1 /*
2 * journal.c --- code for handling the "ext3" journal
3 *
4 * Copyright (C) 2000 Andreas Dilger
5 * Copyright (C) 2000 Theodore Ts'o
6 *
7 * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
8 * Copyright (C) 1999 Red Hat Software
9 *
10 * This file may be redistributed under the terms of the
11 * GNU General Public License version 2 or at your discretion
12 * any later version.
13 */
14
15 #ifdef HAVE_SYS_MOUNT_H
16 #include <sys/param.h>
17 #include <sys/mount.h>
18 #define MNT_FL (MS_MGC_VAL | MS_RDONLY)
19 #endif
20 #ifdef HAVE_SYS_STAT_H
21 #include <sys/stat.h>
22 #endif
23
24 #define E2FSCK_INCLUDE_INLINE_FUNCS
25 #include "jfs_user.h"
26 #include "problem.h"
27 #include "uuid/uuid.h"
28
29 #ifdef CONFIG_JBD_DEBUG /* Enabled by configure --enable-jfs-debug */
30 static int bh_count = 0;
31 #endif
32
33 /*
34 * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
35 * This creates a larger static binary, and a smaller binary using
36 * shared libraries. It's also probably slightly less CPU-efficient,
37 * which is why it's not on by default. But, it's a good way of
38 * testing the functions in inode_io.c and fileio.c.
39 */
40 #undef USE_INODE_IO
41
42 /* Kernel compatibility functions for handling the journal. These allow us
43 * to use the recovery.c file virtually unchanged from the kernel, so we
44 * don't have to do much to keep kernel and user recovery in sync.
45 */
journal_bmap(journal_t * journal,blk_t block,unsigned long * phys)46 int journal_bmap(journal_t *journal, blk_t block, unsigned long *phys)
47 {
48 #ifdef USE_INODE_IO
49 *phys = block;
50 return 0;
51 #else
52 struct inode *inode = journal->j_inode;
53 errcode_t retval;
54 blk_t pblk;
55
56 if (!inode) {
57 *phys = block;
58 return 0;
59 }
60
61 retval= ext2fs_bmap(inode->i_ctx->fs, inode->i_ino,
62 &inode->i_ext2, NULL, 0, block, &pblk);
63 *phys = pblk;
64 return (retval);
65 #endif
66 }
67
getblk(kdev_t kdev,blk_t blocknr,int blocksize)68 struct buffer_head *getblk(kdev_t kdev, blk_t blocknr, int blocksize)
69 {
70 struct buffer_head *bh;
71 int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
72 sizeof(bh->b_data);
73
74 bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
75 if (!bh)
76 return NULL;
77
78 #ifdef CONFIG_JBD_DEBUG
79 if (journal_enable_debug >= 3)
80 bh_count++;
81 #endif
82 jfs_debug(4, "getblk for block %lu (%d bytes)(total %d)\n",
83 (unsigned long) blocknr, blocksize, bh_count);
84
85 bh->b_ctx = kdev->k_ctx;
86 if (kdev->k_dev == K_DEV_FS)
87 bh->b_io = kdev->k_ctx->fs->io;
88 else
89 bh->b_io = kdev->k_ctx->journal_io;
90 bh->b_size = blocksize;
91 bh->b_blocknr = blocknr;
92
93 return bh;
94 }
95
sync_blockdev(kdev_t kdev)96 void sync_blockdev(kdev_t kdev)
97 {
98 io_channel io;
99
100 if (kdev->k_dev == K_DEV_FS)
101 io = kdev->k_ctx->fs->io;
102 else
103 io = kdev->k_ctx->journal_io;
104
105 io_channel_flush(io);
106 }
107
ll_rw_block(int rw,int nr,struct buffer_head * bhp[])108 void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
109 {
110 int retval;
111 struct buffer_head *bh;
112
113 for (; nr > 0; --nr) {
114 bh = *bhp++;
115 if (rw == READ && !bh->b_uptodate) {
116 jfs_debug(3, "reading block %lu/%p\n",
117 (unsigned long) bh->b_blocknr, (void *) bh);
118 retval = io_channel_read_blk(bh->b_io,
119 bh->b_blocknr,
120 1, bh->b_data);
121 if (retval) {
122 com_err(bh->b_ctx->device_name, retval,
123 "while reading block %lu\n",
124 (unsigned long) bh->b_blocknr);
125 bh->b_err = retval;
126 continue;
127 }
128 bh->b_uptodate = 1;
129 } else if (rw == WRITE && bh->b_dirty) {
130 jfs_debug(3, "writing block %lu/%p\n",
131 (unsigned long) bh->b_blocknr, (void *) bh);
132 retval = io_channel_write_blk(bh->b_io,
133 bh->b_blocknr,
134 1, bh->b_data);
135 if (retval) {
136 com_err(bh->b_ctx->device_name, retval,
137 "while writing block %lu\n",
138 (unsigned long) bh->b_blocknr);
139 bh->b_err = retval;
