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/slab.h>
  27 #include <linux/crc32.h>
  28 #include "nilfs.h"
  29 #include "segment.h"
  30 #include "sufile.h"
  31 #include "page.h"
  32 #include "segbuf.h"
  33
  34 /*
  35  * Segment check result
  36  */
  37 enum {
  38         NILFS_SEG_VALID,
  39         NILFS_SEG_NO_SUPER_ROOT,
  40         NILFS_SEG_FAIL_IO,
  41         NILFS_SEG_FAIL_MAGIC,
  42         NILFS_SEG_FAIL_SEQ,
  43         NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT,
  44         NILFS_SEG_FAIL_CHECKSUM_FULL,
  45         NILFS_SEG_FAIL_CONSISTENCY,
  46 };
  47
  48 /* work structure for recovery */
  49 struct nilfs_recovery_block {
  50         ino_t ino;              /* Inode number of the file that this block
  51                                    belongs to */
  52         sector_t blocknr;       /* block number */
  53         __u64 vblocknr;         /* virtual block number */
  54         unsigned long blkoff;   /* File offset of the data block (per block) */
  55         struct list_head list;
  56 };
  57
  58
  59 static int nilfs_warn_segment_error(int err)
  60 {
  61         switch (err) {
  62         case NILFS_SEG_FAIL_IO:
  63                 printk(KERN_WARNING
  64                        "NILFS warning: I/O error on loading last segment\n");
  65                 return -EIO;
  66         case NILFS_SEG_FAIL_MAGIC:
  67                 printk(KERN_WARNING
  68                        "NILFS warning: Segment magic number invalid\n");
  69                 break;
  70         case NILFS_SEG_FAIL_SEQ:
  71                 printk(KERN_WARNING
  72                        "NILFS warning: Sequence number mismatch\n");
  73                 break;
  74         case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
  75                 printk(KERN_WARNING
  76                        "NILFS warning: Checksum error in super root\n");
  77                 break;
  78         case NILFS_SEG_FAIL_CHECKSUM_FULL:
  79                 printk(KERN_WARNING
  80                        "NILFS warning: Checksum error in segment payload\n");
  81                 break;
  82         case NILFS_SEG_FAIL_CONSISTENCY:
  83                 printk(KERN_WARNING
  84                        "NILFS warning: Inconsistent segment\n");
  85                 break;
  86         case NILFS_SEG_NO_SUPER_ROOT:
  87                 printk(KERN_WARNING
  88                        "NILFS warning: No super root in the last segment\n");
  89                 break;
  90         }
  91         return -EINVAL;
  92 }
  93
  94 static void store_segsum_info(struct nilfs_segsum_info *ssi,
  95                               struct nilfs_segment_summary *sum,
  96                               unsigned int blocksize)
  97 {
  98         ssi->flags = le16_to_cpu(sum->ss_flags);
  99         ssi->seg_seq = le64_to_cpu(sum->ss_seq);
 100         ssi->ctime = le64_to_cpu(sum->ss_create);
 101         ssi->next = le64_to_cpu(sum->ss_next);
 102         ssi->nblocks = le32_to_cpu(sum->ss_nblocks);
 103         ssi->nfinfo = le32_to_cpu(sum->ss_nfinfo);
 104         ssi->sumbytes = le32_to_cpu(sum->ss_sumbytes);
 105
 106         ssi->nsumblk = DIV_ROUND_UP(ssi->sumbytes, blocksize);
 107         ssi->nfileblk = ssi->nblocks - ssi->nsumblk - !!NILFS_SEG_HAS_SR(ssi);
 108 }
 109
 110 /**
 111  * calc_crc_cont - check CRC of blocks continuously
 112  * @sbi: nilfs_sb_info
 113  * @bhs: buffer head of start block
 114  * @sum: place to store result
 115  * @offset: offset bytes in the first block
 116  * @check_bytes: number of bytes to be checked
 117  * @start: DBN of start block
 118  * @nblock: number of blocks to be checked
 119  */
 120 static int calc_crc_cont(struct nilfs_sb_info *sbi, struct buffer_head *bhs,
 121                          u32 *sum, unsigned long offset, u64 check_bytes,
 122                          sector_t start, unsigned long nblock)
 123 {
 124         unsigned long blocksize = sbi->s_super->s_blocksize;
 125         unsigned long size;
 126         u32 crc;
