SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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         /* need to verify ->ss_bytes field if read ->ss_cno */
 110 }
 111
 112 /**
 113  * calc_crc_cont - check CRC of blocks continuously
 114  * @sbi: nilfs_sb_info
 115  * @bhs: buffer head of start block
 116  * @sum: place to store result
 117  * @offset: offset bytes in the first block
 118  * @check_bytes: number of bytes to be checked
 119  * @start: DBN of start block
 120  * @nblock: number of blocks to be checked
 121  */
 122 static int calc_crc_cont(struct nilfs_sb_info *sbi, struct buffer_head *bhs,
 123                          u32 *sum, unsigned long offset, u64 check_bytes,
 124                          sector_t start, unsigned long nblock)
 125 {
 126         unsigned long blocksize = sbi->s_super->s_blocksize;
 127         unsigned long size;
 128         u32 crc;
 129
 130         BUG_ON(offset >= blocksize);
 131         check_bytes -= offset;
 132         size = min_t(u64, check_bytes, blocksize - offset);
 133         crc = crc32_le(sbi->s_nilfs->ns_crc_seed,
 134                        (unsigned char *)bhs->b_data + offset, size);
 135         if (--nblock > 0) {
 136                 do {
 137                         struct buffer_head *bh
 138                                 = sb_bread(sbi->s_super, ++start);
 139                         if (!bh)
 140                                 return -EIO;
 141                         check_bytes -= size;
 142                         size = min_t(u64, check_bytes, blocksize);
 143                         crc = crc32_le(crc, bh->b_data, size);
 144                         brelse(bh);
 145                 } while (--nblock > 0);
 146         }
 147         *sum = crc;
 148         return 0;
 149 }
 150
 151 /**
 152  * nilfs_read_super_root_block - read super root block
 153  * @sb: super_block
 154  * @sr_block: disk block number of the super root block
 155  * @pbh: address of a buffer_head pointer to return super root buffer
 156  * @check: CRC check flag
 157  */
 158 int nilfs_read_super_root_block(struct super_block *sb, sector_t sr_block,
 159                                 struct buffer_head **pbh, int check)
 160 {
 161         struct buffer_head *bh_sr;
 162         struct nilfs_super_root *sr;
 163         u32 crc;
 164         int ret;
 165
 166         *pbh = NULL;
 167         bh_sr = sb_bread(sb, sr_block);
 168         if (unlikely(!bh_sr)) {
 169                 ret = NILFS_SEG_FAIL_IO;
 170                 goto failed;
 171         }
 172
 173         sr = (struct nilfs_super_root *)bh_sr->b_data;
 174         if (check) {
 175                 unsigned bytes = le16_to_cpu(sr->sr_bytes);
 176
 177                 if (bytes == 0 || bytes > sb->s_blocksize) {
 178                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 179                         goto failed_bh;
 180                 }
 181                 if (calc_crc_cont(NILFS_SB(sb), bh_sr, &crc,
 182                                   sizeof(sr->sr_sum), bytes, sr_block, 1)) {
 183                         ret = NILFS_SEG_FAIL_IO;
 184                         goto failed_bh;
 185                 }
 186                 if (crc != le32_to_cpu(sr->sr_sum)) {
 187                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 188                         goto failed_bh;
 189                 }
 190         }
 191         *pbh = bh_sr;
 192         return 0;
 193
 194  failed_bh:
 195         brelse(bh_sr);
 196
 197  failed:
 198         return nilfs_warn_segment_error(ret);
 199 }
 200
 201 /**
 202  * load_segment_summary - read segment summary of the specified partial segment
 203  * @sbi: nilfs_sb_info
 204  * @pseg_start: start disk block number of partial segment
 205  * @seg_seq: sequence number requested
 206  * @ssi: pointer to nilfs_segsum_info struct to store information
 207  */
 208 static int
 209 load_segment_summary(struct nilfs_sb_info *sbi, sector_t pseg_start,
 210                      u64 seg_seq, struct nilfs_segsum_info *ssi)
 211 {
 212         struct buffer_head *bh_sum;
 213         struct nilfs_segment_summary *sum;
 214         unsigned long nblock;
 215         u32 crc;
 216         int ret = NILFS_SEG_FAIL_IO;
 217
 218         bh_sum = sb_bread(sbi->s_super, pseg_start);