140 continue;
141 }
142 bh->b_dirty = 0;
143 bh->b_uptodate = 1;
144 } else {
145 jfs_debug(3, "no-op %s for block %lu\n",
146 rw == READ ? "read" : "write",
147 (unsigned long) bh->b_blocknr);
148 }
149 }
150 }
151
mark_buffer_dirty(struct buffer_head * bh)152 void mark_buffer_dirty(struct buffer_head *bh)
153 {
154 bh->b_dirty = 1;
155 }
156
mark_buffer_clean(struct buffer_head * bh)157 static void mark_buffer_clean(struct buffer_head * bh)
158 {
159 bh->b_dirty = 0;
160 }
161
brelse(struct buffer_head * bh)162 void brelse(struct buffer_head *bh)
163 {
164 if (bh->b_dirty)
165 ll_rw_block(WRITE, 1, &bh);
166 jfs_debug(3, "freeing block %lu/%p (total %d)\n",
167 (unsigned long) bh->b_blocknr, (void *) bh, --bh_count);
168 ext2fs_free_mem(&bh);
169 }
170
buffer_uptodate(struct buffer_head * bh)171 int buffer_uptodate(struct buffer_head *bh)
172 {
173 return bh->b_uptodate;
174 }
175
mark_buffer_uptodate(struct buffer_head * bh,int val)176 void mark_buffer_uptodate(struct buffer_head *bh, int val)
177 {
178 bh->b_uptodate = val;
179 }
180
wait_on_buffer(struct buffer_head * bh)181 void wait_on_buffer(struct buffer_head *bh)
182 {
183 if (!bh->b_uptodate)
184 ll_rw_block(READ, 1, &bh);
185 }
186
187
e2fsck_clear_recover(e2fsck_t ctx,int error)188 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
189 {
190 ctx->fs->super->s_feature_incompat &= ~EXT3_FEATURE_INCOMPAT_RECOVER;
191
192 /* if we had an error doing journal recovery, we need a full fsck */
193 if (error)
194 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
195 ext2fs_mark_super_dirty(ctx->fs);
196 }
197
198 /*
199 * This is a helper function to check the validity of the journal.
200 */
201 struct process_block_struct {
202 e2_blkcnt_t last_block;
203 };
204
process_journal_block(ext2_filsys fs,blk_t * block_nr,e2_blkcnt_t blockcnt,blk_t ref_block EXT2FS_ATTR ((unused)),int ref_offset EXT2FS_ATTR ((unused)),void * priv_data)205 static int process_journal_block(ext2_filsys fs,
206 blk_t *block_nr,
207 e2_blkcnt_t blockcnt,
208 blk_t ref_block EXT2FS_ATTR((unused)),
209 int ref_offset EXT2FS_ATTR((unused)),
210 void *priv_data)
211 {
212 struct process_block_struct *p;
213 blk_t blk = *block_nr;
214
215 p = (struct process_block_struct *) priv_data;
216
217 if (blk < fs->super->s_first_data_block ||
218 blk >= fs->super->s_blocks_count)
219 return BLOCK_ABORT;
220
221 if (blockcnt >= 0)
222 p->last_block = blockcnt;
223 return 0;
224 }
225
e2fsck_get_journal(e2fsck_t ctx,journal_t ** ret_journal)226 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
227 {
228 struct process_block_struct pb;
229 struct ext2_super_block *sb = ctx->fs->super;
230 struct ext2_super_block jsuper;
231 struct problem_context pctx;
232 struct buffer_head *bh;
233 struct inode *j_inode = NULL;
234 struct kdev_s *dev_fs = NULL, *dev_journal;
235 const char *journal_name = 0;
236 journal_t *journal = NULL;
237 errcode_t retval = 0;
238 io_manager io_ptr = 0;
239 unsigned long start = 0;
240 int ext_journal = 0;
241 int tried_backup_jnl = 0;
242
243 clear_problem_context(&pctx);
244
245 journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
246 if (!journal) {
247 return EXT2_ET_NO_MEMORY;
248 }
249
250 dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
251 if (!dev_fs) {
252 retval = EXT2_ET_NO_MEMORY;
253 goto errout;
254 }
255 dev_journal = dev_fs+1;
256
257 dev_fs->k_ctx = dev_journal->k_ctx = ctx;
258 dev_fs->k_dev = K_DEV_FS;
259 dev_journal->k_dev = K_DEV_JOURNAL;
260
261 journal->j_dev = dev_journal;
262 journal->j_fs_dev = dev_fs;
263 journal->j_inode = NULL;
264 journal->j_blocksize = ctx->fs->blocksize;
265
266 if (uuid_is_null(sb->s_journal_uuid)) {
267 if (!sb->s_journal_inum) {
268 retval = EXT2_ET_BAD_INODE_NUM;
269 goto errout;
270 }
271 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
272 "journal inode");
273 if (!j_inode) {
274 retval = EXT2_ET_NO_MEMORY;
