Whamcloud - gitweb
e83f3a91c0228914515540df641be3aaba934279
[tools/e2fsprogs.git] / e2fsck / journal.c
1 /*
2  * journal.c --- code for handling the "ext3" journal
3  *
4  * Copyright (C) 2000 Andreas Dilger
5  * Copyright (C) 2000 Theodore Ts'o
6  *
7  * Parts of the code are based on fs/jfs/journal.c by Stephen C. Tweedie
8  * Copyright (C) 1999 Red Hat Software
9  *
10  * This file may be redistributed under the terms of the
11  * GNU General Public License version 2 or at your discretion
12  * any later version.
13  */
14
15 #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 "jfs_user.h"
27 #include "problem.h"
28 #include "uuid/uuid.h"
29
30 #ifdef CONFIG_JBD_DEBUG         /* Enabled by configure --enable-jfs-debug */
31 static int bh_count = 0;
32 #endif
33
34 /*
35  * Define USE_INODE_IO to use the inode_io.c / fileio.c codepaths.
36  * This creates a larger static binary, and a smaller binary using
37  * shared libraries.  It's also probably slightly less CPU-efficient,
38  * which is why it's not on by default.  But, it's a good way of
39  * testing the functions in inode_io.c and fileio.c.
40  */
41 #undef USE_INODE_IO
42
43 /* Checksumming functions */
44 static int e2fsck_journal_verify_csum_type(journal_t *j,
45                                            journal_superblock_t *jsb)
46 {
47         if (!journal_has_csum_v2or3(j))
48                 return 1;
49
50         return jsb->s_checksum_type == JBD2_CRC32C_CHKSUM;
51 }
52
53 static __u32 e2fsck_journal_sb_csum(journal_superblock_t *jsb)
54 {
55         __u32 crc, old_crc;
56
57         old_crc = jsb->s_checksum;
58         jsb->s_checksum = 0;
59         crc = ext2fs_crc32c_le(~0, (unsigned char *)jsb,
60                                sizeof(journal_superblock_t));
61         jsb->s_checksum = old_crc;
62
63         return crc;
64 }
65
66 static int e2fsck_journal_sb_csum_verify(journal_t *j,
67                                          journal_superblock_t *jsb)
68 {
69         __u32 provided, calculated;
70
71         if (!journal_has_csum_v2or3(j))
72                 return 1;
73
74         provided = ext2fs_be32_to_cpu(jsb->s_checksum);
75         calculated = e2fsck_journal_sb_csum(jsb);
76
77         return provided == calculated;
78 }
79
80 static errcode_t e2fsck_journal_sb_csum_set(journal_t *j,
81                                             journal_superblock_t *jsb)
82 {
83         __u32 crc;
84
85         if (!journal_has_csum_v2or3(j))
86                 return 0;
87
88         crc = e2fsck_journal_sb_csum(jsb);
89         jsb->s_checksum = ext2fs_cpu_to_be32(crc);
90         return 0;
91 }
92
93 /* Kernel compatibility functions for handling the journal.  These allow us
94  * to use the recovery.c file virtually unchanged from the kernel, so we
95  * don't have to do much to keep kernel and user recovery in sync.
