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                         retval = EXT2_ET_BAD_INODE_NUM;
 236                         goto errout;
 237                 }
 238                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
 239                                                  "journal inode");
 240                 if (!j_inode) {
 241                         retval = EXT2_ET_NO_MEMORY;
 242                         goto errout;
 243                 }
 244
 245                 j_inode->i_ctx = ctx;
 246                 j_inode->i_ino = sb->s_journal_inum;
 247
 248                 if ((retval = ext2fs_read_inode(ctx->fs,
 249                                                 sb->s_journal_inum,
 250                                                 &j_inode->i_ext2))) {
 251                 try_backup_journal:
 252                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
 253                             tried_backup_jnl)
 254                                 goto errout;
 255                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
 256                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
 257                                EXT2_N_BLOCKS*4);
 258                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
 259                         j_inode->i_ext2.i_links_count = 1;
 260                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
 261                         tried_backup_jnl++;
 262                 }
 263                 if (!j_inode->i_ext2.i_links_count ||
 264                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
 265                         retval = EXT2_ET_NO_JOURNAL;
 266                         goto try_backup_journal;
 267                 }
 268                 if (j_inode->i_ext2.i_size / journal->j_blocksize <
 269                     JFS_MIN_JOURNAL_BLOCKS) {
 270                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 271                         goto try_backup_journal;
 272                 }
 273                 for (i=0; i < EXT2_N_BLOCKS; i++) {
 274                         blk = j_inode->i_ext2.i_block[i];
 275                         if (!blk) {
 276                                 if (i < EXT2_NDIR_BLOCKS) {
 277                                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 278                                         goto try_backup_journal;
 279                                 }
 280                                 continue;
 281                         }
 282                         if (blk < sb->s_first_data_block ||
 283                             blk >= sb->s_blocks_count) {
 284                                 retval = EXT2_ET_BAD_BLOCK_NUM;
 285                                 goto try_backup_journal;
 286                         }
 287                 }
 288                 journal->j_maxlen = j_inode->i_ext2.i_size / journal->j_blocksize;
 289
 290 #ifdef USE_INODE_IO
 291                 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
 292                                                  &j_inode->i_ext2,
 293                                                  &journal_name);
 294                 if (retval)
 295                         goto errout;
 296
 297                 io_ptr = inode_io_manager;
 298 #else
 299                 journal->j_inode = j_inode;
 300                 ctx->journal_io = ctx->fs->io;
 301                 if ((retval = journal_bmap(journal, 0, &start)) != 0)
 302                         goto errout;
 303 #endif
 304         } else {
 305                 ext_journal = 1;
 306                 if (!ctx->journal_name) {
 307                         char uuid[37];
 308
 309                         uuid_unparse(sb->s_journal_uuid, uuid);
 310                         ctx->journal_name = blkid_get_devname(ctx->blkid,
 311                                                               "UUID", uuid);
 312                         if (!ctx->journal_name)
 313                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
 314                 }
 315                 journal_name = ctx->journal_name;
 316
 317                 if (!journal_name) {
 318                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
 319                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 320                         goto errout;
 321                 }
 322
 323                 jfs_debug(1, "Using journal file %s\n", journal_name);
 324                 io_ptr = unix_io_manager;
 325         }
 326
 327 #if 0
 328         test_io_backing_manager = io_ptr;
 329         io_ptr = test_io_manager;
 330 #endif
 331 #ifndef USE_INODE_IO
 332         if (ext_journal)
 333 #endif
 334                 retval = io_ptr->open(journal_name, IO_FLAG_RW,
 335                                       &ctx->journal_io);
 336         if (retval)
 337                 goto errout;
 338
 339         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
 340
 341         if (ext_journal) {
 342                 if (ctx->fs->blocksize == 1024)
 343                         start = 1;
 344                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
 345                 if (!bh) {
 346                         retval = EXT2_ET_NO_MEMORY;
 347                         goto errout;
 348                 }
 349                 ll_rw_block(READ, 1, &bh);
 350                 if ((retval = bh->b_err) != 0) {
 351                         brelse(bh);
 352                         goto errout;
 353                 }
 354                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + 1024,
 355                        sizeof(jsuper));
 356                 brelse(bh);
 357 #ifdef EXT2FS_ENABLE_SWAPFS
 358                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 359                         ext2fs_swap_super(&jsuper);
