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 "seglist.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_SEGSUM,
  44         NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT,
  45         NILFS_SEG_FAIL_CHECKSUM_FULL,
  46         NILFS_SEG_FAIL_CONSISTENCY,
  47 };
  48
  49 /* work structure for recovery */
  50 struct nilfs_recovery_block {
  51         ino_t ino;              /* Inode number of the file that this block
  52                                    belongs to */
  53         sector_t blocknr;       /* block number */
  54         __u64 vblocknr;         /* virtual block number */
  55         unsigned long blkoff;   /* File offset of the data block (per block) */
  56         struct list_head list;
  57 };
  58
  59
  60 static int nilfs_warn_segment_error(int err)
  61 {
  62         switch (err) {
  63         case NILFS_SEG_FAIL_IO:
  64                 printk(KERN_WARNING
  65                        "NILFS warning: I/O error on loading last segment\n");
  66                 return -EIO;
  67         case NILFS_SEG_FAIL_MAGIC:
  68                 printk(KERN_WARNING
  69                        "NILFS warning: Segment magic number invalid\n");
  70                 break;
  71         case NILFS_SEG_FAIL_SEQ:
  72                 printk(KERN_WARNING
  73                        "NILFS warning: Sequence number mismatch\n");
  74                 break;
  75         case NILFS_SEG_FAIL_CHECKSUM_SEGSUM:
  76                 printk(KERN_WARNING
  77                        "NILFS warning: Checksum error in segment summary\n");
  78                 break;
  79         case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT:
  80                 printk(KERN_WARNING
  81                        "NILFS warning: Checksum error in super root\n");
  82                 break;
  83         case NILFS_SEG_FAIL_CHECKSUM_FULL:
  84                 printk(KERN_WARNING
  85                        "NILFS warning: Checksum error in segment payload\n");
  86                 break;
  87         case NILFS_SEG_FAIL_CONSISTENCY:
  88                 printk(KERN_WARNING
  89                        "NILFS warning: Inconsistent segment\n");
  90                 break;
  91         case NILFS_SEG_NO_SUPER_ROOT:
  92                 printk(KERN_WARNING
  93                        "NILFS warning: No super root in the last segment\n");
  94                 break;
  95         case NILFS_SEG_VALID:
  96         default:
  97                 BUG();
  98         }
  99         return -EINVAL;
 100 }
 101
 102 static void store_segsum_info(struct nilfs_segsum_info *ssi,
 103                               struct nilfs_segment_summary *sum,
 104                               unsigned int blocksize)
 105 {
 106         ssi->flags = le16_to_cpu(sum->ss_flags);
 107         ssi->seg_seq = le64_to_cpu(sum->ss_seq);
 108         ssi->ctime = le64_to_cpu(sum->ss_create);
 109         ssi->next = le64_to_cpu(sum->ss_next);
 110         ssi->nblocks = le32_to_cpu(sum->ss_nblocks);
 111         ssi->nfinfo = le32_to_cpu(sum->ss_nfinfo);
 112         ssi->sumbytes = le32_to_cpu(sum->ss_sumbytes);
 113
 114         ssi->nsumblk = DIV_ROUND_UP(ssi->sumbytes, blocksize);
 115         ssi->nfileblk = ssi->nblocks - ssi->nsumblk - !!NILFS_SEG_HAS_SR(ssi);
 116 }
 117
 118 /**
 119  * calc_crc_cont - check CRC of blocks continuously
 120  * @sbi: nilfs_sb_info
 121  * @bhs: buffer head of start block
 122  * @sum: place to store result
 123  * @offset: offset bytes in the first block
 124  * @check_bytes: number of bytes to be checked
 125  * @start: DBN of start block
 126  * @nblock: number of blocks to be checked
 127  */
 128 static int calc_crc_cont(struct nilfs_sb_info *sbi, struct buffer_head *bhs,
 129                          u32 *sum, unsigned long offset, u64 check_bytes,
 130                          sector_t start, unsigned long nblock)
 131 {
 132         unsigned long blocksize = sbi->s_super->s_blocksize;
 133         unsigned long size;
 134         u32 crc;
 135
 136         BUG_ON(offset >= blocksize);
 137         check_bytes -= offset;
 138         size = min_t(u64, check_bytes, blocksize - offset);
 139         crc = crc32_le(sbi->s_nilfs->ns_crc_seed,
 140                        (unsigned char *)bhs->b_data + offset, size);
 141         if (--nblock > 0) {
 142                 do {