275 goto errout;
276 }
277
278 j_inode->i_ctx = ctx;
279 j_inode->i_ino = sb->s_journal_inum;
280
281 if ((retval = ext2fs_read_inode(ctx->fs,
282 sb->s_journal_inum,
283 &j_inode->i_ext2))) {
284 try_backup_journal:
285 if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
286 tried_backup_jnl)
287 goto errout;
288 memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
289 memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
290 EXT2_N_BLOCKS*4);
291 j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
292 j_inode->i_ext2.i_links_count = 1;
293 j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
294 e2fsck_use_inode_shortcuts(ctx, 1);
295 ctx->stashed_ino = j_inode->i_ino;
296 ctx->stashed_inode = &j_inode->i_ext2;
297 tried_backup_jnl++;
298 }
299 if (!j_inode->i_ext2.i_links_count ||
300 !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
301 retval = EXT2_ET_NO_JOURNAL;
302 goto try_backup_journal;
303 }
304 if (j_inode->i_ext2.i_size / journal->j_blocksize <
305 JFS_MIN_JOURNAL_BLOCKS) {
306 retval = EXT2_ET_JOURNAL_TOO_SMALL;
307 goto try_backup_journal;
308 }
309 pb.last_block = -1;
310 retval = ext2fs_block_iterate2(ctx->fs, j_inode->i_ino,
311 BLOCK_FLAG_HOLE, 0,
312 process_journal_block, &pb);
313 if ((pb.last_block+1) * ctx->fs->blocksize <
314 j_inode->i_ext2.i_size) {
315 retval = EXT2_ET_JOURNAL_TOO_SMALL;
316 goto try_backup_journal;
317 }
318 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
319 retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
320 &j_inode->i_ext2);
321 if (retval)
322 goto errout;
323 }
324
325 journal->j_maxlen = j_inode->i_ext2.i_size / journal->j_blocksize;
326
327 #ifdef USE_INODE_IO
328 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
329 &j_inode->i_ext2,
330 &journal_name);
331 if (retval)
332 goto errout;
333
334 io_ptr = inode_io_manager;
335 #else
336 journal->j_inode = j_inode;
337 ctx->journal_io = ctx->fs->io;
338 if ((retval = journal_bmap(journal, 0, &start)) != 0)
339 goto errout;
340 #endif
341 } else {
342 ext_journal = 1;
343 if (!ctx->journal_name) {
344 char uuid[37];
345
346 uuid_unparse(sb->s_journal_uuid, uuid);
347 ctx->journal_name = blkid_get_devname(ctx->blkid,
348 "UUID", uuid);
349 if (!ctx->journal_name)
350 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
351 }
352 journal_name = ctx->journal_name;
353
354 if (!journal_name) {
355 fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
356 retval = EXT2_ET_LOAD_EXT_JOURNAL;
357 goto errout;
358 }
359
360 jfs_debug(1, "Using journal file %s\n", journal_name);
361 io_ptr = unix_io_manager;
362 }
363
364 #if 0
365 test_io_backing_manager = io_ptr;
366 io_ptr = test_io_manager;
367 #endif
368 #ifndef USE_INODE_IO
369 if (ext_journal)
370 #endif
371 retval = io_ptr->open(journal_name, IO_FLAG_RW,
372 &ctx->journal_io);
373 if (retval)
374 goto errout;
375
376 io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
377
378 if (ext_journal) {
379 if (ctx->fs->blocksize == 1024)
380 start = 1;
381 bh = getblk(dev_journal, start, ctx->fs->blocksize);
382 if (!bh) {
383 retval = EXT2_ET_NO_MEMORY;
384 goto errout;
385 }
386 ll_rw_block(READ, 1, &bh);
387 if ((retval = bh->b_err) != 0) {
388 brelse(bh);
389 goto errout;
390 }
391 memcpy(&jsuper, start ? bh->b_data : bh->b_data + 1024,
392 sizeof(jsuper));
393 brelse(bh);
394 #ifdef WORDS_BIGENDIAN
395 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
396 ext2fs_swap_super(&jsuper);
397 #endif
398 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
399 !(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
400 fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
401 retval = EXT2_ET_LOAD_EXT_JOURNAL;
402 goto errout;
403 }
404 /* Make sure the journal UUID is correct */
405 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
406 sizeof(jsuper.s_uuid))) {
407 fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
408 retval = EXT2_ET_LOAD_EXT_JOURNAL;
