fs/nilfs2/recovery.c

   1 /*
   2  * recovery.c - NILFS recovery logic
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  */
  22
  23 #include <linux/buffer_head.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/swap.h>
  26 #include <linux/crc32.h>
  27 #include "nilfs.h"
  28 #include "segment.h"
  29 #include "sufile.h"
  30 #include "page.h"
  31 #include "segbuf.h"
  32
  33 /*
  34  * Segment check result
  35  */
  36 enum {
  37         NILFS_SEG_VALID,
  38         NILFS_SEG_NO_SUPER_ROOT,
  39         NILFS_SEG_FAIL_IO,
  40         NILFS_SEG_FAIL_MAGIC,
  41         NILFS_SEG_FAIL_SEQ,
  42         NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT,
  43         NILFS_SEG_FAIL_CHECKSUM_FULL,
  44         NILFS_SEG_FAIL_CONSISTENCY,
  45 };
  46
  47 /* work structure for recovery */
  48 struct nilfs_recovery_block {
  49         ino_t ino;              /* Inode number of the file that this block
  50                                    belongs to */
  51         sector_t blocknr;       /* block number */
  52         __u64 vblocknr;         /* virtual block number */
  53         unsigned long blkoff;   /* File offset of the data block (per block) */
  54         struct list_head list;
  55 };
  56
  57
  58 static int nilfs_warn_segment_error(int err)
  59 {
  60         switch (err) {
  61         case NILFS_SEG_FAIL_IO:
  62                 printk(KERN_WARNING
  63                        "NILFS warning: I/O error on loading last segment\n");
  64                 return -EIO;
  65         case NILFS_SEG_FAIL_MAGIC:
  66                 printk(KERN_WARNING
  67                        "NILFS warning: Segment magic number invalid\n");
  68                 break;
  69         case NILFS_SEG_FAIL_SEQ:
  70                 printk(KERN_WARNING
  71                        "NILFS warning: Sequence number mismatch\n");
  72                 break;
  73         case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
  74                 printk(KERN_WARNING
  75                        "NILFS warning: Checksum error in super root\n");
  76                 break;
  77         case NILFS_SEG_FAIL_CHECKSUM_FULL:
  78                 printk(KERN_WARNING
  79                        "NILFS warning: Checksum error in segment payload\n");
  80                 break;
  81         case NILFS_SEG_FAIL_CONSISTENCY:
  82                 printk(KERN_WARNING
  83                        "NILFS warning: Inconsistent segment\n");
  84                 break;
  85         case NILFS_SEG_NO_SUPER_ROOT:
  86                 printk(KERN_WARNING
  87                        "NILFS warning: No super root in the last segment\n");
  88                 break;
  89         }
  90         return -EINVAL;
  91 }
  92
  93 static void store_segsum_info(struct nilfs_segsum_info *ssi,
  94                               struct nilfs_segment_summary *sum,
  95                               unsigned int blocksize)
  96 {
  97         ssi->flags = le16_to_cpu(sum->ss_flags);
  98         ssi->seg_seq = le64_to_cpu(sum->ss_seq);
  99         ssi->ctime = le64_to_cpu(sum->ss_create);
 100         ssi->next = le64_to_cpu(sum->ss_next);
 101         ssi->nblocks = le32_to_cpu(sum->ss_nblocks);
 102         ssi->nfinfo = le32_to_cpu(sum->ss_nfinfo);
 103         ssi->sumbytes = le32_to_cpu(sum->ss_sumbytes);
 104
 105         ssi->nsumblk = DIV_ROUND_UP(ssi->sumbytes, blocksize);
 106         ssi->nfileblk = ssi->nblocks - ssi->nsumblk - !!NILFS_SEG_HAS_SR(ssi);
 107 }
 108
 109 /**
 110  * calc_crc_cont - check CRC of blocks continuously
 111  * @sbi: nilfs_sb_info
 112  * @bhs: buffer head of start block
 113  * @sum: place to store result
 114  * @offset: offset bytes in the first block
 115  * @check_bytes: number of bytes to be checked
 116  * @start: DBN of start block
 117  * @nblock: number of blocks to be checked
 118  */
 119 static int calc_crc_cont(struct nilfs_sb_info *sbi, struct buffer_head *bhs,
 120                          u32 *sum, unsigned long offset, u64 check_bytes,
 121                          sector_t start, unsigned long nblock)