 127
 128         BUG_ON(offset >= blocksize);
 129         check_bytes -= offset;
 130         size = min_t(u64, check_bytes, blocksize - offset);
 131         crc = crc32_le(sbi->s_nilfs->ns_crc_seed,
 132                        (unsigned char *)bhs->b_data + offset, size);
 133         if (--nblock > 0) {
 134                 do {
 135                         struct buffer_head *bh
 136                                 = sb_bread(sbi->s_super, ++start);
 137                         if (!bh)
 138                                 return -EIO;
 139                         check_bytes -= size;
 140                         size = min_t(u64, check_bytes, blocksize);
 141                         crc = crc32_le(crc, bh->b_data, size);
 142                         brelse(bh);
 143                 } while (--nblock > 0);
 144         }
 145         *sum = crc;
 146         return 0;
 147 }
 148
 149 /**
 150  * nilfs_read_super_root_block - read super root block
 151  * @sb: super_block
 152  * @sr_block: disk block number of the super root block
 153  * @pbh: address of a buffer_head pointer to return super root buffer
 154  * @check: CRC check flag
 155  */
 156 int nilfs_read_super_root_block(struct super_block *sb, sector_t sr_block,
 157                                 struct buffer_head **pbh, int check)
 158 {
 159         struct buffer_head *bh_sr;
 160         struct nilfs_super_root *sr;
 161         u32 crc;
 162         int ret;
 163
 164         *pbh = NULL;
 165         bh_sr = sb_bread(sb, sr_block);
 166         if (unlikely(!bh_sr)) {
 167                 ret = NILFS_SEG_FAIL_IO;
 168                 goto failed;
 169         }
 170
 171         sr = (struct nilfs_super_root *)bh_sr->b_data;
 172         if (check) {
 173                 unsigned bytes = le16_to_cpu(sr->sr_bytes);
 174
 175                 if (bytes == 0 || bytes > sb->s_blocksize) {
 176                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 177                         goto failed_bh;
 178                 }
 179                 if (calc_crc_cont(NILFS_SB(sb), bh_sr, &crc,
 180                                   sizeof(sr->sr_sum), bytes, sr_block, 1)) {
 181                         ret = NILFS_SEG_FAIL_IO;
 182                         goto failed_bh;
 183                 }
 184                 if (crc != le32_to_cpu(sr->sr_sum)) {
 185                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 186                         goto failed_bh;
 187                 }
 188         }
 189         *pbh = bh_sr;
 190         return 0;
 191
 192  failed_bh:
 193         brelse(bh_sr);
 194
 195  failed:
 196         return nilfs_warn_segment_error(ret);
 197 }
 198
 199 /**
 200  * load_segment_summary - read segment summary of the specified partial segment
 201  * @sbi: nilfs_sb_info
 202  * @pseg_start: start disk block number of partial segment
 203  * @seg_seq: sequence number requested
 204  * @ssi: pointer to nilfs_segsum_info struct to store information
 205  */
 206 static int
 207 load_segment_summary(struct nilfs_sb_info *sbi, sector_t pseg_start,
 208                      u64 seg_seq, struct nilfs_segsum_info *ssi)
 209 {
 210         struct buffer_head *bh_sum;
 211         struct nilfs_segment_summary *sum;
 212         unsigned long nblock;
 213         u32 crc;
 214         int ret = NILFS_SEG_FAIL_IO;
 215
 216         bh_sum = sb_bread(sbi->s_super, pseg_start);
 217         if (!bh_sum)
 218                 goto out;
 219
 220         sum = (struct nilfs_segment_summary *)bh_sum->b_data;
 221
 222         /* Check consistency of segment summary */
 223         if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC) {
 224                 ret = NILFS_SEG_FAIL_MAGIC;
 225                 goto failed;
 226         }
 227         store_segsum_info(ssi, sum, sbi->s_super->s_blocksize);
 228         if (seg_seq != ssi->seg_seq) {
 229                 ret = NILFS_SEG_FAIL_SEQ;
 230                 goto failed;
 231         }
 232
 233         nblock = ssi->nblocks;