96  */
97 int journal_bmap(journal_t *journal, blk64_t block, unsigned long long *phys)
98 {
99 #ifdef USE_INODE_IO
100         *phys = block;
101         return 0;
102 #else
103         struct inode    *inode = journal->j_inode;
104         errcode_t       retval;
105         blk64_t         pblk;
106
107         if (!inode) {
108                 *phys = block;
109                 return 0;
110         }
111
112         retval= ext2fs_bmap2(inode->i_ctx->fs, inode->i_ino,
113                              &inode->i_ext2, NULL, 0, block, 0, &pblk);
114         *phys = pblk;
115         return -1 * ((int) retval);
116 #endif
117 }
118
119 struct buffer_head *getblk(kdev_t kdev, blk64_t blocknr, int blocksize)
120 {
121         struct buffer_head *bh;
122         int bufsize = sizeof(*bh) + kdev->k_ctx->fs->blocksize -
123                 sizeof(bh->b_data);
124
125         bh = e2fsck_allocate_memory(kdev->k_ctx, bufsize, "block buffer");
126         if (!bh)
127                 return NULL;
128
129 #ifdef CONFIG_JBD_DEBUG
130         if (journal_enable_debug >= 3)
131                 bh_count++;
132 #endif
133         jfs_debug(4, "getblk for block %llu (%d bytes)(total %d)\n",
134                   (unsigned long long) blocknr, blocksize, bh_count);
135
136         bh->b_ctx = kdev->k_ctx;
137         if (kdev->k_dev == K_DEV_FS)
138                 bh->b_io = kdev->k_ctx->fs->io;
139         else
140                 bh->b_io = kdev->k_ctx->journal_io;
141         bh->b_size = blocksize;
142         bh->b_blocknr = blocknr;
143
144         return bh;
145 }
146
147 int sync_blockdev(kdev_t kdev)
148 {
149         io_channel      io;
150
151         if (kdev->k_dev == K_DEV_FS)
152                 io = kdev->k_ctx->fs->io;
153         else
154                 io = kdev->k_ctx->journal_io;
155
156         return io_channel_flush(io) ? -EIO : 0;
157 }
158
159 void ll_rw_block(int rw, int nr, struct buffer_head *bhp[])
160 {
161         errcode_t retval;
162         struct buffer_head *bh;
163
164         for (; nr > 0; --nr) {
165                 bh = *bhp++;
166                 if (rw == READ && !bh->b_uptodate) {
167                         jfs_debug(3, "reading block %llu/%p\n",
168                                   bh->b_blocknr, (void *) bh);
169                         retval = io_channel_read_blk64(bh->b_io,
170                                                      bh->b_blocknr,
171                                                      1, bh->b_data);
172                         if (retval) {
173                                 com_err(bh->b_ctx->device_name, retval,
174                                         "while reading block %llu\n",
175                                         bh->b_blocknr);
176                                 bh->b_err = (int) retval;
177                                 continue;
178                         }
179                         bh->b_uptodate = 1;
180                 } else if (rw == WRITE && bh->b_dirty) {
181                         jfs_debug(3, "writing block %llu/%p\n",
182                                   bh->b_blocknr,
183                                   (void *) bh);
184                         retval = io_channel_write_blk64(bh->b_io,
185                                                       bh->b_blocknr,
186                                                       1, bh->b_data);
187                         if (retval) {
188                                 com_err(bh->b_ctx->device_name, retval,
189                                         "while writing block %llu\n",
190                                         bh->b_blocknr);
191                                 bh->b_err = (int) retval;
192                                 continue;
193                         }
194                         bh->b_dirty = 0;
195                         bh->b_uptodate = 1;
196                 } else {
197                         jfs_debug(3, "no-op %s for block %llu\n",
198                                   rw == READ ? "read" : "write",
199                                   bh->b_blocknr);
200                 }
201         }
202 }
203
204 void mark_buffer_dirty(struct buffer_head *bh)
205 {
206         bh->b_dirty = 1;
207 }
208
209 static void mark_buffer_clean(struct buffer_head * bh)
210 {
211         bh->b_dirty = 0;
212 }
213
214 void brelse(struct buffer_head *bh)
215 {
216         if (bh->b_dirty)
217                 ll_rw_block(WRITE, 1, &bh);
218         jfs_debug(3, "freeing block %llu/%p (total %d)\n",
219                   bh->b_blocknr, (void *) bh, --bh_count);
220         ext2fs_free_mem(&bh);
221 }
222
223 int buffer_uptodate(struct buffer_head *bh)
224 {
225         return bh->b_uptodate;
226 }
227
228 void mark_buffer_uptodate(struct buffer_head *bh, int val)
229 {
230         bh->b_uptodate = val;
231 }
232
233 void wait_on_buffer(struct buffer_head *bh)
234 {
235         if (!bh->b_uptodate)
236                 ll_rw_block(READ, 1, &bh);
237 }
238
239
240 static void e2fsck_clear_recover(e2fsck_t ctx, int error)
241 {
242         ext2fs_clear_feature_journal_needs_recovery(ctx->fs->super);
243
244         /* if we had an error doing journal recovery, we need a full fsck */
245         if (error)
246                 ctx->fs->super->s_state &= ~EXT2_VALID_FS;
247         ext2fs_mark_super_dirty(ctx->fs);
248 }
249
250 /*
251  * This is a helper function to check the validity of the journal.