409 goto errout;
410 }
411
412 journal->j_maxlen = jsuper.s_blocks_count;
413 start++;
414 }
415
416 if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
417 retval = EXT2_ET_NO_MEMORY;
418 goto errout;
419 }
420
421 journal->j_sb_buffer = bh;
422 journal->j_superblock = (journal_superblock_t *)bh->b_data;
423
424 #ifdef USE_INODE_IO
425 if (j_inode)
426 ext2fs_free_mem(&j_inode);
427 #endif
428
429 *ret_journal = journal;
430 e2fsck_use_inode_shortcuts(ctx, 0);
431 return 0;
432
433 errout:
434 e2fsck_use_inode_shortcuts(ctx, 0);
435 if (dev_fs)
436 ext2fs_free_mem(&dev_fs);
437 if (j_inode)
438 ext2fs_free_mem(&j_inode);
439 if (journal)
440 ext2fs_free_mem(&journal);
441 return retval;
442 }
443
e2fsck_journal_fix_bad_inode(e2fsck_t ctx,struct problem_context * pctx)444 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
445 struct problem_context *pctx)
446 {
447 struct ext2_super_block *sb = ctx->fs->super;
448 int recover = ctx->fs->super->s_feature_incompat &
449 EXT3_FEATURE_INCOMPAT_RECOVER;
450 int has_journal = ctx->fs->super->s_feature_compat &
451 EXT3_FEATURE_COMPAT_HAS_JOURNAL;
452
453 if (has_journal || sb->s_journal_inum) {
454 /* The journal inode is bogus, remove and force full fsck */
455 pctx->ino = sb->s_journal_inum;
456 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
457 if (has_journal && sb->s_journal_inum)
458 printf("*** ext3 journal has been deleted - "
459 "filesystem is now ext2 only ***\n\n");
460 sb->s_feature_compat &= ~EXT3_FEATURE_COMPAT_HAS_JOURNAL;
461 sb->s_journal_inum = 0;
462 ctx->flags |= E2F_FLAG_JOURNAL_INODE;
463 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
464 e2fsck_clear_recover(ctx, 1);
465 return 0;
466 }
467 return EXT2_ET_BAD_INODE_NUM;
468 } else if (recover) {
469 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
470 e2fsck_clear_recover(ctx, 1);
471 return 0;
472 }
473 return EXT2_ET_UNSUPP_FEATURE;
474 }
475 return 0;
476 }
477
478 #define V1_SB_SIZE 0x0024
clear_v2_journal_fields(journal_t * journal)479 static void clear_v2_journal_fields(journal_t *journal)
480 {
481 e2fsck_t ctx = journal->j_dev->k_ctx;
482 struct problem_context pctx;
483
484 clear_problem_context(&pctx);
485
486 if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
487 return;
488
489 memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
490 ctx->fs->blocksize-V1_SB_SIZE);
491 mark_buffer_dirty(journal->j_sb_buffer);
492 }
493
494
e2fsck_journal_load(journal_t * journal)495 static errcode_t e2fsck_journal_load(journal_t *journal)
496 {
497 e2fsck_t ctx = journal->j_dev->k_ctx;
498 journal_superblock_t *jsb;
499 struct buffer_head *jbh = journal->j_sb_buffer;
500 struct problem_context pctx;
501
502 clear_problem_context(&pctx);
503
504 ll_rw_block(READ, 1, &jbh);
505 if (jbh->b_err) {
506 com_err(ctx->device_name, jbh->b_err,
507 _("reading journal superblock\n"));
508 return jbh->b_err;
509 }
510
511 jsb = journal->j_superblock;
512 /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
513 if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
514 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
515
516 switch (ntohl(jsb->s_header.h_blocktype)) {
517 case JFS_SUPERBLOCK_V1:
518 journal->j_format_version = 1;
519 if (jsb->s_feature_compat ||
520 jsb->s_feature_incompat ||
521 jsb->s_feature_ro_compat ||
522 jsb->s_nr_users)
523 clear_v2_journal_fields(journal);
524 break;
525
526 case JFS_SUPERBLOCK_V2:
527 journal->j_format_version = 2;
528 if (ntohl(jsb->s_nr_users) > 1 &&
529 uuid_is_null(ctx->fs->super->s_journal_uuid))
530 clear_v2_journal_fields(journal);
531 if (ntohl(jsb->s_nr_users) > 1) {
532 fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
533 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
534 }
535 break;
536
537 /*
538 * These should never appear in a journal super block, so if
539 * they do, the journal is badly corrupted.