 143                         struct buffer_head *bh
 144                                 = sb_bread(sbi->s_super, ++start);
 145                         if (!bh)
 146                                 return -EIO;
 147                         check_bytes -= size;
 148                         size = min_t(u64, check_bytes, blocksize);
 149                         crc = crc32_le(crc, bh->b_data, size);
 150                         brelse(bh);
 151                 } while (--nblock > 0);
 152         }
 153         *sum = crc;
 154         return 0;
 155 }
 156
 157 /**
 158  * nilfs_read_super_root_block - read super root block
 159  * @sb: super_block
 160  * @sr_block: disk block number of the super root block
 161  * @pbh: address of a buffer_head pointer to return super root buffer
 162  * @check: CRC check flag
 163  */
 164 int nilfs_read_super_root_block(struct super_block *sb, sector_t sr_block,
 165                                 struct buffer_head **pbh, int check)
 166 {
 167         struct buffer_head *bh_sr;
 168         struct nilfs_super_root *sr;
 169         u32 crc;
 170         int ret;
 171
 172         *pbh = NULL;
 173         bh_sr = sb_bread(sb, sr_block);
 174         if (unlikely(!bh_sr)) {
 175                 ret = NILFS_SEG_FAIL_IO;
 176                 goto failed;
 177         }
 178
 179         sr = (struct nilfs_super_root *)bh_sr->b_data;
 180         if (check) {
 181                 unsigned bytes = le16_to_cpu(sr->sr_bytes);
 182
 183                 if (bytes == 0 || bytes > sb->s_blocksize) {
 184                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 185                         goto failed_bh;
 186                 }
 187                 if (calc_crc_cont(NILFS_SB(sb), bh_sr, &crc,
 188                                   sizeof(sr->sr_sum), bytes, sr_block, 1)) {
 189                         ret = NILFS_SEG_FAIL_IO;
 190                         goto failed_bh;
 191                 }
 192                 if (crc != le32_to_cpu(sr->sr_sum)) {
 193                         ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT;
 194                         goto failed_bh;
 195                 }
 196         }
 197         *pbh = bh_sr;
 198         return 0;
 199
 200  failed_bh:
 201         brelse(bh_sr);
 202
 203  failed:
 204         return nilfs_warn_segment_error(ret);
 205 }
 206
 207 /**
 208  * load_segment_summary - read segment summary of the specified partial segment
 209  * @sbi: nilfs_sb_info
 210  * @pseg_start: start disk block number of partial segment
 211  * @seg_seq: sequence number requested
 212  * @ssi: pointer to nilfs_segsum_info struct to store information
 213  * @full_check: full check flag
 214  *              (0: only checks segment summary CRC, 1: data CRC)
 215  */
 216 static int
 217 load_segment_summary(struct nilfs_sb_info *sbi, sector_t pseg_start,
 218                      u64 seg_seq, struct nilfs_segsum_info *ssi,
 219                      int full_check)
 220 {
 221         struct buffer_head *bh_sum;
 222         struct nilfs_segment_summary *sum;
 223         unsigned long offset, nblock;
 224         u64 check_bytes;
 225         u32 crc, crc_sum;
 226         int ret = NILFS_SEG_FAIL_IO;
 227
 228         bh_sum = sb_bread(sbi->s_super, pseg_start);
 229         if (!bh_sum)
 230                 goto out;
 231
 232         sum = (struct nilfs_segment_summary *)bh_sum->b_data;
 233
 234         /* Check consistency of segment summary */
 235         if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC) {
 236                 ret = NILFS_SEG_FAIL_MAGIC;
 237                 goto failed;
 238         }
 239         store_segsum_info(ssi, sum, sbi->s_super->s_blocksize);
 240         if (seg_seq != ssi->seg_seq) {
 241                 ret = NILFS_SEG_FAIL_SEQ;
 242                 goto failed;
 243         }
 244         if (full_check) {
 245                 offset = sizeof(sum->ss_datasum);
 246                 check_bytes =
 247                         ((u64)ssi->nblocks << sbi->s_super->s_blocksize_bits);
 248                 nblock = ssi->nblocks;
 249                 crc_sum = le32_to_cpu(sum->ss_datasum);
 250                 ret = NILFS_SEG_FAIL_CHECKSUM_FULL;
 251         } else { /* only checks segment summary */
 252                 offset = sizeof(sum->ss_datasum) + sizeof(sum->ss_sumsum);
 253                 check_bytes = ssi->sumbytes;
