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 (!jbd2_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 (!jbd2_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 (!jbd2_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 jbd2_journal_bmap(journal_t *journal, blk64_t block,
 103                       unsigned long long *phys)
 104 {
 105 #ifdef USE_INODE_IO
 106         *phys = block;
 107         return 0;
 108 #else
 109         struct inode    *inode = journal->j_inode;
 110         errcode_t       retval;
 111         blk64_t         pblk;
 112
 113         if (!inode) {
 114                 *phys = block;
 115                 return 0;
 116         }
 117
 118         retval = ext2fs_bmap2(inode->i_fs, inode->i_ino,
 119                               &inode->i_ext2, NULL, 0, block, 0, &pblk);
 120         *phys = pblk;
 121         return (int) retval;
 122 #endif
 123 }
 124
 125 struct buffer_head *getblk(kdev_t kdev, blk64_t blocknr, int blocksize)
 126 {
 127         struct buffer_head *bh;
 128         int bufsize = sizeof(*bh) + kdev->k_fs->blocksize -
 129                 sizeof(bh->b_data);
 130         errcode_t retval;
 131
 132         retval = ext2fs_get_memzero(bufsize, &bh);
 133         if (retval)
 134                 return NULL;
 135
 136 #ifdef CONFIG_JBD_DEBUG
 137         if (journal_enable_debug >= 3)
 138                 bh_count++;
 139 #endif
 140         jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
 141                   (unsigned long long) blocknr, blocksize, bh_count);
 142
 143         bh->b_fs = kdev->k_fs;
 144         if (kdev->k_dev == K_DEV_FS)
 145                 bh->b_io = kdev->k_fs->io;
 146         else
 147                 bh->b_io = kdev->k_fs->journal_io;
 148         bh->b_size = blocksize;
 149         bh->b_blocknr = blocknr;
 150
 151         return bh;
 152 }
 153
 154 int sync_blockdev(kdev_t kdev)
 155 {
 156         io_channel      io;
 157
 158         if (kdev->k_dev == K_DEV_FS)
 159                 io = kdev->k_fs->io;
 160         else
 161                 io = kdev->k_fs->journal_io;
 162
 163         return io_channel_flush(io) ? EIO : 0;
 164 }
 165
 166 void ll_rw_block(int rw, int op_flags, int nr, struct buffer_head *bhp[])
 167 {
 168         errcode_t retval;
 169         struct buffer_head *bh;
 170
 171         for (; nr > 0; --nr) {
 172                 bh = *bhp++;
 173                 if (rw == REQ_OP_READ && !bh->b_uptodate) {
 174                         jfs_debug(3, "reading block %llu/%p\n",
 175                                   bh->b_blocknr, (void *) bh);
 176                         retval = io_channel_read_blk64(bh->b_io,
 177                                                      bh->b_blocknr,
 178                                                      1, bh->b_data);
 179                         if (retval) {
 180                                 com_err(bh->b_fs->device_name, retval,
 181                                         "while reading block %llu\n",
 182                                         bh->b_blocknr);
 183                                 bh->b_err = (int) retval;
 184                                 continue;
 185                         }
 186                         bh->b_uptodate = 1;
 187                 } else if (rw == REQ_OP_WRITE && bh->b_dirty) {
 188                         jfs_debug(3, "writing block %llu/%p\n",
 189                                   bh->b_blocknr,
 190                                   (void *) bh);
 191                         retval = io_channel_write_blk64(bh->b_io,
 192                                                       bh->b_blocknr,
 193                                                       1, bh->b_data);
 194                         if (retval) {
 195                                 com_err(bh->b_fs->device_name, retval,
 196                                         "while writing block %llu\n",
 197                                         bh->b_blocknr);
 198                                 bh->b_err = (int) retval;
 199                                 continue;
 200                         }
 201                         bh->b_dirty = 0;
 202                         bh->b_uptodate = 1;
 203                 } else {
 204                         jfs_debug(3, "no-op %s for block %llu\n",
