e2fsck/journal.c

   1 /*
   2  * journal.c --- code for handling the "ext3" journal
   3  *
   4  * Copyright (C) 2000 Andreas Dilger
   5  * Copyright (C) 2000 Theodore Ts'o
   6  *
   7  * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
   8  * Copyright (C) 1999 Red Hat Software
   9  *
  10  * This file may be redistributed under the terms of the
  11  * GNU General Public License version 2 or at your discretion
  12  * any later version.
  13  */
  14
  15 #ifdef HAVE_SYS_MOUNT_H
  16 #include <sys/param.h>
  17 #include <sys/mount.h>
  18 #define MNT_FL (MS_MGC_VAL | MS_RDONLY)
  19 #endif
  20 #ifdef HAVE_SYS_STAT_H
  21 #include <sys/stat.h>
  22 #endif
  23
  24 #define E2FSCK_INCLUDE_INLINE_FUNCS
  25 #include "jfs_user.h"
  26 #include "problem.h"
  27 #include "uuid/uuid.h"
  28
  29 #ifdef CONFIG_JBD_DEBUG         /* Enabled by configure --enable-jfs-debug */
  30 static int bh_count = 0;
  31 #endif
  32
  33 /*
  34  * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
  35  * This creates a larger static binary, and a smaller binary using
  36  * shared libraries.  It's also probably slightly less CPU-efficient,
  37  * which is why it's not on by default.  But, it's a good way of
  38  * testing the functions in inode_io.c and fileio.c.
  39  */
  40 #undef USE_INODE_IO
  41
  42 /* Kernel compatibility functions for handling the journal.  These allow us
  43  * to use the recovery.c file virtually unchanged from the kernel, so we
  44  * don't have to do much to keep kernel and user recovery in sync.
  45  */
  46 int journal_bmap(journal_t *journal, blk_t block, unsigned long *phys)
  47 {
  48 #ifdef USE_INODE_IO
  49         *phys = block;
  50         return 0;
  51 #else
  52         struct inode    *inode = journal->j_inode;
  53         errcode_t       retval;
  54         blk_t           pblk;
  55
  56         if (!inode) {
  57                 *phys = block;
  58                 return 0;
  59         }
  60
  61         retval= ext2fs_bmap(inode->i_ctx->fs, inode->i_ino,
  62                             &inode->i_ext2, NULL, 0, block, &pblk);
  63         *phys = pblk;
  64         return (retval);
  65 #endif
  66 }
  67
  68 struct buffer_head *getblk(kdev_t kdev, blk_t blocknr, int blocksize)
  69 {
  70         struct buffer_head *bh;
  71
  72         bh = e2fsck_allocate_memory(kdev->k_ctx, sizeof(*bh), "block buffer");
  73         if (!bh)
  74                 return NULL;
  75
  76         jfs_debug(4, "getblk for block %lu (%d bytes)(total %d)\n",
  77                   (unsigned long) blocknr, blocksize, ++bh_count);
  78
  79         bh->b_ctx = kdev->k_ctx;
  80         if (kdev->k_dev == K_DEV_FS)
  81                 bh->b_io = kdev->k_ctx->fs->io;
  82         else
  83                 bh->b_io = kdev->k_ctx->journal_io;
  84         bh->b_size = blocksize;
  85         bh->b_blocknr = blocknr;
  86
  87         return bh;
  88 }
  89
  90 void sync_blockdev(kdev_t kdev)
  91 {
  92         io_channel      io;
  93
  94         if (kdev->k_dev == K_DEV_FS)
  95                 io = kdev->k_ctx->fs->io;
  96         else
  97                 io = kdev->k_ctx->journal_io;
  98
  99         io_channel_flush(io);
 100 }
 101
 102 void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
 103 {
 104         int retval;
 105         struct buffer_head *bh;
 106
 107         for (; nr > 0; --nr) {
 108                 bh = *bhp++;
 109                 if (rw == READ && !bh->b_uptodate) {
 110                         jfs_debug(3, "reading block %lu/%p\n",
 111                                   (unsigned long) bh->b_blocknr, (void *) bh);
 112                         retval = io_channel_read_blk(bh->b_io,
 113                                                      bh->b_blocknr,
 114                                                      1, bh->b_data);
 115                         if (retval) {
 116                                 com_err(bh->b_ctx->device_name, retval,
 117                                         "while reading block %lu\n",
 118                                         (unsigned long) bh->b_blocknr);
 119                                 bh->b_err = retval;
 120                                 continue;
 121                         }
 122                         bh->b_uptodate = 1;
 123                 } else if (rw == WRITE && bh->b_dirty) {
 124                         jfs_debug(3, "writing block %lu/%p\n",
 125                                   (unsigned long) bh->b_blocknr, (void *) bh);
 126                         retval = io_channel_write_blk(bh->b_io,
 127                                                       bh->b_blocknr,
 128                                                       1, bh->b_data);
 129                         if (retval) {
 130                                 com_err(bh->b_ctx->device_name, retval,
 131                                         "while writing block %lu\n",
 132                                         (unsigned long) bh->b_blocknr);
