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                         goto errout;
 349                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 350                        sizeof(jsuper));
 351                 brelse(bh);
 352 #ifdef EXT2FS_ENABLE_SWAPFS
 353                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 354                         ext2fs_swap_super(&jsuper);
 355 #endif
 356                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 357                     !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 358                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 359                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 360                         goto errout;
 361                 }
 362                 /* Make sure the journal UUID is correct */
 363                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 364                            sizeof(jsuper.s_uuid))) {
 365                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 366                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 367                         goto errout;
 368                 }
 369
 370                 journal->j_maxlen = jsuper.s_blocks_count;
 371                 start++;
 372         }
 373
 374         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
 375                 retval = EXT2_ET_NO_MEMORY;
 376                 goto errout;
 377         }
 378
 379         journal->j_sb_buffer = bh;
 380         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 381
 382 #ifdef USE_INODE_IO
 383         if (j_inode)
 384                 ext2fs_free_mem(&j_inode);
 385 #endif
 386
 387         *ret_journal = journal;
 388         return 0;
 389
 390 errout:
 391         if (dev_fs)
 392                 ext2fs_free_mem(&dev_fs);
 393         if (j_inode)
 394                 ext2fs_free_mem(&j_inode);
 395         if (journal)
 396                 ext2fs_free_mem(&journal);
 397         return retval;
 398
 399 }
 400
 401 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 402                                               struct problem_context *pctx)
 403 {
 404         struct ext2_super_block *sb = ctx->fs->super;
 405         int recover = ctx->fs->super->s_feature_incompat &
 406                 EXT3_FEATURE_INCOMPAT_RECOVER;
 407         int has_journal = ctx->fs->super->s_feature_compat &
 408                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 409
 410         if (has_journal || sb->s_journal_inum) {
 411                 /* The journal inode is bogus, remove and force full fsck */
 412                 pctx->ino = sb->s_journal_inum;
 413                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 414                         if (has_journal && sb->s_journal_inum)
 415                                 printf("*** ext3 journal has been deleted - "
 416                                        "filesystem is now ext2 only ***\n\n");
 417                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 418                         sb->s_journal_inum = 0;
 419                         ctx->flags |= E2F_FLAG_JOURNAL_INODE; /* FIXME: todo */
 420                         e2fsck_clear_recover(ctx, 1);
 421                         return 0;
 422                 }
 423                 return EXT2_ET_BAD_INODE_NUM;
 424         } else if (recover) {
 425                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 426                         e2fsck_clear_recover(ctx, 1);
 427                         return 0;
 428                 }
 429                 return EXT2_ET_UNSUPP_FEATURE;
 430         }
 431         return 0;
 432 }
 433
 434 #define V1_SB_SIZE      0x0024
 435 static void clear_v2_journal_fields(journal_t *journal)
 436 {
 437         e2fsck_t ctx = journal->j_dev->k_ctx;
 438         struct problem_context pctx;
 439
 440         clear_problem_context(&pctx);
 441
 442         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 443                 return;
 444
 445         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 446                ctx->fs->blocksize-V1_SB_SIZE);
 447         mark_buffer_dirty(journal->j_sb_buffer);
 448 }
 449
 450
 451 static errcode_t e2fsck_journal_load(journal_t *journal)
 452 {
 453         e2fsck_t ctx = journal->j_dev->k_ctx;
 454         journal_superblock_t *jsb;
 455         struct buffer_head *jbh = journal->j_sb_buffer;
 456         struct problem_context pctx;
 457
 458         clear_problem_context(&pctx);
 459
 460         ll_rw_block(READ, 1, &jbh);
 461         if (jbh->b_err) {
 462                 com_err(ctx->device_name, jbh->b_err,
 463                         _("reading journal superblock\n"));
 464                 return jbh->b_err;
 465         }
 466
 467         jsb = journal->j_superblock;
 468         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 469         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 470                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 471
 472         switch (ntohl(jsb->s_header.h_blocktype)) {
 473         case JFS_SUPERBLOCK_V1:
 474                 journal->j_format_version = 1;
 475                 if (jsb->s_feature_compat ||
 476                     jsb->s_feature_incompat ||
 477                     jsb->s_feature_ro_compat ||
 478                     jsb->s_nr_users)
 479                         clear_v2_journal_fields(journal);
 480                 break;
 481
 482         case JFS_SUPERBLOCK_V2:
 483                 journal->j_format_version = 2;
 484                 if (ntohl(jsb->s_nr_users) > 1 &&
 485                     uuid_is_null(ctx->fs->super->s_journal_uuid))
 486                         clear_v2_journal_fields(journal);
 487                 if (ntohl(jsb->s_nr_users) > 1) {
 488                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 489                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 490                 }
 491                 break;
 492
 493         /*
 494          * These should never appear in a journal super block, so if
 495          * they do, the journal is badly corrupted.