 122 {
 123         unsigned long blocksize = sbi->s_super->s_blocksize;
 124         unsigned long size;
 125         u32 crc;
 126
 127         BUG_ON(offset >= blocksize);
 128         check_bytes -= offset;
 129         size = min_t(u64, check_bytes, blocksize - offset);
 130         crc = crc32_le(sbi->s_nilfs->ns_crc_seed,
 131                        (unsigned char *)bhs->b_data + offset, size);
 132         if (--nblock > 0) {
 133                 do {
 134                         struct buffer_head *bh
 135                                 = sb_bread(sbi->s_super, ++start);
 136                         if (!bh)
 137                                 return -EIO;
 138                         check_bytes -= size;
 139                         size = min_t(u64, check_bytes, blocksize);
 140                         crc = crc32_le(crc, bh->b_data, size);
 141                         brelse(bh);
 142                 } while (--nblock > 0);
 143         }
 144         *sum = crc;
 145         return 0;
 146 }
 147
 148 /**
 149  * nilfs_read_super_root_block - read super root block
 150  * @sb: super_block
 151  * @sr_block: disk block number of the super root block
 152  * @pbh: address of a buffer_head pointer to return super root buffer
 153  * @check: CRC check flag
 154  */
 155 int nilfs_read_super_root_block(struct super_block *sb, sector_t sr_block,
 156                                 struct buffer_head **pbh, int check)
 157 {
 158         struct buffer_head *bh_sr;
 159         struct nilfs_super_root *sr;
 160         u32 crc;
 161         int ret;
 162
 163         *pbh = NULL;
 164         bh_sr = sb_bread(sb, sr_block);
 165         if (unlikely(!bh_sr)) {
 166                 ret = NILFS_SEG_FAIL_IO;
 167                 goto failed;
 168         }
 169
 170         sr = (struct nilfs_super_root *)bh_sr->b_data;
 171         if (check) {
 172                 unsigned bytes = le16_to_cpu(sr->sr_bytes);
 173
 174                 if (bytes == 0 || bytes > sb->s_blocksize) {
 175                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 176                         goto failed_bh;
 177                 }
 178                 if (calc_crc_cont(NILFS_SB(sb), bh_sr, &crc,
 179                                   sizeof(sr->sr_sum), bytes, sr_block, 1)) {
 180                         ret = NILFS_SEG_FAIL_IO;
 181                         goto failed_bh;
 182                 }
 183                 if (crc != le32_to_cpu(sr->sr_sum)) {
 184                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 185                         goto failed_bh;
 186                 }
 187         }
 188         *pbh = bh_sr;
 189         return 0;
 190
 191  failed_bh:
 192         brelse(bh_sr);
 193
 194  failed:
 195         return nilfs_warn_segment_error(ret);
 196 }
 197
 198 /**
 199  * load_segment_summary - read segment summary of the specified partial segment
 200  * @sbi: nilfs_sb_info
 201  * @pseg_start: start disk block number of partial segment
 202  * @seg_seq: sequence number requested
 203  * @ssi: pointer to nilfs_segsum_info struct to store information
 204  */
 205 static int
 206 load_segment_summary(struct nilfs_sb_info *sbi, sector_t pseg_start,
 207                      u64 seg_seq, struct nilfs_segsum_info *ssi)
 208 {
 209         struct buffer_head *bh_sum;
 210         struct nilfs_segment_summary *sum;
 211         unsigned long nblock;
 212         u32 crc;
 213         int ret = NILFS_SEG_FAIL_IO;
 214
 215         bh_sum = sb_bread(sbi->s_super, pseg_start);
 216         if (!bh_sum)
 217                 goto out;
 218
 219         sum = (struct nilfs_segment_summary *)bh_sum->b_data;
 220
 221         /* Check consistency of segment summary */
 222         if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC) {
 223                 ret = NILFS_SEG_FAIL_MAGIC;
 224                 goto failed;
 225         }
 226         store_segsum_info(ssi, sum, sbi->s_super->s_blocksize);
 227         if (seg_seq != ssi->seg_seq) {
 228                 ret = NILFS_SEG_FAIL_SEQ;