 133                                 bh->b_err = retval;
 134                                 continue;
 135                         }
 136                         bh->b_dirty = 0;
 137                         bh->b_uptodate = 1;
 138                 } else {
 139                         jfs_debug(3, "no-op %s for block %lu\n",
 140                                   rw == READ ? "read" : "write",
 141                                   (unsigned long) bh->b_blocknr);
 142                 }
 143         }
 144 }
 145
 146 void mark_buffer_dirty(struct buffer_head *bh)
 147 {
 148         bh->b_dirty = 1;
 149 }
 150
 151 static void mark_buffer_clean(struct buffer_head * bh)
 152 {
 153         bh->b_dirty = 0;
 154 }
 155
 156 void brelse(struct buffer_head *bh)
 157 {
 158         if (bh->b_dirty)
 159                 ll_rw_block(WRITE, 1, &bh);
 160         jfs_debug(3, "freeing block %lu/%p (total %d)\n",
 161                   (unsigned long) bh->b_blocknr, (void *) bh, --bh_count);
 162         ext2fs_free_mem(&bh);
 163 }
 164
 165 int buffer_uptodate(struct buffer_head *bh)
 166 {
 167         return bh->b_uptodate;
 168 }
 169
 170 void mark_buffer_uptodate(struct buffer_head *bh, int val)
 171 {
 172         bh->b_uptodate = val;
 173 }
 174
 175 void wait_on_buffer(struct buffer_head *bh)
 176 {
 177         if (!bh->b_uptodate)
 178                 ll_rw_block(READ, 1, &bh);
 179 }
 180
 181
 182 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
 183 {
 184         ctx->fs->super->s_feature_incompat &= ~EXT3_FEATURE_INCOMPAT_RECOVER;
 185
 186         /* if we had an error doing journal recovery, we need a full fsck */
 187         if (error)
 188                 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
 189         ext2fs_mark_super_dirty(ctx->fs);
 190 }
 191
 192 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
 193 {
 194         struct ext2_super_block *sb = ctx->fs->super;
 195         struct ext2_super_block jsuper;
 196         struct problem_context  pctx;
 197         struct buffer_head      *bh;
 198         struct inode            *j_inode = NULL;
 199         struct kdev_s           *dev_fs = NULL, *dev_journal;
 200         const char              *journal_name = 0;
 201         journal_t               *journal = NULL;
 202         errcode_t               retval = 0;
 203         io_manager              io_ptr = 0;
 204         unsigned long           start = 0;
 205         blk_t                   blk;
 206         int                     ext_journal = 0;
 207         int                     tried_backup_jnl = 0;
 208         int                     i;
 209
 210         clear_problem_context(&pctx);
 211
 212         journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
 213         if (!journal) {
 214                 return EXT2_ET_NO_MEMORY;
 215         }
 216
 217         dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
 218         if (!dev_fs) {
 219                 retval = EXT2_ET_NO_MEMORY;
 220                 goto errout;
 221         }
 222         dev_journal = dev_fs+1;
 223
 224         dev_fs->k_ctx = dev_journal->k_ctx = ctx;
 225         dev_fs->k_dev = K_DEV_FS;
 226         dev_journal->k_dev = K_DEV_JOURNAL;
 227
 228         journal->j_dev = dev_journal;
 229         journal->j_fs_dev = dev_fs;
 230         journal->j_inode = NULL;
 231         journal->j_blocksize = ctx->fs->blocksize;
 232
 233         if (uuid_is_null(sb->s_journal_uuid)) {
 234                 if (!sb->s_journal_inum)
 235                         return EXT2_ET_BAD_INODE_NUM;
 236                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
 237                                                  "journal inode");
 238                 if (!j_inode) {
 239                         retval = EXT2_ET_NO_MEMORY;
 240                         goto errout;
 241                 }
 242
 243                 j_inode->i_ctx = ctx;
 244                 j_inode->i_ino = sb->s_journal_inum;
 245
 246                 if ((retval = ext2fs_read_inode(ctx->fs,
 247                                                 sb->s_journal_inum,
 248                                                 &j_inode->i_ext2))) {
 249                 try_backup_journal:
 250                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
 251                             tried_backup_jnl)
 252                                 goto errout;
 253                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
 254                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
 255                                EXT2_N_BLOCKS*4);
 256                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
 257                         j_inode->i_ext2.i_links_count = 1;
 258                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
 259                         tried_backup_jnl++;
 260                 }
 261                 if (!j_inode->i_ext2.i_links_count ||