 234         if (unlikely(nblock == 0 ||
 235                      nblock > sbi->s_nilfs->ns_blocks_per_segment)) {
 236                 /* This limits the number of blocks read in the CRC check */
 237                 ret = NILFS_SEG_FAIL_CONSISTENCY;
 238                 goto failed;
 239         }
 240         if (calc_crc_cont(sbi, bh_sum, &crc, sizeof(sum->ss_datasum),
 241                           ((u64)nblock << sbi->s_super->s_blocksize_bits),
 242                           pseg_start, nblock)) {
 243                 ret = NILFS_SEG_FAIL_IO;
 244                 goto failed;
 245         }
 246         if (crc == le32_to_cpu(sum->ss_datasum))
 247                 ret = 0;
 248         else
 249                 ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
 250  failed:
 251         brelse(bh_sum);
 252  out:
 253         return ret;
 254 }
 255
 256 static void *segsum_get(struct super_block *sb, struct buffer_head **pbh,
 257                         unsigned int *offset, unsigned int bytes)
 258 {
 259         void *ptr;
 260         sector_t blocknr;
 261
 262         BUG_ON((*pbh)->b_size < *offset);
 263         if (bytes > (*pbh)->b_size - *offset) {
 264                 blocknr = (*pbh)->b_blocknr;
 265                 brelse(*pbh);
 266                 *pbh = sb_bread(sb, blocknr + 1);
 267                 if (unlikely(!*pbh))
 268                         return NULL;
 269                 *offset = 0;
 270         }
 271         ptr = (*pbh)->b_data + *offset;
 272         *offset += bytes;
 273         return ptr;
 274 }
 275
 276 static void segsum_skip(struct super_block *sb, struct buffer_head **pbh,
 277                         unsigned int *offset, unsigned int bytes,
 278                         unsigned long count)
 279 {
 280         unsigned int rest_item_in_current_block
 281                 = ((*pbh)->b_size - *offset) / bytes;
 282
 283         if (count <= rest_item_in_current_block) {
 284                 *offset += bytes * count;
 285         } else {
 286                 sector_t blocknr = (*pbh)->b_blocknr;
 287                 unsigned int nitem_per_block = (*pbh)->b_size / bytes;
 288                 unsigned int bcnt;
 289
 290                 count -= rest_item_in_current_block;
 291                 bcnt = DIV_ROUND_UP(count, nitem_per_block);
 292                 *offset = bytes * (count - (bcnt - 1) * nitem_per_block);
 293
 294                 brelse(*pbh);
 295                 *pbh = sb_bread(sb, blocknr + bcnt);
 296         }
 297 }
 298
 299 static int
 300 collect_blocks_from_segsum(struct nilfs_sb_info *sbi, sector_t sum_blocknr,
 301                            struct nilfs_segsum_info *ssi,
 302                            struct list_head *head)
 303 {
 304         struct buffer_head *bh;
 305         unsigned int offset;
 306         unsigned long nfinfo = ssi->nfinfo;
 307         sector_t blocknr = sum_blocknr + ssi->nsumblk;
 308         ino_t ino;
 309         int err = -EIO;
 310
 311         if (!nfinfo)
 312                 return 0;
 313
 314         bh = sb_bread(sbi->s_super, sum_blocknr);
 315         if (unlikely(!bh))
 316                 goto out;
 317
 318         offset = le16_to_cpu(
 319                 ((struct nilfs_segment_summary *)bh->b_data)->ss_bytes);
 320         for (;;) {
 321                 unsigned long nblocks, ndatablk, nnodeblk;
 322                 struct nilfs_finfo *finfo;
 323
 324                 finfo = segsum_get(sbi->s_super, &bh, &offset, sizeof(*finfo));
 325                 if (unlikely(!finfo))
 326                         goto out;
 327
 328                 ino = le64_to_cpu(finfo->fi_ino);
 329                 nblocks = le32_to_cpu(finfo->fi_nblocks);
 330                 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
 331                 nnodeblk = nblocks - ndatablk;
 332
 333                 while (ndatablk-- > 0) {
 334                         struct nilfs_recovery_block *rb;
 335                         struct nilfs_binfo_v *binfo;
 336