 229                 goto failed;
 230         }
 231
 232         nblock = ssi->nblocks;
 233         if (unlikely(nblock == 0 ||
 234                      nblock > sbi->s_nilfs->ns_blocks_per_segment)) {
 235                 /* This limits the number of blocks read in the CRC check */
 236                 ret = NILFS_SEG_FAIL_CONSISTENCY;
 237                 goto failed;
 238         }
 239         if (calc_crc_cont(sbi, bh_sum, &crc, sizeof(sum->ss_datasum),
 240                           ((u64)nblock << sbi->s_super->s_blocksize_bits),
 241                           pseg_start, nblock)) {
 242                 ret = NILFS_SEG_FAIL_IO;
 243                 goto failed;
 244         }
 245         if (crc == le32_to_cpu(sum->ss_datasum))
 246                 ret = 0;
 247         else
 248                 ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
 249  failed:
 250         brelse(bh_sum);
 251  out:
 252         return ret;
 253 }
 254
 255 static void *segsum_get(struct super_block *sb, struct buffer_head **pbh,
 256                         unsigned int *offset, unsigned int bytes)
 257 {
 258         void *ptr;
 259         sector_t blocknr;
 260
 261         BUG_ON((*pbh)->b_size < *offset);
 262         if (bytes > (*pbh)->b_size - *offset) {
 263                 blocknr = (*pbh)->b_blocknr;
 264                 brelse(*pbh);
 265                 *pbh = sb_bread(sb, blocknr + 1);
 266                 if (unlikely(!*pbh))
 267                         return NULL;
 268                 *offset = 0;
 269         }
 270         ptr = (*pbh)->b_data + *offset;
 271         *offset += bytes;
 272         return ptr;
 273 }
 274
 275 static void segsum_skip(struct super_block *sb, struct buffer_head **pbh,
 276                         unsigned int *offset, unsigned int bytes,
 277                         unsigned long count)
 278 {
 279         unsigned int rest_item_in_current_block
 280                 = ((*pbh)->b_size - *offset) / bytes;
 281
 282         if (count <= rest_item_in_current_block) {
 283                 *offset += bytes * count;
 284         } else {
 285                 sector_t blocknr = (*pbh)->b_blocknr;
 286                 unsigned int nitem_per_block = (*pbh)->b_size / bytes;
 287                 unsigned int bcnt;
 288
 289                 count -= rest_item_in_current_block;
 290                 bcnt = DIV_ROUND_UP(count, nitem_per_block);
 291                 *offset = bytes * (count - (bcnt - 1) * nitem_per_block);
 292
 293                 brelse(*pbh);
 294                 *pbh = sb_bread(sb, blocknr + bcnt);
 295         }
 296 }
 297
 298 static int
 299 collect_blocks_from_segsum(struct nilfs_sb_info *sbi, sector_t sum_blocknr,
 300                            struct nilfs_segsum_info *ssi,
 301                            struct list_head *head)
 302 {
 303         struct buffer_head *bh;
 304         unsigned int offset;
 305         unsigned long nfinfo = ssi->nfinfo;
 306         sector_t blocknr = sum_blocknr + ssi->nsumblk;
 307         ino_t ino;
 308         int err = -EIO;
 309
 310         if (!nfinfo)
 311                 return 0;
 312
 313         bh = sb_bread(sbi->s_super, sum_blocknr);
 314         if (unlikely(!bh))
 315                 goto out;
 316
 317         offset = le16_to_cpu(
 318                 ((struct nilfs_segment_summary *)bh->b_data)->ss_bytes);
 319         for (;;) {
 320                 unsigned long nblocks, ndatablk, nnodeblk;
 321                 struct nilfs_finfo *finfo;
 322
 323                 finfo = segsum_get(sbi->s_super, &bh, &offset, sizeof(*finfo));
 324                 if (unlikely(!finfo))
 325                         goto out;
 326
 327                 ino = le64_to_cpu(finfo->fi_ino);
 328                 nblocks = le32_to_cpu(finfo->fi_nblocks);
 329                 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
 330                 nnodeblk = nblocks - ndatablk;
 331
 332                 while (ndatablk-- > 0) {
 333                         struct nilfs_recovery_block *rb;
 334                         struct nilfs_binfo_v *binfo;