 360 #endif
 361                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 362                     !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
 363                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
 364                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 365                         goto errout;
 366                 }
 367                 /* Make sure the journal UUID is correct */
 368                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
 369                            sizeof(jsuper.s_uuid))) {
 370                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
 371                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 372                         goto errout;
 373                 }
 374
 375                 journal->j_maxlen = jsuper.s_blocks_count;
 376                 start++;
 377         }
 378
 379         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
 380                 retval = EXT2_ET_NO_MEMORY;
 381                 goto errout;
 382         }
 383
 384         journal->j_sb_buffer = bh;
 385         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 386
 387 #ifdef USE_INODE_IO
 388         if (j_inode)
 389                 ext2fs_free_mem(&j_inode);
 390 #endif
 391
 392         *ret_journal = journal;
 393         return 0;
 394
 395 errout:
 396         if (dev_fs)
 397                 ext2fs_free_mem(&dev_fs);
 398         if (j_inode)
 399                 ext2fs_free_mem(&j_inode);
 400         if (journal)
 401                 ext2fs_free_mem(&journal);
 402         return retval;
 403 }
 404
 405 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
 406                                               struct problem_context *pctx)
 407 {
 408         struct ext2_super_block *sb = ctx->fs->super;
 409         int recover = ctx->fs->super->s_feature_incompat &
 410                 EXT3_FEATURE_INCOMPAT_RECOVER;
 411         int has_journal = ctx->fs->super->s_feature_compat &
 412                 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 413
 414         if (has_journal || sb->s_journal_inum) {
 415                 /* The journal inode is bogus, remove and force full fsck */
 416                 pctx->ino = sb->s_journal_inum;
 417                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
 418                         if (has_journal && sb->s_journal_inum)
 419                                 printf("*** ext3 journal has been deleted - "
 420                                        "filesystem is now ext2 only ***\n\n");
 421                         sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 422                         sb->s_journal_inum = 0;
 423                         ctx->flags |= E2F_FLAG_JOURNAL_INODE; /* FIXME: todo */
 424                         e2fsck_clear_recover(ctx, 1);
 425                         return 0;
 426                 }
 427                 return EXT2_ET_BAD_INODE_NUM;
 428         } else if (recover) {
 429                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
 430                         e2fsck_clear_recover(ctx, 1);
 431                         return 0;
 432                 }
 433                 return EXT2_ET_UNSUPP_FEATURE;
 434         }
 435         return 0;
 436 }
 437
 438 #define V1_SB_SIZE      0x0024
 439 static void clear_v2_journal_fields(journal_t *journal)
 440 {
 441         e2fsck_t ctx = journal->j_dev->k_ctx;
 442         struct problem_context pctx;
 443
 444         clear_problem_context(&pctx);
 445
 446         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
 447                 return;
 448
 449         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 450                ctx->fs->blocksize-V1_SB_SIZE);
 451         mark_buffer_dirty(journal->j_sb_buffer);
 452 }
 453
 454
 455 static errcode_t e2fsck_journal_load(journal_t *journal)
 456 {
 457         e2fsck_t ctx = journal->j_dev->k_ctx;
 458         journal_superblock_t *jsb;
 459         struct buffer_head *jbh = journal->j_sb_buffer;
 460         struct problem_context pctx;
 461
 462         clear_problem_context(&pctx);
 463
 464         ll_rw_block(READ, 1, &jbh);
 465         if (jbh->b_err) {
 466                 com_err(ctx->device_name, jbh->b_err,
 467                         _("reading journal superblock\n"));
 468                 return jbh->b_err;
 469         }
 470
 471         jsb = journal->j_superblock;
 472         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 473         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 474                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 475
 476         switch (ntohl(jsb->s_header.h_blocktype)) {
 477         case JFS_SUPERBLOCK_V1:
 478                 journal->j_format_version = 1;
 479                 if (jsb->s_feature_compat ||
 480                     jsb->s_feature_incompat ||
 481                     jsb->s_feature_ro_compat ||
 482                     jsb->s_nr_users)
 483                         clear_v2_journal_fields(journal);
 484                 break;
 485
 486         case JFS_SUPERBLOCK_V2:
 487                 journal->j_format_version = 2;
 488                 if (ntohl(jsb->s_nr_users) > 1 &&
 489                     uuid_is_null(ctx->fs->super->s_journal_uuid))
 490                         clear_v2_journal_fields(journal);
 491                 if (ntohl(jsb->s_nr_users) > 1) {
 492                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
 493                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 494                 }
 495                 break;
 496
 497         /*
 498          * These should never appear in a journal super block, so if
 499          * they do, the journal is badly corrupted.