 254                 nblock = ssi->nsumblk;
 255                 crc_sum = le32_to_cpu(sum->ss_sumsum);
 256                 ret = NILFS_SEG_FAIL_CHECKSUM_SEGSUM;
 257         }
 258
 259         if (unlikely(nblock == 0 ||
 260                      nblock > sbi->s_nilfs->ns_blocks_per_segment)) {
 261                 /* This limits the number of blocks read in the CRC check */
 262                 ret = NILFS_SEG_FAIL_CONSISTENCY;
 263                 goto failed;
 264         }
 265         if (calc_crc_cont(sbi, bh_sum, &crc, offset, check_bytes,
 266                           pseg_start, nblock)) {
 267                 ret = NILFS_SEG_FAIL_IO;
 268                 goto failed;
 269         }
 270         if (crc == crc_sum)
 271                 ret = 0;
 272  failed:
 273         brelse(bh_sum);
 274  out:
 275         return ret;
 276 }
 277
 278 static void *segsum_get(struct super_block *sb, struct buffer_head **pbh,
 279                         unsigned int *offset, unsigned int bytes)
 280 {
 281         void *ptr;
 282         sector_t blocknr;
 283
 284         BUG_ON((*pbh)->b_size < *offset);
 285         if (bytes > (*pbh)->b_size - *offset) {
 286                 blocknr = (*pbh)->b_blocknr;
 287                 brelse(*pbh);
 288                 *pbh = sb_bread(sb, blocknr + 1);
 289                 if (unlikely(!*pbh))
 290                         return NULL;
 291                 *offset = 0;
 292         }
 293         ptr = (*pbh)->b_data + *offset;
 294         *offset += bytes;
 295         return ptr;
 296 }
 297
 298 static void segsum_skip(struct super_block *sb, struct buffer_head **pbh,
 299                         unsigned int *offset, unsigned int bytes,
 300                         unsigned long count)
 301 {
 302         unsigned int rest_item_in_current_block
 303                 = ((*pbh)->b_size - *offset) / bytes;
 304
 305         if (count <= rest_item_in_current_block) {
 306                 *offset += bytes * count;
 307         } else {
 308                 sector_t blocknr = (*pbh)->b_blocknr;
 309                 unsigned int nitem_per_block = (*pbh)->b_size / bytes;
 310                 unsigned int bcnt;
 311
 312                 count -= rest_item_in_current_block;
 313                 bcnt = DIV_ROUND_UP(count, nitem_per_block);
 314                 *offset = bytes * (count - (bcnt - 1) * nitem_per_block);
 315
 316                 brelse(*pbh);
 317                 *pbh = sb_bread(sb, blocknr + bcnt);
 318         }
 319 }
 320
 321 static int
 322 collect_blocks_from_segsum(struct nilfs_sb_info *sbi, sector_t sum_blocknr,
 323                            struct nilfs_segsum_info *ssi,
 324                            struct list_head *head)
 325 {
 326         struct buffer_head *bh;
 327         unsigned int offset;
 328         unsigned long nfinfo = ssi->nfinfo;
 329         sector_t blocknr = sum_blocknr + ssi->nsumblk;
 330         ino_t ino;
 331         int err = -EIO;
 332
 333         if (!nfinfo)
 334                 return 0;
 335
 336         bh = sb_bread(sbi->s_super, sum_blocknr);
 337         if (unlikely(!bh))
 338                 goto out;
 339
 340         offset = le16_to_cpu(
 341                 ((struct nilfs_segment_summary *)bh->b_data)->ss_bytes);
 342         for (;;) {
 343                 unsigned long nblocks, ndatablk, nnodeblk;
 344                 struct nilfs_finfo *finfo;
 345
 346                 finfo = segsum_get(sbi->s_super, &bh, &offset, sizeof(*finfo));
 347                 if (unlikely(!finfo))
 348                         goto out;
 349
 350                 ino = le64_to_cpu(finfo->fi_ino);
 351                 nblocks = le32_to_cpu(finfo->fi_nblocks);
 352                 ndatablk = le32_to_cpu(finfo->fi_ndatablk);
 353                 nnodeblk = nblocks - ndatablk;
 354
 355                 while (ndatablk-- > 0) {
 356                         struct nilfs_recovery_block *rb;
 357                         struct nilfs_binfo_v *binfo;
 358
 359                         binfo = segsum_get(sbi->s_super, &bh, &offset,
 360                                            sizeof(*binfo));
 361                         if (unlikely(!binfo))
 362                                 goto out;
 363
 364                         rb = kmalloc(sizeof(*rb), GFP_NOFS);
 365                         if (unlikely(!rb)) {