 219         if (!bh_sum)
 220                 goto out;
 221
 222         sum = (struct nilfs_segment_summary *)bh_sum->b_data;
 223
 224         /* Check consistency of segment summary */
 225         if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC) {
 226                 ret = NILFS_SEG_FAIL_MAGIC;
 227                 goto failed;
 228         }
 229         store_segsum_info(ssi, sum, sbi->s_super->s_blocksize);
 230         if (seg_seq != ssi->seg_seq) {
 231                 ret = NILFS_SEG_FAIL_SEQ;
 232                 goto failed;
 233         }
 234
 235         nblock = ssi->nblocks;
 236         if (unlikely(nblock == 0 ||
 237                      nblock > sbi->s_nilfs->ns_blocks_per_segment)) {
 238                 /* This limits the number of blocks read in the CRC check */
 239                 ret = NILFS_SEG_FAIL_CONSISTENCY;
 240                 goto failed;
 241         }
 242         if (calc_crc_cont(sbi, bh_sum, &crc, sizeof(sum->ss_datasum),
 243                           ((u64)nblock << sbi->s_super->s_blocksize_bits),
 244                           pseg_start, nblock)) {
 245                 ret = NILFS_SEG_FAIL_IO;
 246                 goto failed;
 247         }
 248         if (crc == le32_to_cpu(sum->ss_datasum))
 249                 ret = 0;
 250         else
 251                 ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
 252  failed:
 253         brelse(bh_sum);
 254  out:
 255         return ret;
 256 }
 257
 258 static void *segsum_get(struct super_block *sb, struct buffer_head **pbh,
 259                         unsigned int *offset, unsigned int bytes)
 260 {
 261         void *ptr;
 262         sector_t blocknr;
 263
 264         BUG_ON((*pbh)->b_size < *offset);
 265         if (bytes > (*pbh)->b_size - *offset) {
 266                 blocknr = (*pbh)->b_blocknr;
 267                 brelse(*pbh);
 268                 *pbh = sb_bread(sb, blocknr + 1);
 269                 if (unlikely(!*pbh))
 270                         return NULL;
 271                 *offset = 0;
 272         }
 273         ptr = (*pbh)->b_data + *offset;
 274         *offset += bytes;
 275         return ptr;
 276 }
 277
 278 static void segsum_skip(struct super_block *sb, struct buffer_head **pbh,
 279                         unsigned int *offset, unsigned int bytes,
 280                         unsigned long count)
 281 {
 282         unsigned int rest_item_in_current_block
 283                 = ((*pbh)->b_size - *offset) / bytes;
 284
 285         if (count <= rest_item_in_current_block) {
 286                 *offset += bytes * count;
 287         } else {
 288                 sector_t blocknr = (*pbh)->b_blocknr;
 289                 unsigned int nitem_per_block = (*pbh)->b_size / bytes;
 290                 unsigned int bcnt;
 291
 292                 count -= rest_item_in_current_block;
 293                 bcnt = DIV_ROUND_UP(count, nitem_per_block);
 294                 *offset = bytes * (count - (bcnt - 1) * nitem_per_block);
 295
 296                 brelse(*pbh);
 297                 *pbh = sb_bread(sb, blocknr + bcnt);
 298         }
 299 }
 300
 301 static int
 302 collect_blocks_from_segsum(struct nilfs_sb_info *sbi, sector_t sum_blocknr,
 303                            struct nilfs_segsum_info *ssi,
 304                            struct list_head *head)
 305 {
 306         struct buffer_head *bh;
 307         unsigned int offset;
 308         unsigned long nfinfo = ssi->nfinfo;
 309         sector_t blocknr = sum_blocknr + ssi->nsumblk;
 310         ino_t ino;
 311         int err = -EIO;
 312
 313         if (!nfinfo)
 314                 return 0;
 315
 316         bh = sb_bread(sbi->s_super, sum_blocknr);
 317         if (unlikely(!bh))
 318                 goto out;
 319
 320         offset = le16_to_cpu(
 321                 ((struct nilfs_segment_summary *)bh->b_data)->ss_bytes);
 322         for (;;) {
 323                 unsigned long nblocks, ndatablk, nnodeblk;
 324                 struct nilfs_finfo *finfo;
 325
 326                 finfo = segsum_get(sbi->s_super, &bh, &offset, sizeof(*finfo));
 327                 if (unlikely(!finfo))
 328                         goto out;
 329
 330                 ino = le64_to_cpu(finfo->fi_ino);
 331                 nblocks = le32_to_cpu(finfo->fi_nblocks);