 500          */
 501         case JFS_DESCRIPTOR_BLOCK:
 502         case JFS_COMMIT_BLOCK:
 503         case JFS_REVOKE_BLOCK:
 504                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 505
 506         /* If we don't understand the superblock major type, but there
 507          * is a magic number, then it is likely to be a new format we
 508          * just don't understand, so leave it alone. */
 509         default:
 510                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 511         }
 512
 513         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 514                 return EXT2_ET_UNSUPP_FEATURE;
 515
 516         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 517                 return EXT2_ET_RO_UNSUPP_FEATURE;
 518
 519         /* We have now checked whether we know enough about the journal
 520          * format to be able to proceed safely, so any other checks that
 521          * fail we should attempt to recover from. */
 522         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
 523                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 524                         _("%s: no valid journal superblock found\n"),
 525                         ctx->device_name);
 526                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 527         }
 528
 529         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 530                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 531         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
 532                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
 533                         _("%s: journal too short\n"),
 534                         ctx->device_name);
 535                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 536         }
 537
 538         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 539         journal->j_transaction_sequence = journal->j_tail_sequence;
 540         journal->j_tail = ntohl(jsb->s_start);
 541         journal->j_first = ntohl(jsb->s_first);
 542         journal->j_last = ntohl(jsb->s_maxlen);
 543
 544         return 0;
 545 }
 546
 547 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
 548                                        journal_t *journal)
 549 {
 550         char *p;
 551         union {
 552                 uuid_t uuid;
 553                 __u32 val[4];
 554         } u;
 555         __u32 new_seq = 0;
 556         int i;
 557
 558         /* Leave a valid existing V1 superblock signature alone.
 559          * Anything unrecognisable we overwrite with a new V2
 560          * signature. */
 561
 562         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
 563             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
 564                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
 565                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
 566         }
 567
 568         /* Zero out everything else beyond the superblock header */
 569
 570         p = ((char *) jsb) + sizeof(journal_header_t);
 571         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
 572
 573         jsb->s_blocksize = htonl(ctx->fs->blocksize);
 574         jsb->s_maxlen = htonl(journal->j_maxlen);
 575         jsb->s_first = htonl(1);
 576
 577         /* Initialize the journal sequence number so that there is "no"
 578          * chance we will find old "valid" transactions in the journal.
 579          * This avoids the need to zero the whole journal (slow to do,
 580          * and risky when we are just recovering the filesystem).