 366                                 err = -ENOMEM;
 367                                 goto out;
 368                         }
 369                         rb->ino = ino;
 370                         rb->blocknr = blocknr++;
 371                         rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr);
 372                         rb->blkoff = le64_to_cpu(binfo->bi_blkoff);
 373                         /* INIT_LIST_HEAD(&rb->list); */
 374                         list_add_tail(&rb->list, head);
 375                 }
 376                 if (--nfinfo == 0)
 377                         break;
 378                 blocknr += nnodeblk; /* always 0 for the data sync segments */
 379                 segsum_skip(sbi->s_super, &bh, &offset, sizeof(__le64),
 380                             nnodeblk);
 381                 if (unlikely(!bh))
 382                         goto out;
 383         }
 384         err = 0;
 385  out:
 386         brelse(bh);   /* brelse(NULL) is just ignored */
 387         return err;
 388 }
 389
 390 static void dispose_recovery_list(struct list_head *head)
 391 {
 392         while (!list_empty(head)) {
 393                 struct nilfs_recovery_block *rb
 394                         = list_entry(head->next,
 395                                      struct nilfs_recovery_block, list);
 396                 list_del(&rb->list);
 397                 kfree(rb);
 398         }
 399 }
 400
 401 void nilfs_dispose_segment_list(struct list_head *head)
 402 {
 403         while (!list_empty(head)) {
 404                 struct nilfs_segment_entry *ent
 405                         = list_entry(head->next,
 406                                      struct nilfs_segment_entry, list);
 407                 list_del(&ent->list);
 408                 nilfs_free_segment_entry(ent);
 409         }
 410 }
 411
 412 static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
 413                                               struct nilfs_recovery_info *ri)
 414 {
 415         struct list_head *head = &ri->ri_used_segments;
 416         struct nilfs_segment_entry *ent, *n;
 417         struct inode *sufile = nilfs->ns_sufile;
 418         __u64 segnum[4];
 419         time_t mtime;
 420         int err;
 421         int i;
 422
 423         segnum[0] = nilfs->ns_segnum;
 424         segnum[1] = nilfs->ns_nextnum;
 425         segnum[2] = ri->ri_segnum;
 426         segnum[3] = ri->ri_nextnum;
 427
 428         /*
 429          * Releasing the next segment of the latest super root.
 430          * The next segment is invalidated by this recovery.
 431          */
 432         err = nilfs_sufile_free(sufile, segnum[1]);
 433         if (unlikely(err))
 434                 goto failed;
 435
 436         err = -ENOMEM;
 437         for (i = 1; i < 4; i++) {
 438                 ent = nilfs_alloc_segment_entry(segnum[i]);
 439                 if (unlikely(!ent))
 440                         goto failed;
 441                 list_add_tail(&ent->list, head);
 442         }
 443
 444         /*
 445          * Collecting segments written after the latest super root.
 446          * These are marked dirty to avoid being reallocated in the next write.
 447          */
 448         mtime = get_seconds();
 449         list_for_each_entry_safe(ent, n, head, list) {
 450                 if (ent->segnum == segnum[0]) {
 451                         list_del(&ent->list);
 452                         nilfs_free_segment_entry(ent);
 453                         continue;
 454                 }
 455                 err = nilfs_open_segment_entry(ent, sufile);
 456                 if (unlikely(err))
 457                         goto failed;
 458                 if (!nilfs_segment_usage_dirty(ent->raw_su)) {
 459                         /* make the segment garbage */
 460                         ent->raw_su->su_nblocks = cpu_to_le32(0);
 461                         ent->raw_su->su_lastmod = cpu_to_le32(mtime);
 462                         nilfs_segment_usage_set_dirty(ent->raw_su);
 463                 }
 464                 list_del(&ent->list);
 465                 nilfs_close_segment_entry(ent, sufile);
 466                 nilfs_free_segment_entry(ent);
 467         }
 468
 469         /*
 470          * The segment having the latest super root is active, and
 471          * should be deactivated on the next construction for recovery.
