journal.c revision 2eae093081a097a0a611fe2817e1533825dfdd86
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 */ 30static 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 */ 46int 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 68struct 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 90void 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 102void 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 146void mark_buffer_dirty(struct buffer_head *bh) 147{ 148 bh->b_dirty = 1; 149} 150 151static void mark_buffer_clean(struct buffer_head * bh) 152{ 153 bh->b_dirty = 0; 154} 155 156void 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 165int buffer_uptodate(struct buffer_head *bh) 166{ 167 return bh->b_uptodate; 168} 169 170void mark_buffer_uptodate(struct buffer_head *bh, int val) 171{ 172 bh->b_uptodate = val; 173} 174 175void wait_on_buffer(struct buffer_head *bh) 176{ 177 if (!bh->b_uptodate) 178 ll_rw_block(READ, 1, &bh); 179} 180 181 182static 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/* 193 * This is a helper function to check the validity of the journal. 194 */ 195struct process_block_struct { 196 e2_blkcnt_t last_block; 197}; 198 199static int process_journal_block(ext2_filsys fs, 200 blk_t *block_nr, 201 e2_blkcnt_t blockcnt, 202 blk_t ref_block EXT2FS_ATTR((unused)), 203 int ref_offset EXT2FS_ATTR((unused)), 204 void *priv_data) 205{ 206 struct process_block_struct *p; 207 blk_t blk = *block_nr; 208 209 p = (struct process_block_struct *) priv_data; 210 211 if (blk < fs->super->s_first_data_block || 212 blk >= fs->super->s_blocks_count) 213 return BLOCK_ABORT; 214 215 if (blockcnt >= 0) 216 p->last_block = blockcnt; 217 return 0; 218} 219 220static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal) 221{ 222 struct process_block_struct pb; 223 struct ext2_super_block *sb = ctx->fs->super; 224 struct ext2_super_block jsuper; 225 struct problem_context pctx; 226 struct buffer_head *bh; 227 struct inode *j_inode = NULL; 228 struct kdev_s *dev_fs = NULL, *dev_journal; 229 const char *journal_name = 0; 230 journal_t *journal = NULL; 231 errcode_t retval = 0; 232 io_manager io_ptr = 0; 233 unsigned long start = 0; 234 int ext_journal = 0; 235 int tried_backup_jnl = 0; 236 237 clear_problem_context(&pctx); 238 239 journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal"); 240 if (!journal) { 241 return EXT2_ET_NO_MEMORY; 242 } 243 244 dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev"); 245 if (!dev_fs) { 246 retval = EXT2_ET_NO_MEMORY; 247 goto errout; 248 } 249 dev_journal = dev_fs+1; 250 251 dev_fs->k_ctx = dev_journal->k_ctx = ctx; 252 dev_fs->k_dev = K_DEV_FS; 253 dev_journal->k_dev = K_DEV_JOURNAL; 254 255 journal->j_dev = dev_journal; 256 journal->j_fs_dev = dev_fs; 257 journal->j_inode = NULL; 258 journal->j_blocksize = ctx->fs->blocksize; 259 260 if (uuid_is_null(sb->s_journal_uuid)) { 261 if (!sb->s_journal_inum) { 262 retval = EXT2_ET_BAD_INODE_NUM; 263 goto errout; 264 } 265 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode), 266 "journal inode"); 267 if (!j_inode) { 268 retval = EXT2_ET_NO_MEMORY; 269 goto errout; 270 } 271 272 j_inode->i_ctx = ctx; 273 j_inode->i_ino = sb->s_journal_inum; 274 275 if ((retval = ext2fs_read_inode(ctx->fs, 276 sb->s_journal_inum, 277 &j_inode->i_ext2))) { 278 try_backup_journal: 279 if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS || 280 tried_backup_jnl) 281 goto errout; 282 memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode)); 283 memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks, 284 EXT2_N_BLOCKS*4); 285 j_inode->i_ext2.i_size = sb->s_jnl_blocks[16]; 286 j_inode->i_ext2.i_links_count = 1; 287 j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600; 288 e2fsck_use_inode_shortcuts(ctx, 1); 289 ctx->stashed_ino = j_inode->i_ino; 290 ctx->stashed_inode = &j_inode->i_ext2; 291 tried_backup_jnl++; 292 } 293 if (!j_inode->i_ext2.i_links_count || 294 !