 262                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
 263                         retval = EXT2_ET_NO_JOURNAL;
 264                         goto try_backup_journal;
 265                 }
 266                 if (j_inode->i_ext2.i_size / journal->j_blocksize <
 267                     JFS_MIN_JOURNAL_BLOCKS) {
 268                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 269                         goto try_backup_journal;
 270                 }
 271                 for (i=0; i < EXT2_N_BLOCKS; i++) {
 272                         blk = j_inode->i_ext2.i_block[i];
 273                         if (!blk) {
 274                                 if (i < EXT2_NDIR_BLOCKS) {
 275                                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 276                                         goto try_backup_journal;
 277                                 }
 278                                 continue;
 279                         }
 280                         if (blk < sb->s_first_data_block ||
 281                             blk >= sb->s_blocks_count) {
 282                                 retval = EXT2_ET_BAD_BLOCK_NUM;
 283                                 goto try_backup_journal;
 284                         }
 285                 }
 286                 journal->j_maxlen = j_inode->i_ext2.i_size / journal->j_blocksize;
 287
 288 #ifdef USE_INODE_IO
 289                 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
 290                                                  &j_inode->i_ext2,
 291                                                  &journal_name);
 292                 if (retval)
 293                         goto errout;
 294
 295                 io_ptr = inode_io_manager;
 296 #else
 297                 journal->j_inode = j_inode;
 298                 ctx->journal_io = ctx->fs->io;
 299                 if ((retval = journal_bmap(journal, 0, &start)) != 0)
 300                         goto errout;
 301 #endif
 302         } else {
 303                 ext_journal = 1;
 304                 if (!ctx->journal_name) {
 305                         char uuid[37];
 306
 307                         uuid_unparse(sb->s_journal_uuid, uuid);
 308                         ctx->journal_name = blkid_get_devname(ctx->blkid,
 309                                                               "UUID", uuid);
 310                         if (!ctx->journal_name)
 311                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
 312                 }
 313                 journal_name = ctx->journal_name;
 314
 315                 if (!journal_name) {
 316                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
 317                         return EXT2_ET_LOAD_EXT_JOURNAL;
 318                 }
 319
 320                 jfs_debug(1, "Using journal file %s\n", journal_name);
 321                 io_ptr = unix_io_manager;
 322         }
 323
 324 #if 0
 325         test_io_backing_manager = io_ptr;
 326         io_ptr = test_io_manager;
 327 #endif
 328 #ifndef USE_INODE_IO
 329         if (ext_journal)
 330 #endif
 331                 retval = io_ptr->open(journal_name, IO_FLAG_RW,
 332                                       &ctx->journal_io);
 333         if (retval)
 334                 goto errout;
 335
 336         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
 337
 338         if (ext_journal) {
 339                 if (ctx->fs->blocksize == 1024)
 340                         start = 1;
 341                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
 342                 if (!bh) {
 343                         retval = EXT2_ET_NO_MEMORY;
 344                         goto errout;
 345                 }
 346                 ll_rw_block(READ, 1, &bh);
 347                 if ((retval = bh->b_err) != 0) {
 348                         brelse(bh);
 349                         goto errout;
 350                 }
 351                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 352                        sizeof(jsuper));
 353                 brelse(bh);
 354 #ifdef EXT2FS_ENABLE_SWAPFS
 355                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 356                         ext2fs_swap_super(&jsuper);
 357 #endif
 358                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 359                     !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 360                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 361                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 362                         goto errout;
 363                 }
 364                 /* Make sure the journal UUID is correct */
 365                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 366                            sizeof(jsuper.s_uuid))) {
 367                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 368                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 369                         goto errout;
 370                 }
