debugfs/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 #include "config.h"
  16 #ifdef HAVE_SYS_MOUNT_H
  17 #include <sys/param.h>
  18 #include <sys/mount.h>
  19 #define MNT_FL (MS_MGC_VAL | MS_RDONLY)
  20 #endif
  21 #ifdef HAVE_SYS_STAT_H
  22 #include <sys/stat.h>
  23 #endif
  24
  25 #define E2FSCK_INCLUDE_INLINE_FUNCS
  26 #include "uuid/uuid.h"
  27 #include "journal.h"
  28
  29 #ifdef CONFIG_JBD_DEBUG         /* Enabled by configure --enable-jfs-debug */
  30 static int bh_count = 0;
  31 #endif
  32
  33 #if EXT2_FLAT_INCLUDES
  34 #include "blkid.h"
  35 #else
  36 #include "blkid/blkid.h"
  37 #endif
  38
  39 /*
  40  * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
  41  * This creates a larger static binary, and a smaller binary using
  42  * shared libraries.  It's also probably slightly less CPU-efficient,
  43  * which is why it's not on by default.  But, it's a good way of
  44  * testing the functions in inode_io.c and fileio.c.
  45  */
  46 #undef USE_INODE_IO
  47
  48 /* Checksumming functions */
  49 static int ext2fs_journal_verify_csum_type(journal_t *j,
  50                                            journal_superblock_t *jsb)
  51 {
  52         if (!journal_has_csum_v2or3(j))
  53                 return 1;
  54
  55         return jsb->s_checksum_type == JBD2_CRC32C_CHKSUM;
  56 }
  57
  58 static __u32 ext2fs_journal_sb_csum(journal_superblock_t *jsb)
  59 {
  60         __u32 crc, old_crc;
  61
  62         old_crc = jsb->s_checksum;
  63         jsb->s_checksum = 0;
  64         crc = ext2fs_crc32c_le(~0, (unsigned char *)jsb,
  65                                sizeof(journal_superblock_t));
  66         jsb->s_checksum = old_crc;
  67
  68         return crc;
  69 }
  70
  71 static int ext2fs_journal_sb_csum_verify(journal_t *j,
  72                                          journal_superblock_t *jsb)
  73 {
  74         __u32 provided, calculated;
  75
  76         if (!journal_has_csum_v2or3(j))
  77                 return 1;
  78
  79         provided = ext2fs_be32_to_cpu(jsb->s_checksum);
  80         calculated = ext2fs_journal_sb_csum(jsb);
  81
  82         return provided == calculated;
  83 }
  84
  85 static errcode_t ext2fs_journal_sb_csum_set(journal_t *j,
  86                                             journal_superblock_t *jsb)
  87 {
  88         __u32 crc;
  89
  90         if (!journal_has_csum_v2or3(j))
  91                 return 0;
  92
  93         crc = ext2fs_journal_sb_csum(jsb);
  94         jsb->s_checksum = ext2fs_cpu_to_be32(crc);
  95         return 0;
  96 }
  97
  98 /* Kernel compatibility functions for handling the journal.  These allow us
  99  * to use the recovery.c file virtually unchanged from the kernel, so we
 100  * don't have to do much to keep kernel and user recovery in sync.