 337                         binfo = segsum_get(sbi->s_super, &bh, &offset,
 338                                            sizeof(*binfo));
 339                         if (unlikely(!binfo))
 340                                 goto out;
 341
 342                         rb = kmalloc(sizeof(*rb), GFP_NOFS);
 343                         if (unlikely(!rb)) {
 344                                 err = -ENOMEM;
 345                                 goto out;
 346                         }
 347                         rb->ino = ino;
 348                         rb->blocknr = blocknr++;
 349                         rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
 350                         rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
 351                         /* INIT_LIST_HEAD(&rb->list); */
 352                         list_add_tail(&rb->list, head);
 353                 }
 354                 if (--nfinfo == 0)
 355                         break;
 356                 blocknr += nnodeblk; /* always 0 for the data sync segments */
 357                 segsum_skip(sbi->s_super, &bh, &offset, sizeof(__le64),
 358                             nnodeblk);
 359                 if (unlikely(!bh))
 360                         goto out;
 361         }
 362         err = 0;
 363  out:
 364         brelse(bh);   /* brelse(NULL) is just ignored */
 365         return err;
 366 }
 367
 368 static void dispose_recovery_list(struct list_head *head)
 369 {
 370         while (!list_empty(head)) {
 371                 struct nilfs_recovery_block *rb
 372                         = list_entry(head->next,
 373                                      struct nilfs_recovery_block, list);
 374                 list_del(&rb->list);
 375                 kfree(rb);
 376         }
 377 }
 378
 379 struct nilfs_segment_entry {
 380         struct list_head        list;
 381         __u64                   segnum;
 382 };
 383
 384 static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
 385 {
 386         struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
 387
 388         if (unlikely(!ent))
 389                 return -ENOMEM;
 390
 391         ent->segnum = segnum;
 392         INIT_LIST_HEAD(&ent->list);
 393         list_add_tail(&ent->list, head);
 394         return 0;
 395 }
 396
 397 void nilfs_dispose_segment_list(struct list_head *head)
 398 {
 399         while (!list_empty(head)) {
 400                 struct nilfs_segment_entry *ent
 401                         = list_entry(head->next,
 402                                      struct nilfs_segment_entry, list);
 403                 list_del(&ent->list);
 404                 kfree(ent);
 405         }
 406 }
 407
 408 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
 409                                               struct nilfs_sb_info *sbi,
 410                                               struct nilfs_recovery_info *ri)
 411 {
 412         struct list_head *head = &ri->ri_used_segments;
 413         struct nilfs_segment_entry *ent, *n;
 414         struct inode *sufile = nilfs->ns_sufile;
 415         __u64 segnum[4];
 416         int err;
 417         int i;
 418
 419         segnum[0] = nilfs->ns_segnum;
 420         segnum[1] = nilfs->ns_nextnum;
 421         segnum[2] = ri->ri_segnum;
 422         segnum[3] = ri->ri_nextnum;
 423
 424         nilfs_attach_writer(nilfs, sbi);
 425         /*
 426          * Releasing the next segment of the latest super root.
 427          * The next segment is invalidated by this recovery.
 428          */
 429         err = nilfs_sufile_free(sufile, segnum[1]);
 430         if (unlikely(err))
 431                 goto failed;
 432
 433         for (i = 1; i < 4; i++) {
 434                 err = nilfs_segment_list_add(head, segnum[i]);
 435                 if (unlikely(err))
 436                         goto failed;
 437         }
 438
 439         /*
 440          * Collecting segments written after the latest super root.
 441          * These are marked dirty to avoid being reallocated in the next write.