252  */
253 struct process_block_struct {
254         e2_blkcnt_t     last_block;
255 };
256
257 static int process_journal_block(ext2_filsys fs,
258                                  blk64_t        *block_nr,
259                                  e2_blkcnt_t blockcnt,
260                                  blk64_t ref_block EXT2FS_ATTR((unused)),
261                                  int ref_offset EXT2FS_ATTR((unused)),
262                                  void *priv_data)
263 {
264         struct process_block_struct *p;
265         blk64_t blk = *block_nr;
266
267         p = (struct process_block_struct *) priv_data;
268
269         if (!blk || blk < fs->super->s_first_data_block ||
270             blk >= ext2fs_blocks_count(fs->super))
271                 return BLOCK_ABORT;
272
273         if (blockcnt >= 0)
274                 p->last_block = blockcnt;
275         return 0;
276 }
277
278 static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
279 {
280         struct process_block_struct pb;
281         struct ext2_super_block *sb = ctx->fs->super;
282         struct ext2_super_block jsuper;
283         struct problem_context  pctx;
284         struct buffer_head      *bh;
285         struct inode            *j_inode = NULL;
286         struct kdev_s           *dev_fs = NULL, *dev_journal;
287         const char              *journal_name = 0;
288         journal_t               *journal = NULL;
289         errcode_t               retval = 0;
290         io_manager              io_ptr = 0;
291         unsigned long long      start = 0;
292         int                     ret;
293         int                     ext_journal = 0;
294         int                     tried_backup_jnl = 0;
295
296         clear_problem_context(&pctx);
297
298         journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
299         if (!journal) {
300                 return EXT2_ET_NO_MEMORY;
301         }
302
303         dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
304         if (!dev_fs) {
305                 retval = EXT2_ET_NO_MEMORY;
306                 goto errout;
307         }
308         dev_journal = dev_fs+1;
309
310         dev_fs->k_ctx = dev_journal->k_ctx = ctx;
311         dev_fs->k_dev = K_DEV_FS;
312         dev_journal->k_dev = K_DEV_JOURNAL;
313
314         journal->j_dev = dev_journal;
315         journal->j_fs_dev = dev_fs;
316         journal->j_inode = NULL;
317         journal->j_blocksize = ctx->fs->blocksize;
318
319         if (uuid_is_null(sb->s_journal_uuid)) {
320                 if (!sb->s_journal_inum) {
321                         retval = EXT2_ET_BAD_INODE_NUM;
322                         goto errout;
323                 }
324                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
325                                                  "journal inode");
326                 if (!j_inode) {
327                         retval = EXT2_ET_NO_MEMORY;
328                         goto errout;
329                 }
330
331                 j_inode->i_ctx = ctx;
332                 j_inode->i_ino = sb->s_journal_inum;
333
334                 if ((retval = ext2fs_read_inode(ctx->fs,
335                                                 sb->s_journal_inum,
336                                                 &j_inode->i_ext2))) {
337                 try_backup_journal:
338                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
339                             tried_backup_jnl)
340                                 goto errout;
341                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
342                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
343                                EXT2_N_BLOCKS*4);
344                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
345                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
346                         j_inode->i_ext2.i_links_count = 1;
347                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
348                         e2fsck_use_inode_shortcuts(ctx, 1);
349                         ctx->stashed_ino = j_inode->i_ino;
350                         ctx->stashed_inode = &j_inode->i_ext2;
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                     JFS_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(ctx->fs, j_inode->i_ino,
365                                                BLOCK_FLAG_HOLE, 0,
366                                                process_journal_block, &pb);
367                 if ((pb.last_block + 1) * ctx->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 && !(ctx->options & E2F_OPT_READONLY)) {
373                         retval = ext2fs_write_inode(ctx->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(ctx->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                 ctx->journal_io = ctx->fs->io;
393                 if ((ret = journal_bmap(journal, 0, &start)) != 0) {