 101  */
 102 int journal_bmap(journal_t *journal, blk64_t block, unsigned long long *phys)
 103 {
 104 #ifdef USE_INODE_IO
 105         *phys = block;
 106         return 0;
 107 #else
 108         struct inode    *inode = journal->j_inode;
 109         errcode_t       retval;
 110         blk64_t         pblk;
 111
 112         if (!inode) {
 113                 *phys = block;
 114                 return 0;
 115         }
 116
 117         retval = ext2fs_bmap2(inode->i_fs, inode->i_ino,
 118                               &inode->i_ext2, NULL, 0, block, 0, &pblk);
 119         *phys = pblk;
 120         return (int) retval;
 121 #endif
 122 }
 123
 124 struct buffer_head *getblk(kdev_t kdev, blk64_t blocknr, int blocksize)
 125 {
 126         struct buffer_head *bh;
 127         int bufsize = sizeof(*bh) + kdev->k_fs->blocksize -
 128                 sizeof(bh->b_data);
 129         errcode_t retval;
 130
 131         retval = ext2fs_get_memzero(bufsize, &bh);
 132         if (retval)
 133                 return NULL;
 134
 135 #ifdef CONFIG_JBD_DEBUG
 136         if (journal_enable_debug >= 3)
 137                 bh_count++;
 138 #endif
 139         jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
 140                   (unsigned long long) blocknr, blocksize, bh_count);
 141
 142         bh->b_fs = kdev->k_fs;
 143         if (kdev->k_dev == K_DEV_FS)
 144                 bh->b_io = kdev->k_fs->io;
 145         else
 146                 bh->b_io = kdev->k_fs->journal_io;
 147         bh->b_size = blocksize;
 148         bh->b_blocknr = blocknr;
 149
 150         return bh;
 151 }
 152
 153 int sync_blockdev(kdev_t kdev)
 154 {
 155         io_channel      io;
 156
 157         if (kdev->k_dev == K_DEV_FS)
 158                 io = kdev->k_fs->io;
 159         else
 160                 io = kdev->k_fs->journal_io;
 161
 162         return io_channel_flush(io) ? EIO : 0;
 163 }
 164
 165 void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
 166 {
 167         errcode_t retval;
 168         struct buffer_head *bh;
 169
 170         for (; nr > 0; --nr) {
 171                 bh = *bhp++;
 172                 if (rw == READ && !bh->b_uptodate) {
 173                         jfs_debug(3, "reading block %llu/%p\n",
 174                                   bh->b_blocknr, (void *) bh);
 175                         retval = io_channel_read_blk64(bh->b_io,
 176                                                      bh->b_blocknr,
 177                                                      1, bh->b_data);
 178                         if (retval) {
 179                                 com_err(bh->b_fs->device_name, retval,
 180                                         "while reading block %llu\n",
 181                                         bh->b_blocknr);
 182                                 bh->b_err = (int) retval;
 183                                 continue;
 184                         }
 185                         bh->b_uptodate = 1;
 186                 } else if (rw == WRITE && bh->b_dirty) {
 187                         jfs_debug(3, "writing block %llu/%p\n",
 188                                   bh->b_blocknr,
 189                                   (void *) bh);
 190                         retval = io_channel_write_blk64(bh->b_io,
 191                                                       bh->b_blocknr,
 192                                                       1, bh->b_data);
 193                         if (retval) {
 194                                 com_err(bh->b_fs->device_name, retval,
 195                                         "while writing block %llu\n",
 196                                         bh->b_blocknr);
 197                                 bh->b_err = (int) retval;
 198                                 continue;
 199                         }
 200                         bh->b_dirty = 0;
 201                         bh->b_uptodate = 1;
 202                 } else {
 203                         jfs_debug(3, "no-op %s for block %llu\n",
 204                                   rw == READ ? "read" : "write",
 205                                   bh->b_blocknr);
 206                 }
 207         }
 208 }
 209
 210 void mark_buffer_dirty(struct buffer_head *bh)
 211 {
 212         bh->b_dirty = 1;
 213 }
 214
 215 static void mark_buffer_clean(struct buffer_head *bh)
 216 {
 217         bh->b_dirty = 0;
 218 }
 219
 220 void brelse(struct buffer_head *bh)
 221 {
 222         if (bh->b_dirty)
 223                 ll_rw_block(WRITE, 1, &bh);
 224         jfs_debug(3, "freeing block %llu/%p (total %d)\n",
 225                   bh->b_blocknr, (void *) bh, --bh_count);
 226         ext2fs_free_mem(&bh);
 227 }
 228
 229 int buffer_uptodate(struct buffer_head *bh)
 230 {
 231         return bh->b_uptodate;
 232 }
 233
 234 void mark_buffer_uptodate(struct buffer_head *bh, int val)
 235 {
 236         bh->b_uptodate = val;
 237 }
 238
 239 void wait_on_buffer(struct buffer_head *bh)
 240 {
 241         if (!bh->b_uptodate)
 242                 ll_rw_block(READ, 1, &bh);
 243 }
 244
 245
 246 static void ext2fs_clear_recover(ext2_filsys fs, int error)
 247 {
 248         ext2fs_clear_feature_journal_needs_recovery(fs->super);
 249
 250         /* if we had an error doing journal recovery, we need a full fsck */
 251         if (error)
 252                 fs->super->s_state &= ~EXT2_VALID_FS;
 253         /*
 254          * If we replayed the journal by definition the file system
 255          * was mounted since the last time it was checked
 256          */
 257         if (fs->super->s_lastcheck >= fs->super->s_mtime)
 258                 fs->super->s_lastcheck = fs->super->s_mtime - 1;
 259         ext2fs_mark_super_dirty(fs);
 260 }
 261
 262 /*
 263  * This is a helper function to check the validity of the journal.