 442          */
 443         list_for_each_entry_safe(ent, n, head, list) {
 444                 if (ent->segnum != segnum[0]) {
 445                         err = nilfs_sufile_scrap(sufile, ent->segnum);
 446                         if (unlikely(err))
 447                                 goto failed;
 448                 }
 449                 list_del(&ent->list);
 450                 kfree(ent);
 451         }
 452
 453         /* Allocate new segments for recovery */
 454         err = nilfs_sufile_alloc(sufile, &segnum[0]);
 455         if (unlikely(err))
 456                 goto failed;
 457
 458         nilfs->ns_pseg_offset = 0;
 459         nilfs->ns_seg_seq = ri->ri_seq + 2;
 460         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
 461
 462  failed:
 463         /* No need to recover sufile because it will be destroyed on error */
 464         nilfs_detach_writer(nilfs, sbi);
 465         return err;
 466 }
 467
 468 static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi,
 469                                      struct nilfs_recovery_block *rb,
 470                                      struct page *page)
 471 {
 472         struct buffer_head *bh_org;
 473         void *kaddr;
 474
 475         bh_org = sb_bread(sbi->s_super, rb->blocknr);
 476         if (unlikely(!bh_org))
 477                 return -EIO;
 478
 479         kaddr = kmap_atomic(page, KM_USER0);
 480         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
 481         kunmap_atomic(kaddr, KM_USER0);
 482         brelse(bh_org);
 483         return 0;
 484 }
 485
 486 static int recover_dsync_blocks(struct nilfs_sb_info *sbi,
 487                                 struct list_head *head,
 488                                 unsigned long *nr_salvaged_blocks)
 489 {
 490         struct inode *inode;
 491         struct nilfs_recovery_block *rb, *n;
 492         unsigned blocksize = sbi->s_super->s_blocksize;
 493         struct page *page;
 494         loff_t pos;
 495         int err = 0, err2 = 0;
 496
 497         list_for_each_entry_safe(rb, n, head, list) {
 498                 inode = nilfs_iget(sbi->s_super, rb->ino);
 499                 if (IS_ERR(inode)) {
 500                         err = PTR_ERR(inode);
 501                         inode = NULL;
 502                         goto failed_inode;
 503                 }
 504
 505                 pos = rb->blkoff << inode->i_blkbits;
 506                 page = NULL;
 507                 err = block_write_begin(NULL, inode->i_mapping, pos, blocksize,
 508                                         0, &page, NULL, nilfs_get_block);
 509                 if (unlikely(err))
 510                         goto failed_inode;
 511
 512                 err = nilfs_recovery_copy_block(sbi, rb, page);
 513                 if (unlikely(err))
 514                         goto failed_page;
 515
 516                 err = nilfs_set_file_dirty(sbi, inode, 1);
 517                 if (unlikely(err))
 518                         goto failed_page;
 519
 520                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
 521                                 blocksize, page, NULL);
 522
 523                 unlock_page(page);
 524                 page_cache_release(page);
 525
 526                 (*nr_salvaged_blocks)++;
 527                 goto next;
 528
 529  failed_page:
 530                 unlock_page(page);
 531                 page_cache_release(page);
 532
 533  failed_inode:
 534                 printk(KERN_WARNING
 535                        "NILFS warning: error recovering data block "
 536                        "(err=%d, ino=%lu, block-offset=%llu)\n",
 537                        err, (unsigned long)rb->ino,
 538                        (unsigned long long)rb->blkoff);
 539                 if (!err2)
 540                         err2 = err;
 541  next:
 542                 iput(inode); /* iput(NULL) is just ignored */
 543                 list_del_init(&rb->list);
 544                 kfree(rb);
 545         }
 546         return err2;
 547 }
 548
 549 /**
 550  * nilfs_do_roll_forward - salvage logical segments newer than the latest
 551  * checkpoint
 552  * @sbi: nilfs_sb_info
 553  * @nilfs: the_nilfs
 554  * @ri: pointer to a nilfs_recovery_info
 555  */
 556 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
 557                                  struct nilfs_sb_info *sbi,
 558                                  struct nilfs_recovery_info *ri)
 559 {
 560         struct nilfs_segsum_info ssi;
 561         sector_t pseg_start;
 562         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
 563         unsigned long nsalvaged_blocks = 0;
 564         u64 seg_seq;
 565         __u64 segnum, nextnum = 0;
 566         int empty_seg = 0;
 567         int err = 0, ret;
 568         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
 569         enum {
 570                 RF_INIT_ST,
 571                 RF_DSYNC_ST,   /* scanning data-sync segments */
 572         };
 573         int state = RF_INIT_ST;
 574
 575         nilfs_attach_writer(nilfs, sbi);