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) { 295 retval = EXT2_ET_NO_JOURNAL; 296 goto try_backup_journal; 297 } 298 if (j_inode->i_ext2.i_size / journal->j_blocksize < 299 JFS_MIN_JOURNAL_BLOCKS) { 300 retval = EXT2_ET_JOURNAL_TOO_SMALL; 301 goto try_backup_journal; 302 } 303 pb.last_block = -1; 304 retval = ext2fs_block_iterate2(ctx->fs, j_inode->i_ino, 305 BLOCK_FLAG_HOLE, 0, 306 process_journal_block, &pb); 307 if ((pb.last_block+1) * ctx->fs->blocksize < 308 j_inode->i_ext2.i_size) { 309 retval = EXT2_ET_JOURNAL_TOO_SMALL; 310 goto try_backup_journal; 311 } 312 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) { 313 retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum, 314 &j_inode->i_ext2); 315 if (retval) 316 goto errout; 317 } 318 319 journal->j_maxlen = j_inode->i_ext2.i_size / journal->j_blocksize; 320 321#ifdef USE_INODE_IO 322 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum, 323 &j_inode->i_ext2, 324 &journal_name); 325 if (retval) 326 goto errout; 327 328 io_ptr = inode_io_manager; 329#else 330 journal->j_inode = j_inode; 331 ctx->journal_io = ctx->fs->io; 332 if ((retval = journal_bmap(journal, 0, &start)) != 0) 333 goto errout; 334#endif 335 } else { 336 ext_journal = 1; 337 if (!ctx->journal_name) { 338 char uuid[37]; 339 340 uuid_unparse(sb->s_journal_uuid, uuid); 341 ctx->journal_name = blkid_get_devname(ctx->blkid, 342 "UUID", uuid); 343 if (!ctx->journal_name) 344 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev); 345 } 346 journal_name = ctx->journal_name; 347 348 if (!journal_name) { 349 fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx); 350 retval = EXT2_ET_LOAD_EXT_JOURNAL; 351 goto errout; 352 } 353 354 jfs_debug(1, "Using journal file %s\n", journal_name); 355 io_ptr = unix_io_manager; 356 } 357 358#if 0 359 test_io_backing_manager = io_ptr; 360 io_ptr = test_io_manager; 361#endif 362#ifndef USE_INODE_IO 363 if (ext_journal) 364#endif 365 retval = io_ptr->open(journal_name, IO_FLAG_RW, 366 &ctx->journal_io); 367 if (retval) 368 goto errout; 369 370 io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize); 371 372 if (ext_journal) { 373 if (ctx->fs->blocksize == 1024) 374 start = 1; 375 bh = getblk(dev_journal, start, ctx->fs->blocksize); 376 if (!bh) { 377 retval = EXT2_ET_NO_MEMORY; 378 goto errout; 379 } 380 ll_rw_block(READ, 1, &bh); 381 if ((retval = bh->b_err) != 0) { 382 brelse(bh); 383 goto errout; 384 } 385 memcpy(&jsuper, start ? bh->b_data : bh->b_data + 1024, 386 sizeof(jsuper)); 387 brelse(bh); 388#ifdef WORDS_BIGENDIAN 389 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC)) 390 ext2fs_swap_super(&jsuper); 391#endif 392 if (jsuper.s_magic != EXT2_SUPER_MAGIC || 393 !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) { 394 fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx); 395 retval = EXT2_ET_LOAD_EXT_JOURNAL; 396 goto errout; 397 } 398 /* Make sure the journal UUID is correct */ 399 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid, 400 sizeof(jsuper.s_uuid))) { 401 fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx); 402 retval = EXT2_ET_LOAD_EXT_JOURNAL; 403 goto errout; 404 } 405 406 journal->j_maxlen = jsuper.s_blocks_count; 407 start++; 408 } 409 410 if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) { 411 retval = EXT2_ET_NO_MEMORY; 412 goto errout; 413 } 414 415 journal->j_sb_buffer = bh; 416 journal->j_superblock = (journal_superblock_t *)bh->b_data; 417 418#ifdef USE_INODE_IO 