 371
 372                 journal->j_maxlen = jsuper.s_blocks_count;
 373                 start++;
 374         }
 375
 376         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
 377                 retval = EXT2_ET_NO_MEMORY;
 378                 goto errout;
 379         }
 380
 381         journal->j_sb_buffer = bh;
 382         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 383
 384 #ifdef USE_INODE_IO
 385         if (j_inode)
 386                 ext2fs_free_mem(&j_inode);
 387 #endif
 388
 389         *ret_journal = journal;
 390         return 0;
 391
 392 errout:
 393         if (dev_fs)
 394                 ext2fs_free_mem(&dev_fs);
 395         if (j_inode)
 396                 ext2fs_free_mem(&j_inode);
 397         if (journal)
 398                 ext2fs_free_mem(&journal);
 399         return retval;
 400
 401 }
 402
 403 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 404                                               struct problem_context *pctx)
 405 {
 406         struct ext2_super_block *sb = ctx->fs->super;
 407         int recover = ctx->fs->super->s_feature_incompat &
 408                 EXT3_FEATURE_INCOMPAT_RECOVER;
 409         int has_journal = ctx->fs->super->s_feature_compat &
 410                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 411
 412         if (has_journal || sb->s_journal_inum) {
 413                 /* The journal inode is bogus, remove and force full fsck */
 414                 pctx->ino = sb->s_journal_inum;
 415                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 416                         if (has_journal && sb->s_journal_inum)
 417                                 printf("*** ext3 journal has been deleted - "
 418                                        "filesystem is now ext2 only ***\n\n");
 419                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 420                         sb->s_journal_inum = 0;
 421                         ctx->flags |= E2F_FLAG_JOURNAL_INODE; /* FIXME: todo */
 422                         e2fsck_clear_recover(ctx, 1);
 423                         return 0;
 424                 }
 425                 return EXT2_ET_BAD_INODE_NUM;
 426         } else if (recover) {
 427                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 428                         e2fsck_clear_recover(ctx, 1);
 429                         return 0;
 430                 }
 431                 return EXT2_ET_UNSUPP_FEATURE;
 432         }
 433         return 0;
 434 }
 435
 436 #define V1_SB_SIZE      0x0024
 437 static void clear_v2_journal_fields(journal_t *journal)
 438 {
 439         e2fsck_t ctx = journal->j_dev->k_ctx;
 440         struct problem_context pctx;
 441
 442         clear_problem_context(&pctx);
 443
 444         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 445                 return;
 446
 447         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 448                ctx->fs->blocksize-V1_SB_SIZE);
 449         mark_buffer_dirty(journal->j_sb_buffer);
 450 }
 451
 452
 453 static errcode_t e2fsck_journal_load(journal_t *journal)
 454 {
 455         e2fsck_t ctx = journal->j_dev->k_ctx;
 456         journal_superblock_t *jsb;
 457         struct buffer_head *jbh = journal->j_sb_buffer;
 458         struct problem_context pctx;
 459
 460         clear_problem_context(&pctx);
 461
 462         ll_rw_block(READ, 1, &jbh);
 463         if (jbh->b_err) {
 464                 com_err(ctx->device_name, jbh->b_err,
 465                         _("reading journal superblock\n"));
 466                 return jbh->b_err;
 467         }
 468
 469         jsb = journal->j_superblock;
 470         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 471         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 472                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 473
 474         switch (ntohl(jsb->s_header.h_blocktype)) {
 475         case JFS_SUPERBLOCK_V1:
 476                 journal->j_format_version = 1;
 477                 if (jsb->s_feature_compat ||
 478                     jsb->s_feature_incompat ||
 479                     jsb->s_feature_ro_compat ||
 480                     jsb->s_nr_users)
 481                         clear_v2_journal_fields(journal);
 482                 break;
 483
 484         case JFS_SUPERBLOCK_V2:
 485                 journal->j_format_version = 2;
 486                 if (ntohl(jsb->s_nr_users) > 1 &&
 487                     uuid_is_null(ctx->fs->super->s_journal_uuid))
 488                         clear_v2_journal_fields(journal);
 489                 if (ntohl(jsb->s_nr_users) > 1) {
 490                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 491                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 492                 }
 493                 break;
 494
 495         /*
 496          * These should never appear in a journal super block, so if
 497          * they do, the journal is badly corrupted.