 332                 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
 333                 nnodeblk = nblocks - ndatablk;
 334
 335                 while (ndatablk-- > 0) {
 336                         struct nilfs_recovery_block *rb;
 337                         struct nilfs_binfo_v *binfo;
 338
 339                         binfo = segsum_get(sbi->s_super, &bh, &offset,
 340                                            sizeof(*binfo));
 341                         if (unlikely(!binfo))
 342                                 goto out;
 343
 344                         rb = kmalloc(sizeof(*rb), GFP_NOFS);
 345                         if (unlikely(!rb)) {
 346                                 err = -ENOMEM;
 347                                 goto out;
 348                         }
 349                         rb->ino = ino;
 350                         rb->blocknr = blocknr++;
 351                         rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
 352                         rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
 353                         /* INIT_LIST_HEAD(&rb->list); */
 354                         list_add_tail(&rb->list, head);
 355                 }
 356                 if (--nfinfo == 0)
 357                         break;
 358                 blocknr += nnodeblk; /* always 0 for the data sync segments */
 359                 segsum_skip(sbi->s_super, &bh, &offset, sizeof(__le64),
 360                             nnodeblk);
 361                 if (unlikely(!bh))
 362                         goto out;
 363         }
 364         err = 0;
 365  out:
 366         brelse(bh);   /* brelse(NULL) is just ignored */
 367         return err;
 368 }
 369
 370 static void dispose_recovery_list(struct list_head *head)
 371 {
 372         while (!list_empty(head)) {
 373                 struct nilfs_recovery_block *rb
 374                         = list_entry(head->next,
 375                                      struct nilfs_recovery_block, list);
 376                 list_del(&rb->list);
 377                 kfree(rb);
 378         }
 379 }
 380
 381 struct nilfs_segment_entry {
 382         struct list_head        list;
 383         __u64                   segnum;
 384 };
 385
 386 static int nilfs_segment_list_add(struct list_head *head, __u64 segnum)
 387 {
 388         struct nilfs_segment_entry *ent = kmalloc(sizeof(*ent), GFP_NOFS);
 389
 390         if (unlikely(!ent))
 391                 return -ENOMEM;
 392
 393         ent->segnum = segnum;
 394         INIT_LIST_HEAD(&ent->list);
 395         list_add_tail(&ent->list, head);
 396         return 0;
 397 }
 398
 399 void nilfs_dispose_segment_list(struct list_head *head)
 400 {
 401         while (!list_empty(head)) {
 402                 struct nilfs_segment_entry *ent
 403                         = list_entry(head->next,
 404                                      struct nilfs_segment_entry, list);
 405                 list_del(&ent->list);
 406                 kfree(ent);
 407         }
 408 }
 409
 410 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
 411                                               struct nilfs_sb_info *sbi,
 412                                               struct nilfs_recovery_info *ri)
 413 {
 414         struct list_head *head = &ri->ri_used_segments;
 415         struct nilfs_segment_entry *ent, *n;
 416         struct inode *sufile = nilfs->ns_sufile;
 417         __u64 segnum[4];
 418         int err;
 419         int i;
 420
 421         segnum[0] = nilfs->ns_segnum;
 422         segnum[1] = nilfs->ns_nextnum;
 423         segnum[2] = ri->ri_segnum;
 424         segnum[3] = ri->ri_nextnum;
 425
 426         nilfs_attach_writer(nilfs, sbi);
 427         /*
 428          * Releasing the next segment of the latest super root.
 429          * The next segment is invalidated by this recovery.
 430          */
 431         err = nilfs_sufile_free(sufile, segnum[1]);
 432         if (unlikely(err))
 433                 goto failed;
 434
 435         for (i = 1; i < 4; i++) {
 436                 err = nilfs_segment_list_add(head, segnum[i]);
 437                 if (unlikely(err))
 438                         goto failed;
 439         }
 440
 441         /*
 442          * Collecting segments written after the latest super root.