 581          */
 582         uuid_generate(u.uuid);
 583         for (i = 0; i < 4; i ++)
 584                 new_seq ^= u.val[i];
 585         jsb->s_sequence = htonl(new_seq);
 586
 587         mark_buffer_dirty(journal->j_sb_buffer);
 588         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
 589 }
 590
 591 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
 592                                                   journal_t *journal,
 593                                                   struct problem_context *pctx)
 594 {
 595         struct ext2_super_block *sb = ctx->fs->super;
 596         int recover = ctx->fs->super->s_feature_incompat &
 597                 EXT3_FEATURE_INCOMPAT_RECOVER;
 598
 599         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
 600                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
 601                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
 602                                                    journal);
 603                         journal->j_transaction_sequence = 1;
 604                         e2fsck_clear_recover(ctx, recover);
 605                         return 0;
 606                 }
 607                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 608         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
 609                 return EXT2_ET_CORRUPT_SUPERBLOCK;
 610
 611         return 0;
 612 }
 613
 614 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
 615                                    int reset, int drop)
 616 {
 617         journal_superblock_t *jsb;
 618
 619         if (drop)
 620                 mark_buffer_clean(journal->j_sb_buffer);
 621         else if (!(ctx->options & E2F_OPT_READONLY)) {
 622                 jsb = journal->j_superblock;
 623                 jsb->s_sequence = htonl(journal->j_transaction_sequence);
 624                 if (reset)
 625                         jsb->s_start = 0; /* this marks the journal as empty */
 626                 mark_buffer_dirty(journal->j_sb_buffer);
 627         }
 628         brelse(journal->j_sb_buffer);
 629
 630         if (ctx->journal_io) {
 631                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
 632                         io_channel_close(ctx->journal_io);
 633                 ctx->journal_io = 0;
 634         }
 635
 636 #ifndef USE_INODE_IO
 637         if (journal->j_inode)
 638                 ext2fs_free_mem(&journal->j_inode);
 639 #endif
 640         if (journal->j_fs_dev)
 641                 ext2fs_free_mem(&journal->j_fs_dev);
 642         ext2fs_free_mem(&journal);
 643 }
 644
 645 /*
 646  * This function makes sure that the superblock fields regarding the
 647  * journal are consistent.
 648  */
 649 int e2fsck_check_ext3_journal(e2fsck_t ctx)
 650 {
 651         struct ext2_super_block *sb = ctx->fs->super;
 652         journal_t *journal;
 653         int recover = ctx->fs->super->s_feature_incompat &
 654                 EXT3_FEATURE_INCOMPAT_RECOVER;
 655         struct problem_context pctx;
 656         problem_t problem;
 657         int reset = 0, force_fsck = 0;
 658         int retval;
 659
 660         /* If we don't have any journal features, don't do anything more */
 661         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
 662             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 663             uuid_is_null(sb->s_journal_uuid))
 664                 return 0;
 665
 666         clear_problem_context(&pctx);
 667         pctx.num = sb->s_journal_inum;
 668
 669         retval = e2fsck_get_journal(ctx, &journal);
 670         if (retval) {
 671                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
 672                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
 673                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
 674                     (retval == EXT2_ET_NO_JOURNAL))
 675                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
 676                 return retval;
 677         }
 678
 679         retval = e2fsck_journal_load(journal);
 680         if (retval) {
 681                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
 682                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
 683                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
 684                                   &pctx))) ||
 685                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
 686                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
 687                                   &pctx))) ||
 688                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
 689                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
 690                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
 691                                                                   &pctx);
 692                 e2fsck_journal_release(ctx, journal, 0, 1);
 693                 return retval;
 694         }
 695
 696         /*
 697          * We want to make the flags consistent here.  We will not leave with
 698          * needs_recovery set but has_journal clear.  We can't get in a loop
 699          * with -y, -n, or -p, only if a user isn't making up their mind.
 700          */
 701 no_has_journal:
 702         if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
 703                 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
 704                 pctx.str = "inode";
 705                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
 706                         if (recover &&
 707                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
 708                                 goto no_has_journal;
 709                         /*
 710                          * Need a full fsck if we are releasing a
 711                          * journal stored on a reserved inode.