 498          */
 499         case JFS_DESCRIPTOR_BLOCK:
 500         case JFS_COMMIT_BLOCK:
 501         case JFS_REVOKE_BLOCK:
 502                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 503
 504         /* If we don't understand the superblock major type, but there
 505          * is a magic number, then it is likely to be a new format we
 506          * just don't understand, so leave it alone. */
 507         default:
 508                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 509         }
 510
 511         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 512                 return EXT2_ET_UNSUPP_FEATURE;
 513
 514         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 515                 return EXT2_ET_RO_UNSUPP_FEATURE;
 516
 517         /* We have now checked whether we know enough about the journal
 518          * format to be able to proceed safely, so any other checks that
 519          * fail we should attempt to recover from. */
 520         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 521                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 522                         _("%s: no valid journal superblock found\n"),
 523                         ctx->device_name);
 524                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 525         }
 526
 527         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 528                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 529         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 530                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 531                         _("%s: journal too short\n"),
 532                         ctx->device_name);
 533                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 534         }
 535
 536         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 537         journal->j_transaction_sequence = journal->j_tail_sequence;
 538         journal->j_tail = ntohl(jsb->s_start);
 539         journal->j_first = ntohl(jsb->s_first);
 540         journal->j_last = ntohl(jsb->s_maxlen);
 541
 542         return 0;
 543 }
 544
 545 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 546                                        journal_t *journal)
 547 {
 548         char *p;
 549         union {
 550                 uuid_t uuid;
 551                 __u32 val[4];
 552         } u;
 553         __u32 new_seq = 0;
 554         int i;
 555
 556         /* Leave a valid existing V1 superblock signature alone.
 557          * Anything unrecognisable we overwrite with a new V2
 558          * signature. */
 559
 560         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 561             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 562                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 563                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 564         }
 565
 566         /* Zero out everything else beyond the superblock header */
 567
 568         p = ((char *) jsb) + sizeof(journal_header_t);
 569         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 570
 571         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 572         jsb->s_maxlen = htonl(journal->j_maxlen);
 573         jsb->s_first = htonl(1);
 574
 575         /* Initialize the journal sequence number so that there is "no"
 576          * chance we will find old "valid" transactions in the journal.
 577          * This avoids the need to zero the whole journal (slow to do,
 578          * and risky when we are just recovering the filesystem).