 205                                   rw == REQ_OP_READ ? "read" : "write",
 206                                   bh->b_blocknr);
 207                 }
 208         }
 209 }
 210
 211 void mark_buffer_dirty(struct buffer_head *bh)
 212 {
 213         bh->b_dirty = 1;
 214 }
 215
 216 static void mark_buffer_clean(struct buffer_head *bh)
 217 {
 218         bh->b_dirty = 0;
 219 }
 220
 221 void brelse(struct buffer_head *bh)
 222 {
 223         if (bh->b_dirty)
 224                 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 225         jfs_debug(3, "freeing block %llu/%p (total %d)\n",
 226                   bh->b_blocknr, (void *) bh, --bh_count);
 227         ext2fs_free_mem(&bh);
 228 }
 229
 230 int buffer_uptodate(struct buffer_head *bh)
 231 {
 232         return bh->b_uptodate;
 233 }
 234
 235 void mark_buffer_uptodate(struct buffer_head *bh, int val)
 236 {
 237         bh->b_uptodate = val;
 238 }
 239
 240 void wait_on_buffer(struct buffer_head *bh)
 241 {
 242         if (!bh->b_uptodate)
 243                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
 244 }
 245
 246
 247 static void ext2fs_clear_recover(ext2_filsys fs, int error)
 248 {
 249         ext2fs_clear_feature_journal_needs_recovery(fs->super);
 250
 251         /* if we had an error doing journal recovery, we need a full fsck */
 252         if (error)
 253                 fs->super->s_state &= ~EXT2_VALID_FS;
 254         /*
 255          * If we replayed the journal by definition the file system
 256          * was mounted since the last time it was checked
 257          */
 258         if (fs->super->s_lastcheck >= fs->super->s_mtime)
 259                 fs->super->s_lastcheck = fs->super->s_mtime - 1;
 260         ext2fs_mark_super_dirty(fs);
 261 }
 262
 263 /*
 264  * This is a helper function to check the validity of the journal.
 265  */
 266 struct process_block_struct {
 267         e2_blkcnt_t     last_block;
 268 };
 269
 270 static int process_journal_block(ext2_filsys fs,
 271                                  blk64_t        *block_nr,
 272                                  e2_blkcnt_t blockcnt,
 273                                  blk64_t ref_block EXT2FS_ATTR((unused)),
 274                                  int ref_offset EXT2FS_ATTR((unused)),
 275                                  void *priv_data)
 276 {
 277         struct process_block_struct *p;
 278         blk64_t blk = *block_nr;
 279
 280         p = (struct process_block_struct *) priv_data;
 281
 282         if (!blk || blk < fs->super->s_first_data_block ||
 283             blk >= ext2fs_blocks_count(fs->super))
 284                 return BLOCK_ABORT;
 285
 286         if (blockcnt >= 0)
 287                 p->last_block = blockcnt;
 288         return 0;
 289 }
 290
 291 static errcode_t ext2fs_get_journal(ext2_filsys fs, journal_t **ret_journal)
 292 {
 293         struct process_block_struct pb;
 294         struct ext2_super_block *sb = fs->super;
 295         struct ext2_super_block jsuper;
 296         struct buffer_head      *bh;
 297         struct inode            *j_inode = NULL;
 298         struct kdev_s           *dev_fs = NULL, *dev_journal;
 299         const char              *journal_name = 0;
 300         journal_t               *journal = NULL;
 301         errcode_t               retval = 0;
 302         io_manager              io_ptr = 0;
 303         unsigned long long      start = 0;
 304         int                     ext_journal = 0;
 305         int                     tried_backup_jnl = 0;
 306
 307         retval = ext2fs_get_memzero(sizeof(journal_t), &journal);
 308         if (retval)
 309                 return retval;
 310
 311         retval = ext2fs_get_memzero(2 * sizeof(struct kdev_s), &dev_fs);
 312         if (retval)
 313                 goto errout;
 314         dev_journal = dev_fs+1;
 315
 316         dev_fs->k_fs = dev_journal->k_fs = fs;
 317         dev_fs->k_dev = K_DEV_FS;
 318         dev_journal->k_dev = K_DEV_JOURNAL;
 319
 320         journal->j_dev = dev_journal;
 321         journal->j_fs_dev = dev_fs;
 322         journal->j_inode = NULL;
 323         journal->j_blocksize = fs->blocksize;
 324
 325         if (uuid_is_null(sb->s_journal_uuid)) {
 326                 if (!sb->s_journal_inum) {