 264  */
 265 struct process_block_struct {
 266         e2_blkcnt_t     last_block;
 267 };
 268
 269 static int process_journal_block(ext2_filsys fs,
 270                                  blk64_t        *block_nr,
 271                                  e2_blkcnt_t blockcnt,
 272                                  blk64_t ref_block EXT2FS_ATTR((unused)),
 273                                  int ref_offset EXT2FS_ATTR((unused)),
 274                                  void *priv_data)
 275 {
 276         struct process_block_struct *p;
 277         blk64_t blk = *block_nr;
 278
 279         p = (struct process_block_struct *) priv_data;
 280
 281         if (!blk || blk < fs->super->s_first_data_block ||
 282             blk >= ext2fs_blocks_count(fs->super))
 283                 return BLOCK_ABORT;
 284
 285         if (blockcnt >= 0)
 286                 p->last_block = blockcnt;
 287         return 0;
 288 }
 289
 290 static errcode_t ext2fs_get_journal(ext2_filsys fs, journal_t **ret_journal)
 291 {
 292         struct process_block_struct pb;
 293         struct ext2_super_block *sb = fs->super;
 294         struct ext2_super_block jsuper;
 295         struct buffer_head      *bh;
 296         struct inode            *j_inode = NULL;
 297         struct kdev_s           *dev_fs = NULL, *dev_journal;
 298         const char              *journal_name = 0;
 299         journal_t               *journal = NULL;
 300         errcode_t               retval = 0;
 301         io_manager              io_ptr = 0;
 302         unsigned long long      start = 0;
 303         int                     ext_journal = 0;
 304         int                     tried_backup_jnl = 0;
 305
 306         retval = ext2fs_get_memzero(sizeof(journal_t), &journal);
 307         if (retval)
 308                 return retval;
 309
 310         retval = ext2fs_get_memzero(2 * sizeof(struct kdev_s), &dev_fs);
 311         if (retval)
 312                 goto errout;
 313         dev_journal = dev_fs+1;
 314
 315         dev_fs->k_fs = dev_journal->k_fs = fs;
 316         dev_fs->k_dev = K_DEV_FS;
 317         dev_journal->k_dev = K_DEV_JOURNAL;
 318
 319         journal->j_dev = dev_journal;
 320         journal->j_fs_dev = dev_fs;
 321         journal->j_inode = NULL;
 322         journal->j_blocksize = fs->blocksize;
 323
 324         if (uuid_is_null(sb->s_journal_uuid)) {
 325                 if (!sb->s_journal_inum) {
 326                         retval = EXT2_ET_BAD_INODE_NUM;
 327                         goto errout;
 328                 }
 329                 retval = ext2fs_get_memzero(sizeof(*j_inode), &j_inode);
 330                 if (retval)
 331                         goto errout;
 332
 333                 j_inode->i_fs = fs;
 334                 j_inode->i_ino = sb->s_journal_inum;
 335
 336                 retval = ext2fs_read_inode(fs, sb->s_journal_inum,
 337                                            &j_inode->i_ext2);
 338                 if (retval) {
 339 try_backup_journal:
 340                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
 341                             tried_backup_jnl)
 342                                 goto errout;
 343                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
 344                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
 345                                EXT2_N_BLOCKS*4);
 346                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
 347                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
 348                         j_inode->i_ext2.i_links_count = 1;
 349                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
 350                         tried_backup_jnl++;
 351                 }
 352                 if (!j_inode->i_ext2.i_links_count ||
 353                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
 354                         retval = EXT2_ET_NO_JOURNAL;
 355                         goto try_backup_journal;
 356                 }
 357                 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
 358                     JFS_MIN_JOURNAL_BLOCKS) {
 359                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 360                         goto try_backup_journal;
 361                 }
 362                 pb.last_block = -1;
 363                 retval = ext2fs_block_iterate3(fs, j_inode->i_ino,
 364                                                BLOCK_FLAG_HOLE, 0,
 365                                                process_journal_block, &pb);