419 if (j_inode) 420 ext2fs_free_mem(&j_inode); 421#endif 422 423 *ret_journal = journal; 424 e2fsck_use_inode_shortcuts(ctx, 0); 425 return 0; 426 427errout: 428 e2fsck_use_inode_shortcuts(ctx, 0); 429 if (dev_fs) 430 ext2fs_free_mem(&dev_fs); 431 if (j_inode) 432 ext2fs_free_mem(&j_inode); 433 if (journal) 434 ext2fs_free_mem(&journal); 435 return retval; 436} 437 438static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx, 439 struct problem_context *pctx) 440{ 441 struct ext2_super_block *sb = ctx->fs->super; 442 int recover = ctx->fs->super->s_feature_incompat & 443 EXT3_FEATURE_INCOMPAT_RECOVER; 444 int has_journal = ctx->fs->super->s_feature_compat & 445 EXT3_FEATURE_COMPAT_HAS_JOURNAL; 446 447 if (has_journal || sb->s_journal_inum) { 448 /* The journal inode is bogus, remove and force full fsck */ 449 pctx->ino = sb->s_journal_inum; 450 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) { 451 if (has_journal && sb->s_journal_inum) 452 printf("*** ext3 journal has been deleted - " 453 "filesystem is now ext2 only ***\n\n"); 454 sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL; 455 sb->s_journal_inum = 0; 456 ctx->flags |= E2F_FLAG_JOURNAL_INODE; 457 e2fsck_clear_recover(ctx, 1); 458 return 0; 459 } 460 return EXT2_ET_BAD_INODE_NUM; 461 } else if (recover) { 462 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) { 463 e2fsck_clear_recover(ctx, 1); 464 return 0; 465 } 466 return EXT2_ET_UNSUPP_FEATURE; 467 } 468 return 0; 469} 470 471#define V1_SB_SIZE 0x0024 472static void clear_v2_journal_fields(journal_t *journal) 473{ 474 e2fsck_t ctx = journal->j_dev->k_ctx; 475 struct problem_context pctx; 476 477 clear_problem_context(&pctx); 478 479 if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx)) 480 return; 481 482 memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0, 483 ctx->fs->blocksize-V1_SB_SIZE); 484 mark_buffer_dirty(journal->j_sb_buffer); 485} 486 487 488static errcode_t e2fsck_journal_load(journal_t *journal) 489{ 490 e2fsck_t ctx = journal->j_dev->k_ctx; 491 journal_superblock_t *jsb; 492 struct buffer_head *jbh = journal->j_sb_buffer; 493 struct problem_context pctx; 494 495 clear_problem_context(&pctx); 496 497 ll_rw_block(READ, 1, &jbh); 498 if (jbh->b_err) { 499 com_err(ctx->device_name, jbh->b_err, 500 _("reading journal superblock\n")); 501 return jbh->b_err; 502 } 503 504 jsb = journal->j_superblock; 505 /* If we don't even have JFS_MAGIC, we probably have a wrong inode */ 506 if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER)) 507 return e2fsck_journal_fix_bad_inode(ctx, &pctx); 508 509 switch (ntohl(jsb->s_header.h_blocktype)) { 510 case JFS_SUPERBLOCK_V1: 511 journal->j_format_version = 1; 512 if (jsb->s_feature_compat || 513 jsb->s_feature_incompat || 514 jsb->s_feature_ro_compat || 515 jsb->s_nr_users) 516 clear_v2_journal_fields(journal); 517 break; 518 519 case JFS_SUPERBLOCK_V2: 520 journal->j_format_version = 2; 521 if (ntohl(jsb->s_nr_users) > 1 && 522 uuid_is_null(ctx->fs->super->s_journal_uuid)) 523 clear_v2_journal_fields(journal); 524 if (ntohl(jsb->s_nr_users) > 1) { 525 fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx); 526 return EXT2_ET_JOURNAL_UNSUPP_VERSION; 527 } 528 break; 529 530 /* 531 * These should never appear in a journal super block, so if 532 * they do, the journal is badly corrupted. 