 327                         retval = EXT2_ET_BAD_INODE_NUM;
 328                         goto errout;
 329                 }
 330                 retval = ext2fs_get_memzero(sizeof(*j_inode), &j_inode);
 331                 if (retval)
 332                         goto errout;
 333
 334                 j_inode->i_fs = fs;
 335                 j_inode->i_ino = sb->s_journal_inum;
 336
 337                 retval = ext2fs_read_inode(fs, sb->s_journal_inum,
 338                                            &j_inode->i_ext2);
 339                 if (retval) {
 340 try_backup_journal:
 341                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
 342                             tried_backup_jnl)
 343                                 goto errout;
 344                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
 345                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
 346                                EXT2_N_BLOCKS*4);
 347                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
 348                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
 349                         j_inode->i_ext2.i_links_count = 1;
 350                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
 351                         tried_backup_jnl++;
 352                 }
 353                 if (!j_inode->i_ext2.i_links_count ||
 354                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
 355                         retval = EXT2_ET_NO_JOURNAL;
 356                         goto try_backup_journal;
 357                 }
 358                 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
 359                     JBD2_MIN_JOURNAL_BLOCKS) {
 360                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 361                         goto try_backup_journal;
 362                 }
 363                 pb.last_block = -1;
 364                 retval = ext2fs_block_iterate3(fs, j_inode->i_ino,
 365                                                BLOCK_FLAG_HOLE, 0,
 366                                                process_journal_block, &pb);
 367                 if ((pb.last_block + 1) * fs->blocksize <
 368                     (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
 369                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
 370                         goto try_backup_journal;
 371                 }
 372                 if (tried_backup_jnl && (fs->flags & EXT2_FLAG_RW)) {
 373                         retval = ext2fs_write_inode(fs, sb->s_journal_inum,
 374                                                     &j_inode->i_ext2);
 375                         if (retval)
 376                                 goto errout;
 377                 }
 378
 379                 journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
 380                         journal->j_blocksize;
 381
 382 #ifdef USE_INODE_IO
 383                 retval = ext2fs_inode_io_intern2(fs, sb->s_journal_inum,
 384                                                  &j_inode->i_ext2,
 385                                                  &journal_name);
 386                 if (retval)
 387                         goto errout;
 388
 389                 io_ptr = inode_io_manager;
 390 #else
 391                 journal->j_inode = j_inode;
 392                 fs->journal_io = fs->io;
 393                 retval = (errcode_t) jbd2_journal_bmap(journal, 0, &start);
 394                 if (retval)
 395                         goto errout;
 396 #endif
 397         } else {
 398                 ext_journal = 1;
 399                 if (!fs->journal_name) {
 400                         char uuid[37];
 401                         blkid_cache blkid;
 402
 403                         blkid_get_cache(&blkid, NULL);
 404                         uuid_unparse(sb->s_journal_uuid, uuid);
 405                         fs->journal_name = blkid_get_devname(blkid,
 406                                                               "UUID", uuid);
 407                         if (!fs->journal_name)
 408                                 fs->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
 409                         blkid_put_cache(blkid);
 410                 }
 411                 journal_name = fs->journal_name;