 712                          */
 713                         force_fsck = recover ||
 714                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
 715                         /* Clear all of the journal fields */
 716                         sb->s_journal_inum = 0;
 717                         sb->s_journal_dev = 0;
 718                         memset(sb->s_journal_uuid, 0,
 719                                sizeof(sb->s_journal_uuid));
 720                         e2fsck_clear_recover(ctx, force_fsck);
 721                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
 722                         sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
 723                         ext2fs_mark_super_dirty(ctx->fs);
 724                 }
 725         }
 726
 727         if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
 728             !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
 729             journal->j_superblock->s_start != 0) {
 730                 /* Print status information */
 731                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
 732                 if (ctx->superblock)
 733                         problem = PR_0_JOURNAL_RUN_DEFAULT;
 734                 else
 735                         problem = PR_0_JOURNAL_RUN;
 736                 if (fix_problem(ctx, problem, &pctx)) {
 737                         ctx->options |= E2F_OPT_FORCE;
 738                         sb->s_feature_incompat |=
 739                                 EXT3_FEATURE_INCOMPAT_RECOVER;
 740                         ext2fs_mark_super_dirty(ctx->fs);
 741                 } else if (fix_problem(ctx,
 742                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
 743                         reset = 1;
 744                         sb->s_state &= ~EXT2_VALID_FS;
 745                         ext2fs_mark_super_dirty(ctx->fs);
 746                 }
 747                 /*
 748                  * If the user answers no to the above question, we
 749                  * ignore the fact that journal apparently has data;
 750                  * accidentally replaying over valid data would be far
 751                  * worse than skipping a questionable recovery.
 752                  *
 753                  * XXX should we abort with a fatal error here?  What
 754                  * will the ext3 kernel code do if a filesystem with
 755                  * !NEEDS_RECOVERY but with a non-zero
 756                  * journal->j_superblock->s_start is mounted?
 757                  */
 758         }
 759
 760         e2fsck_journal_release(ctx, journal, reset, 0);
 761         return retval;
 762 }
 763
 764 static errcode_t recover_ext3_journal(e2fsck_t ctx)
 765 {
 766         journal_t *journal;
 767         int retval;
 768
 769         journal_init_revoke_caches();
 770         retval = e2fsck_get_journal(ctx, &journal);
 771         if (retval)
 772                 return retval;
 773
 774         retval = e2fsck_journal_load(journal);
 775         if (retval)
 776                 goto errout;
 777
 778         retval = journal_init_revoke(journal, 1024);
 779         if (retval)
 780                 goto errout;
 781
 782         retval = -journal_recover(journal);
 783         if (retval)
 784                 goto errout;
 785
 786         if (journal->j_superblock->s_errno) {
 787                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
 788                 ext2fs_mark_super_dirty(ctx->fs);
 789                 journal->j_superblock->s_errno = 0;
 790                 mark_buffer_dirty(journal->j_sb_buffer);
 791         }
 792
 793 errout:
 794         journal_destroy_revoke(journal);
 795         journal_destroy_revoke_caches();
 796         e2fsck_journal_release(ctx, journal, 1, 0);
 797         return retval;
 798 }
 799
 800 int e2fsck_run_ext3_journal(e2fsck_t ctx)
 801 {
 802         io_manager io_ptr = ctx->fs->io->manager;
 803         int blocksize = ctx->fs->blocksize;
 804         errcode_t       retval, recover_retval;
 805
 806         printf(_("%s: recovering journal\n"), ctx->device_name);
 807         if (ctx->options & E2F_OPT_READONLY) {
 808                 printf(_("%s: won't do journal recovery while read-only\n"),
 809                        ctx->device_name);
 810                 return EXT2_ET_FILE_RO;
 811         }
 812
 813         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
 814                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
 815
 816         recover_retval = recover_ext3_journal(ctx);
 817
 818         /*
 819          * Reload the filesystem context to get up-to-date data from disk
 820          * because journal recovery will change the filesystem under us.
 821          */
 822         ext2fs_close(ctx->fs);
 823         retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
 824                              ctx->superblock, blocksize, io_ptr,
 825                              &ctx->fs);
 826
 827         if (retval) {
 828                 com_err(ctx->program_name, retval,
 829                         _("while trying to re-open %s"),
 830                         ctx->device_name);
 831                 fatal_error(ctx, 0);
 832         }
 833         ctx->fs->priv_data = ctx;
 834         ctx->fs->now = ctx->now;
 835
 836         /* Set the superblock flags */
 837         e2fsck_clear_recover(ctx, recover_retval);
 838         return recover_retval;
 839 }
 840
 841 /*
 842  * This function will move the journal inode from a visible file in
 843  * the filesystem directory hierarchy to the reserved inode if necessary.