533 */ 534 case JFS_DESCRIPTOR_BLOCK: 535 case JFS_COMMIT_BLOCK: 536 case JFS_REVOKE_BLOCK: 537 return EXT2_ET_CORRUPT_SUPERBLOCK; 538 539 /* If we don't understand the superblock major type, but there 540 * is a magic number, then it is likely to be a new format we 541 * just don't understand, so leave it alone. */ 542 default: 543 return EXT2_ET_JOURNAL_UNSUPP_VERSION; 544 } 545 546 if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES)) 547 return EXT2_ET_UNSUPP_FEATURE; 548 549 if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES)) 550 return EXT2_ET_RO_UNSUPP_FEATURE; 551 552 /* We have now checked whether we know enough about the journal 553 * format to be able to proceed safely, so any other checks that 554 * fail we should attempt to recover from. */ 555 if (jsb->s_blocksize != htonl(journal->j_blocksize)) { 556 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK, 557 _("%s: no valid journal superblock found\n"), 558 ctx->device_name); 559 return EXT2_ET_CORRUPT_SUPERBLOCK; 560 } 561 562 if (ntohl(jsb->s_maxlen) < journal->j_maxlen) 563 journal->j_maxlen = ntohl(jsb->s_maxlen); 564 else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) { 565 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK, 566 _("%s: journal too short\n"), 567 ctx->device_name); 568 return EXT2_ET_CORRUPT_SUPERBLOCK; 569 } 570 571 journal->j_tail_sequence = ntohl(jsb->s_sequence); 572 journal->j_transaction_sequence = journal->j_tail_sequence; 573 journal->j_tail = ntohl(jsb->s_start); 574 journal->j_first = ntohl(jsb->s_first); 575 journal->j_last = ntohl(jsb->s_maxlen); 576 577 return 0; 578} 579 580static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb, 581 journal_t *journal) 582{ 583 char *p; 584 union { 585 uuid_t uuid; 586 __u32 val[4]; 587 } u; 588 __u32 new_seq = 0; 589 int i; 590 591 /* Leave a valid existing V1 superblock signature alone. 592 * Anything unrecognisable we overwrite with a new V2 593 * signature. */ 594 595 if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) || 596 jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) { 597 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER); 598 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2); 599 } 600 601 /* Zero out everything else beyond the superblock header */ 602 603 p = ((char *) jsb) + sizeof(journal_header_t); 604 memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t)); 605 606 jsb->s_blocksize = htonl(ctx->fs->blocksize); 607 jsb->s_maxlen = htonl(journal->j_maxlen); 608 jsb->s_first = htonl(1); 609 610 /* Initialize the journal sequence number so that there is "no" 611 * chance we will find old "valid" transactions in the journal. 612 * This avoids the need to zero the whole journal (slow to do, 613 * and risky when we are just recovering the filesystem). 614 */ 615 uuid_generate(u.uuid); 616 for (i = 0; i < 4; i ++) 617 new_seq ^= u.val[i]; 618 jsb->s_sequence = htonl(new_seq); 619 620 mark_buffer_dirty(journal->j_sb_buffer); 621 ll_rw_block(WRITE, 1, &journal->j_sb_buffer); 622} 623 624static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx, 625 journal_t *journal, 626 struct problem_context *pctx) 627{ 628 struct ext2_super_block *sb = ctx->fs->super; 629 int recover = ctx->fs->super->s_feature_incompat & 630 EXT3_FEATURE_INCOMPAT_RECOVER; 631 632 if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) { 633 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) { 634 e2fsck_journal_reset_super(ctx, journal->j_superblock, 635 journal); 636 journal->j_transaction_sequence = 1; 637 e2fsck_clear_recover(ctx, recover); 638 return 0; 639 } 640 return EXT2_ET_CORRUPT_SUPERBLOCK; 641 } else if (e2fsck_journal_fix_bad_inode(ctx, pctx)) 642 return EXT2_ET_CORRUPT_SUPERBLOCK; 643 644 return 0; 645} 646 647static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal, 648 int reset, int drop) 649{ 650 journal_superblock_t *jsb; 651 652 if (drop) 653 mark_buffer_clean(journal->j_sb_buffer); 654 else if (!