394                         retval = (errcode_t) (-1 * ret);
395                         goto errout;
396                 }
397 #endif
398         } else {
399                 ext_journal = 1;
400                 if (!ctx->journal_name) {
401                         char uuid[37];
402
403                         uuid_unparse(sb->s_journal_uuid, uuid);
404                         ctx->journal_name = blkid_get_devname(ctx->blkid,
405                                                               "UUID", uuid);
406                         if (!ctx->journal_name)
407                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
408                 }
409                 journal_name = ctx->journal_name;
410
411                 if (!journal_name) {
412                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
413                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
414                         goto errout;
415                 }
416
417                 jfs_debug(1, "Using journal file %s\n", journal_name);
418                 io_ptr = unix_io_manager;
419         }
420
421 #if 0
422         test_io_backing_manager = io_ptr;
423         io_ptr = test_io_manager;
424 #endif
425 #ifndef USE_INODE_IO
426         if (ext_journal)
427 #endif
428         {
429                 int flags = IO_FLAG_RW;
430                 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
431                       ctx->mount_flags & EXT2_MF_READONLY))
432                         flags |= IO_FLAG_EXCLUSIVE;
433                 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
434                     (ctx->options & E2F_OPT_FORCE))
435                         flags &= ~IO_FLAG_EXCLUSIVE;
436
437
438                 retval = io_ptr->open(journal_name, flags,
439                                       &ctx->journal_io);
440         }
441         if (retval)
442                 goto errout;
443
444         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
445
446         if (ext_journal) {
447                 blk64_t maxlen;
448
449                 start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
450                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
451                 if (!bh) {
452                         retval = EXT2_ET_NO_MEMORY;
453                         goto errout;
454                 }
455                 ll_rw_block(READ, 1, &bh);
456                 if ((retval = bh->b_err) != 0) {
457                         brelse(bh);
458                         goto errout;
459                 }
460                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + SUPERBLOCK_OFFSET,
461                        sizeof(jsuper));
462 #ifdef WORDS_BIGENDIAN
463                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
464                         ext2fs_swap_super(&jsuper);
465 #endif
466                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
467                     !ext2fs_has_feature_journal_dev(&jsuper)) {
468                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
469                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
470                         brelse(bh);
471                         goto errout;
472                 }
473                 /* Make sure the journal UUID is correct */
474                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
475                            sizeof(jsuper.s_uuid))) {
476                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
477                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
478                         brelse(bh);
479                         goto errout;
480                 }
481
482                 /* Check the superblock checksum */
483                 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
484                         struct struct_ext2_filsys fsx;
485                         struct ext2_super_block superx;
486                         void *p;
487
488                         p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
489                         memcpy(&fsx, ctx->fs, sizeof(fsx));
490                         memcpy(&superx, ctx->fs->super, sizeof(superx));
491                         fsx.super = &superx;
492                         ext2fs_set_feature_metadata_csum(fsx.super);
493                         if (!ext2fs_superblock_csum_verify(&fsx, p) &&
494                             fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
495                                         &pctx)) {
496                                 ext2fs_superblock_csum_set(&fsx, p);
497                                 mark_buffer_dirty(bh);
498                         }
499                 }
500                 brelse(bh);
501
502                 maxlen = ext2fs_blocks_count(&jsuper);
503                 journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