 412
 413                 if (!journal_name) {
 414                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 415                         goto errout;
 416                 }
 417
 418                 jfs_debug(1, "Using journal file %s\n", journal_name);
 419                 io_ptr = unix_io_manager;
 420         }
 421
 422 #if 0
 423         test_io_backing_manager = io_ptr;
 424         io_ptr = test_io_manager;
 425 #endif
 426 #ifndef USE_INODE_IO
 427         if (ext_journal)
 428 #endif
 429         {
 430                 retval = io_ptr->open(journal_name, fs->flags & EXT2_FLAG_RW,
 431                                       &fs->journal_io);
 432         }
 433         if (retval)
 434                 goto errout;
 435
 436         io_channel_set_blksize(fs->journal_io, fs->blocksize);
 437
 438         if (ext_journal) {
 439                 blk64_t maxlen;
 440
 441                 start = ext2fs_journal_sb_start(fs->blocksize) - 1;
 442                 bh = getblk(dev_journal, start, fs->blocksize);
 443                 if (!bh) {
 444                         retval = EXT2_ET_NO_MEMORY;
 445                         goto errout;
 446                 }
 447                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
 448                 retval = bh->b_err;
 449                 if (retval) {
 450                         brelse(bh);
 451                         goto errout;
 452                 }
 453                 memcpy(&jsuper, start ? bh->b_data :
 454                                 bh->b_data + SUPERBLOCK_OFFSET,
 455                        sizeof(jsuper));
 456 #ifdef WORDS_BIGENDIAN
 457                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
 458                         ext2fs_swap_super(&jsuper);
 459 #endif
 460                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
 461                     !ext2fs_has_feature_journal_dev(&jsuper)) {
 462                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 463                         brelse(bh);
 464                         goto errout;
 465                 }
 466                 /* Make sure the journal UUID is correct */
 467                 if (memcmp(jsuper.s_uuid, fs->super->s_journal_uuid,
 468                            sizeof(jsuper.s_uuid))) {
 469                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
 470                         brelse(bh);
 471                         goto errout;
 472                 }
 473
 474                 /* Check the superblock checksum */
 475                 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
 476                         struct struct_ext2_filsys fsx;
 477                         struct ext2_super_block superx;
 478                         void *p;
 479
 480                         p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
 481                         memcpy(&fsx, fs, sizeof(fsx));
 482                         memcpy(&superx, fs->super, sizeof(superx));
 483                         fsx.super = &superx;
 484                         ext2fs_set_feature_metadata_csum(fsx.super);
 485                         if (!ext2fs_superblock_csum_verify(&fsx, p)) {
 486                                 retval = EXT2_ET_LOAD_EXT_JOURNAL;
 487                                 brelse(bh);
 488                                 goto errout;
 489                         }
 490                 }
 491                 brelse(bh);
 492
 493                 maxlen = ext2fs_blocks_count(&jsuper);
 494                 journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen :
 495                                     (1ULL << 32) - 1;
 496                 start++;
 497         }
 498
 499         bh = getblk(dev_journal, start, journal->j_blocksize);
 500         if (!bh) {
 501                 retval = EXT2_ET_NO_MEMORY;
 502                 goto errout;
 503         }
 504
 505         journal->j_sb_buffer = bh;
 506         journal->j_superblock = (journal_superblock_t *)bh->b_data;
 507
 508 #ifdef USE_INODE_IO
 509         if (j_inode)
 510                 ext2fs_free_mem(&j_inode);
 511 #endif
 512
 513         *ret_journal = journal;
 514         return 0;
 515
 516 errout:
 517         if (dev_fs)
 518                 ext2fs_free_mem(&dev_fs);