(ctx->options & E2F_OPT_READONLY)) { 655 jsb = journal->j_superblock; 656 jsb->s_sequence = htonl(journal->j_transaction_sequence); 657 if (reset) 658 jsb->s_start = 0; /* this marks the journal as empty */ 659 mark_buffer_dirty(journal->j_sb_buffer); 660 } 661 brelse(journal->j_sb_buffer); 662 663 if (ctx->journal_io) { 664 if (ctx->fs && ctx->fs->io != ctx->journal_io) 665 io_channel_close(ctx->journal_io); 666 ctx->journal_io = 0; 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 */ 682int e2fsck_check_ext3_journal(e2fsck_t ctx) 683{ 684 struct ext2_super_block *sb = ctx->fs->super; 685 journal_t *journal; 686 int recover = ctx->fs->super->s_feature_incompat & 687 EXT3_FEATURE_INCOMPAT_RECOVER; 688 struct problem_context pctx; 689 problem_t problem; 690 int reset = 0, force_fsck = 0; 691 int retval; 692 693 /* If we don't have any journal features, don't do anything more */ 694 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) && 695 !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 && 696 uuid_is_null(sb->s_journal_uuid)) 697 return 0; 698 699 clear_problem_context(&pctx); 700 pctx.num = sb->s_journal_inum; 701 702 retval = e2fsck_get_journal(ctx, &journal); 703 if (retval) { 704 if ((retval == EXT2_ET_BAD_INODE_NUM) || 705 (retval == EXT2_ET_BAD_BLOCK_NUM) || 706 (retval == EXT2_ET_JOURNAL_TOO_SMALL) || 707 (retval == EXT2_ET_NO_JOURNAL)) 708 return e2fsck_journal_fix_bad_inode(ctx, &pctx); 709 return retval; 710 } 711 712 retval = e2fsck_journal_load(journal); 713 if (retval) { 714 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) || 715 ((retval == EXT2_ET_UNSUPP_FEATURE) && 716 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT, 717 &pctx))) || 718 ((retval == EXT2_ET_RO_UNSUPP_FEATURE) && 719 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT, 720 &pctx))) || 721 ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) && 722 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx)))) 723 retval = e2fsck_journal_fix_corrupt_super(ctx, journal, 724 &pctx); 725 e2fsck_journal_release(ctx, journal, 0, 1); 726 return retval; 727 } 728 729 /* 730 * We want to make the flags consistent here. We will not leave with 731 * needs_recovery set but has_journal clear. We can't get in a loop 732 * with -y, -n, or -p, only if a user isn't making up their mind. 733 */ 734no_has_journal: 735 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) { 736 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER; 737 pctx.str = "inode"; 738 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) { 739 if (recover && 740 !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx)) 741 goto no_has_journal; 742 /* 743 * Need a full fsck if we are releasing a 744 * journal stored on a reserved inode. 745 */ 746 force_fsck = recover || 747 (sb->s_journal_inum < EXT2_FIRST_INODE(sb)); 748 /* Clear all of the journal fields */ 749 sb->s_journal_inum = 0; 750 sb->s_journal_dev = 0; 751 memset(sb->s_journal_uuid, 0, 752 sizeof(sb->s_journal_uuid)); 753 e2fsck_clear_recover(ctx, force_fsck); 754 } else if (!(ctx->options & E2F_OPT_READONLY)) { 755 sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL; 756 ext2fs_mark_super_dirty(ctx->fs); 757 } 758 } 759 760 if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL && 761 !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) && 762 journal->j_superblock->s_start != 0) { 763 /* Print status information */ 764 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx); 765 if (ctx->superblock) 766 problem = PR_0_JOURNAL_RUN_DEFAULT; 767 else 768 problem = PR_0_JOURNAL_RUN; 769 if (fix_problem(ctx, problem, &pctx)) { 770 ctx->options |= E2F_OPT_FORCE; 771 sb->s_feature_incompat |= 772 EXT3_FEATURE_INCOMPAT_RECOVER; 773 ext2fs_mark_super_dirty(ctx->fs); 774 } else if (fix_problem(ctx, 775 PR_0_JOURNAL_RESET_JOURNAL, &pctx)) { 776 reset = 1; 777 sb->s_state &= ~EXT2_VALID_FS; 778 ext2fs_mark_super_dirty(ctx->fs); 779 } 780 /* 781 * If the user answers no to the above question, we 782 * ignore the fact that journal apparently has data; 783 * accidentally replaying over valid data would be far 784 * worse than skipping a questionable recovery. 