 335
 336                         binfo = segsum_get(sbi->s_super, &bh, &offset,
 337                                            sizeof(*binfo));
 338                         if (unlikely(!binfo))
 339                                 goto out;
 340
 341                         rb = kmalloc(sizeof(*rb), GFP_NOFS);
 342                         if (unlikely(!rb)) {
 343                                 err = -ENOMEM;
 344                                 goto out;
 345                         }
 346                         rb->ino = ino;
 347                         rb->blocknr = blocknr++;
 348                         rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
 349                         rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
 350                         /* INIT_LIST_HEAD(&rb->list); */
 351                         list_add_tail(&rb->list, head);
 352                 }
 353                 if (--nfinfo == 0)
 354                         break;
 355                 blocknr += nnodeblk; /* always 0 for the data sync segments */
 356                 segsum_skip(sbi->s_super, &bh, &offset, sizeof(__le64),
 357                             nnodeblk);
 358                 if (unlikely(!bh))
 359                         goto out;
 360         }
 361         err = 0;
 362  out:
 363         brelse(bh);   /* brelse(NULL) is just ignored */
 364         return err;
 365 }
 366
 367 static void dispose_recovery_list(struct list_head *head)
 368 {
 369         while (!list_empty(head)) {
 370                 struct nilfs_recovery_block *rb
 371                         = list_entry(head->next,
 372                                      struct nilfs_recovery_block, list);
 373                 list_del(&rb->list);
 374                 kfree(rb);
 375         }
 376 }
 377
 378 struct nilfs_segment_entry {
 379         struct list_head        list;
 380         __u64                   segnum;
 381 };
 382
 383 static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
 384 {
 385         struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
 386
 387         if (unlikely(!ent))
 388                 return -ENOMEM;
 389
 390         ent->segnum = segnum;
 391         INIT_LIST_HEAD(&ent->list);
 392         list_add_tail(&ent->list, head);
 393         return 0;
 394 }
 395
 396 void nilfs_dispose_segment_list(struct list_head *head)
 397 {
 398         while (!list_empty(head)) {
 399                 struct nilfs_segment_entry *ent
 400                         = list_entry(head->next,
 401                                      struct nilfs_segment_entry, list);
 402                 list_del(&ent->list);
 403                 kfree(ent);
 404         }
 405 }
 406
 407 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
 408                                               struct nilfs_sb_info *sbi,
 409                                               struct nilfs_recovery_info *ri)
 410 {
 411         struct list_head *head = &ri->ri_used_segments;
 412         struct nilfs_segment_entry *ent, *n;
 413         struct inode *sufile = nilfs->ns_sufile;
 414         __u64 segnum[4];
 415         int err;
 416         int i;
 417
 418         segnum[0] = nilfs->ns_segnum;
 419         segnum[1] = nilfs->ns_nextnum;
 420         segnum[2] = ri->ri_segnum;
 421         segnum[3] = ri->ri_nextnum;
 422
 423         nilfs_attach_writer(nilfs, sbi);
 424         /*
 425          * Releasing the next segment of the latest super root.
 426          * The next segment is invalidated by this recovery.
 427          */
 428         err = nilfs_sufile_free(sufile, segnum[1]);
 429         if (unlikely(err))
 430                 goto failed;
 431
 432         for (i = 1; i < 4; i++) {
 433                 err = nilfs_segment_list_add(head, segnum[i]);
 434                 if (unlikely(err))
 435                         goto failed;
 436         }
 437
 438         /*
 439          * Collecting segments written after the latest super root.
 440          * These are marked dirty to avoid being reallocated in the next write.