 579          */
 580         uuid_generate(u.uuid);
 581         for (i = 0; i < 4; i ++)
 582                 new_seq ^= u.val[i];
 583         jsb->s_sequence = htonl(new_seq);
 584
 585         mark_buffer_dirty(journal->j_sb_buffer);
 586         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 587 }
 588
 589 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 590                                                   journal_t *journal,
 591                                                   struct problem_context *pctx)
 592 {
 593         struct ext2_super_block *sb = ctx->fs->super;
 594         int recover = ctx->fs->super->s_feature_incompat &
 595                 EXT3_FEATURE_INCOMPAT_RECOVER;
 596
 597         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 598                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 599                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 600                                                    journal);
 601                         journal->j_transaction_sequence = 1;
 602                         e2fsck_clear_recover(ctx, recover);
 603                         return 0;
 604                 }
 605                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 606         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 607                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 608
 609         return 0;
 610 }
 611
 612 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 613                                    int reset, int drop)
 614 {
 615         journal_superblock_t *jsb;
 616
 617         if (drop)
 618                 mark_buffer_clean(journal->j_sb_buffer);
 619         else if (!(ctx->options & E2F_OPT_READONLY)) {
 620                 jsb = journal->j_superblock;
 621                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 622                 if (reset)
 623                         jsb->s_start = 0; /* this marks the journal as empty */
 624                 mark_buffer_dirty(journal->j_sb_buffer);
 625         }
 626         brelse(journal->j_sb_buffer);
 627
 628         if (ctx->journal_io) {
 629                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 630                         io_channel_close(ctx->journal_io);
 631                 ctx->journal_io = 0;
 632         }
 633
 634 #ifndef USE_INODE_IO
 635         if (journal->j_inode)
 636                 ext2fs_free_mem(&journal->j_inode);
 637 #endif
 638         if (journal->j_fs_dev)
 639                 ext2fs_free_mem(&journal->j_fs_dev);
 640         ext2fs_free_mem(&journal);
 641 }
 642
 643 /*
 644  * This function makes sure that the superblock fields regarding the
 645  * journal are consistent.
 646  */
 647 int e2fsck_check_ext3_journal(e2fsck_t ctx)
 648 {
 649         struct ext2_super_block *sb = ctx->fs->super;
 650         journal_t *journal;
 651         int recover = ctx->fs->super->s_feature_incompat &
 652                 EXT3_FEATURE_INCOMPAT_RECOVER;
 653         struct problem_context pctx;
 654         problem_t problem;
 655         int reset = 0, force_fsck = 0;
 656         int retval;
 657
 658         /* If we don't have any journal features, don't do anything more */
 659         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 660             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 661             uuid_is_null(sb->s_journal_uuid))
 662                 return 0;
 663
 664         clear_problem_context(&pctx);
 665         pctx.num = sb->s_journal_inum;
 666
 667         retval = e2fsck_get_journal(ctx, &journal);
 668         if (retval) {
 669                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
 670                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
 671                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
 672                     (retval == EXT2_ET_NO_JOURNAL))
 673                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 674                 return retval;
 675         }
 676
 677         retval = e2fsck_journal_load(journal);
 678         if (retval) {
 679                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 680                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 681                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 682                                   &pctx))) ||
 683                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 684                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 685                                   &pctx))) ||
 686                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 687                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 688                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 689                                                                   &pctx);
 690                 e2fsck_journal_release(ctx, journal, 0, 1);
 691                 return retval;
 692         }
 693
 694         /*
 695          * We want to make the flags consistent here.  We will not leave with
 696          * needs_recovery set but has_journal clear.  We can't get in a loop
 697          * with -y, -n, or -p, only if a user isn't making up their mind.
 698          */
 699 no_has_journal:
 700         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 701                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 702                 pctx.str = "inode";
 703                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 704                         if (recover &&
 705                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 706                                 goto no_has_journal;
 707                         /*
 708                          * Need a full fsck if we are releasing a
 709                          * journal stored on a reserved inode.
 710                          */
 711                         force_fsck = recover ||
 712                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 713                         /* Clear all of the journal fields */
 714                         sb->s_journal_inum = 0;
 715                         sb->s_journal_dev = 0;
 716                         memset(sb->s_journal_uuid, 0,
 717                                sizeof(sb->s_journal_uuid));
 718                         e2fsck_clear_recover(ctx, force_fsck);
 719                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 720                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 721                         ext2fs_mark_super_dirty(ctx->fs);
 722                 }
 723         }
 724
 725         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 726             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 727             journal->j_superblock->s_start != 0) {
 728                 /* Print status information */
 729                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
 730                 if (ctx->superblock)
 731                         problem = PR_0_JOURNAL_RUN_DEFAULT;
 732                 else
 733                         problem = PR_0_JOURNAL_RUN;
 734                 if (fix_problem(ctx, problem, &pctx)) {
 735                         ctx->options |= E2F_OPT_FORCE;
 736                         sb->s_feature_incompat |=
 737                                 EXT3_FEATURE_INCOMPAT_RECOVER;
 738                         ext2fs_mark_super_dirty(ctx->fs);
 739                 } else if (fix_problem(ctx,
 740                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 741                         reset = 1;
 742                         sb->s_state &= ~EXT2_VALID_FS;
 743                         ext2fs_mark_super_dirty(ctx->fs);
 744                 }
 745                 /*
 746                  * If the user answers no to the above question, we
 747                  * ignore the fact that journal apparently has data;
 748                  * accidentally replaying over valid data would be far
 749                  * worse than skipping a questionable recovery.