785 * 786 * XXX should we abort with a fatal error here? What 787 * will the ext3 kernel code do if a filesystem with 788 * !NEEDS_RECOVERY but with a non-zero 789 * journal->j_superblock->s_start is mounted? 790 */ 791 } 792 793 e2fsck_journal_release(ctx, journal, reset, 0); 794 return retval; 795} 796 797static errcode_t recover_ext3_journal(e2fsck_t ctx) 798{ 799 journal_t *journal; 800 int retval; 801 802 journal_init_revoke_caches(); 803 retval = e2fsck_get_journal(ctx, &journal); 804 if (retval) 805 return retval; 806 807 retval = e2fsck_journal_load(journal); 808 if (retval) 809 goto errout; 810 811 retval = journal_init_revoke(journal, 1024); 812 if (retval) 813 goto errout; 814 815 retval = -journal_recover(journal); 816 if (retval) 817 goto errout; 818 819 if (journal->j_superblock->s_errno) { 820 ctx->fs->super->s_state |= EXT2_ERROR_FS; 821 ext2fs_mark_super_dirty(ctx->fs); 822 journal->j_superblock->s_errno = 0; 823 mark_buffer_dirty(journal->j_sb_buffer); 824 } 825 826errout: 827 journal_destroy_revoke(journal); 828 journal_destroy_revoke_caches(); 829 e2fsck_journal_release(ctx, journal, 1, 0); 830 return retval; 831} 832 833int e2fsck_run_ext3_journal(e2fsck_t ctx) 834{ 835 io_manager io_ptr = ctx->fs->io->manager; 836 int blocksize = ctx->fs->blocksize; 837 errcode_t retval, recover_retval; 838 839 printf(_("%s: recovering journal\n"), ctx->device_name); 840 if (ctx->options & E2F_OPT_READONLY) { 841 printf(_("%s: won't do journal recovery while read-only\n"), 842 ctx->device_name); 843 return EXT2_ET_FILE_RO; 844 } 845 846 if (ctx->fs->flags & EXT2_FLAG_DIRTY) 847 ext2fs_flush(ctx->fs); /* Force out any modifications */ 848 849 recover_retval = recover_ext3_journal(ctx); 850 851 /* 852 * Reload the filesystem context to get up-to-date data from disk 853 * because journal recovery will change the filesystem under us. 854 */ 855 ext2fs_close(ctx->fs); 856 retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW, 857 ctx->superblock, blocksize, io_ptr, 858 &ctx->fs); 859 860 if (retval) { 861 com_err(ctx->program_name, retval, 862 _("while trying to re-open %s"), 863 ctx->device_name); 864 fatal_error(ctx, 0); 865 } 866 ctx->fs->priv_data = ctx; 867 ctx->fs->now = ctx->now; 868 ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY; 869 870 /* Set the superblock flags */ 871 e2fsck_clear_recover(ctx, recover_retval); 872 return recover_retval; 873} 874 875/* 876 * This function will move the journal inode from a visible file in 877 * the filesystem directory hierarchy to the reserved inode if necessary. 878 */ 879static const char * const journal_names[] = { 880 ".journal", "journal", ".journal.dat", "journal.dat", 0 }; 881 882void e2fsck_move_ext3_journal(e2fsck_t ctx) 883{ 884 struct ext2_super_block *sb = ctx->fs->super; 885 struct problem_context pctx; 886 struct ext2_inode inode; 887 ext2_filsys fs = ctx->fs; 888 ext2_ino_t ino; 889 errcode_t retval; 890 const char * const * cpp; 891 int group, mount_flags; 892 893 clear_problem_context(&pctx); 894 895 /* 896 * If the filesystem is opened read-only, or there is no 897 * journal, then do nothing. 898 */ 899 if ((ctx->options & E2F_OPT_READONLY) || 900 (sb->s_journal_inum == 0) || 901 !