri) 692{ 693 struct buffer_head *bh; 694 int err; 695 696 if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) != 697 nilfs_get_segnum_of_block(nilfs, ri->ri_super_root)) 698 return; 699 700 bh = __getblk(nilfs->ns_bdev, ri->ri_lsegs_start, nilfs->ns_blocksize); 701 BUG_ON(!bh); 702 memset(bh->b_data, 0, bh->b_size); 703 set_buffer_dirty(bh); 704 err = sync_dirty_buffer(bh); 705 if (unlikely(err)) 706 nilfs_warn(nilfs->ns_sb, 707 "buffer sync write failed during post-cleaning of recovery."); 708 brelse(bh); 709} 710 711/** 712 * nilfs_salvage_orphan_logs - salvage logs written after the latest checkpoint 713 * @nilfs: nilfs object 714 * @sb: super block instance 715 * @ri: pointer to a nilfs_recovery_info struct to store search results. 716 * 717 * Return Value: On success, 0 is returned. On error, one of the following 718 * negative error code is returned. 719 * 720 * %-EINVAL - Inconsistent filesystem state. 721 * 722 * %-EIO - I/O error 723 * 724 * %-ENOSPC - No space left on device (only in a panic state). 725 * 726 * %-ERESTARTSYS - Interrupted. 727 * 728 * %-ENOMEM - Insufficient memory available. 729 */ 730int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs, 731 struct super_block *sb, 732 struct nilfs_recovery_info *ri) 733{ 734 struct nilfs_root *root; 735 int err; 736 737 if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0) 738 return 0; 739 740 err = nilfs_attach_checkpoint(sb, ri->ri_cno, true, &root); 741 if (unlikely(err)) { 742 nilfs_err(sb, "error %d loading the latest checkpoint", err); 743 return err; 744 } 745 746 err = nilfs_do_roll_forward(nilfs, sb, root, ri); 747 if (unlikely(err)) 748 goto failed; 749 750 if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) { 751 err = nilfs_prepare_segment_for_recovery(nilfs, sb, ri); 752 if (unlikely(err)) { 753 nilfs_err(sb, "error %d preparing segment for recovery", 754 err); 755 goto failed; 756 } 757 758 err = nilfs_attach_log_writer(sb, root); 759 if (unlikely(err)) 760 goto failed; 761 762 set_nilfs_discontinued(nilfs); 763 err = nilfs_construct_segment(sb); 764 nilfs_detach_log_writer(sb); 765 766 if (unlikely(err)) { 767 nilfs_err(sb, "error %d writing segment for recovery", 768 err); 769 goto failed; 770 } 771 772 nilfs_finish_roll_forward(nilfs, ri); 773 } 774 775 failed: 776 nilfs_put_root(root); 777 return err; 778} 779 780/** 781 * nilfs_search_super_root - search the latest valid super root 782 * @nilfs: the_nilfs 783 * @ri: pointer to a nilfs_recovery_info struct to store search results. 784 * 785 * nilfs_search_super_root() looks for the latest super-root from a partial 786 * segment pointed by the superblock. It sets up struct the_nilfs through 787 * this search. It fills nilfs_recovery_info (ri) required for recovery. 788 * 789 * Return Value: On success, 0 is returned. On error, one of the following 790 * negative error code is returned. 791 * 792 * %-EINVAL - No valid segment found 793 * 794 * %-EIO - I/O error 795 * 796 * %-ENOMEM - Insufficient memory available. 797 */ 798int nilfs_search_super_root(struct the_nilfs *nilfs, 799 struct nilfs_recovery_info *ri) 800{ 801 struct buffer_head *bh_sum = NULL; 802 struct nilfs_segment_summary *sum = NULL; 803 sector_t pseg_start, pseg_end, sr_pseg_start = 0; 804 sector_t seg_start, seg_end; /* range of full segment (block number) */ 805 sector_t b, end; 806 unsigned long nblocks; 807 unsigned int flags; 808 u64 seg_seq; 809 __u64 segnum, nextnum = 0; 810 __u64 cno; 811 LIST_HEAD(segments); 812 int empty_seg = 0, scan_newer = 0; 813 int ret; 814 815 pseg_start = nilfs->ns_last_pseg; 816 seg_seq = nilfs->ns_last_seq; 817 cno = nilfs->ns_last_cno; 818 segnum = nilfs_get_segnum_of_block(nilfs, pseg_start); 819 820 /* Calculate range of segment */ 821 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); 822 823 /* Read ahead segment */ 824 b = seg_start; 825 while (b <= seg_end) 826 __breadahead(nilfs->ns_bdev, b++, nilfs->ns_blocksize); 827 828 for (;;) { 829 brelse(bh_sum); 830 ret = NILFS_SEG_FAIL_IO; 831 bh_sum = nilfs_read_log_header(nilfs, pseg_start, &sum); 832 if (!bh_sum) 833 goto failed; 834 835 ret = nilfs_validate_log(nilfs, seg_seq, bh_sum, sum); 836 if (ret) { 837 if (ret == NILFS_SEG_FAIL_IO) 838 goto failed; 839 goto strayed; 840 } 841 842 nblocks = le32_to_cpu(sum->ss_nblocks); 843 pseg_end = pseg_start + nblocks - 1; 844 if (unlikely(pseg_end > seg_end)) { 845 ret = NILFS_SEG_FAIL_CONSISTENCY; 846 goto strayed; 847 } 848 849 /* A valid partial segment */ 850 ri->ri_pseg_start = pseg_start; 851 ri->ri_seq = seg_seq; 852 ri->ri_segnum = segnum; 853 nextnum = nilfs_get_segnum_of_block(nilfs, 854 le64_to_cpu(sum->ss_next)); 855 ri->ri_nextnum = nextnum; 856 empty_seg = 0; 857 858 flags = le16_to_cpu(sum->ss_flags); 859 if (!(flags & NILFS_SS_SR) && !scan_newer) { 860 /* 861 * This will never happen because a superblock 862 * (last_segment) always points to a pseg with 863 * a super root. 864 */ 865 ret = NILFS_SEG_FAIL_CONSISTENCY; 866 goto failed; 867 } 868 869 if (pseg_start == seg_start) { 870 nilfs_get_segment_range(nilfs, nextnum, &b, &end); 871 while (b <= end) 872 __breadahead(nilfs->ns_bdev, b++, 873 nilfs->ns_blocksize); 874 } 875 if (!(flags & NILFS_SS_SR)) { 876 if (!ri->ri_lsegs_start && (flags & NILFS_SS_LOGBGN)) { 877 ri->ri_lsegs_start = pseg_start; 878 ri->ri_lsegs_start_seq = seg_seq; 879 } 880 if (flags & NILFS_SS_LOGEND) 881 ri->ri_lsegs_end = pseg_start; 882 goto try_next_pseg; 883 } 884 885 /* A valid super root was found. */ 886 ri->ri_cno = cno++; 887 ri->ri_super_root = pseg_end; 888 ri->ri_lsegs_start = ri->ri_lsegs_end = 0; 889 890 nilfs_dispose_segment_list(&segments); 891 sr_pseg_start = pseg_start; 892 nilfs->ns_pseg_offset = pseg_start + nblocks - seg_start; 893 nilfs->ns_seg_seq = seg_seq; 894 nilfs->ns_segnum = segnum; 895 nilfs->ns_cno = cno; /* nilfs->ns_cno = ri->ri_cno + 1 */ 896 nilfs->ns_ctime = le64_to_cpu(sum->ss_create); 897 nilfs->ns_nextnum = nextnum; 898 899 if (scan_newer) 900 ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED; 901 else { 902 if (nilfs->ns_mount_state & NILFS_VALID_FS) 903 goto super_root_found; 904 scan_newer = 1; 905 } 906 907 try_next_pseg: 908 /* Standing on a course, or met an inconsistent state */ 909 pseg_start += nblocks; 910 if (pseg_start < seg_end) 911 continue; 912 goto feed_segment; 913 914 strayed: 915 /* Off the trail */ 916 if (!scan_newer) 917 /* 918 * This can happen if a checkpoint was written without 919 * barriers, or as a result of an I/O failure. 920 */ 921 goto failed; 922 923 feed_segment: 924 /* Looking to the next full segment */ 925 if (empty_seg++) 926 goto super_root_found; /* found a valid super root */ 927 928 ret = nilfs_segment_list_add(&segments, segnum); 929 if (unlikely(ret)) 930 goto failed; 931 932 seg_seq++; 933 segnum = nextnum; 934 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); 935 pseg_start = seg_start; 936 } 937 938 super_root_found: 939 /* Updating pointers relating to the latest checkpoint */ 940 brelse(bh_sum); 941 list_splice_tail(&segments, &ri->ri_used_segments); 942 nilfs->ns_last_pseg = sr_pseg_start; 943 nilfs->ns_last_seq = nilfs->ns_seg_seq; 944 nilfs->ns_last_cno = ri->ri_cno; 945 return 0; 946 947 failed: 948 brelse(bh_sum); 949 nilfs_dispose_segment_list(&segments); 950 return ret < 0 ? ret : nilfs_warn_segment_error(nilfs->ns_sb, ret); 951} 952