 750                  *
 751                  * XXX should we abort with a fatal error here?  What
 752                  * will the ext3 kernel code do if a filesystem with
 753                  * !NEEDS_RECOVERY but with a non-zero
 754                  * journal->j_superblock->s_start is mounted?
 755                  */
 756         }
 757
 758         e2fsck_journal_release(ctx, journal, reset, 0);
 759         return retval;
 760 }
 761
 762 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 763 {
 764         journal_t *journal;
 765         int retval;
 766
 767         journal_init_revoke_caches();
 768         retval = e2fsck_get_journal(ctx, &journal);
 769         if (retval)
 770                 return retval;
 771
 772         retval = e2fsck_journal_load(journal);
 773         if (retval)
 774                 goto errout;
 775
 776         retval = journal_init_revoke(journal, 1024);
 777         if (retval)
 778                 goto errout;
 779
 780         retval = -journal_recover(journal);
 781         if (retval)
 782                 goto errout;
 783
 784         if (journal->j_superblock->s_errno) {
 785                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 786                 ext2fs_mark_super_dirty(ctx->fs);
 787                 journal->j_superblock->s_errno = 0;
 788                 mark_buffer_dirty(journal->j_sb_buffer);
 789         }
 790
 791 errout:
 792         journal_destroy_revoke(journal);
 793         journal_destroy_revoke_caches();
 794         e2fsck_journal_release(ctx, journal, 1, 0);
 795         return retval;
 796 }
 797
 798 int e2fsck_run_ext3_journal(e2fsck_t ctx)
 799 {
 800         io_manager io_ptr = ctx->fs->io->manager;
 801         int blocksize = ctx->fs->blocksize;
 802         errcode_t       retval, recover_retval;
 803
 804         printf(_("%s: recovering journal\n"), ctx->device_name);
 805         if (ctx->options & E2F_OPT_READONLY) {
 806                 printf(_("%s: won't do journal recovery while read-only\n"),
 807                        ctx->device_name);
 808                 return EXT2_ET_FILE_RO;
 809         }
 810
 811         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 812                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 813
 814         recover_retval = recover_ext3_journal(ctx);
 815
 816         /*
 817          * Reload the filesystem context to get up-to-date data from disk
 818          * because journal recovery will change the filesystem under us.
 819          */
 820         ext2fs_close(ctx->fs);
 821         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 822                              ctx->superblock, blocksize, io_ptr,
 823                              &ctx->fs);
 824
 825         if (retval) {
 826                 com_err(ctx->program_name, retval,
 827                         _("while trying to re-open %s"),
 828                         ctx->device_name);
 829                 fatal_error(ctx, 0);
 830         }
 831         ctx->fs->priv_data = ctx;
 832         ctx->fs->now = ctx->now;
 833
 834         /* Set the superblock flags */
 835         e2fsck_clear_recover(ctx, recover_retval);
 836         return recover_retval;
 837 }
 838
 839 /*
 840  * This function will move the journal inode from a visible file in
 841  * the filesystem directory hierarchy to the reserved inode if necessary.
 842  */
 843 static const char * const journal_names[] = {
 844         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 845
 846 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 847 {
 848         struct ext2_super_block *sb = ctx->fs->super;
 849         struct problem_context  pctx;
 850         struct ext2_inode       inode;
 851         ext2_filsys             fs = ctx->fs;
 852         ext2_ino_t              ino;
 853         errcode_t               retval;
 854         const char * const *    cpp;
 855         int                     group, mount_flags;
 856
 857         clear_problem_context(&pctx);
 858
 859         /*
 860          * If the filesystem is opened read-only, or there is no
 861          * journal, then do nothing.