 519         if (j_inode)
 520                 ext2fs_free_mem(&j_inode);
 521         if (journal)
 522                 ext2fs_free_mem(&journal);
 523         return retval;
 524 }
 525
 526 static errcode_t ext2fs_journal_fix_bad_inode(ext2_filsys fs)
 527 {
 528         struct ext2_super_block *sb = fs->super;
 529         int recover = ext2fs_has_feature_journal_needs_recovery(fs->super);
 530         int has_journal = ext2fs_has_feature_journal(fs->super);
 531
 532         if (has_journal || sb->s_journal_inum) {
 533                 /* The journal inode is bogus, remove and force full fsck */
 534                 return EXT2_ET_BAD_INODE_NUM;
 535         } else if (recover) {
 536                 return EXT2_ET_UNSUPP_FEATURE;
 537         }
 538         return 0;
 539 }
 540
 541 #define V1_SB_SIZE      0x0024
 542 static void clear_v2_journal_fields(journal_t *journal)
 543 {
 544         ext2_filsys fs = journal->j_dev->k_fs;
 545
 546         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
 547                fs->blocksize-V1_SB_SIZE);
 548         mark_buffer_dirty(journal->j_sb_buffer);
 549 }
 550
 551
 552 static errcode_t ext2fs_journal_load(journal_t *journal)
 553 {
 554         ext2_filsys fs = journal->j_dev->k_fs;
 555         journal_superblock_t *jsb;
 556         struct buffer_head *jbh = journal->j_sb_buffer;
 557
 558         ll_rw_block(REQ_OP_READ, 0, 1, &jbh);
 559         if (jbh->b_err)
 560                 return jbh->b_err;
 561
 562         jsb = journal->j_superblock;
 563         /* If we don't even have JBD2_MAGIC, we probably have a wrong inode */
 564         if (jsb->s_header.h_magic != htonl(JBD2_MAGIC_NUMBER))
 565                 return ext2fs_journal_fix_bad_inode(fs);
 566
 567         switch (ntohl(jsb->s_header.h_blocktype)) {
 568         case JBD2_SUPERBLOCK_V1:
 569                 journal->j_format_version = 1;
 570                 if (jsb->s_feature_compat ||
 571                     jsb->s_feature_incompat ||
 572                     jsb->s_feature_ro_compat ||
 573                     jsb->s_nr_users)
 574                         clear_v2_journal_fields(journal);
 575                 break;
 576
 577         case JBD2_SUPERBLOCK_V2:
 578                 journal->j_format_version = 2;
 579                 if (ntohl(jsb->s_nr_users) > 1 &&
 580                     uuid_is_null(fs->super->s_journal_uuid))
 581                         clear_v2_journal_fields(journal);
 582                 if (ntohl(jsb->s_nr_users) > 1)
 583                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 584                 break;
 585
 586         /*
 587          * These should never appear in a journal super block, so if
 588          * they do, the journal is badly corrupted.
 589          */
 590         case JBD2_DESCRIPTOR_BLOCK:
 591         case JBD2_COMMIT_BLOCK:
 592         case JBD2_REVOKE_BLOCK:
 593                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 594
 595         /* If we don't understand the superblock major type, but there
 596          * is a magic number, then it is likely to be a new format we
 597          * just don't understand, so leave it alone. */
 598         default:
 599                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
 600         }
 601
 602         if (JBD2_HAS_INCOMPAT_FEATURE(journal, ~JBD2_KNOWN_INCOMPAT_FEATURES))
 603                 return EXT2_ET_UNSUPP_FEATURE;
 604
 605         if (JBD2_HAS_RO_COMPAT_FEATURE(journal, ~JBD2_KNOWN_ROCOMPAT_FEATURES))
 606                 return EXT2_ET_RO_UNSUPP_FEATURE;
 607
 608         /* Checksum v1-3 are mutually exclusive features. */
 609         if (jbd2_has_feature_csum2(journal) && jbd2_has_feature_csum3(journal))
 610                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 611
 612         if (jbd2_journal_has_csum_v2or3(journal) &&
 613             jbd2_has_feature_checksum(journal))
 614                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 615
 616         if (!ext2fs_journal_verify_csum_type(journal, jsb) ||
 617             !ext2fs_journal_sb_csum_verify(journal, jsb))
 618                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 619
 620         if (jbd2_journal_has_csum_v2or3(journal))