540 */
541 case JFS_DESCRIPTOR_BLOCK:
542 case JFS_COMMIT_BLOCK:
543 case JFS_REVOKE_BLOCK:
544 return EXT2_ET_CORRUPT_SUPERBLOCK;
545
546 /* If we don't understand the superblock major type, but there
547 * is a magic number, then it is likely to be a new format we
548 * just don't understand, so leave it alone. */
549 default:
550 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
551 }
552
553 if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
554 return EXT2_ET_UNSUPP_FEATURE;
555
556 if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
557 return EXT2_ET_RO_UNSUPP_FEATURE;
558
559 /* We have now checked whether we know enough about the journal
560 * format to be able to proceed safely, so any other checks that
561 * fail we should attempt to recover from. */
562 if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
563 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
564 _("%s: no valid journal superblock found\n"),
565 ctx->device_name);
566 return EXT2_ET_CORRUPT_SUPERBLOCK;
567 }
568
569 if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
570 journal->j_maxlen = ntohl(jsb->s_maxlen);
571 else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
572 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
573 _("%s: journal too short\n"),
574 ctx->device_name);
575 return EXT2_ET_CORRUPT_SUPERBLOCK;
576 }
577
578 journal->j_tail_sequence = ntohl(jsb->s_sequence);
579 journal->j_transaction_sequence = journal->j_tail_sequence;
580 journal->j_tail = ntohl(jsb->s_start);
581 journal->j_first = ntohl(jsb->s_first);
582 journal->j_last = ntohl(jsb->s_maxlen);
583
584 return 0;
585 }
586
e2fsck_journal_reset_super(e2fsck_t ctx,journal_superblock_t * jsb,journal_t * journal)587 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
588 journal_t *journal)
589 {
590 char *p;
591 union {
592 uuid_t uuid;
593 __u32 val[4];
594 } u;
595 __u32 new_seq = 0;
596 int i;
597
598 /* Leave a valid existing V1 superblock signature alone.
599 * Anything unrecognisable we overwrite with a new V2
600 * signature. */
601
602 if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
603 jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
604 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
605 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
606 }
607
608 /* Zero out everything else beyond the superblock header */
609
610 p = ((char *) jsb) + sizeof(journal_header_t);
611 memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
612
613 jsb->s_blocksize = htonl(ctx->fs->blocksize);
614 jsb->s_maxlen = htonl(journal->j_maxlen);
615 jsb->s_first = htonl(1);
616
617 /* Initialize the journal sequence number so that there is "no"
618 * chance we will find old "valid" transactions in the journal.
619 * This avoids the need to zero the whole journal (slow to do,
620 * and risky when we are just recovering the filesystem).