 496          */
 497         case JFS_DESCRIPTOR_BLOCK:
 498         case JFS_COMMIT_BLOCK:
 499         case JFS_REVOKE_BLOCK:
 500                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 501
 502         /* If we don't understand the superblock major type, but there
 503          * is a magic number, then it is likely to be a new format we
 504          * just don't understand, so leave it alone. */
 505         default:
 506                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 507         }
 508
 509         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 510                 return EXT2_ET_UNSUPP_FEATURE;
 511
 512         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 513                 return EXT2_ET_RO_UNSUPP_FEATURE;
 514
 515         /* We have now checked whether we know enough about the journal
 516          * format to be able to proceed safely, so any other checks that
 517          * fail we should attempt to recover from. */
 518         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 519                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 520                         _("%s: no valid journal superblock found\n"),
 521                         ctx->device_name);
 522                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 523         }
 524
 525         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 526                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 527         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 528                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 529                         _("%s: journal too short\n"),
 530                         ctx->device_name);
 531                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 532         }
 533
 534         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 535         journal->j_transaction_sequence = journal->j_tail_sequence;
 536         journal->j_tail = ntohl(jsb->s_start);
 537         journal->j_first = ntohl(jsb->s_first);
 538         journal->j_last = ntohl(jsb->s_maxlen);
 539
 540         return 0;
 541 }
 542
 543 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 544                                        journal_t *journal)
 545 {
 546         char *p;
 547         union {
 548                 uuid_t uuid;
 549                 __u32 val[4];
 550         } u;
 551         __u32 new_seq = 0;
 552         int i;
 553
 554         /* Leave a valid existing V1 superblock signature alone.
 555          * Anything unrecognisable we overwrite with a new V2
 556          * signature. */
 557
 558         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 559             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 560                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 561                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 562         }
 563
 564         /* Zero out everything else beyond the superblock header */
 565
 566         p = ((char *) jsb) + sizeof(journal_header_t);
 567         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 568
 569         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 570         jsb->s_maxlen = htonl(journal->j_maxlen);
 571         jsb->s_first = htonl(1);
 572
 573         /* Initialize the journal sequence number so that there is "no"
 574          * chance we will find old "valid" transactions in the journal.
 575          * This avoids the need to zero the whole journal (slow to do,
 576          * and risky when we are just recovering the filesystem).