 472          */
 473         err = -ENOMEM;
 474         ent = nilfs_alloc_segment_entry(segnum[0]);
 475         if (unlikely(!ent))
 476                 goto failed;
 477         list_add_tail(&ent->list, &ri->ri_used_segments);
 478
 479         /* Allocate new segments for recovery */
 480         err = nilfs_sufile_alloc(sufile, &segnum[0]);
 481         if (unlikely(err))
 482                 goto failed;
 483
 484         nilfs->ns_pseg_offset = 0;
 485         nilfs->ns_seg_seq = ri->ri_seq + 2;
 486         nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0];
 487         return 0;
 488
 489  failed:
 490         /* No need to recover sufile because it will be destroyed on error */
 491         return err;
 492 }
 493
 494 static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi,
 495                                      struct nilfs_recovery_block *rb,
 496                                      struct page *page)
 497 {
 498         struct buffer_head *bh_org;
 499         void *kaddr;
 500
 501         bh_org = sb_bread(sbi->s_super, rb->blocknr);
 502         if (unlikely(!bh_org))
 503                 return -EIO;
 504
 505         kaddr = kmap_atomic(page, KM_USER0);
 506         memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size);
 507         kunmap_atomic(kaddr, KM_USER0);
 508         brelse(bh_org);
 509         return 0;
 510 }
 511
 512 static int recover_dsync_blocks(struct nilfs_sb_info *sbi,
 513                                 struct list_head *head,
 514                                 unsigned long *nr_salvaged_blocks)
 515 {
 516         struct inode *inode;
 517         struct nilfs_recovery_block *rb, *n;
 518         unsigned blocksize = sbi->s_super->s_blocksize;
 519         struct page *page;
 520         loff_t pos;
 521         int err = 0, err2 = 0;
 522
 523         list_for_each_entry_safe(rb, n, head, list) {
 524                 inode = nilfs_iget(sbi->s_super, rb->ino);
 525                 if (IS_ERR(inode)) {
 526                         err = PTR_ERR(inode);
 527                         inode = NULL;
 528                         goto failed_inode;
 529                 }
 530
 531                 pos = rb->blkoff << inode->i_blkbits;
 532                 page = NULL;
 533                 err = block_write_begin(NULL, inode->i_mapping, pos, blocksize,
 534                                         0, &page, NULL, nilfs_get_block);
 535                 if (unlikely(err))
 536                         goto failed_inode;
 537
 538                 err = nilfs_recovery_copy_block(sbi, rb, page);
 539                 if (unlikely(err))
 540                         goto failed_page;
 541
 542                 err = nilfs_set_file_dirty(sbi, inode, 1);
 543                 if (unlikely(err))
 544                         goto failed_page;
 545
 546                 block_write_end(NULL, inode->i_mapping, pos, blocksize,
 547                                 blocksize, page, NULL);
 548
 549                 unlock_page(page);
 550                 page_cache_release(page);
 551
 552                 (*nr_salvaged_blocks)++;
 553                 goto next;
 554
 555  failed_page:
 556                 unlock_page(page);
 557                 page_cache_release(page);
 558
 559  failed_inode:
 560                 printk(KERN_WARNING
 561                        "NILFS warning: error recovering data block "
 562                        "(err=%d, ino=%lu, block-offset=%llu)\n",
 563                        err, rb->ino, (unsigned long long)rb->blkoff);
 564                 if (!err2)
 565                         err2 = err;
 566  next:
 567                 iput(inode); /* iput(NULL) is just ignored */
 568                 list_del_init(&rb->list);
 569                 kfree(rb);
 570         }
 571         return err2;
 572 }
 573
 574 /**
 575  * nilfs_do_roll_forward - salvage logical segments newer than the latest
 576  * checkpoint
 577  * @sbi: nilfs_sb_info
 578  * @nilfs: the_nilfs
 579  * @ri: pointer to a nilfs_recovery_info
 580  */
 581 static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
 582                                  struct nilfs_sb_info *sbi,
 583                                  struct nilfs_recovery_info *ri)
 584 {
 585         struct nilfs_segsum_info ssi;
 586         sector_t pseg_start;
 587         sector_t seg_start, seg_end;  /* Starting/ending DBN of full segment */
 588         unsigned long nsalvaged_blocks = 0;
 589         u64 seg_seq;
 590         __u64 segnum, nextnum = 0;
 591         int empty_seg = 0;
 592         int err = 0, ret;
 593         LIST_HEAD(dsync_blocks);  /* list of data blocks to be recovered */
 594         enum {
 595                 RF_INIT_ST,
 596                 RF_DSYNC_ST,   /* scanning data-sync segments */
 597         };
 598         int state = RF_INIT_ST;
 599
 600         nilfs_attach_writer(nilfs, sbi);
 601         pseg_start = ri->ri_lsegs_start;
 602         seg_seq = ri->ri_lsegs_start_seq;