 621                 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
 622                                                    sizeof(jsb->s_uuid));
 623
 624         /* We have now checked whether we know enough about the journal
 625          * format to be able to proceed safely, so any other checks that
 626          * fail we should attempt to recover from. */
 627         if (jsb->s_blocksize != htonl(journal->j_blocksize))
 628                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 629
 630         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
 631                 journal->j_maxlen = ntohl(jsb->s_maxlen);
 632         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen)
 633                 return EXT2_ET_CORRUPT_JOURNAL_SB;
 634
 635         journal->j_tail_sequence = ntohl(jsb->s_sequence);
 636         journal->j_transaction_sequence = journal->j_tail_sequence;
 637         journal->j_tail = ntohl(jsb->s_start);
 638         journal->j_first = ntohl(jsb->s_first);
 639         journal->j_last = ntohl(jsb->s_maxlen);
 640
 641         return 0;
 642 }
 643
 644 static void ext2fs_journal_release(ext2_filsys fs, journal_t *journal,
 645                                    int reset, int drop)
 646 {
 647         journal_superblock_t *jsb;
 648
 649         if (drop)
 650                 mark_buffer_clean(journal->j_sb_buffer);
 651         else if (fs->flags & EXT2_FLAG_RW) {
 652                 jsb = journal->j_superblock;
 653                 jsb->s_sequence = htonl(journal->j_tail_sequence);
 654                 if (reset)
 655                         jsb->s_start = 0; /* this marks the journal as empty */
 656                 ext2fs_journal_sb_csum_set(journal, jsb);
 657                 mark_buffer_dirty(journal->j_sb_buffer);
 658         }
 659         brelse(journal->j_sb_buffer);
 660
 661         if (fs && fs->journal_io) {
 662                 if (fs->io != fs->journal_io)
 663                         io_channel_close(fs->journal_io);
 664                 fs->journal_io = NULL;
 665                 free(fs->journal_name);
 666                 fs->journal_name = NULL;
 667         }
 668
 669 #ifndef USE_INODE_IO
 670         if (journal->j_inode)
 671                 ext2fs_free_mem(&journal->j_inode);
 672 #endif
 673         if (journal->j_fs_dev)
 674                 ext2fs_free_mem(&journal->j_fs_dev);
 675         ext2fs_free_mem(&journal);
 676 }
 677
 678 /*
 679  * This function makes sure that the superblock fields regarding the
 680  * journal are consistent.
 681  */
 682 static errcode_t ext2fs_check_ext3_journal(ext2_filsys fs)
 683 {
 684         struct ext2_super_block *sb = fs->super;
 685         journal_t *journal;
 686         int recover = ext2fs_has_feature_journal_needs_recovery(fs->super);
 687         errcode_t retval;
 688
 689         /* If we don't have any journal features, don't do anything more */
 690         if (!ext2fs_has_feature_journal(sb) &&
 691             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
 692             uuid_is_null(sb->s_journal_uuid))
 693                 return 0;
 694
 695         retval = ext2fs_get_journal(fs, &journal);
 696         if (retval)
 697                 return retval;
 698
 699         retval = ext2fs_journal_load(journal);
 700         if (retval)
 701                 goto err;
 702
 703         /*
 704          * We want to make the flags consistent here.  We will not leave with
 705          * needs_recovery set but has_journal clear.  We can't get in a loop
 706          * with -y, -n, or -p, only if a user isn't making up their mind.
 707          */
 708         if (!ext2fs_has_feature_journal(sb)) {
 709                 retval = EXT2_ET_JOURNAL_FLAGS_WRONG;
 710                 goto err;
 711         }
 712
 713         if (ext2fs_has_feature_journal(sb) &&
 714             !ext2fs_has_feature_journal_needs_recovery(sb) &&
 715             journal->j_superblock->s_start != 0) {
 716                 retval = EXT2_ET_JOURNAL_FLAGS_WRONG;
 717                 goto err;
 718         }
 719
 720         /*
 721          * If we don't need to do replay the journal, check to see if
 722          * the journal's errno is set; if so, we need to mark the file
 723          * system as being corrupt and clear the journal's s_errno.