 443          * These are marked dirty to avoid being reallocated in the next write.
 444          */
 445         list_for_each_entry_safe(ent, n, head, list) {
 446                 if (ent->segnum != segnum[0]) {
 447                         err = nilfs_sufile_scrap(sufile, ent->segnum);
 448                         if (unlikely(err))
 449                                 goto failed;
 450                 }
 451                 list_del(&ent->list);
 452                 kfree(ent);
 453         }
 454
 455         /* Allocate new segments for recovery */
 456         err = nilfs_sufile_alloc(sufile, &segnum[0]);
 457         if (unlikely(err))
 458                 goto failed;
 459
 460         nilfs->ns_pseg_offset = 0;
 461         nilfs->ns_seg_seq = ri->ri_seq + 2;
 462         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
 463
 464  failed:
 465         /* No need to recover sufile because it will be destroyed on error */
 466         nilfs_detach_writer(nilfs, sbi);
 467         return err;
 468 }
 469
 470 static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi,
 471                                      struct nilfs_recovery_block *rb,
 472                                      struct page *page)
 473 {
 474         struct buffer_head *bh_org;
 475         void *kaddr;
 476
 477         bh_org = sb_bread(sbi->s_super, rb->blocknr);
 478         if (unlikely(!bh_org))
 479                 return -EIO;
 480
 481         kaddr = kmap_atomic(page, KM_USER0);
 482         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
 483         kunmap_atomic(kaddr, KM_USER0);
 484         brelse(bh_org);
 485         return 0;
 486 }
 487
 488 static int recover_dsync_blocks(struct nilfs_sb_info *sbi,
 489                                 struct list_head *head,
 490                                 unsigned long *nr_salvaged_blocks)
 491 {
 492         struct inode *inode;
 493         struct nilfs_recovery_block *rb, *n;
 494         unsigned blocksize = sbi->s_super->s_blocksize;
 495         struct page *page;
 496         loff_t pos;
 497         int err = 0, err2 = 0;
 498
 499         list_for_each_entry_safe(rb, n, head, list) {
 500                 inode = nilfs_iget(sbi->s_super, rb->ino);
 501                 if (IS_ERR(inode)) {
 502                         err = PTR_ERR(inode);
 503                         inode = NULL;
 504                         goto failed_inode;
 505                 }
 506
 507                 pos = rb->blkoff << inode->i_blkbits;
 508                 page = NULL;
 509                 err = block_write_begin(NULL, inode->i_mapping, pos, blocksize,
 510                                         0, &page, NULL, nilfs_get_block);
 511                 if (unlikely(err))
 512                         goto failed_inode;
 513
 514                 err = nilfs_recovery_copy_block(sbi, rb, page);
 515                 if (unlikely(err))
 516                         goto failed_page;
 517
 518                 err = nilfs_set_file_dirty(sbi, inode, 1);
 519                 if (unlikely(err))
 520                         goto failed_page;
 521
 522                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
 523                                 blocksize, page, NULL);
 524
 525                 unlock_page(page);
 526                 page_cache_release(page);
 527
 528                 (*nr_salvaged_blocks)++;
 529                 goto next;
 530
 531  failed_page:
 532                 unlock_page(page);
 533                 page_cache_release(page);
 534
 535  failed_inode:
 536                 printk(KERN_WARNING
 537                        "NILFS warning: error recovering data block "
 538                        "(err=%d, ino=%lu, block-offset=%llu)\n",
 539                        err, (unsigned long)rb->ino,
 540                        (unsigned long long)rb->blkoff);
 541                 if (!err2)
 542                         err2 = err;
 543  next:
 544                 iput(inode); /* iput(NULL) is just ignored */
 545                 list_del_init(&rb->list);
 546                 kfree(rb);
 547         }
 548         return err2;
 549 }
 550
 551 /**
 552  * nilfs_do_roll_forward - salvage logical segments newer than the latest
 553  * checkpoint
 554  * @sbi: nilfs_sb_info
 555  * @nilfs: the_nilfs
 556  * @ri: pointer to a nilfs_recovery_info
 557  */
 558 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
 559                                  struct nilfs_sb_info *sbi,
 560                                  struct nilfs_recovery_info *ri)
 561 {
 562         struct nilfs_segsum_info ssi;
 563         sector_t pseg_start;
 564         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
 565         unsigned long nsalvaged_blocks = 0;
 566         u64 seg_seq;
 567         __u64 segnum, nextnum = 0;
 568         int empty_seg = 0;
 569         int err = 0, ret;
 570         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
 571         enum {
 572                 RF_INIT_ST,
 573                 RF_DSYNC_ST,   /* scanning data-sync segments */
 574         };
 575         int state = RF_INIT_ST;
 576