 366                 if ((pb.last_block + 1) * fs->blocksize <
 367                     (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
 368                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 369                         goto try_backup_journal;
 370                 }
 371                 if (tried_backup_jnl && (fs->flags & EXT2_FLAG_RW)) {
 372                         retval = ext2fs_write_inode(fs, sb->s_journal_inum,
 373                                                     &j_inode->i_ext2);
 374                         if (retval)
 375                                 goto errout;
 376                 }
 377
 378                 journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
 379                         journal->j_blocksize;
 380
 381 #ifdef USE_INODE_IO
 382                 retval = ext2fs_inode_io_intern2(fs, sb->s_journal_inum,
 383                                                  &j_inode->i_ext2,
 384                                                  &journal_name);
 385                 if (retval)
 386                         goto errout;
 387
 388                 io_ptr = inode_io_manager;
 389 #else
 390                 journal->j_inode = j_inode;
 391                 fs->journal_io = fs->io;
 392                 retval = (errcode_t)journal_bmap(journal, 0, &start);
 393                 if (retval)
 394                         goto errout;
 395 #endif
 396         } else {
 397                 ext_journal = 1;
 398                 if (!fs->journal_name) {
 399                         char uuid[37];
 400                         blkid_cache blkid;
 401
 402                         blkid_get_cache(&blkid, NULL);
 403                         uuid_unparse(sb->s_journal_uuid, uuid);
 404                         fs->journal_name = blkid_get_devname(blkid,
 405                                                               "UUID", uuid);
 406                         if (!fs->journal_name)
 407                                 fs->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
 408                         blkid_put_cache(blkid);
 409                 }
 410                 journal_name = fs->journal_name;
 411
 412                 if (!journal_name) {
 413                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 414                         goto errout;
 415                 }
 416
 417                 jfs_debug(1, "Using journal file %s\n", journal_name);
 418                 io_ptr = unix_io_manager;
 419         }
 420
 421 #if 0
 422         test_io_backing_manager = io_ptr;
 423         io_ptr = test_io_manager;
 424 #endif
 425 #ifndef USE_INODE_IO
 426         if (ext_journal)
 427 #endif
 428         {
 429                 retval = io_ptr->open(journal_name, fs->flags & EXT2_FLAG_RW,
 430                                       &fs->journal_io);
 431         }
 432         if (retval)
 433                 goto errout;
 434
 435         io_channel_set_blksize(fs->journal_io, fs->blocksize);
 436
 437         if (ext_journal) {
 438                 blk64_t maxlen;
 439
 440                 start = ext2fs_journal_sb_start(fs->blocksize) - 1;
 441                 bh = getblk(dev_journal, start, fs->blocksize);
 442                 if (!bh) {
 443                         retval = EXT2_ET_NO_MEMORY;
 444                         goto errout;
 445                 }
 446                 ll_rw_block(READ, 1, &bh);
 447                 retval = bh->b_err;
 448                 if (retval) {
 449                         brelse(bh);
 450                         goto errout;
 451                 }
 452                 memcpy(&jsuper, start ? bh->b_data :
 453                                 bh->b_data + SUPERBLOCK_OFFSET,
 454                        sizeof(jsuper));
 455 #ifdef WORDS_BIGENDIAN
 456                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 457                         ext2fs_swap_super(&jsuper);
 458 #endif
 459                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 460                     !ext2fs_has_feature_journal_dev(&jsuper)) {
 461                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 462                         brelse(bh);
 463                         goto errout;
 464                 }
 465                 /* Make sure the journal UUID is correct */
 466                 if (memcmp(jsuper.s_uuid, fs->super->s_journal_uuid,
 467                            sizeof(jsuper.s_uuid))) {
 468                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 469                         brelse(bh);
 470                         goto errout;