 724          */
 725         if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
 726             journal->j_superblock->s_errno) {
 727                 fs->super->s_state |= EXT2_ERROR_FS;
 728                 ext2fs_mark_super_dirty(fs);
 729                 journal->j_superblock->s_errno = 0;
 730                 ext2fs_journal_sb_csum_set(journal, journal->j_superblock);
 731                 mark_buffer_dirty(journal->j_sb_buffer);
 732         }
 733
 734 err:
 735         ext2fs_journal_release(fs, journal, 0, retval ? 1 : 0);
 736         return retval;
 737 }
 738
 739 static errcode_t recover_ext3_journal(ext2_filsys fs)
 740 {
 741         journal_t *journal;
 742         errcode_t retval;
 743
 744         retval = jbd2_journal_init_revoke_record_cache();
 745         if (retval)
 746                 return retval;
 747
 748         retval = jbd2_journal_init_revoke_table_cache();
 749         if (retval)
 750                 return retval;
 751
 752         retval = ext2fs_get_journal(fs, &journal);
 753         if (retval)
 754                 return retval;
 755
 756         retval = ext2fs_journal_load(journal);
 757         if (retval)
 758                 goto errout;
 759
 760         retval = jbd2_journal_init_revoke(journal, 1024);
 761         if (retval)
 762                 goto errout;
 763
 764         retval = -jbd2_journal_recover(journal);
 765         if (retval)
 766                 goto errout;
 767
 768         if (journal->j_failed_commit) {
 769                 journal->j_superblock->s_errno = -EINVAL;
 770                 mark_buffer_dirty(journal->j_sb_buffer);
 771         }
 772
 773 errout:
 774         jbd2_journal_destroy_revoke(journal);
 775         jbd2_journal_destroy_revoke_record_cache();
 776         jbd2_journal_destroy_revoke_table_cache();
 777         ext2fs_journal_release(fs, journal, 1, 0);
 778         return retval;
 779 }
 780
 781 errcode_t ext2fs_run_ext3_journal(ext2_filsys *fsp)
 782 {
 783         ext2_filsys fs = *fsp;
 784         io_manager io_ptr = fs->io->manager;
 785         errcode_t       retval, recover_retval;
 786         io_stats        stats = 0;
 787         unsigned long long kbytes_written = 0;
 788         char *fsname;
 789         int fsflags;
 790         int fsblocksize;
 791
 792         if (!(fs->flags & EXT2_FLAG_RW))
 793                 return EXT2_ET_FILE_RO;
 794
 795         if (fs->flags & EXT2_FLAG_DIRTY)
 796                 ext2fs_flush(fs);       /* Force out any modifications */
 797
 798         recover_retval = recover_ext3_journal(fs);
 799
 800         /*
 801          * Reload the filesystem context to get up-to-date data from disk
 802          * because journal recovery will change the filesystem under us.
 803          */
 804         if (fs->super->s_kbytes_written &&
 805             fs->io->manager->get_stats)
 806                 fs->io->manager->get_stats(fs->io, &stats);
 807         if (stats && stats->bytes_written)
 808                 kbytes_written = stats->bytes_written >> 10;
 809
 810         ext2fs_mmp_stop(fs);
 811         fsname = fs->device_name;
 812         fs->device_name = NULL;
 813         fsflags = fs->flags;
 814         fsblocksize = fs->blocksize;
 815         ext2fs_free(fs);
 816         *fsp = NULL;
 817         retval = ext2fs_open(fsname, fsflags, 0, fsblocksize, io_ptr, fsp);
 818         ext2fs_free_mem(&fsname);
 819         if (retval)
 820                 return retval;
 821
 822         fs = *fsp;
 823         fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
 824         fs->super->s_kbytes_written += kbytes_written;
 825
 826         /* Set the superblock flags */
 827         ext2fs_clear_recover(fs, recover_retval != 0);
 828
 829         /*
 830          * Do one last sanity check, and propagate journal->s_errno to
 831          * the EXT2_ERROR_FS flag in the fs superblock if needed.