 577         nilfs_attach_writer(nilfs, sbi);
 578         pseg_start = ri->ri_lsegs_start;
 579         seg_seq = ri->ri_lsegs_start_seq;
 580         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 581         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 582
 583         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
 584
 585                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi);
 586                 if (ret) {
 587                         if (ret == NILFS_SEG_FAIL_IO) {
 588                                 err = -EIO;
 589                                 goto failed;
 590                         }
 591                         goto strayed;
 592                 }
 593                 if (unlikely(NILFS_SEG_HAS_SR(&ssi)))
 594                         goto confused;
 595
 596                 /* Found a valid partial segment; do recovery actions */
 597                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 598                 empty_seg = 0;
 599                 nilfs->ns_ctime = ssi.ctime;
 600                 if (!(ssi.flags & NILFS_SS_GC))
 601                         nilfs->ns_nongc_ctime = ssi.ctime;
 602
 603                 switch (state) {
 604                 case RF_INIT_ST:
 605                         if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi))
 606                                 goto try_next_pseg;
 607                         state = RF_DSYNC_ST;
 608                         /* Fall through */
 609                 case RF_DSYNC_ST:
 610                         if (!NILFS_SEG_DSYNC(&ssi))
 611                                 goto confused;
 612
 613                         err = collect_blocks_from_segsum(
 614                                 sbi, pseg_start, &ssi, &dsync_blocks);
 615                         if (unlikely(err))
 616                                 goto failed;
 617                         if (NILFS_SEG_LOGEND(&ssi)) {
 618                                 err = recover_dsync_blocks(
 619                                         sbi, &dsync_blocks, &nsalvaged_blocks);
 620                                 if (unlikely(err))
 621                                         goto failed;
 622                                 state = RF_INIT_ST;
 623                         }
 624                         break; /* Fall through to try_next_pseg */
 625                 }
 626
 627  try_next_pseg:
 628                 if (pseg_start == ri->ri_lsegs_end)
 629                         break;
 630                 pseg_start += ssi.nblocks;
 631                 if (pseg_start < seg_end)
 632                         continue;
 633                 goto feed_segment;
 634
 635  strayed:
 636                 if (pseg_start == ri->ri_lsegs_end)
 637                         break;
 638
 639  feed_segment:
 640                 /* Looking to the next full segment */
 641                 if (empty_seg++)
 642                         break;
 643                 seg_seq++;
 644                 segnum = nextnum;
 645                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 646                 pseg_start = seg_start;
 647         }
 648
 649         if (nsalvaged_blocks) {
 650                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
 651                        sbi->s_super->s_id, nsalvaged_blocks);
 652                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
 653         }
 654  out:
 655         dispose_recovery_list(&dsync_blocks);
 656         nilfs_detach_writer(sbi->s_nilfs, sbi);
 657         return err;
 658
 659  confused:
 660         err = -EINVAL;
 661  failed:
 662         printk(KERN_ERR
 663                "NILFS (device %s): Error roll-forwarding "
 664                "(err=%d, pseg block=%llu). ",
 665                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
 666         goto out;
 667 }
 668
 669 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
 670                                       struct nilfs_sb_info *sbi,
 671                                       struct nilfs_recovery_info *ri)
 672 {
 673         struct buffer_head *bh;
 674         int err;
 675
 676         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
 677             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
 678                 return;
 679
 680         bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start);
 681         BUG_ON(!bh);
 682         memset(bh->b_data, 0, bh->b_size);
 683         set_buffer_dirty(bh);
 684         err = sync_dirty_buffer(bh);
 685         if (unlikely(err))
 686                 printk(KERN_WARNING
 687                        "NILFS warning: buffer sync write failed during "
 688                        "post-cleaning of recovery.\n");
 689         brelse(bh);
 690 }
 691
 692 /**
 693  * nilfs_recover_logical_segments - salvage logical segments written after
 694  * the latest super root
 695  * @nilfs: the_nilfs
 696  * @sbi: nilfs_sb_info
 697  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 698  *
 699  * Return Value: On success, 0 is returned.  On error, one of the following
 700  * negative error code is returned.