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) 902 return; 903 904 /* 905 * Read in the journal inode 906 */ 907 if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0) 908 return; 909 910 /* 911 * If it's necessary to backup the journal inode, do so. 912 */ 913 if ((sb->s_jnl_backup_type == 0) || 914 ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) && 915 memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) { 916 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) { 917 memcpy(sb->s_jnl_blocks, inode.i_block, 918 EXT2_N_BLOCKS*4); 919 sb->s_jnl_blocks[16] = inode.i_size; 920 sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS; 921 ext2fs_mark_super_dirty(fs); 922 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY; 923 } 924 } 925 926 /* 927 * If the journal is already the hidden inode, then do nothing 928 */ 929 if (sb->s_journal_inum == EXT2_JOURNAL_INO) 930 return; 931 932 /* 933 * The journal inode had better have only one link and not be readable. 934 */ 935 if (inode.i_links_count != 1) 936 return; 937 938 /* 939 * If the filesystem is mounted, or we can't tell whether 940 * or not it's mounted, do nothing. 941 */ 942 retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags); 943 if (retval || (mount_flags & EXT2_MF_MOUNTED)) 944 return; 945 946 /* 947 * If we can't find the name of the journal inode, then do 948 * nothing. 949 */ 950 for (cpp = journal_names; *cpp; cpp++) { 951 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp, 952 strlen(*cpp), 0, &ino); 953 if ((retval == 0) && (ino == sb->s_journal_inum)) 954 break; 955 } 956 if (*cpp == 0) 957 return; 958 959 /* We need the inode bitmap to be loaded */ 960 retval = ext2fs_read_bitmaps(fs); 961 if (retval) 962 return; 963 964 pctx.str = *cpp; 965 if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx)) 966 return; 967 968 /* 969 * OK, we've done all the checks, let's actually move the 970 * journal inode. Errors at this point mean we need to force 971 * an ext2 filesystem check. 972 */ 973 if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0) 974 goto err_out; 975 if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0) 976 goto err_out; 977 sb->s_journal_inum = EXT2_JOURNAL_INO; 978 ext2fs_mark_super_dirty(fs); 979 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY; 980 inode.i_links_count = 0; 981 inode.i_dtime = ctx->now; 982 if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0) 983 goto err_out; 984 985 group = ext2fs_group_of_ino(fs, ino); 986 ext2fs_unmark_inode_bitmap(fs->inode_map, ino); 987 ext2fs_mark_ib_dirty(fs); 988 fs->group_desc[group].bg_free_inodes_count++; 989 fs->super->s_free_inodes_count++; 990 return; 991 992err_out: 993 pctx.errcode = retval; 994 fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx); 995 fs->super->s_state &= ~EXT2_VALID_FS; 996 ext2fs_mark_super_dirty(fs); 997 return; 998} 999 1000/* 1001 * This function makes sure the superblock hint for the external 1002 * journal is correct. 1003 */ 1004int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx) 1005{ 1006 struct ext2_super_block *sb = ctx->fs->super; 1007 struct problem_context pctx; 1008 char uuid[37], *journal_name; 1009 struct stat st; 1010 1011 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) || 1012 uuid_is_null(sb->s_journal_uuid)) 1013 return 0; 1014 1015 uuid_unparse(sb->s_journal_uuid, uuid); 1016 journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid); 1017 if (!journal_name) 1018 return 0; 1019 1020 if (stat(journal_name, &st) < 0) 1021 return 0; 1022 1023 if (st.st_rdev != sb->s_journal_dev) { 1024 clear_problem_context(&pctx); 1025 pctx.num = st.st_rdev; 1026 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) { 1027 sb->s_journal_dev = st.st_rdev; 1028 ext2fs_mark_super_dirty(ctx->fs); 1029 } 1030 } 1031 1032 free(journal_name); 1033 return 0; 1034} 1035