 576         pseg_start = ri->ri_lsegs_start;
 577         seg_seq = ri->ri_lsegs_start_seq;
 578         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 579         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 580
 581         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
 582
 583                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi);
 584                 if (ret) {
 585                         if (ret == NILFS_SEG_FAIL_IO) {
 586                                 err = -EIO;
 587                                 goto failed;
 588                         }
 589                         goto strayed;
 590                 }
 591                 if (unlikely(NILFS_SEG_HAS_SR(&ssi)))
 592                         goto confused;
 593
 594                 /* Found a valid partial segment; do recovery actions */
 595                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 596                 empty_seg = 0;
 597                 nilfs->ns_ctime = ssi.ctime;
 598                 if (!(ssi.flags & NILFS_SS_GC))
 599                         nilfs->ns_nongc_ctime = ssi.ctime;
 600
 601                 switch (state) {
 602                 case RF_INIT_ST:
 603                         if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi))
 604                                 goto try_next_pseg;
 605                         state = RF_DSYNC_ST;
 606                         /* Fall through */
 607                 case RF_DSYNC_ST:
 608                         if (!NILFS_SEG_DSYNC(&ssi))
 609                                 goto confused;
 610
 611                         err = collect_blocks_from_segsum(
 612                                 sbi, pseg_start, &ssi, &dsync_blocks);
 613                         if (unlikely(err))
 614                                 goto failed;
 615                         if (NILFS_SEG_LOGEND(&ssi)) {
 616                                 err = recover_dsync_blocks(
 617                                         sbi, &dsync_blocks, &nsalvaged_blocks);
 618                                 if (unlikely(err))
 619                                         goto failed;
 620                                 state = RF_INIT_ST;
 621                         }
 622                         break; /* Fall through to try_next_pseg */
 623                 }
 624
 625  try_next_pseg:
 626                 if (pseg_start == ri->ri_lsegs_end)
 627                         break;
 628                 pseg_start += ssi.nblocks;
 629                 if (pseg_start < seg_end)
 630                         continue;
 631                 goto feed_segment;
 632
 633  strayed:
 634                 if (pseg_start == ri->ri_lsegs_end)
 635                         break;
 636
 637  feed_segment:
 638                 /* Looking to the next full segment */
 639                 if (empty_seg++)
 640                         break;
 641                 seg_seq++;
 642                 segnum = nextnum;
 643                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 644                 pseg_start = seg_start;
 645         }
 646
 647         if (nsalvaged_blocks) {
 648                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
 649                        sbi->s_super->s_id, nsalvaged_blocks);
 650                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
 651         }
 652  out:
 653         dispose_recovery_list(&dsync_blocks);
 654         nilfs_detach_writer(sbi->s_nilfs, sbi);
 655         return err;
 656
 657  confused:
 658         err = -EINVAL;
 659  failed:
 660         printk(KERN_ERR
 661                "NILFS (device %s): Error roll-forwarding "
 662                "(err=%d, pseg block=%llu). ",
 663                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
 664         goto out;
 665 }
 666
 667 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
 668                                       struct nilfs_sb_info *sbi,
 669                                       struct nilfs_recovery_info *ri)
 670 {
 671         struct buffer_head *bh;
 672         int err;
 673
 674         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
 675             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
 676                 return;
 677
 678         bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start);
 679         BUG_ON(!bh);
 680         memset(bh->b_data, 0, bh->b_size);
 681         set_buffer_dirty(bh);
 682         err = sync_dirty_buffer(bh);
 683         if (unlikely(err))
 684                 printk(KERN_WARNING
 685                        "NILFS warning: buffer sync write failed during "
 686                        "post-cleaning of recovery.\n");
 687         brelse(bh);
 688 }
 689
 690 /**
 691  * nilfs_recover_logical_segments - salvage logical segments written after
 692  * the latest super root
 693  * @nilfs: the_nilfs
 694  * @sbi: nilfs_sb_info
 695  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 696  *
 697  * Return Value: On success, 0 is returned.  On error, one of the following
 698  * negative error code is returned.