621 */
622 uuid_generate(u.uuid);
623 for (i = 0; i < 4; i ++)
624 new_seq ^= u.val[i];
625 jsb->s_sequence = htonl(new_seq);
626
627 mark_buffer_dirty(journal->j_sb_buffer);
628 ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
629 }
630
e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,journal_t * journal,struct problem_context * pctx)631 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
632 journal_t *journal,
633 struct problem_context *pctx)
634 {
635 struct ext2_super_block *sb = ctx->fs->super;
636 int recover = ctx->fs->super->s_feature_incompat &
637 EXT3_FEATURE_INCOMPAT_RECOVER;
638
639 if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) {
640 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
641 e2fsck_journal_reset_super(ctx, journal->j_superblock,
642 journal);
643 journal->j_transaction_sequence = 1;
644 e2fsck_clear_recover(ctx, recover);
645 return 0;
646 }
647 return EXT2_ET_CORRUPT_SUPERBLOCK;
648 } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
649 return EXT2_ET_CORRUPT_SUPERBLOCK;
650
651 return 0;
652 }
653
e2fsck_journal_release(e2fsck_t ctx,journal_t * journal,int reset,int drop)654 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
655 int reset, int drop)
656 {
657 journal_superblock_t *jsb;
658
659 if (drop)
660 mark_buffer_clean(journal->j_sb_buffer);
661 else if (!(ctx->options & E2F_OPT_READONLY)) {
662 jsb = journal->j_superblock;
663 jsb->s_sequence = htonl(journal->j_transaction_sequence);
664 if (reset)
665 jsb->s_start = 0; /* this marks the journal as empty */
666 mark_buffer_dirty(journal->j_sb_buffer);
667 }
668 brelse(journal->j_sb_buffer);
669
670 if (ctx->journal_io) {
671 if (ctx->fs && ctx->fs->io != ctx->journal_io)
672 io_channel_close(ctx->journal_io);
673 ctx->journal_io = 0;
674 }
675
676 #ifndef USE_INODE_IO
677 if (journal->j_inode)
678 ext2fs_free_mem(&journal->j_inode);
679 #endif
680 if (journal->j_fs_dev)
681 ext2fs_free_mem(&journal->j_fs_dev);
682 ext2fs_free_mem(&journal);
683 }
684
685 /*
686 * This function makes sure that the superblock fields regarding the
687 * journal are consistent.
688 */
e2fsck_check_ext3_journal(e2fsck_t ctx)689 int e2fsck_check_ext3_journal(e2fsck_t ctx)
690 {
691 struct ext2_super_block *sb = ctx->fs->super;
692 journal_t *journal;
693 int recover = ctx->fs->super->s_feature_incompat &
694 EXT3_FEATURE_INCOMPAT_RECOVER;
695 struct problem_context pctx;
696 problem_t problem;
697 int reset = 0, force_fsck = 0;
698 int retval;
699
700 /* If we don't have any journal features, don't do anything more */
701 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) &&
702 !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
703 uuid_is_null(sb->s_journal_uuid))
704 return 0;
705
706 clear_problem_context(&pctx);
707 pctx.num = sb->s_journal_inum;
708
709 retval = e2fsck_get_journal(ctx, &journal);
710 if (retval) {
711 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
712 (retval == EXT2_ET_BAD_BLOCK_NUM) ||
713 (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
714 (retval == EXT2_ET_NO_JOURNAL))
715 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
716 return retval;
717 }
718
719 retval = e2fsck_journal_load(journal);
720 if (retval) {
721 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
722 ((retval == EXT2_ET_UNSUPP_FEATURE) &&
723 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
724 &pctx))) ||
725 ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
726 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
727 &pctx))) ||
728 ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
729 (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
730 retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
731 &pctx);
732 e2fsck_journal_release(ctx, journal, 0, 1);
733 return retval;
734 }
735
736 /*
737 * We want to make the flags consistent here. We will not leave with
738 * needs_recovery set but has_journal clear. We can't get in a loop
739 * with -y, -n, or -p, only if a user isn't making up their mind.
740 */
741 no_has_journal:
742 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
743 recover = sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER;
744 pctx.str = "inode";
745 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
746 if (recover &&
747 !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
748 goto no_has_journal;
749 /*
750 * Need a full fsck if we are releasing a
751 * journal stored on a reserved inode.
752 */
753 force_fsck = recover ||
754 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
755 /* Clear all of the journal fields */
756 sb->s_journal_inum = 0;
757 sb->s_journal_dev = 0;
758 memset(sb->s_journal_uuid, 0,
759 sizeof(sb->s_journal_uuid));
760 e2fsck_clear_recover(ctx, force_fsck);
761 } else if (!(ctx->options & E2F_OPT_READONLY)) {
762 sb->s_feature_compat |= EXT3_FEATURE_COMPAT_HAS_JOURNAL;
763 ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
764 ext2fs_mark_super_dirty(ctx->fs);
765 }
766 }
767
768 if (sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL &&
769 !(sb->s_feature_incompat & EXT3_FEATURE_INCOMPAT_RECOVER) &&
770 journal->j_superblock->s_start != 0) {
771 /* Print status information */
772 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
773 if (ctx->superblock)
774 problem = PR_0_JOURNAL_RUN_DEFAULT;
775 else
776 problem = PR_0_JOURNAL_RUN;
777 if (fix_problem(ctx, problem, &pctx)) {
778 ctx->options |= E2F_OPT_FORCE;
779 sb->s_feature_incompat |=
780 EXT3_FEATURE_INCOMPAT_RECOVER;
781 ext2fs_mark_super_dirty(ctx->fs);
782 } else if (fix_problem(ctx,
783 PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
784 reset = 1;
785 sb->s_state &= ~EXT2_VALID_FS;
786 ext2fs_mark_super_dirty(ctx->fs);
787 }
788 /*
789 * If the user answers no to the above question, we
790 * ignore the fact that journal apparently has data;
791 * accidentally replaying over valid data would be far
792 * worse than skipping a questionable recovery.