 441          */
 442         list_for_each_entry_safe(ent, n, head, list) {
 443                 if (ent->segnum != segnum[0]) {
 444                         err = nilfs_sufile_scrap(sufile, ent->segnum);
 445                         if (unlikely(err))
 446                                 goto failed;
 447                 }
 448                 list_del(&ent->list);
 449                 kfree(ent);
 450         }
 451
 452         /* Allocate new segments for recovery */
 453         err = nilfs_sufile_alloc(sufile, &segnum[0]);
 454         if (unlikely(err))
 455                 goto failed;
 456
 457         nilfs->ns_pseg_offset = 0;
 458         nilfs->ns_seg_seq = ri->ri_seq + 2;
 459         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
 460
 461  failed:
 462         /* No need to recover sufile because it will be destroyed on error */
 463         nilfs_detach_writer(nilfs, sbi);
 464         return err;
 465 }
 466
 467 static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi,
 468                                      struct nilfs_recovery_block *rb,
 469                                      struct page *page)
 470 {
 471         struct buffer_head *bh_org;
 472         void *kaddr;
 473
 474         bh_org = sb_bread(sbi->s_super, rb->blocknr);
 475         if (unlikely(!bh_org))
 476                 return -EIO;
 477
 478         kaddr = kmap_atomic(page, KM_USER0);
 479         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
 480         kunmap_atomic(kaddr, KM_USER0);
 481         brelse(bh_org);
 482         return 0;
 483 }
 484
 485 static int recover_dsync_blocks(struct nilfs_sb_info *sbi,
 486                                 struct list_head *head,
 487                                 unsigned long *nr_salvaged_blocks)
 488 {
 489         struct inode *inode;
 490         struct nilfs_recovery_block *rb, *n;
 491         unsigned blocksize = sbi->s_super->s_blocksize;
 492         struct page *page;
 493         loff_t pos;
 494         int err = 0, err2 = 0;
 495
 496         list_for_each_entry_safe(rb, n, head, list) {
 497                 inode = nilfs_iget(sbi->s_super, rb->ino);
 498                 if (IS_ERR(inode)) {
 499                         err = PTR_ERR(inode);
 500                         inode = NULL;
 501                         goto failed_inode;
 502                 }
 503
 504                 pos = rb->blkoff << inode->i_blkbits;
 505                 page = NULL;
 506                 err = block_write_begin(NULL, inode->i_mapping, pos, blocksize,
 507                                         0, &page, NULL, nilfs_get_block);
 508                 if (unlikely(err))
 509                         goto failed_inode;
 510
 511                 err = nilfs_recovery_copy_block(sbi, rb, page);
 512                 if (unlikely(err))
 513                         goto failed_page;
 514
 515                 err = nilfs_set_file_dirty(sbi, inode, 1);
 516                 if (unlikely(err))
 517                         goto failed_page;
 518
 519                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
 520                                 blocksize, page, NULL);
 521
 522                 unlock_page(page);
 523                 page_cache_release(page);
 524
 525                 (*nr_salvaged_blocks)++;
 526                 goto next;
 527
 528  failed_page:
 529                 unlock_page(page);
 530                 page_cache_release(page);
 531
 532  failed_inode:
 533                 printk(KERN_WARNING
 534                        "NILFS warning: error recovering data block "
 535                        "(err=%d, ino=%lu, block-offset=%llu)\n",
 536                        err, (unsigned long)rb->ino,
 537                        (unsigned long long)rb->blkoff);
 538                 if (!err2)
 539                         err2 = err;
 540  next:
 541                 iput(inode); /* iput(NULL) is just ignored */
 542                 list_del_init(&rb->list);
 543                 kfree(rb);
 544         }
 545         return err2;
 546 }
 547
 548 /**
 549  * nilfs_do_roll_forward - salvage logical segments newer than the latest
 550  * checkpoint
 551  * @sbi: nilfs_sb_info
 552  * @nilfs: the_nilfs
 553  * @ri: pointer to a nilfs_recovery_info
 554  */
 555 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
 556                                  struct nilfs_sb_info *sbi,
 557                                  struct nilfs_recovery_info *ri)
 558 {
 559         struct nilfs_segsum_info ssi;
 560         sector_t pseg_start;
 561         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
 562         unsigned long nsalvaged_blocks = 0;
 563         u64 seg_seq;
 564         __u64 segnum, nextnum = 0;
 565         int empty_seg = 0;
 566         int err = 0, ret;
 567         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
 568         enum {
 569                 RF_INIT_ST,
 570                 RF_DSYNC_ST,   /* scanning data-sync segments */
 571         };
 572         int state = RF_INIT_ST;
 573
 574         nilfs_attach_writer(nilfs, sbi);