 471                 }
 472
 473                 /* Check the superblock checksum */
 474                 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
 475                         struct struct_ext2_filsys fsx;
 476                         struct ext2_super_block superx;
 477                         void *p;
 478
 479                         p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
 480                         memcpy(&fsx, fs, sizeof(fsx));
 481                         memcpy(&superx, fs->super, sizeof(superx));
 482                         fsx.super = &superx;
 483                         ext2fs_set_feature_metadata_csum(fsx.super);
 484                         if (!ext2fs_superblock_csum_verify(&fsx, p)) {
 485                                 retval = EXT2_ET_LOAD_EXT_JOURNAL;
 486                                 brelse(bh);
 487                                 goto errout;
 488                         }
 489                 }
 490                 brelse(bh);
 491
 492                 maxlen = ext2fs_blocks_count(&jsuper);
 493                 journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen :
 494                                     (1ULL << 32) - 1;
 495                 start++;
 496         }
 497
 498         bh = getblk(dev_journal, start, journal->j_blocksize);
 499         if (!bh) {
 500                 retval = EXT2_ET_NO_MEMORY;
 501                 goto errout;
 502         }
 503
 504         journal->j_sb_buffer = bh;
 505         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 506
 507 #ifdef USE_INODE_IO
 508         if (j_inode)
 509                 ext2fs_free_mem(&j_inode);
 510 #endif
 511
 512         *ret_journal = journal;
 513         return 0;
 514
 515 errout:
 516         if (dev_fs)
 517                 ext2fs_free_mem(&dev_fs);
 518         if (j_inode)
 519                 ext2fs_free_mem(&j_inode);
 520         if (journal)
 521                 ext2fs_free_mem(&journal);
 522         return retval;
 523 }
 524
 525 static errcode_t ext2fs_journal_fix_bad_inode(ext2_filsys fs)
 526 {
 527         struct ext2_super_block *sb = fs->super;
 528         int recover = ext2fs_has_feature_journal_needs_recovery(fs->super);
 529         int has_journal = ext2fs_has_feature_journal(fs->super);
 530
 531         if (has_journal || sb->s_journal_inum) {
 532                 /* The journal inode is bogus, remove and force full fsck */
 533                 return EXT2_ET_BAD_INODE_NUM;
 534         } else if (recover) {
 535                 return EXT2_ET_UNSUPP_FEATURE;
 536         }
 537         return 0;
 538 }
 539
 540 #define V1_SB_SIZE      0x0024
 541 static void clear_v2_journal_fields(journal_t *journal)
 542 {
 543         ext2_filsys fs = journal->j_dev->k_fs;
 544
 545         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 546                fs->blocksize-V1_SB_SIZE);
 547         mark_buffer_dirty(journal->j_sb_buffer);
 548 }
 549
 550
 551 static errcode_t ext2fs_journal_load(journal_t *journal)
 552 {
 553         ext2_filsys fs = journal->j_dev->k_fs;
 554         journal_superblock_t *jsb;
 555         struct buffer_head *jbh = journal->j_sb_buffer;
 556
 557         ll_rw_block(READ, 1, &jbh);
 558         if (jbh->b_err)
 559                 return jbh->b_err;
 560
 561         jsb = journal->j_superblock;
 562         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
 563         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
 564                 return ext2fs_journal_fix_bad_inode(fs);
 565
 566         switch (ntohl(jsb->s_header.h_blocktype)) {
 567         case JFS_SUPERBLOCK_V1:
 568                 journal->j_format_version = 1;
 569                 if (jsb->s_feature_compat ||
 570                     jsb->s_feature_incompat ||
 571                     jsb->s_feature_ro_compat ||
 572                     jsb->s_nr_users)
 573                         clear_v2_journal_fields(journal);
 574                 break;
 575
 576         case JFS_SUPERBLOCK_V2:
 577                 journal->j_format_version = 2;
 578                 if (ntohl(jsb->s_nr_users) > 1 &&
 579                     uuid_is_null(fs->super->s_journal_uuid))
 580                         clear_v2_journal_fields(journal);
 581                 if (ntohl(jsb->s_nr_users) > 1)
 582                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 583                 break;
 584
 585         /*
 586          * These should never appear in a journal super block, so if
 587          * they do, the journal is badly corrupted.