 701  *
 702  * %-EINVAL - Inconsistent filesystem state.
 703  *
 704  * %-EIO - I/O error
 705  *
 706  * %-ENOSPC - No space left on device (only in a panic state).
 707  *
 708  * %-ERESTARTSYS - Interrupted.
 709  *
 710  * %-ENOMEM - Insufficient memory available.
 711  */
 712 int nilfs_recover_logical_segments(struct the_nilfs *nilfs,
 713                                    struct nilfs_sb_info *sbi,
 714                                    struct nilfs_recovery_info *ri)
 715 {
 716         int err;
 717
 718         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
 719                 return 0;
 720
 721         err = nilfs_attach_checkpoint(sbi, ri->ri_cno);
 722         if (unlikely(err)) {
 723                 printk(KERN_ERR
 724                        "NILFS: error loading the latest checkpoint.\n");
 725                 return err;
 726         }
 727
 728         err = nilfs_do_roll_forward(nilfs, sbi, ri);
 729         if (unlikely(err))
 730                 goto failed;
 731
 732         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
 733                 err = nilfs_prepare_segment_for_recovery(nilfs, sbi, ri);
 734                 if (unlikely(err)) {
 735                         printk(KERN_ERR "NILFS: Error preparing segments for "
 736                                "recovery.\n");
 737                         goto failed;
 738                 }
 739
 740                 err = nilfs_attach_segment_constructor(sbi);
 741                 if (unlikely(err))
 742                         goto failed;
 743
 744                 set_nilfs_discontinued(nilfs);
 745                 err = nilfs_construct_segment(sbi->s_super);
 746                 nilfs_detach_segment_constructor(sbi);
 747
 748                 if (unlikely(err)) {
 749                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
 750                                "(err=%d)\n", err);
 751                         goto failed;
 752                 }
 753
 754                 nilfs_finish_roll_forward(nilfs, sbi, ri);
 755         }
 756
 757  failed:
 758         nilfs_detach_checkpoint(sbi);
 759         return err;
 760 }
 761
 762 /**
 763  * nilfs_search_super_root - search the latest valid super root
 764  * @nilfs: the_nilfs
 765  * @sbi: nilfs_sb_info
 766  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 767  *
 768  * nilfs_search_super_root() looks for the latest super-root from a partial
 769  * segment pointed by the superblock.  It sets up struct the_nilfs through
 770  * this search. It fills nilfs_recovery_info (ri) required for recovery.
 771  *
 772  * Return Value: On success, 0 is returned.  On error, one of the following
 773  * negative error code is returned.