504                 start++;
505         }
506
507         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
508                 retval = EXT2_ET_NO_MEMORY;
509                 goto errout;
510         }
511
512         journal->j_sb_buffer = bh;
513         journal->j_superblock = (journal_superblock_t *)bh->b_data;
514
515 #ifdef USE_INODE_IO
516         if (j_inode)
517                 ext2fs_free_mem(&j_inode);
518 #endif
519
520         *ret_journal = journal;
521         e2fsck_use_inode_shortcuts(ctx, 0);
522         return 0;
523
524 errout:
525         e2fsck_use_inode_shortcuts(ctx, 0);
526         if (dev_fs)
527                 ext2fs_free_mem(&dev_fs);
528         if (j_inode)
529                 ext2fs_free_mem(&j_inode);
530         if (journal)
531                 ext2fs_free_mem(&journal);
532         return retval;
533 }
534
535 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
536                                               struct problem_context *pctx)
537 {
538         struct ext2_super_block *sb = ctx->fs->super;
539         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
540         int has_journal = ext2fs_has_feature_journal(ctx->fs->super);
541
542         if (has_journal || sb->s_journal_inum) {
543                 /* The journal inode is bogus, remove and force full fsck */
544                 pctx->ino = sb->s_journal_inum;
545                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
546                         if (has_journal && sb->s_journal_inum)
547                                 printf("*** journal has been deleted ***\n\n");
548                         ext2fs_clear_feature_journal(sb);
549                         sb->s_journal_inum = 0;
550                         memset(sb->s_jnl_blocks, 0, sizeof(sb->s_jnl_blocks));
551                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
552                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
553                         e2fsck_clear_recover(ctx, 1);
554                         return 0;
555                 }
556                 return EXT2_ET_CORRUPT_JOURNAL_SB;
557         } else if (recover) {
558                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
559                         e2fsck_clear_recover(ctx, 1);
560                         return 0;
561                 }
562                 return EXT2_ET_UNSUPP_FEATURE;
563         }
564         return 0;
565 }
566
567 #define V1_SB_SIZE      0x0024
568 static void clear_v2_journal_fields(journal_t *journal)
569 {
570         e2fsck_t ctx = journal->j_dev->k_ctx;
571         struct problem_context pctx;
572
573         clear_problem_context(&pctx);
574
575         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
576                 return;
577
578         ctx->flags |= E2F_FLAG_PROBLEMS_FIXED;
579         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
580                ctx->fs->blocksize-V1_SB_SIZE);
581         mark_buffer_dirty(journal->j_sb_buffer);
582 }
583
584
585 static errcode_t e2fsck_journal_load(journal_t *journal)
586 {
587         e2fsck_t ctx = journal->j_dev->k_ctx;
588         journal_superblock_t *jsb;
589         struct buffer_head *jbh = journal->j_sb_buffer;
590         struct problem_context pctx;
591
592         clear_problem_context(&pctx);
593
594         ll_rw_block(READ, 1, &jbh);
595         if (jbh->b_err) {
596                 com_err(ctx->device_name, jbh->b_err, "%s",
597                         _("reading journal superblock\n"));
598                 return jbh->b_err;
599         }
600
601         jsb = journal->j_superblock;
602         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
603         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
604                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
605
606         switch (ntohl(jsb->s_header.h_blocktype)) {
607         case JFS_SUPERBLOCK_V1:
608                 journal->j_format_version = 1;
609                 if (jsb->s_feature_compat ||
610                     jsb->s_feature_incompat ||
611                     jsb->s_feature_ro_compat ||
612                     jsb->s_nr_users)
613                         clear_v2_journal_fields(journal);
614                 break;
615
616         case JFS_SUPERBLOCK_V2:
617                 journal->j_format_version = 2;
618                 if (ntohl(jsb->s_nr_users) > 1 &&
619                     uuid_is_null(ctx->fs->super->s_journal_uuid))
620                         clear_v2_journal_fields(journal);
621                 if (ntohl(jsb->s_nr_users) > 1) {
622                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
623                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
624                 }
625                 break;
626
627         /*
628          * These should never appear in a journal super block, so if
629          * they do, the journal is badly corrupted.