 588          */
 589         case JFS_DESCRIPTOR_BLOCK:
 590         case JFS_COMMIT_BLOCK:
 591         case JFS_REVOKE_BLOCK:
 592                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 593
 594         /* If we don't understand the superblock major type, but there
 595          * is a magic number, then it is likely to be a new format we
 596          * just don't understand, so leave it alone. */
 597         default:
 598                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 599         }
 600
 601         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
 602                 return EXT2_ET_UNSUPP_FEATURE;
 603
 604         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
 605                 return EXT2_ET_RO_UNSUPP_FEATURE;
 606
 607         /* Checksum v1-3 are mutually exclusive features. */
 608         if (jfs_has_feature_csum2(journal) && jfs_has_feature_csum3(journal))
 609                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 610
 611         if (journal_has_csum_v2or3(journal) &&
 612             jfs_has_feature_checksum(journal))
 613                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 614
 615         if (!ext2fs_journal_verify_csum_type(journal, jsb) ||
 616             !ext2fs_journal_sb_csum_verify(journal, jsb))
 617                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 618
 619         if (journal_has_csum_v2or3(journal))
 620                 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
 621                                                    sizeof(jsb->s_uuid));
 622
 623         /* We have now checked whether we know enough about the journal
 624          * format to be able to proceed safely, so any other checks that
 625          * fail we should attempt to recover from. */
 626         if (jsb->s_blocksize != htonl(journal->j_blocksize))
 627                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 628
 629         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 630                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 631         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen)
 632                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 633
 634         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 635         journal->j_transaction_sequence = journal->j_tail_sequence;
 636         journal->j_tail = ntohl(jsb->s_start);
 637         journal->j_first = ntohl(jsb->s_first);
 638         journal->j_last = ntohl(jsb->s_maxlen);
 639
 640         return 0;
 641 }
 642
 643 static void ext2fs_journal_release(ext2_filsys fs, journal_t *journal,
 644                                    int reset, int drop)
 645 {
 646         journal_superblock_t *jsb;
 647
 648         if (drop)
 649                 mark_buffer_clean(journal->j_sb_buffer);
 650         else if (fs->flags & EXT2_FLAG_RW) {
 651                 jsb = journal->j_superblock;
 652                 jsb->s_sequence = htonl(journal->j_tail_sequence);
 653                 if (reset)
 654                         jsb->s_start = 0; /* this marks the journal as empty */
 655                 ext2fs_journal_sb_csum_set(journal, jsb);
 656                 mark_buffer_dirty(journal->j_sb_buffer);
 657         }
 658         brelse(journal->j_sb_buffer);
 659
 660         if (fs && fs->journal_io) {
 661                 if (fs->io != fs->journal_io)
 662                         io_channel_close(fs->journal_io);
 663                 fs->journal_io = NULL;
 664                 free(fs->journal_name);
 665                 fs->journal_name = NULL;
 666         }
 667
 668 #ifndef USE_INODE_IO
 669         if (journal->j_inode)
 670                 ext2fs_free_mem(&journal->j_inode);
 671 #endif
 672         if (journal->j_fs_dev)
 673                 ext2fs_free_mem(&journal->j_fs_dev);
 674         ext2fs_free_mem(&journal);
 675 }
 676
 677 /*
 678  * This function makes sure that the superblock fields regarding the
 679  * journal are consistent.
 680  */
 681 static errcode_t ext2fs_check_ext3_journal(ext2_filsys fs)
 682 {
 683         struct ext2_super_block *sb = fs->super;
 684         journal_t *journal;
 685         int recover = ext2fs_has_feature_journal_needs_recovery(fs->super);
 686         errcode_t retval;
 687
 688         /* If we don't have any journal features, don't do anything more */
 689         if (!ext2fs_has_feature_journal(sb) &&
 690             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 691             uuid_is_null(sb->s_journal_uuid))
 692                 return 0;
 693
 694         retval = ext2fs_get_journal(fs, &journal);
 695         if (retval)
 696                 return retval;
 697
 698         retval = ext2fs_journal_load(journal);
 699         if (retval)
 700                 goto err;
 701
 702         /*
 703          * We want to make the flags consistent here.  We will not leave with
 704          * needs_recovery set but has_journal clear.  We can't get in a loop
 705          * with -y, -n, or -p, only if a user isn't making up their mind.