 603         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 604         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 605
 606         while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) {
 607
 608                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
 609                 if (ret) {
 610                         if (ret == NILFS_SEG_FAIL_IO) {
 611                                 err = -EIO;
 612                                 goto failed;
 613                         }
 614                         goto strayed;
 615                 }
 616                 if (unlikely(NILFS_SEG_HAS_SR(&ssi)))
 617                         goto confused;
 618
 619                 /* Found a valid partial segment; do recovery actions */
 620                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 621                 empty_seg = 0;
 622                 nilfs->ns_ctime = ssi.ctime;
 623                 if (!(ssi.flags & NILFS_SS_GC))
 624                         nilfs->ns_nongc_ctime = ssi.ctime;
 625
 626                 switch (state) {
 627                 case RF_INIT_ST:
 628                         if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi))
 629                                 goto try_next_pseg;
 630                         state = RF_DSYNC_ST;
 631                         /* Fall through */
 632                 case RF_DSYNC_ST:
 633                         if (!NILFS_SEG_DSYNC(&ssi))
 634                                 goto confused;
 635
 636                         err = collect_blocks_from_segsum(
 637                                 sbi, pseg_start, &ssi, &dsync_blocks);
 638                         if (unlikely(err))
 639                                 goto failed;
 640                         if (NILFS_SEG_LOGEND(&ssi)) {
 641                                 err = recover_dsync_blocks(
 642                                         sbi, &dsync_blocks, &nsalvaged_blocks);
 643                                 if (unlikely(err))
 644                                         goto failed;
 645                                 state = RF_INIT_ST;
 646                         }
 647                         break; /* Fall through to try_next_pseg */
 648                 }
 649
 650  try_next_pseg:
 651                 if (pseg_start == ri->ri_lsegs_end)
 652                         break;
 653                 pseg_start += ssi.nblocks;
 654                 if (pseg_start < seg_end)
 655                         continue;
 656                 goto feed_segment;
 657
 658  strayed:
 659                 if (pseg_start == ri->ri_lsegs_end)
 660                         break;
 661
 662  feed_segment:
 663                 /* Looking to the next full segment */
 664                 if (empty_seg++)
 665                         break;
 666                 seg_seq++;
 667                 segnum = nextnum;
 668                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 669                 pseg_start = seg_start;
 670         }
 671
 672         if (nsalvaged_blocks) {
 673                 printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n",
 674                        sbi->s_super->s_id, nsalvaged_blocks);
 675                 ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE;
 676         }
 677  out:
 678         dispose_recovery_list(&dsync_blocks);
 679         nilfs_detach_writer(sbi->s_nilfs, sbi);
 680         return err;
 681
 682  confused:
 683         err = -EINVAL;
 684  failed:
 685         printk(KERN_ERR
 686                "NILFS (device %s): Error roll-forwarding "
 687                "(err=%d, pseg block=%llu). ",
 688                sbi->s_super->s_id, err, (unsigned long long)pseg_start);
 689         goto out;
 690 }
 691
 692 static void nilfs_finish_roll_forward(struct the_nilfs *nilfs,
 693                                       struct nilfs_sb_info *sbi,
 694                                       struct nilfs_recovery_info *ri)
 695 {
 696         struct buffer_head *bh;
 697         int err;
 698
 699         if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) !=
 700             nilfs_get_segnum_of_block(nilfs, ri->ri_super_root))
 701                 return;
 702
 703         bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start);
 704         BUG_ON(!bh);
 705         memset(bh->b_data, 0, bh->b_size);
 706         set_buffer_dirty(bh);
 707         err = sync_dirty_buffer(bh);
 708         if (unlikely(err))
 709                 printk(KERN_WARNING
 710                        "NILFS warning: buffer sync write failed during "
 711                        "post-cleaning of recovery.\n");
 712         brelse(bh);
 713 }
 714
 715 /**
 716  * nilfs_recover_logical_segments - salvage logical segments written after
 717  * the latest super root
 718  * @nilfs: the_nilfs
 719  * @sbi: nilfs_sb_info
 720  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 721  *
 722  * Return Value: On success, 0 is returned.  On error, one of the following
 723  * negative error code is returned.