630          */
631         case JFS_DESCRIPTOR_BLOCK:
632         case JFS_COMMIT_BLOCK:
633         case JFS_REVOKE_BLOCK:
634                 return EXT2_ET_CORRUPT_JOURNAL_SB;
635
636         /* If we don't understand the superblock major type, but there
637          * is a magic number, then it is likely to be a new format we
638          * just don't understand, so leave it alone. */
639         default:
640                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
641         }
642
643         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
644                 return EXT2_ET_UNSUPP_FEATURE;
645
646         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
647                 return EXT2_ET_RO_UNSUPP_FEATURE;
648
649         /* Checksum v1-3 are mutually exclusive features. */
650         if (jfs_has_feature_csum2(journal) && jfs_has_feature_csum3(journal))
651                 return EXT2_ET_CORRUPT_JOURNAL_SB;
652
653         if (journal_has_csum_v2or3(journal) &&
654             jfs_has_feature_checksum(journal))
655                 return EXT2_ET_CORRUPT_JOURNAL_SB;
656
657         if (!e2fsck_journal_verify_csum_type(journal, jsb) ||
658             !e2fsck_journal_sb_csum_verify(journal, jsb))
659                 return EXT2_ET_CORRUPT_JOURNAL_SB;
660
661         if (journal_has_csum_v2or3(journal))
662                 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
663                                                    sizeof(jsb->s_uuid));
664
665         /* We have now checked whether we know enough about the journal
666          * format to be able to proceed safely, so any other checks that
667          * fail we should attempt to recover from. */
668         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
669                 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
670                         _("%s: no valid journal superblock found\n"),
671                         ctx->device_name);
672                 return EXT2_ET_CORRUPT_JOURNAL_SB;
673         }
674
675         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
676                 journal->j_maxlen = ntohl(jsb->s_maxlen);
677         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
678                 com_err(ctx->program_name, EXT2_ET_CORRUPT_JOURNAL_SB,
679                         _("%s: journal too short\n"),
680                         ctx->device_name);
681                 return EXT2_ET_CORRUPT_JOURNAL_SB;
682         }
683
684         journal->j_tail_sequence = ntohl(jsb->s_sequence);
685         journal->j_transaction_sequence = journal->j_tail_sequence;
686         journal->j_tail = ntohl(jsb->s_start);
687         journal->j_first = ntohl(jsb->s_first);
688         journal->j_last = ntohl(jsb->s_maxlen);
689
690         return 0;
691 }
692
693 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
694                                        journal_t *journal)
695 {
696         char *p;
697         union {
698                 uuid_t uuid;
699                 __u32 val[4];
700         } u;
701         __u32 new_seq = 0;
702         int i;
703
704         /* Leave a valid existing V1 superblock signature alone.
705          * Anything unrecognisable we overwrite with a new V2
706          * signature. */
707
708         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
709             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
710                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
711                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
712         }
713
714         /* Zero out everything else beyond the superblock header */
715
716         p = ((char *) jsb) + sizeof(journal_header_t);
717         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
718
719         jsb->s_blocksize = htonl(ctx->fs->blocksize);
720         jsb->s_maxlen = htonl(journal->j_maxlen);
721         jsb->s_first = htonl(1);
722
723         /* Initialize the journal sequence number so that there is "no"
724          * chance we will find old "valid" transactions in the journal.
725          * This avoids the need to zero the whole journal (slow to do,
726          * and risky when we are just recovering the filesystem).
727          */
728         uuid_generate(u.uuid);
729         for (i = 0; i < 4; i ++)
730                 new_seq ^= u.val[i];
731         jsb->s_sequence = htonl(new_seq);
732         e2fsck_journal_sb_csum_set(journal, jsb);
733
734         mark_buffer_dirty(journal->j_sb_buffer);
735         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
736 }
737
738 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
739                                                   journal_t *journal,
740                                                   struct problem_context *pctx)
741 {
742         struct ext2_super_block *sb = ctx->fs->super;
743         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
744
745         if (ext2fs_has_feature_journal(sb)) {
746                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
747                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
748                                                    journal);
749                         journal->j_transaction_sequence = 1;
750                         e2fsck_clear_recover(ctx, recover);
751                         return 0;
752                 }
753                 return EXT2_ET_CORRUPT_JOURNAL_SB;
754         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
755                 return EXT2_ET_CORRUPT_JOURNAL_SB;
756
757         return 0;
758 }
759
760 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
761                                    int reset, int drop)
762 {
763         journal_superblock_t *jsb;
764
765         if (drop)
766                 mark_buffer_clean(journal->j_sb_buffer);
767         else if (!(ctx->options & E2F_OPT_READONLY)) {
768                 jsb = journal->j_superblock;
769                 jsb->s_sequence = htonl(journal->j_tail_sequence);
770                 if (reset)
771                         jsb->s_start = 0; /* this marks the journal as empty */
772                 e2fsck_journal_sb_csum_set(journal, jsb);
773                 mark_buffer_dirty(journal->j_sb_buffer);
774         }
775         brelse(journal->j_sb_buffer);
776
777         if (ctx->journal_io) {
778                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
779                         io_channel_close(ctx->journal_io);
780                 ctx->journal_io = 0;
781         }
782
783 #ifndef USE_INODE_IO
784         if (journal->j_inode)
785                 ext2fs_free_mem(&journal->j_inode);
786 #endif
787         if (journal->j_fs_dev)
788                 ext2fs_free_mem(&journal->j_fs_dev);
789         ext2fs_free_mem(&journal);
790 }
791
792 /*
793  * This function makes sure that the superblock fields regarding the
794  * journal are consistent.