 706          */
 707         if (!ext2fs_has_feature_journal(sb)) {
 708                 retval = EXT2_ET_JOURNAL_FLAGS_WRONG;
 709                 goto err;
 710         }
 711
 712         if (ext2fs_has_feature_journal(sb) &&
 713             !ext2fs_has_feature_journal_needs_recovery(sb) &&
 714             journal->j_superblock->s_start != 0) {
 715                 retval = EXT2_ET_JOURNAL_FLAGS_WRONG;
 716                 goto err;
 717         }
 718
 719         /*
 720          * If we don't need to do replay the journal, check to see if
 721          * the journal's errno is set; if so, we need to mark the file
 722          * system as being corrupt and clear the journal's s_errno.
 723          */
 724         if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
 725             journal->j_superblock->s_errno) {
 726                 fs->super->s_state |= EXT2_ERROR_FS;
 727                 ext2fs_mark_super_dirty(fs);
 728                 journal->j_superblock->s_errno = 0;
 729                 ext2fs_journal_sb_csum_set(journal, journal->j_superblock);
 730                 mark_buffer_dirty(journal->j_sb_buffer);
 731         }
 732
 733 err:
 734         ext2fs_journal_release(fs, journal, 0, retval ? 1 : 0);
 735         return retval;
 736 }
 737
 738 static errcode_t recover_ext3_journal(ext2_filsys fs)
 739 {
 740         journal_t *journal;
 741         errcode_t retval;
 742
 743         journal_init_revoke_caches();
 744         retval = ext2fs_get_journal(fs, &journal);
 745         if (retval)
 746                 return retval;
 747
 748         retval = ext2fs_journal_load(journal);
 749         if (retval)
 750                 goto errout;
 751
 752         retval = journal_init_revoke(journal, 1024);
 753         if (retval)
 754                 goto errout;
 755
 756         retval = -journal_recover(journal);
 757         if (retval)
 758                 goto errout;
 759
 760         if (journal->j_failed_commit) {
 761                 journal->j_superblock->s_errno = -EINVAL;
 762                 mark_buffer_dirty(journal->j_sb_buffer);
 763         }
 764
 765 errout:
 766         journal_destroy_revoke(journal);
 767         journal_destroy_revoke_caches();
 768         ext2fs_journal_release(fs, journal, 1, 0);
 769         return retval;
 770 }
 771
 772 errcode_t ext2fs_run_ext3_journal(ext2_filsys *fsp)
 773 {
 774         ext2_filsys fs = *fsp;
 775         io_manager io_ptr = fs->io->manager;
 776         errcode_t       retval, recover_retval;
 777         io_stats        stats = 0;
 778         unsigned long long kbytes_written = 0;
 779         char *fsname;
 780         int fsflags;
 781         int fsblocksize;
 782
 783         if (!(fs->flags & EXT2_FLAG_RW))
 784                 return EXT2_ET_FILE_RO;
 785
 786         if (fs->flags & EXT2_FLAG_DIRTY)
 787                 ext2fs_flush(fs);       /* Force out any modifications */
 788
 789         recover_retval = recover_ext3_journal(fs);
 790
 791         /*
 792          * Reload the filesystem context to get up-to-date data from disk
 793          * because journal recovery will change the filesystem under us.
 794          */
 795         if (fs->super->s_kbytes_written &&
 796             fs->io->manager->get_stats)
 797                 fs->io->manager->get_stats(fs->io, &stats);
 798         if (stats && stats->bytes_written)
 799                 kbytes_written = stats->bytes_written >> 10;
 800
 801         ext2fs_mmp_stop(fs);
 802         fsname = fs->device_name;
 803         fs->device_name = NULL;
 804         fsflags = fs->flags;
 805         fsblocksize = fs->blocksize;
 806         ext2fs_free(fs);
 807         *fsp = NULL;
 808         retval = ext2fs_open(fsname, fsflags, 0, fsblocksize, io_ptr, fsp);
 809         ext2fs_free_mem(&fsname);
 810         if (retval)
 811                 return retval;
 812
 813         fs = *fsp;
 814         fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
 815         fs->super->s_kbytes_written += kbytes_written;
 816
 817         /* Set the superblock flags */
 818         ext2fs_clear_recover(fs, recover_retval != 0);
 819
 820         /*
 821          * Do one last sanity check, and propagate journal->s_errno to
 822          * the EXT2_ERROR_FS flag in the fs superblock if needed.