 724  *
 725  * %-EINVAL - Inconsistent filesystem state.
 726  *
 727  * %-EIO - I/O error
 728  *
 729  * %-ENOSPC - No space left on device (only in a panic state).
 730  *
 731  * %-ERESTARTSYS - Interrupted.
 732  *
 733  * %-ENOMEM - Insufficient memory available.
 734  */
 735 int nilfs_recover_logical_segments(struct the_nilfs *nilfs,
 736                                    struct nilfs_sb_info *sbi,
 737                                    struct nilfs_recovery_info *ri)
 738 {
 739         int err;
 740
 741         if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
 742                 return 0;
 743
 744         err = nilfs_attach_checkpoint(sbi, ri->ri_cno);
 745         if (unlikely(err)) {
 746                 printk(KERN_ERR
 747                        "NILFS: error loading the latest checkpoint.\n");
 748                 return err;
 749         }
 750
 751         err = nilfs_do_roll_forward(nilfs, sbi, ri);
 752         if (unlikely(err))
 753                 goto failed;
 754
 755         if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) {
 756                 err = nilfs_prepare_segment_for_recovery(nilfs, ri);
 757                 if (unlikely(err)) {
 758                         printk(KERN_ERR "NILFS: Error preparing segments for "
 759                                "recovery.\n");
 760                         goto failed;
 761                 }
 762
 763                 err = nilfs_attach_segment_constructor(sbi, ri);
 764                 if (unlikely(err))
 765                         goto failed;
 766
 767                 set_nilfs_discontinued(nilfs);
 768                 err = nilfs_construct_segment(sbi->s_super);
 769                 nilfs_detach_segment_constructor(sbi);
 770
 771                 if (unlikely(err)) {
 772                         printk(KERN_ERR "NILFS: Oops! recovery failed. "
 773                                "(err=%d)\n", err);
 774                         goto failed;
 775                 }
 776
 777                 nilfs_finish_roll_forward(nilfs, sbi, ri);
 778         }
 779
 780         nilfs_detach_checkpoint(sbi);
 781         return 0;
 782
 783  failed:
 784         nilfs_detach_checkpoint(sbi);
 785         nilfs_mdt_clear(nilfs->ns_cpfile);
 786         nilfs_mdt_clear(nilfs->ns_sufile);
 787         nilfs_mdt_clear(nilfs->ns_dat);
 788         return err;
 789 }
 790
 791 /**
 792  * nilfs_search_super_root - search the latest valid super root
 793  * @nilfs: the_nilfs
 794  * @sbi: nilfs_sb_info
 795  * @ri: pointer to a nilfs_recovery_info struct to store search results.
 796  *
 797  * nilfs_search_super_root() looks for the latest super-root from a partial
 798  * segment pointed by the superblock.  It sets up struct the_nilfs through
 799  * this search. It fills nilfs_recovery_info (ri) required for recovery.
 800  *
 801  * Return Value: On success, 0 is returned.  On error, one of the following
 802  * negative error code is returned.