 575         pseg_start = ri->ri_lsegs_start;
 576         seg_seq = ri->ri_lsegs_start_seq;
 577         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 578         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 579
 580         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
 581
 582                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi);
 583                 if (ret) {
 584                         if (ret == NILFS_SEG_FAIL_IO) {
 585                                 err = -EIO;
 586                                 goto failed;
 587                         }
 588                         goto strayed;
 589                 }
 590                 if (unlikely(NILFS_SEG_HAS_SR(&ssi)))
 591                         goto confused;
 592
 593                 /* Found a valid partial segment; do recovery actions */
 594                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 595                 empty_seg = 0;
 596                 nilfs->ns_ctime = ssi.ctime;
 597                 if (!(ssi.flags & NILFS_SS_GC))
 598                         nilfs->ns_nongc_ctime = ssi.ctime;
 599
 600                 switch (state) {
 601                 case RF_INIT_ST:
 602                         if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi))
 603                                 goto try_next_pseg;
 604                         state = RF_DSYNC_ST;
 605                         /* Fall through */
 606                 case RF_DSYNC_ST:
 607                         if (!NILFS_SEG_DSYNC(&ssi))
 608                                 goto confused;
 609
 610                         err = collect_blocks_from_segsum(
 611                                 sbi, pseg_start, &ssi, &dsync_blocks);
 612                         if (unlikely(err))
 613                                 goto failed;
 614                         if (NILFS_SEG_LOGEND(&ssi)) {
 615                                 err = recover_dsync_blocks(
 616                                         sbi, &dsync_blocks, &nsalvaged_blocks);
 617                                 if (unlikely(err))
 618                                         goto failed;
 619                                 state = RF_INIT_ST;
 620                         }
 621                         break; /* Fall through to try_next_pseg */
 622                 }
 623
 624  try_next_pseg:
 625                 if (pseg_start == ri->ri_lsegs_end)
 626                         break;
 627                 pseg_start += ssi.nblocks;
 628                 if (pseg_start < seg_end)
 629                         continue;
 630                 goto feed_segment;
 631
 632  strayed:
 633                 if (pseg_start == ri->ri_lsegs_end)
 634                         break;
 635
 636  feed_segment:
 637                 /* Looking to the next full segment */
 638                 if (empty_seg++)
 639                         break;
 640                 seg_seq++;
 641                 segnum = nextnum;
 642                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 643                 pseg_start = seg_start;
 644         }
 645
 646         if (nsalvaged_blocks) {
 647                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
 648                        sbi->s_super->s_id, nsalvaged_blocks);
 649                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
 650         }
 651  out:
 652         dispose_recovery_list(&dsync_blocks);
 653         nilfs_detach_writer(sbi->s_nilfs, sbi);
 654         return err;
 655
 656  confused:
 657         err = -EINVAL;
 658  failed:
 659         printk(KERN_ERR
 660                "NILFS (device %s): Error roll-forwarding "
 661                "(err=%d, pseg block=%llu). ",
 662                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
 663         goto out;
 664 }
 665
 666 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
 667                                       struct nilfs_sb_info *sbi,
 668                                       struct nilfs_recovery_info *ri)
 669 {
 670         struct buffer_head *bh;
 671         int err;
 672
 673         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
 674             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
 675                 return;
 676
 677         bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start);
 678         BUG_ON(!bh);
 679         memset(bh->b_data, 0, bh->b_size);
 680         set_buffer_dirty(bh);
 681         err = sync_dirty_buffer(bh);
 682         if (unlikely(err))
 683                 printk(KERN_WARNING
 684                        "NILFS warning: buffer sync write failed during "
 685                        "post-cleaning of recovery.\n");
 686         brelse(bh);
 687 }
 688
 689 /**
 690  * nilfs_recover_logical_segments - salvage logical segments written after
 691  * the latest super root
 692  * @nilfs: the_nilfs
 693  * @sbi: nilfs_sb_info
 694  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 695  *
 696  * Return Value: On success, 0 is returned.  On error, one of the following
 697  * negative error code is returned.