793 *
794 * XXX should we abort with a fatal error here? What
795 * will the ext3 kernel code do if a filesystem with
796 * !NEEDS_RECOVERY but with a non-zero
797 * journal->j_superblock->s_start is mounted?
798 */
799 }
800
801 e2fsck_journal_release(ctx, journal, reset, 0);
802 return retval;
803 }
804
recover_ext3_journal(e2fsck_t ctx)805 static errcode_t recover_ext3_journal(e2fsck_t ctx)
806 {
807 struct problem_context pctx;
808 journal_t *journal;
809 int retval;
810
811 clear_problem_context(&pctx);
812
813 journal_init_revoke_caches();
814 retval = e2fsck_get_journal(ctx, &journal);
815 if (retval)
816 return retval;
817
818 retval = e2fsck_journal_load(journal);
819 if (retval)
820 goto errout;
821
822 retval = journal_init_revoke(journal, 1024);
823 if (retval)
824 goto errout;
825
826 retval = -journal_recover(journal);
827 if (retval)
828 goto errout;
829
830 if (journal->j_failed_commit) {
831 pctx.ino = journal->j_failed_commit;
832 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
833 ctx->fs->super->s_state |= EXT2_ERROR_FS;
834 ext2fs_mark_super_dirty(ctx->fs);
835 }
836
837
838 if (journal->j_superblock->s_errno) {
839 ctx->fs->super->s_state |= EXT2_ERROR_FS;
840 ext2fs_mark_super_dirty(ctx->fs);
841 journal->j_superblock->s_errno = 0;
842 mark_buffer_dirty(journal->j_sb_buffer);
843 }
844
845 errout:
846 journal_destroy_revoke(journal);
847 journal_destroy_revoke_caches();
848 e2fsck_journal_release(ctx, journal, 1, 0);
849 return retval;
850 }
851
e2fsck_run_ext3_journal(e2fsck_t ctx)852 int e2fsck_run_ext3_journal(e2fsck_t ctx)
853 {
854 io_manager io_ptr = ctx->fs->io->manager;
855 int blocksize = ctx->fs->blocksize;
856 errcode_t retval, recover_retval;
857 io_stats stats = 0;
858 unsigned long long kbytes_written = 0;
859
860 printf(_("%s: recovering journal\n"), ctx->device_name);
861 if (ctx->options & E2F_OPT_READONLY) {
862 printf(_("%s: won't do journal recovery while read-only\n"),
863 ctx->device_name);
864 return EXT2_ET_FILE_RO;
865 }
866
867 if (ctx->fs->flags & EXT2_FLAG_DIRTY)
868 ext2fs_flush(ctx->fs); /* Force out any modifications */
869
870 recover_retval = recover_ext3_journal(ctx);
871
872 /*
873 * Reload the filesystem context to get up-to-date data from disk
874 * because journal recovery will change the filesystem under us.
875 */
876 if (ctx->fs->super->s_kbytes_written &&
877 ctx->fs->io->manager->get_stats)
878 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
879 if (stats && stats->bytes_written)
880 kbytes_written = stats->bytes_written >> 10;
881 ext2fs_free(ctx->fs);
882 retval = ext2fs_open(ctx->filesystem_name, EXT2_FLAG_RW,
883 ctx->superblock, blocksize, io_ptr,
884 &ctx->fs);
885 if (retval) {
886 com_err(ctx->program_name, retval,
887 _("while trying to re-open %s"),
888 ctx->device_name);
889 fatal_error(ctx, 0);
890 }
891 ctx->fs->priv_data = ctx;
892 ctx->fs->now = ctx->now;
893 ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
894 ctx->fs->super->s_kbytes_written += kbytes_written;
895
896 /* Set the superblock flags */
897 e2fsck_clear_recover(ctx, recover_retval);
898 return recover_retval;
899 }
900
901 /*
902 * This function will move the journal inode from a visible file in
903 * the filesystem directory hierarchy to the reserved inode if necessary.