 699  *
 700  * %-EINVAL - Inconsistent filesystem state.
 701  *
 702  * %-EIO - I/O error
 703  *
 704  * %-ENOSPC - No space left on device (only in a panic state).
 705  *
 706  * %-ERESTARTSYS - Interrupted.
 707  *
 708  * %-ENOMEM - Insufficient memory available.
 709  */
 710 int nilfs_recover_logical_segments(struct the_nilfs *nilfs,
 711                                    struct nilfs_sb_info *sbi,
 712                                    struct nilfs_recovery_info *ri)
 713 {
 714         int err;
 715
 716         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
 717                 return 0;
 718
 719         err = nilfs_attach_checkpoint(sbi, ri->ri_cno);
 720         if (unlikely(err)) {
 721                 printk(KERN_ERR
 722                        "NILFS: error loading the latest checkpoint.\n");
 723                 return err;
 724         }
 725
 726         err = nilfs_do_roll_forward(nilfs, sbi, ri);
 727         if (unlikely(err))
 728                 goto failed;
 729
 730         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
 731                 err = nilfs_prepare_segment_for_recovery(nilfs, sbi, ri);
 732                 if (unlikely(err)) {
 733                         printk(KERN_ERR "NILFS: Error preparing segments for "
 734                                "recovery.\n");
 735                         goto failed;
 736                 }
 737
 738                 err = nilfs_attach_segment_constructor(sbi);
 739                 if (unlikely(err))
 740                         goto failed;
 741
 742                 set_nilfs_discontinued(nilfs);
 743                 err = nilfs_construct_segment(sbi->s_super);
 744                 nilfs_detach_segment_constructor(sbi);
 745
 746                 if (unlikely(err)) {
 747                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
 748                                "(err=%d)\n", err);
 749                         goto failed;
 750                 }
 751
 752                 nilfs_finish_roll_forward(nilfs, sbi, ri);
 753         }
 754
 755  failed:
 756         nilfs_detach_checkpoint(sbi);
 757         return err;
 758 }
 759
 760 /**
 761  * nilfs_search_super_root - search the latest valid super root
 762  * @nilfs: the_nilfs
 763  * @sbi: nilfs_sb_info
 764  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 765  *
 766  * nilfs_search_super_root() looks for the latest super-root from a partial
 767  * segment pointed by the superblock.  It sets up struct the_nilfs through
 768  * this search. It fills nilfs_recovery_info (ri) required for recovery.
 769  *
 770  * Return Value: On success, 0 is returned.  On error, one of the following
 771  * negative error code is returned.