 844  */
 845 static const char * const journal_names[] = {
 846         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
 847
 848 void e2fsck_move_ext3_journal(e2fsck_t ctx)
 849 {
 850         struct ext2_super_block *sb = ctx->fs->super;
 851         struct problem_context  pctx;
 852         struct ext2_inode       inode;
 853         ext2_filsys             fs = ctx->fs;
 854         ext2_ino_t              ino;
 855         errcode_t               retval;
 856         const char * const *    cpp;
 857         int                     group, mount_flags;
 858
 859         clear_problem_context(&pctx);
 860
 861         /*
 862          * If the filesystem is opened read-only, or there is no
 863          * journal, then do nothing.
 864          */
 865         if ((ctx->options & E2F_OPT_READONLY) ||
 866             (sb->s_journal_inum == 0) ||
 867             !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
 868                 return;
 869
 870         /*
 871          * Read in the journal inode
 872          */
 873         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
 874                 return;
 875
 876         /*
 877          * If it's necessary to backup the journal inode, do so.
 878          */
 879         if ((sb->s_jnl_backup_type == 0) ||
 880             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
 881              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
 882                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
 883                         memcpy(sb->s_jnl_blocks, inode.i_block,
 884                                EXT2_N_BLOCKS*4);
 885                         sb->s_jnl_blocks[16] = inode.i_size;
 886                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
 887                         ext2fs_mark_super_dirty(fs);
 888                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 889                 }
 890         }
 891
 892         /*
 893          * If the journal is already the hidden inode, then do nothing
 894          */
 895         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
 896                 return;
 897
 898         /*
 899          * The journal inode had better have only one link and not be readable.
 900          */
 901         if (inode.i_links_count != 1)
 902                 return;
 903
 904         /*
 905          * If the filesystem is mounted, or we can't tell whether
 906          * or not it's mounted, do nothing.
 907          */
 908         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
 909         if (retval || (mount_flags & EXT2_MF_MOUNTED))
 910                 return;
 911
 912         /*
 913          * If we can't find the name of the journal inode, then do
 914          * nothing.
 915          */
 916         for (cpp = journal_names; *cpp; cpp++) {
 917                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
 918                                        strlen(*cpp), 0, &ino);
 919                 if ((retval == 0) && (ino == sb->s_journal_inum))
 920                         break;
 921         }
 922         if (*cpp == 0)
 923                 return;
 924
 925         /* We need the inode bitmap to be loaded */
 926         retval = ext2fs_read_bitmaps(fs);
 927         if (retval)
 928                 return;
 929
 930         pctx.str = *cpp;
 931         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
 932                 return;
 933
 934         /*
 935          * OK, we've done all the checks, let's actually move the
 936          * journal inode.  Errors at this point mean we need to force
 937          * an ext2 filesystem check.
 938          */
 939         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
 940                 goto err_out;
 941         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
 942                 goto err_out;
 943         sb->s_journal_inum = EXT2_JOURNAL_INO;
 944         ext2fs_mark_super_dirty(fs);
 945         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
 946         inode.i_links_count = 0;
 947         inode.i_dtime = ctx->now;
 948         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
 949                 goto err_out;
 950
 951         group = ext2fs_group_of_ino(fs, ino);
 952         ext2fs_unmark_inode_bitmap(fs->inode_map, ino);
 953         ext2fs_mark_ib_dirty(fs);
 954         fs->group_desc[group].bg_free_inodes_count++;
 955         fs->super->s_free_inodes_count++;
 956         return;
 957
 958 err_out:
 959         pctx.errcode = retval;
 960         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
 961         fs->super->s_state &= ~EXT2_VALID_FS;
 962         ext2fs_mark_super_dirty(fs);
 963         return;
 964 }
 965
 966 /*
 967  * This function makes sure the superblock hint for the external
 968  * journal is correct.
 969  */
 970 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
 971 {
 972         struct ext2_super_block *sb = ctx->fs->super;
 973         struct problem_context pctx;
 974         char uuid[37], *journal_name;
 975         struct stat st;
 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 }