 803  *
 804  * %-EINVAL - No valid segment found
 805  *
 806  * %-EIO - I/O error
 807  */
 808 int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi,
 809                             struct nilfs_recovery_info *ri)
 810 {
 811         struct nilfs_segsum_info ssi;
 812         sector_t pseg_start, pseg_end, sr_pseg_start = 0;
 813         sector_t seg_start, seg_end; /* range of full segment (block number) */
 814         u64 seg_seq;
 815         __u64 segnum, nextnum = 0;
 816         __u64 cno;
 817         struct nilfs_segment_entry *ent;
 818         LIST_HEAD(segments);
 819         int empty_seg = 0, scan_newer = 0;
 820         int ret;
 821
 822         pseg_start = nilfs->ns_last_pseg;
 823         seg_seq = nilfs->ns_last_seq;
 824         cno = nilfs->ns_last_cno;
 825         segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
 826
 827         /* Calculate range of segment */
 828         nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 829
 830         for (;;) {
 831                 /* Load segment summary */
 832                 ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1);
 833                 if (ret) {
 834                         if (ret == NILFS_SEG_FAIL_IO)
 835                                 goto failed;
 836                         goto strayed;
 837                 }
 838                 pseg_end = pseg_start + ssi.nblocks - 1;
 839                 if (unlikely(pseg_end > seg_end)) {
 840                         ret = NILFS_SEG_FAIL_CONSISTENCY;
 841                         goto strayed;
 842                 }
 843
 844                 /* A valid partial segment */
 845                 ri->ri_pseg_start = pseg_start;
 846                 ri->ri_seq = seg_seq;
 847                 ri->ri_segnum = segnum;
 848                 nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next);
 849                 ri->ri_nextnum = nextnum;
 850                 empty_seg = 0;
 851
 852                 if (!NILFS_SEG_HAS_SR(&ssi)) {
 853                         if (!scan_newer) {
 854                                 /* This will never happen because a superblock
 855                                    (last_segment) always points to a pseg
 856                                    having a super root. */
 857                                 ret = NILFS_SEG_FAIL_CONSISTENCY;
 858                                 goto failed;
 859                         }
 860                         if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) {
 861                                 ri->ri_lsegs_start = pseg_start;
 862                                 ri->ri_lsegs_start_seq = seg_seq;
 863                         }
 864                         if (NILFS_SEG_LOGEND(&ssi))
 865                                 ri->ri_lsegs_end = pseg_start;
 866                         goto try_next_pseg;
 867                 }
 868
 869                 /* A valid super root was found. */
 870                 ri->ri_cno = cno++;
 871                 ri->ri_super_root = pseg_end;
 872                 ri->ri_lsegs_start = ri->ri_lsegs_end = 0;
 873
 874                 nilfs_dispose_segment_list(&segments);
 875                 nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start)
 876                         + ssi.nblocks - seg_start;
 877                 nilfs->ns_seg_seq = seg_seq;
 878                 nilfs->ns_segnum = segnum;
 879                 nilfs->ns_cno = cno;  /* nilfs->ns_cno = ri->ri_cno + 1 */
 880                 nilfs->ns_ctime = ssi.ctime;
 881                 nilfs->ns_nextnum = nextnum;
 882
 883                 if (scan_newer)
 884                         ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED;
 885                 else {
 886                         nilfs->ns_prot_seq = ssi.seg_seq;
 887                         if (nilfs->ns_mount_state & NILFS_VALID_FS)
 888                                 goto super_root_found;
 889                         scan_newer = 1;
 890                 }
 891
 892                 /* reset region for roll-forward */
 893                 pseg_start += ssi.nblocks;
 894                 if (pseg_start < seg_end)
 895                         continue;
 896                 goto feed_segment;
 897
 898  try_next_pseg:
 899                 /* Standing on a course, or met an inconsistent state */
 900                 pseg_start += ssi.nblocks;
 901                 if (pseg_start < seg_end)
 902                         continue;
 903                 goto feed_segment;
 904
 905  strayed:
 906                 /* Off the trail */
 907                 if (!scan_newer)
 908                         /*
 909                          * This can happen if a checkpoint was written without
 910                          * barriers, or as a result of an I/O failure.
 911                          */
 912                         goto failed;
 913
 914  feed_segment:
 915                 /* Looking to the next full segment */
 916                 if (empty_seg++)
 917                         goto super_root_found; /* found a valid super root */
 918
 919                 ent = nilfs_alloc_segment_entry(segnum);
 920                 if (unlikely(!ent)) {
 921                         ret = -ENOMEM;
 922                         goto failed;
 923                 }
 924                 list_add_tail(&ent->list, &segments);
 925
 926                 seg_seq++;
 927                 segnum = nextnum;
 928                 nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end);
 929                 pseg_start = seg_start;
 930         }
 931
 932  super_root_found:
 933         /* Updating pointers relating to the latest checkpoint */
 934         list_splice(&segments, ri->ri_used_segments.prev);
 935         nilfs->ns_last_pseg = sr_pseg_start;
 936         nilfs->ns_last_seq = nilfs->ns_seg_seq;
 937         nilfs->ns_last_cno = ri->ri_cno;
 938         return 0;
 939
 940  failed:
 941         nilfs_dispose_segment_list(&segments);
 942         return (ret < 0) ? ret : nilfs_warn_segment_error(ret);
 943 }