 577          */
 578         uuid_generate(u.uuid);
 579         for (i = 0; i < 4; i ++)
 580                 new_seq ^= u.val[i];
 581         jsb->s_sequence = htonl(new_seq);
 582
 583         mark_buffer_dirty(journal->j_sb_buffer);
 584         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 585 }
 586
 587 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 588                                                   journal_t *journal,
 589                                                   struct problem_context *pctx)
 590 {
 591         struct ext2_super_block *sb = ctx->fs->super;
 592         int recover = ctx->fs->super->s_feature_incompat &
 593                 EXT3_FEATURE_INCOMPAT_RECOVER;
 594
 595         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 596                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 597                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 598                                                    journal);
 599                         journal->j_transaction_sequence = 1;
 600                         e2fsck_clear_recover(ctx, recover);
 601                         return 0;
 602                 }
 603                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 604         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 605                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 606
 607         return 0;
 608 }
 609
 610 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 611                                    int reset, int drop)
 612 {
 613         journal_superblock_t *jsb;
 614
 615         if (drop)
 616                 mark_buffer_clean(journal->j_sb_buffer);
 617         else if (!(ctx->options & E2F_OPT_READONLY)) {
 618                 jsb = journal->j_superblock;
 619                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 620                 if (reset)
 621                         jsb->s_start = 0; /* this marks the journal as empty */
 622                 mark_buffer_dirty(journal->j_sb_buffer);
 623         }
 624         brelse(journal->j_sb_buffer);
 625
 626         if (ctx->journal_io) {
 627                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 628                         io_channel_close(ctx->journal_io);
 629                 ctx->journal_io = 0;
 630         }
 631
 632 #ifndef USE_INODE_IO
 633         if (journal->j_inode)
 634                 ext2fs_free_mem(&journal->j_inode);
 635 #endif
 636         if (journal->j_fs_dev)
 637                 ext2fs_free_mem(&journal->j_fs_dev);
 638         ext2fs_free_mem(&journal);
 639 }
 640
 641 /*
 642  * This function makes sure that the superblock fields regarding the
 643  * journal are consistent.
 644  */
 645 int e2fsck_check_ext3_journal(e2fsck_t ctx)
 646 {
 647         struct ext2_super_block *sb = ctx->fs->super;
 648         journal_t *journal;
 649         int recover = ctx->fs->super->s_feature_incompat &
 650                 EXT3_FEATURE_INCOMPAT_RECOVER;
 651         struct problem_context pctx;
 652         problem_t problem;
 653         int reset = 0, force_fsck = 0;
 654         int retval;
 655
 656         /* If we don't have any journal features, don't do anything more */
 657         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 658             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 659             uuid_is_null(sb->s_journal_uuid))
 660                 return 0;
 661
 662         clear_problem_context(&pctx);
 663         pctx.num = sb->s_journal_inum;
 664
 665         retval = e2fsck_get_journal(ctx, &journal);
 666         if (retval) {
 667                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
 668                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
 669                     (retval == EXT2_ET_NO_JOURNAL))
 670                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 671                 return retval;
 672         }
 673
 674         retval = e2fsck_journal_load(journal);
 675         if (retval) {
 676                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 677                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 678                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 679                                   &pctx))) ||
 680                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 681                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 682                                   &pctx))) ||
 683                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 684                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 685                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 686                                                                   &pctx);
 687                 e2fsck_journal_release(ctx, journal, 0, 1);
 688                 return retval;
 689         }
 690
 691         /*
 692          * We want to make the flags consistent here.  We will not leave with
 693          * needs_recovery set but has_journal clear.  We can't get in a loop
 694          * with -y, -n, or -p, only if a user isn't making up their mind.
 695          */
 696 no_has_journal:
 697         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 698                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 699                 pctx.str = "inode";
 700                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 701                         if (recover &&
 702                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 703                                 goto no_has_journal;
 704                         /*
 705                          * Need a full fsck if we are releasing a
 706                          * journal stored on a reserved inode.