 774  *
 775  * %-EINVAL - No valid segment found
 776  *
 777  * %-EIO - I/O error
 778  */
 779 int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi,
 780                             struct nilfs_recovery_info *ri)
 781 {
 782         struct nilfs_segsum_info ssi;
 783         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
 784         sector_t seg_start, seg_end; /* range of full segment (block number) */
 785         sector_t b, end;
 786         u64 seg_seq;
 787         __u64 segnum, nextnum = 0;
 788         __u64 cno;
 789         LIST_HEAD(segments);
 790         int empty_seg = 0, scan_newer = 0;
 791         int ret;
 792
 793         pseg_start = nilfs->ns_last_pseg;
 794         seg_seq = nilfs->ns_last_seq;
 795         cno = nilfs->ns_last_cno;
 796         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 797
 798         /* Calculate range of segment */
 799         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 800
 801         /* Read ahead segment */
 802         b = seg_start;
 803         while (b <= seg_end)
 804                 sb_breadahead(sbi->s_super, b++);
 805
 806         for (;;) {
 807                 /* Load segment summary */
 808                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi);
 809                 if (ret) {
 810                         if (ret == NILFS_SEG_FAIL_IO)
 811                                 goto failed;
 812                         goto strayed;
 813                 }
 814                 pseg_end = pseg_start + ssi.nblocks - 1;
 815                 if (unlikely(pseg_end > seg_end)) {
 816                         ret = NILFS_SEG_FAIL_CONSISTENCY;
 817                         goto strayed;
 818                 }
 819
 820                 /* A valid partial segment */
 821                 ri->ri_pseg_start = pseg_start;
 822                 ri->ri_seq = seg_seq;
 823                 ri->ri_segnum = segnum;
 824                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 825                 ri->ri_nextnum = nextnum;
 826                 empty_seg = 0;
 827
 828                 if (!NILFS_SEG_HAS_SR(&ssi) && !scan_newer) {
 829                         /* This will never happen because a superblock
 830                            (last_segment) always points to a pseg
 831                            having a super root. */
 832                         ret = NILFS_SEG_FAIL_CONSISTENCY;
 833                         goto failed;
 834                 }
 835
 836                 if (pseg_start == seg_start) {
 837                         nilfs_get_segment_range(nilfs, nextnum, &b, &end);
 838                         while (b <= end)
 839                                 sb_breadahead(sbi->s_super, b++);
 840                 }
 841                 if (!NILFS_SEG_HAS_SR(&ssi)) {
 842                         if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) {
 843                                 ri->ri_lsegs_start = pseg_start;
 844                                 ri->ri_lsegs_start_seq = seg_seq;
 845                         }
 846                         if (NILFS_SEG_LOGEND(&ssi))
 847                                 ri->ri_lsegs_end = pseg_start;
 848                         goto try_next_pseg;
 849                 }
 850
 851                 /* A valid super root was found. */
 852                 ri->ri_cno = cno++;
 853                 ri->ri_super_root = pseg_end;
 854                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
 855
 856                 nilfs_dispose_segment_list(&segments);
 857                 nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start)
 858                         + ssi.nblocks - seg_start;
 859                 nilfs->ns_seg_seq = seg_seq;
 860                 nilfs->ns_segnum = segnum;
 861                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
 862                 nilfs->ns_ctime = ssi.ctime;
 863                 nilfs->ns_nextnum = nextnum;
 864
 865                 if (scan_newer)
 866                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
 867                 else {
 868                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
 869                                 goto super_root_found;
 870                         scan_newer = 1;
 871                 }
 872
 873                 /* reset region for roll-forward */
 874                 pseg_start += ssi.nblocks;
 875                 if (pseg_start < seg_end)
 876                         continue;
 877                 goto feed_segment;
 878
 879  try_next_pseg:
 880                 /* Standing on a course, or met an inconsistent state */
 881                 pseg_start += ssi.nblocks;
 882                 if (pseg_start < seg_end)
 883                         continue;
 884                 goto feed_segment;
 885
 886  strayed:
 887                 /* Off the trail */
 888                 if (!scan_newer)
 889                         /*
 890                          * This can happen if a checkpoint was written without
 891                          * barriers, or as a result of an I/O failure.
 892                          */
 893                         goto failed;
 894
 895  feed_segment:
 896                 /* Looking to the next full segment */
 897                 if (empty_seg++)
 898                         goto super_root_found; /* found a valid super root */
 899
 900                 ret = nilfs_segment_list_add(&segments, segnum);
 901                 if (unlikely(ret))
 902                         goto failed;
 903
 904                 seg_seq++;
 905                 segnum = nextnum;
 906                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 907                 pseg_start = seg_start;
 908         }
 909
 910  super_root_found:
 911         /* Updating pointers relating to the latest checkpoint */
 912         list_splice_tail(&segments, &ri->ri_used_segments);
 913         nilfs->ns_last_pseg = sr_pseg_start;
 914         nilfs->ns_last_seq = nilfs->ns_seg_seq;
 915         nilfs->ns_last_cno = ri->ri_cno;
 916         return 0;
 917
 918  failed:
 919         nilfs_dispose_segment_list(&segments);
 920         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
 921 }