904 */
905 static const char * const journal_names[] = {
906 ".journal", "journal", ".journal.dat", "journal.dat", 0 };
907
e2fsck_move_ext3_journal(e2fsck_t ctx)908 void e2fsck_move_ext3_journal(e2fsck_t ctx)
909 {
910 struct ext2_super_block *sb = ctx->fs->super;
911 struct problem_context pctx;
912 struct ext2_inode inode;
913 ext2_filsys fs = ctx->fs;
914 ext2_ino_t ino;
915 errcode_t retval;
916 const char * const * cpp;
917 int group, mount_flags;
918
919 clear_problem_context(&pctx);
920
921 /*
922 * If the filesystem is opened read-only, or there is no
923 * journal, then do nothing.
924 */
925 if ((ctx->options & E2F_OPT_READONLY) ||
926 (sb->s_journal_inum == 0) ||
927 !(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
928 return;
929
930 /*
931 * Read in the journal inode
932 */
933 if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
934 return;
935
936 /*
937 * If it's necessary to backup the journal inode, do so.
938 */
939 if ((sb->s_jnl_backup_type == 0) ||
940 ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
941 memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
942 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
943 memcpy(sb->s_jnl_blocks, inode.i_block,
944 EXT2_N_BLOCKS*4);
945 sb->s_jnl_blocks[16] = inode.i_size;
946 sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
947 ext2fs_mark_super_dirty(fs);
948 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
949 }
950 }
951
952 /*
953 * If the journal is already the hidden inode, then do nothing
954 */
955 if (sb->s_journal_inum == EXT2_JOURNAL_INO)
956 return;
957
958 /*
959 * The journal inode had better have only one link and not be readable.
960 */
961 if (inode.i_links_count != 1)
962 return;
963
964 /*
965 * If the filesystem is mounted, or we can't tell whether
966 * or not it's mounted, do nothing.
967 */
968 retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
969 if (retval || (mount_flags & EXT2_MF_MOUNTED))
970 return;
971
972 /*
973 * If we can't find the name of the journal inode, then do
974 * nothing.
975 */
976 for (cpp = journal_names; *cpp; cpp++) {
977 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
978 strlen(*cpp), 0, &ino);
979 if ((retval == 0) && (ino == sb->s_journal_inum))
980 break;
981 }
982 if (*cpp == 0)
983 return;
984
985 /* We need the inode bitmap to be loaded */
986 retval = ext2fs_read_bitmaps(fs);
987 if (retval)
988 return;
989
990 pctx.str = *cpp;
991 if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
992 return;
993
994 /*
995 * OK, we've done all the checks, let's actually move the
996 * journal inode. Errors at this point mean we need to force
997 * an ext2 filesystem check.
998 */
999 if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1000 goto err_out;
1001 if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1002 goto err_out;
1003 sb->s_journal_inum = EXT2_JOURNAL_INO;
1004 ext2fs_mark_super_dirty(fs);
1005 fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1006 inode.i_links_count = 0;
1007 inode.i_dtime = ctx->now;
1008 if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1009 goto err_out;
1010
1011 group = ext2fs_group_of_ino(fs, ino);
1012 ext2fs_unmark_inode_bitmap(fs->inode_map, ino);
1013 ext2fs_mark_ib_dirty(fs);
1014 fs->group_desc[group].bg_free_inodes_count++;
1015 ext2fs_group_desc_csum_set(fs, group);
1016 fs->super->s_free_inodes_count++;
1017 return;
1018
1019 err_out:
1020 pctx.errcode = retval;
1021 fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1022 fs->super->s_state &= ~EXT2_VALID_FS;
1023 ext2fs_mark_super_dirty(fs);
1024 return;
1025 }
1026
1027 /*
1028 * This function makes sure the superblock hint for the external
1029 * journal is correct.
1030 */
e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)1031 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1032 {
1033 struct ext2_super_block *sb = ctx->fs->super;
1034 struct problem_context pctx;
1035 char uuid[37], *journal_name;
1036 struct stat st;
1037
1038 if (!(sb->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL) ||
1039 uuid_is_null(sb->s_journal_uuid))
1040 return 0;
1041
1042 uuid_unparse(sb->s_journal_uuid, uuid);
1043 journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1044 if (!journal_name)
1045 return 0;
1046
1047 if (stat(journal_name, &st) < 0)
1048 return 0;
1049
1050 if (st.st_rdev != sb->s_journal_dev) {
1051 clear_problem_context(&pctx);
1052 pctx.num = st.st_rdev;
1053 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1054 sb->s_journal_dev = st.st_rdev;
1055 ext2fs_mark_super_dirty(ctx->fs);
1056 }
1057 }
1058
1059 free(journal_name);
1060 return 0;
1061 }
1062