795  */
796 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
797 {
798         struct ext2_super_block *sb = ctx->fs->super;
799         journal_t *journal;
800         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
801         struct problem_context pctx;
802         problem_t problem;
803         int reset = 0, force_fsck = 0;
804         errcode_t retval;
805
806         /* If we don't have any journal features, don't do anything more */
807         if (!ext2fs_has_feature_journal(sb) &&
808             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
809             uuid_is_null(sb->s_journal_uuid))
810                 return 0;
811
812         clear_problem_context(&pctx);
813         pctx.num = sb->s_journal_inum;
814
815         retval = e2fsck_get_journal(ctx, &journal);
816         if (retval) {
817                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
818                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
819                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
820                     (retval == EXT2_ET_NO_JOURNAL))
821                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
822                 return retval;
823         }
824
825         retval = e2fsck_journal_load(journal);
826         if (retval) {
827                 if ((retval == EXT2_ET_CORRUPT_JOURNAL_SB) ||
828                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
829                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
830                                   &pctx))) ||
831                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
832                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
833                                   &pctx))) ||
834                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
835                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
836                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
837                                                                   &pctx);
838                 e2fsck_journal_release(ctx, journal, 0, 1);
839                 return retval;
840         }
841
842         /*
843          * We want to make the flags consistent here.  We will not leave with
844          * needs_recovery set but has_journal clear.  We can't get in a loop
845          * with -y, -n, or -p, only if a user isn't making up their mind.
846          */
847 no_has_journal:
848         if (!ext2fs_has_feature_journal(sb)) {
849                 recover = ext2fs_has_feature_journal_needs_recovery(sb);
850                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
851                         if (recover &&
852                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
853                                 goto no_has_journal;
854                         /*
855                          * Need a full fsck if we are releasing a
856                          * journal stored on a reserved inode.
857                          */
858                         force_fsck = recover ||
859                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
860                         /* Clear all of the journal fields */
861                         sb->s_journal_inum = 0;
862                         sb->s_journal_dev = 0;
863                         memset(sb->s_journal_uuid, 0,
864                                sizeof(sb->s_journal_uuid));
865                         e2fsck_clear_recover(ctx, force_fsck);
866                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
867                         ext2fs_set_feature_journal(sb);
868                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
869                         ext2fs_mark_super_dirty(ctx->fs);
870                 }
871         }
872
873         if (ext2fs_has_feature_journal(sb) &&
874             !ext2fs_has_feature_journal_needs_recovery(sb) &&
875             journal->j_superblock->s_start != 0) {
876                 /* Print status information */
877                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
878                 if (ctx->superblock)
879                         problem = PR_0_JOURNAL_RUN_DEFAULT;
880                 else
881                         problem = PR_0_JOURNAL_RUN;
882                 if (fix_problem(ctx, problem, &pctx)) {
883                         ctx->options |= E2F_OPT_FORCE;
884                         ext2fs_set_feature_journal_needs_recovery(sb);
885                         ext2fs_mark_super_dirty(ctx->fs);
886                 } else if (fix_problem(ctx,
887                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
888                         reset = 1;
889                         sb->s_state &= ~EXT2_VALID_FS;
890                         ext2fs_mark_super_dirty(ctx->fs);
891                 }
892                 /*
893                  * If the user answers no to the above question, we
894                  * ignore the fact that journal apparently has data;
895                  * accidentally replaying over valid data would be far
896                  * worse than skipping a questionable recovery.
897                  *
898                  * XXX should we abort with a fatal error here?  What
899                  * will the ext3 kernel code do if a filesystem with
900                  * !NEEDS_RECOVERY but with a non-zero
901                  * journal->j_superblock->s_start is mounted?