 698  *
 699  * %-EINVAL - Inconsistent filesystem state.
 700  *
 701  * %-EIO - I/O error
 702  *
 703  * %-ENOSPC - No space left on device (only in a panic state).
 704  *
 705  * %-ERESTARTSYS - Interrupted.
 706  *
 707  * %-ENOMEM - Insufficient memory available.
 708  */
 709 int nilfs_recover_logical_segments(struct the_nilfs *nilfs,
 710                                    struct nilfs_sb_info *sbi,
 711                                    struct nilfs_recovery_info *ri)
 712 {
 713         int err;
 714
 715         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
 716                 return 0;
 717
 718         err = nilfs_attach_checkpoint(sbi, ri->ri_cno);
 719         if (unlikely(err)) {
 720                 printk(KERN_ERR
 721                        "NILFS: error loading the latest checkpoint.\n");
 722                 return err;
 723         }
 724
 725         err = nilfs_do_roll_forward(nilfs, sbi, ri);
 726         if (unlikely(err))
 727                 goto failed;
 728
 729         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
 730                 err = nilfs_prepare_segment_for_recovery(nilfs, sbi, ri);
 731                 if (unlikely(err)) {
 732                         printk(KERN_ERR "NILFS: Error preparing segments for "
 733                                "recovery.\n");
 734                         goto failed;
 735                 }
 736
 737                 err = nilfs_attach_segment_constructor(sbi);
 738                 if (unlikely(err))
 739                         goto failed;
 740
 741                 set_nilfs_discontinued(nilfs);
 742                 err = nilfs_construct_segment(sbi->s_super);
 743                 nilfs_detach_segment_constructor(sbi);
 744
 745                 if (unlikely(err)) {
 746                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
 747                                "(err=%d)\n", err);
 748                         goto failed;
 749                 }
 750
 751                 nilfs_finish_roll_forward(nilfs, sbi, ri);
 752         }
 753
 754  failed:
 755         nilfs_detach_checkpoint(sbi);
 756         return err;
 757 }
 758
 759 /**
 760  * nilfs_search_super_root - search the latest valid super root
 761  * @nilfs: the_nilfs
 762  * @sbi: nilfs_sb_info
 763  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 764  *
 765  * nilfs_search_super_root() looks for the latest super-root from a partial
 766  * segment pointed by the superblock.  It sets up struct the_nilfs through
 767  * this search. It fills nilfs_recovery_info (ri) required for recovery.
 768  *
 769  * Return Value: On success, 0 is returned.  On error, one of the following
 770  * negative error code is returned.