 862          */
 863         if ((ctx->options & E2F_OPT_READONLY) ||
 864             (sb->s_journal_inum == 0) ||
 865             !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
 866                 return;
 867
 868         /*
 869          * Read in the journal inode
 870          */
 871         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
 872                 return;
 873
 874         /*
 875          * If it's necessary to backup the journal inode, do so.
 876          */
 877         if ((sb->s_jnl_backup_type == 0) ||
 878             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
 879              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
 880                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
 881                         memcpy(sb->s_jnl_blocks, inode.i_block,
 882                                EXT2_N_BLOCKS*4);
 883                         sb->s_jnl_blocks[16] = inode.i_size;
 884                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 885                         ext2fs_mark_super_dirty(fs);
 886                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 887                 }
 888         }
 889
 890         /*
 891          * If the journal is already the hidden inode, then do nothing
 892          */
 893         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
 894                 return;
 895
 896         /*
 897          * The journal inode had better have only one link and not be readable.
 898          */
 899         if (inode.i_links_count != 1)
 900                 return;
 901
 902         /*
 903          * If the filesystem is mounted, or we can't tell whether
 904          * or not it's mounted, do nothing.
 905          */
 906         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
 907         if (retval || (mount_flags & EXT2_MF_MOUNTED))
 908                 return;
 909
 910         /*
 911          * If we can't find the name of the journal inode, then do
 912          * nothing.
 913          */
 914         for (cpp = journal_names; *cpp; cpp++) {
 915                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
 916                                        strlen(*cpp), 0, &ino);
 917                 if ((retval == 0) && (ino == sb->s_journal_inum))
 918                         break;
 919         }
 920         if (*cpp == 0)
 921                 return;
 922
 923         /* We need the inode bitmap to be loaded */
 924         retval = ext2fs_read_bitmaps(fs);
 925         if (retval)
 926                 return;
 927
 928         pctx.str = *cpp;
 929         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
 930                 return;
 931
 932         /*
 933          * OK, we've done all the checks, let's actually move the
 934          * journal inode.  Errors at this point mean we need to force
 935          * an ext2 filesystem check.
 936          */
 937         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
 938                 goto err_out;
 939         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
 940                 goto err_out;
 941         sb->s_journal_inum = EXT2_JOURNAL_INO;
 942         ext2fs_mark_super_dirty(fs);
 943         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 944         inode.i_links_count = 0;
 945         inode.i_dtime = ctx->now;
 946         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
 947                 goto err_out;
 948
 949         group = ext2fs_group_of_ino(fs, ino);
 950         ext2fs_unmark_inode_bitmap(fs->inode_map, ino);
 951         ext2fs_mark_ib_dirty(fs);
 952         fs->group_desc[group].bg_free_inodes_count++;
 953         fs->super->s_free_inodes_count++;
 954         return;
 955
 956 err_out:
 957         pctx.errcode = retval;
 958         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
 959         fs->super->s_state &= ~EXT2_VALID_FS;
 960         ext2fs_mark_super_dirty(fs);
 961         return;
 962 }
 963
 964 /*
 965  * This function makes sure the superblock hint for the external
 966  * journal is correct.
 967  */
 968 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
 969 {
 970         struct ext2_super_block *sb = ctx->fs->super;
 971         struct problem_context pctx;
 972         char uuid[37], *journal_name;
 973         struct stat st;
 974         problem_t problem;
 975         int retval;
 976
 977         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) ||
 978             uuid_is_null(sb->s_journal_uuid))
 979                 return 0;
 980
 981         uuid_unparse(sb->s_journal_uuid, uuid);
 982         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
 983         if (!journal_name)
 984                 return 0;
 985
 986         if (stat(journal_name, &st) < 0)
 987                 return 0;
 988
 989         if (st.st_rdev != sb->s_journal_dev) {
 990                 clear_problem_context(&pctx);
 991                 pctx.num = st.st_rdev;
 992                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
 993                         sb->s_journal_dev = st.st_rdev;
 994                         ext2fs_mark_super_dirty(ctx->fs);
 995                 }
 996         }
 997
 998         free(journal_name);
 999         return 0;
1000 }