 832          */
 833         retval = ext2fs_check_ext3_journal(fs);
 834         return retval ? retval : recover_retval;
 835 }
 836
 837 errcode_t ext2fs_open_journal(ext2_filsys fs, journal_t **j)
 838 {
 839         journal_t *journal;
 840         errcode_t retval;
 841
 842         retval = jbd2_journal_init_revoke_record_cache();
 843         if (retval)
 844                 return retval;
 845
 846         retval = jbd2_journal_init_revoke_table_cache();
 847         if (retval)
 848                 return retval;
 849
 850         retval = ext2fs_get_journal(fs, &journal);
 851         if (retval)
 852                 return retval;
 853
 854         retval = ext2fs_journal_load(journal);
 855         if (retval)
 856                 goto errout;
 857
 858         retval = jbd2_journal_init_revoke(journal, 1024);
 859         if (retval)
 860                 goto errout;
 861
 862         if (journal->j_failed_commit) {
 863                 journal->j_superblock->s_errno = -EINVAL;
 864                 mark_buffer_dirty(journal->j_sb_buffer);
 865         }
 866
 867         *j = journal;
 868         return 0;
 869
 870 errout:
 871         jbd2_journal_destroy_revoke(journal);
 872         jbd2_journal_destroy_revoke_record_cache();
 873         jbd2_journal_destroy_revoke_table_cache();
 874         ext2fs_journal_release(fs, journal, 1, 0);
 875         return retval;
 876 }
 877
 878 errcode_t ext2fs_close_journal(ext2_filsys fs, journal_t **j)
 879 {
 880         journal_t *journal = *j;
 881
 882         jbd2_journal_destroy_revoke(journal);
 883         jbd2_journal_destroy_revoke_record_cache();
 884         jbd2_journal_destroy_revoke_table_cache();
 885         ext2fs_journal_release(fs, journal, 0, 0);
 886         *j = NULL;
 887
 888         return 0;
 889 }
 890
 891 void jbd2_commit_block_csum_set(journal_t *j, struct buffer_head *bh)
 892 {
 893         struct commit_header *h;
 894         __u32 csum;
 895
 896         if (!jbd2_journal_has_csum_v2or3(j))
 897                 return;
 898
 899         h = (struct commit_header *)(bh->b_data);
 900         h->h_chksum_type = 0;
 901         h->h_chksum_size = 0;
 902         h->h_chksum[0] = 0;
 903         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 904         h->h_chksum[0] = ext2fs_cpu_to_be32(csum);
 905 }
 906
 907 void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh)
 908 {
 909         jbd2_descr_block_csum_set(j, bh);
 910 }
 911
 912 void jbd2_descr_block_csum_set(journal_t *j, struct buffer_head *bh)
 913 {
 914         struct jbd2_journal_block_tail *tail;
 915         __u32 csum;
 916
 917         if (!jbd2_journal_has_csum_v2or3(j))
 918                 return;
 919
 920         tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
 921                         sizeof(struct jbd2_journal_block_tail));
 922         tail->t_checksum = 0;
 923         csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 924         tail->t_checksum = ext2fs_cpu_to_be32(csum);
 925 }
 926
 927 void jbd2_block_tag_csum_set(journal_t *j, journal_block_tag_t *tag,
 928                              struct buffer_head *bh, __u32 sequence)
 929 {
 930         journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
 931         __u32 csum32;
 932         __be32 seq;
 933
 934         if (!jbd2_journal_has_csum_v2or3(j))
 935                 return;
 936
 937         seq = ext2fs_cpu_to_be32(sequence);
 938         csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
 939         csum32 = jbd2_chksum(j, csum32, bh->b_data, bh->b_size);
 940
 941         if (jbd2_has_feature_csum3(j))
 942                 tag3->t_checksum = ext2fs_cpu_to_be32(csum32);
 943         else
 944                 tag->t_checksum = ext2fs_cpu_to_be16(csum32);
 945 }
 946