 707                          */
 708                         force_fsck = recover ||
 709                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 710                         /* Clear all of the journal fields */
 711                         sb->s_journal_inum = 0;
 712                         sb->s_journal_dev = 0;
 713                         memset(sb->s_journal_uuid, 0,
 714                                sizeof(sb->s_journal_uuid));
 715                         e2fsck_clear_recover(ctx, force_fsck);
 716                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 717                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 718                         ext2fs_mark_super_dirty(ctx->fs);
 719                 }
 720         }
 721
 722         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 723             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 724             journal->j_superblock->s_start != 0) {
 725                 /* Print status information */
 726                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
 727                 if (ctx->superblock)
 728                         problem = PR_0_JOURNAL_RUN_DEFAULT;
 729                 else
 730                         problem = PR_0_JOURNAL_RUN;
 731                 if (fix_problem(ctx, problem, &pctx)) {
 732                         ctx->options |= E2F_OPT_FORCE;
 733                         sb->s_feature_incompat |=
 734                                 EXT3_FEATURE_INCOMPAT_RECOVER;
 735                         ext2fs_mark_super_dirty(ctx->fs);
 736                 } else if (fix_problem(ctx,
 737                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 738                         reset = 1;
 739                         sb->s_state &= ~EXT2_VALID_FS;
 740                         ext2fs_mark_super_dirty(ctx->fs);
 741                 }
 742                 /*
 743                  * If the user answers no to the above question, we
 744                  * ignore the fact that journal apparently has data;
 745                  * accidentally replaying over valid data would be far
 746                  * worse than skipping a questionable recovery.
 747                  *
 748                  * XXX should we abort with a fatal error here?  What
 749                  * will the ext3 kernel code do if a filesystem with
 750                  * !NEEDS_RECOVERY but with a non-zero
 751                  * journal->j_superblock->s_start is mounted?
 752                  */
 753         }
 754
 755         e2fsck_journal_release(ctx, journal, reset, 0);
 756         return retval;
 757 }
 758
 759 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 760 {
 761         journal_t *journal;
 762         int retval;
 763
 764         journal_init_revoke_caches();
 765         retval = e2fsck_get_journal(ctx, &journal);
 766         if (retval)
 767                 return retval;
 768
 769         retval = e2fsck_journal_load(journal);
 770         if (retval)
 771                 goto errout;
 772
 773         retval = journal_init_revoke(journal, 1024);
 774         if (retval)
 775                 goto errout;
 776
 777         retval = -journal_recover(journal);
 778         if (retval)
 779                 goto errout;
 780
 781         if (journal->j_superblock->s_errno) {
 782                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 783                 ext2fs_mark_super_dirty(ctx->fs);
 784                 journal->j_superblock->s_errno = 0;
 785                 mark_buffer_dirty(journal->j_sb_buffer);
 786         }
 787
 788 errout:
 789         journal_destroy_revoke(journal);
 790         journal_destroy_revoke_caches();
 791         e2fsck_journal_release(ctx, journal, 1, 0);
 792         return retval;
 793 }
 794
 795 int e2fsck_run_ext3_journal(e2fsck_t ctx)
 796 {
 797         io_manager io_ptr = ctx->fs->io->manager;
 798         int blocksize = ctx->fs->blocksize;
 799         errcode_t       retval, recover_retval;
 800
 801         printf(_("%s: recovering journal\n"), ctx->device_name);
 802         if (ctx->options & E2F_OPT_READONLY) {
 803                 printf(_("%s: won't do journal recovery while read-only\n"),
 804                        ctx->device_name);
 805                 return EXT2_ET_FILE_RO;
 806         }
 807
 808         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 809                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 810
 811         recover_retval = recover_ext3_journal(ctx);
 812
 813         /*
 814          * Reload the filesystem context to get up-to-date data from disk
 815          * because journal recovery will change the filesystem under us.
 816          */
 817         ext2fs_close(ctx->fs);
 818         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 819                              ctx->superblock, blocksize, io_ptr,
 820                              &ctx->fs);
 821
 822         if (retval) {
 823                 com_err(ctx->program_name, retval,
 824                         _("while trying to re-open %s"),
 825                         ctx->device_name);
 826                 fatal_error(ctx, 0);
 827         }
 828         ctx->fs->priv_data = ctx;
 829
 830         /* Set the superblock flags */
 831         e2fsck_clear_recover(ctx, recover_retval);
 832         return recover_retval;
 833 }
 834
 835 /*
 836  * This function will move the journal inode from a visible file in
 837  * the filesystem directory hierarchy to the reserved inode if necessary.