 772  *
 773  * %-EINVAL - No valid segment found
 774  *
 775  * %-EIO - I/O error
 776  */
 777 int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi,
 778                             struct nilfs_recovery_info *ri)
 779 {
 780         struct nilfs_segsum_info ssi;
 781         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
 782         sector_t seg_start, seg_end; /* range of full segment (block number) */
 783         sector_t b, end;
 784         u64 seg_seq;
 785         __u64 segnum, nextnum = 0;
 786         __u64 cno;
 787         LIST_HEAD(segments);
 788         int empty_seg = 0, scan_newer = 0;
 789         int ret;
 790
 791         pseg_start = nilfs->ns_last_pseg;
 792         seg_seq = nilfs->ns_last_seq;
 793         cno = nilfs->ns_last_cno;
 794         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 795
 796         /* Calculate range of segment */
 797         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 798
 799         /* Read ahead segment */
 800         b = seg_start;
 801         while (b <= seg_end)
 802                 sb_breadahead(sbi->s_super, b++);
 803
 804         for (;;) {
 805                 /* Load segment summary */
 806                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi);
 807                 if (ret) {
 808                         if (ret == NILFS_SEG_FAIL_IO)
 809                                 goto failed;
 810                         goto strayed;
 811                 }
 812                 pseg_end = pseg_start + ssi.nblocks - 1;
 813                 if (unlikely(pseg_end > seg_end)) {
 814                         ret = NILFS_SEG_FAIL_CONSISTENCY;
 815                         goto strayed;
 816                 }
 817
 818                 /* A valid partial segment */
 819                 ri->ri_pseg_start = pseg_start;
 820                 ri->ri_seq = seg_seq;
 821                 ri->ri_segnum = segnum;
 822                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 823                 ri->ri_nextnum = nextnum;
 824                 empty_seg = 0;
 825
 826                 if (!NILFS_SEG_HAS_SR(&ssi) && !scan_newer) {
 827                         /* This will never happen because a superblock
 828                            (last_segment) always points to a pseg
 829                            having a super root. */
 830                         ret = NILFS_SEG_FAIL_CONSISTENCY;
 831                         goto failed;
 832                 }
 833
 834                 if (pseg_start == seg_start) {
 835                         nilfs_get_segment_range(nilfs, nextnum, &b, &end);
 836                         while (b <= end)
 837                                 sb_breadahead(sbi->s_super, b++);
 838                 }
 839                 if (!NILFS_SEG_HAS_SR(&ssi)) {
 840                         if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) {
 841                                 ri->ri_lsegs_start = pseg_start;
 842                                 ri->ri_lsegs_start_seq = seg_seq;
 843                         }
 844                         if (NILFS_SEG_LOGEND(&ssi))
 845                                 ri->ri_lsegs_end = pseg_start;
 846                         goto try_next_pseg;
 847                 }
 848
 849                 /* A valid super root was found. */
 850                 ri->ri_cno = cno++;
 851                 ri->ri_super_root = pseg_end;
 852                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
 853
 854                 nilfs_dispose_segment_list(&segments);
 855                 nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start)
 856                         + ssi.nblocks - seg_start;
 857                 nilfs->ns_seg_seq = seg_seq;
 858                 nilfs->ns_segnum = segnum;
 859                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
 860                 nilfs->ns_ctime = ssi.ctime;
 861                 nilfs->ns_nextnum = nextnum;
 862
 863                 if (scan_newer)
 864                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
 865                 else {
 866                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
 867                                 goto super_root_found;
 868                         scan_newer = 1;
 869                 }
 870
 871                 /* reset region for roll-forward */
 872                 pseg_start += ssi.nblocks;
 873                 if (pseg_start < seg_end)
 874                         continue;
 875                 goto feed_segment;
 876
 877  try_next_pseg:
 878                 /* Standing on a course, or met an inconsistent state */
 879                 pseg_start += ssi.nblocks;
 880                 if (pseg_start < seg_end)
 881                         continue;
 882                 goto feed_segment;
 883
 884  strayed:
 885                 /* Off the trail */
 886                 if (!scan_newer)
 887                         /*
 888                          * This can happen if a checkpoint was written without
 889                          * barriers, or as a result of an I/O failure.
 890                          */
 891                         goto failed;
 892
 893  feed_segment:
 894                 /* Looking to the next full segment */
 895                 if (empty_seg++)
 896                         goto super_root_found; /* found a valid super root */
 897
 898                 ret = nilfs_segment_list_add(&segments, segnum);
 899                 if (unlikely(ret))
 900                         goto failed;
 901
 902                 seg_seq++;
 903                 segnum = nextnum;
 904                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 905                 pseg_start = seg_start;
 906         }
 907
 908  super_root_found:
 909         /* Updating pointers relating to the latest checkpoint */
 910         list_splice_tail(&segments, &ri->ri_used_segments);
 911         nilfs->ns_last_pseg = sr_pseg_start;
 912         nilfs->ns_last_seq = nilfs->ns_seg_seq;
 913         nilfs->ns_last_cno = ri->ri_cno;
 914         return 0;
 915
 916  failed:
 917         nilfs_dispose_segment_list(&segments);
 918         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
 919 }