 823          */
 824         retval = ext2fs_check_ext3_journal(fs);
 825         return retval ? retval : recover_retval;
 826 }
 827
 828 errcode_t ext2fs_open_journal(ext2_filsys fs, journal_t **j)
 829 {
 830         journal_t *journal;
 831         errcode_t retval;
 832
 833         journal_init_revoke_caches();
 834         retval = ext2fs_get_journal(fs, &journal);
 835         if (retval)
 836                 return retval;
 837
 838         retval = ext2fs_journal_load(journal);
 839         if (retval)
 840                 goto errout;
 841
 842         retval = journal_init_revoke(journal, 1024);
 843         if (retval)
 844                 goto errout;
 845
 846         if (journal->j_failed_commit) {
 847                 journal->j_superblock->s_errno = -EINVAL;
 848                 mark_buffer_dirty(journal->j_sb_buffer);
 849         }
 850
 851         *j = journal;
 852         return 0;
 853
 854 errout:
 855         journal_destroy_revoke(journal);
 856         journal_destroy_revoke_caches();
 857         ext2fs_journal_release(fs, journal, 1, 0);
 858         return retval;
 859 }
 860
 861 errcode_t ext2fs_close_journal(ext2_filsys fs, journal_t **j)
 862 {
 863         journal_t *journal = *j;
 864
 865         journal_destroy_revoke(journal);
 866         journal_destroy_revoke_caches();
 867         ext2fs_journal_release(fs, journal, 0, 0);
 868         *j = NULL;
 869
 870         return 0;
 871 }
 872
 873 void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
 874 {
 875         struct commit_header *h;
 876         __u32 csum;
 877
 878         if (!journal_has_csum_v2or3(j))
 879                 return;
 880
 881         h = (struct commit_header *)(bh->b_data);
 882         h->h_chksum_type = 0;
 883         h->h_chksum_size = 0;
 884         h->h_chksum[0] = 0;
 885         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 886         h->h_chksum[0] = ext2fs_cpu_to_be32(csum);
 887 }
 888
 889 void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh)
 890 {
 891         struct journal_revoke_tail *tail;
 892         __u32 csum;
 893
 894         if (!journal_has_csum_v2or3(j))
 895                 return;
 896
 897         tail = (struct journal_revoke_tail *)(bh->b_data + j->j_blocksize -
 898                         sizeof(struct journal_revoke_tail));
 899         tail->r_checksum = 0;
 900         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 901         tail->r_checksum = ext2fs_cpu_to_be32(csum);
 902 }
 903
 904 void jbd2_descr_block_csum_set(journal_t *j, struct buffer_head *bh)
 905 {
 906         struct journal_block_tail *tail;
 907         __u32 csum;
 908
 909         if (!journal_has_csum_v2or3(j))
 910                 return;
 911
 912         tail = (struct journal_block_tail *)(bh->b_data + j->j_blocksize -
 913                         sizeof(struct journal_block_tail));
 914         tail->t_checksum = 0;
 915         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 916         tail->t_checksum = ext2fs_cpu_to_be32(csum);
 917 }
 918
 919 void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
 920                              struct buffer_head *bh, __u32 sequence)
 921 {
 922         journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
 923         __u32 csum32;
 924         __be32 seq;
 925
 926         if (!journal_has_csum_v2or3(j))
 927                 return;
 928
 929         seq = ext2fs_cpu_to_be32(sequence);
 930         csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
 931         csum32 = jbd2_chksum(j, csum32, bh->b_data, bh->b_size);
 932
 933         if (jfs_has_feature_csum3(j))
 934                 tag3->t_checksum = ext2fs_cpu_to_be32(csum32);
 935         else
 936                 tag->t_checksum = ext2fs_cpu_to_be16(csum32);
 937 }
 938