 771  *
 772  * %-EINVAL - No valid segment found
 773  *
 774  * %-EIO - I/O error
 775  */
 776 int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi,
 777                             struct nilfs_recovery_info *ri)
 778 {
 779         struct nilfs_segsum_info ssi;
 780         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
 781         sector_t seg_start, seg_end; /* range of full segment (block number) */
 782         sector_t b, end;
 783         u64 seg_seq;
 784         __u64 segnum, nextnum = 0;
 785         __u64 cno;
 786         LIST_HEAD(segments);
 787         int empty_seg = 0, scan_newer = 0;
 788         int ret;
 789
 790         pseg_start = nilfs->ns_last_pseg;
 791         seg_seq = nilfs->ns_last_seq;
 792         cno = nilfs->ns_last_cno;
 793         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 794
 795         /* Calculate range of segment */
 796         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 797
 798         /* Read ahead segment */
 799         b = seg_start;
 800         while (b <= seg_end)
 801                 sb_breadahead(sbi->s_super, b++);
 802
 803         for (;;) {
 804                 /* Load segment summary */
 805                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi);
 806                 if (ret) {
 807                         if (ret == NILFS_SEG_FAIL_IO)
 808                                 goto failed;
 809                         goto strayed;
 810                 }
 811                 pseg_end = pseg_start + ssi.nblocks - 1;
 812                 if (unlikely(pseg_end > seg_end)) {
 813                         ret = NILFS_SEG_FAIL_CONSISTENCY;
 814                         goto strayed;
 815                 }
 816
 817                 /* A valid partial segment */
 818                 ri->ri_pseg_start = pseg_start;
 819                 ri->ri_seq = seg_seq;
 820                 ri->ri_segnum = segnum;
 821                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 822                 ri->ri_nextnum = nextnum;
 823                 empty_seg = 0;
 824
 825                 if (!NILFS_SEG_HAS_SR(&ssi) && !scan_newer) {
 826                         /* This will never happen because a superblock
 827                            (last_segment) always points to a pseg
 828                            having a super root. */
 829                         ret = NILFS_SEG_FAIL_CONSISTENCY;
 830                         goto failed;
 831                 }
 832
 833                 if (pseg_start == seg_start) {
 834                         nilfs_get_segment_range(nilfs, nextnum, &b, &end);
 835                         while (b <= end)
 836                                 sb_breadahead(sbi->s_super, b++);
 837                 }
 838                 if (!NILFS_SEG_HAS_SR(&ssi)) {
 839                         if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) {
 840                                 ri->ri_lsegs_start = pseg_start;
 841                                 ri->ri_lsegs_start_seq = seg_seq;
 842                         }
 843                         if (NILFS_SEG_LOGEND(&ssi))
 844                                 ri->ri_lsegs_end = pseg_start;
 845                         goto try_next_pseg;
 846                 }
 847
 848                 /* A valid super root was found. */
 849                 ri->ri_cno = cno++;
 850                 ri->ri_super_root = pseg_end;
 851                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
 852
 853                 nilfs_dispose_segment_list(&segments);
 854                 nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start)
 855                         + ssi.nblocks - seg_start;
 856                 nilfs->ns_seg_seq = seg_seq;
 857                 nilfs->ns_segnum = segnum;
 858                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
 859                 nilfs->ns_ctime = ssi.ctime;
 860                 nilfs->ns_nextnum = nextnum;
 861
 862                 if (scan_newer)
 863                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
 864                 else {
 865                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
 866                                 goto super_root_found;
 867                         scan_newer = 1;
 868                 }
 869
 870                 /* reset region for roll-forward */
 871                 pseg_start += ssi.nblocks;
 872                 if (pseg_start < seg_end)
 873                         continue;
 874                 goto feed_segment;
 875
 876  try_next_pseg:
 877                 /* Standing on a course, or met an inconsistent state */
 878                 pseg_start += ssi.nblocks;
 879                 if (pseg_start < seg_end)
 880                         continue;
 881                 goto feed_segment;
 882
 883  strayed:
 884                 /* Off the trail */
 885                 if (!scan_newer)
 886                         /*
 887                          * This can happen if a checkpoint was written without
 888                          * barriers, or as a result of an I/O failure.
 889                          */
 890                         goto failed;
 891
 892  feed_segment:
 893                 /* Looking to the next full segment */
 894                 if (empty_seg++)
 895                         goto super_root_found; /* found a valid super root */
 896
 897                 ret = nilfs_segment_list_add(&segments, segnum);
 898                 if (unlikely(ret))
 899                         goto failed;
 900
 901                 seg_seq++;
 902                 segnum = nextnum;
 903                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 904                 pseg_start = seg_start;
 905         }
 906
 907  super_root_found:
 908         /* Updating pointers relating to the latest checkpoint */
 909         list_splice_tail(&segments, &ri->ri_used_segments);
 910         nilfs->ns_last_pseg = sr_pseg_start;
 911         nilfs->ns_last_seq = nilfs->ns_seg_seq;
 912         nilfs->ns_last_cno = ri->ri_cno;
 913         return 0;
 914
 915  failed:
 916         nilfs_dispose_segment_list(&segments);
 917         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
 918 }