 838  */
 839 static const char * const journal_names[] = {
 840         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 841
 842 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 843 {
 844         struct ext2_super_block *sb = ctx->fs->super;
 845         struct problem_context  pctx;
 846         struct ext2_inode       inode;
 847         ext2_filsys             fs = ctx->fs;
 848         ext2_ino_t              ino;
 849         errcode_t               retval;
 850         const char * const *    cpp;
 851         int                     group, mount_flags;
 852
 853         clear_problem_context(&pctx);
 854
 855         /*
 856          * If the filesystem is opened read-only, or there is no
 857          * journal, then do nothing.
 858          */
 859         if ((ctx->options & E2F_OPT_READONLY) ||
 860             (sb->s_journal_inum == 0) ||
 861             !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
 862                 return;
 863
 864         /*
 865          * Read in the journal inode
 866          */
 867         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
 868                 return;
 869
 870         /*
 871          * If it's necessary to backup the journal inode, do so.
 872          */
 873         if ((sb->s_jnl_backup_type == 0) ||
 874             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
 875              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
 876                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
 877                         memcpy(sb->s_jnl_blocks, inode.i_block,
 878                                EXT2_N_BLOCKS*4);
 879                         sb->s_jnl_blocks[16] = inode.i_size;
 880                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 881                         ext2fs_mark_super_dirty(fs);
 882                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 883                 }
 884         }
 885
 886         /*
 887          * If the journal is already the hidden inode, then do nothing
 888          */
 889         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
 890                 return;
 891
 892         /*
 893          * The journal inode had better have only one link and not be readable.
 894          */
 895         if (inode.i_links_count != 1)
 896                 return;
 897
 898         /*
 899          * If the filesystem is mounted, or we can't tell whether
 900          * or not it's mounted, do nothing.
 901          */
 902         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
 903         if (retval || (mount_flags & EXT2_MF_MOUNTED))
 904                 return;
 905
 906         /*
 907          * If we can't find the name of the journal inode, then do
 908          * nothing.
 909          */
 910         for (cpp = journal_names; *cpp; cpp++) {
 911                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
 912                                        strlen(*cpp), 0, &ino);
 913                 if ((retval == 0) && (ino == sb->s_journal_inum))
 914                         break;
 915         }
 916         if (*cpp == 0)
 917                 return;
 918
 919         /* We need the inode bitmap to be loaded */
 920         retval = ext2fs_read_bitmaps(fs);
 921         if (retval)
 922                 return;
 923
 924         pctx.str = *cpp;
 925         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
 926                 return;
 927
 928         /*
 929          * OK, we've done all the checks, let's actually move the
 930          * journal inode.  Errors at this point mean we need to force
 931          * an ext2 filesystem check.
 932          */
 933         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
 934                 goto err_out;
 935         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
 936                 goto err_out;
 937         sb->s_journal_inum = EXT2_JOURNAL_INO;
 938         ext2fs_mark_super_dirty(fs);
 939         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 940         inode.i_links_count = 0;
 941         inode.i_dtime = time(0);
 942         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
 943                 goto err_out;
 944
 945         group = ext2fs_group_of_ino(fs, ino);
 946         ext2fs_unmark_inode_bitmap(fs->inode_map, ino);
 947         ext2fs_mark_ib_dirty(fs);
 948         fs->group_desc[group].bg_free_inodes_count++;
 949         fs->super->s_free_inodes_count++;
 950         return;
 951
 952 err_out:
 953         pctx.errcode = retval;
 954         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
 955         fs->super->s_state &= ~EXT2_VALID_FS;
 956         ext2fs_mark_super_dirty(fs);
 957         return;
 958 }
 959