902                  */
903         }
904
905         /*
906          * If we don't need to do replay the journal, check to see if
907          * the journal's errno is set; if so, we need to mark the file
908          * system as being corrupt and clear the journal's s_errno.
909          */
910         if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
911             journal->j_superblock->s_errno) {
912                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
913                 ext2fs_mark_super_dirty(ctx->fs);
914                 journal->j_superblock->s_errno = 0;
915                 e2fsck_journal_sb_csum_set(journal, journal->j_superblock);
916                 mark_buffer_dirty(journal->j_sb_buffer);
917         }
918
919         e2fsck_journal_release(ctx, journal, reset, 0);
920         return retval;
921 }
922
923 static errcode_t recover_ext3_journal(e2fsck_t ctx)
924 {
925         struct problem_context  pctx;
926         journal_t *journal;
927         errcode_t retval;
928
929         clear_problem_context(&pctx);
930
931         journal_init_revoke_caches();
932         retval = e2fsck_get_journal(ctx, &journal);
933         if (retval)
934                 return retval;
935
936         retval = e2fsck_journal_load(journal);
937         if (retval)
938                 goto errout;
939
940         retval = journal_init_revoke(journal, 1024);
941         if (retval)
942                 goto errout;
943
944         retval = -journal_recover(journal);
945         if (retval)
946                 goto errout;
947
948         if (journal->j_failed_commit) {
949                 pctx.ino = journal->j_failed_commit;
950                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
951                 journal->j_superblock->s_errno = -EINVAL;
952                 mark_buffer_dirty(journal->j_sb_buffer);
953         }
954
955         journal->j_tail_sequence = journal->j_transaction_sequence;
956
957 errout:
958         journal_destroy_revoke(journal);
959         journal_destroy_revoke_caches();
960         e2fsck_journal_release(ctx, journal, 1, 0);
961         return retval;
962 }
963
964 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
965 {
966         io_manager io_ptr = ctx->fs->io->manager;
967         int blocksize = ctx->fs->blocksize;
968         errcode_t       retval, recover_retval;
969         io_stats        stats = 0;
970         unsigned long long kbytes_written = 0;
971
972         printf(_("%s: recovering journal\n"), ctx->device_name);
973         if (ctx->options & E2F_OPT_READONLY) {
974                 printf(_("%s: won't do journal recovery while read-only\n"),
975                        ctx->device_name);
976                 return EXT2_ET_FILE_RO;
977         }
978
979         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
980                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
981
982         recover_retval = recover_ext3_journal(ctx);
983
984         /*
985          * Reload the filesystem context to get up-to-date data from disk
986          * because journal recovery will change the filesystem under us.
987          */
988         if (ctx->fs->super->s_kbytes_written &&
989             ctx->fs->io->manager->get_stats)
990                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
991         if (stats && stats->bytes_written)
992                 kbytes_written = stats->bytes_written >> 10;
993
994         ext2fs_mmp_stop(ctx->fs);
995         ext2fs_free(ctx->fs);
996         retval = ext2fs_open(ctx->filesystem_name, ctx->openfs_flags,
997                              ctx->superblock, blocksize, io_ptr,
998                              &ctx->fs);
999         if (retval) {
1000                 com_err(ctx->program_name, retval,
1001                         _("while trying to re-open %s"),
1002                         ctx->device_name);
1003                 fatal_error(ctx, 0);
1004         }
1005         ctx->fs->priv_data = ctx;
1006         ctx->fs->now = ctx->now;
1007         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1008         ctx->fs->super->s_kbytes_written += kbytes_written;
1009
1010         /* Set the superblock flags */
1011         e2fsck_clear_recover(ctx, recover_retval != 0);
1012
1013         /*
1014          * Do one last sanity check, and propagate journal->s_errno to
1015          * the EXT2_ERROR_FS flag in the fs superblock if needed.
1016          */
1017         retval = e2fsck_check_ext3_journal(ctx);
1018         return retval ? retval : recover_retval;
1019 }
1020
1021 /*
1022  * This function will move the journal inode from a visible file in
1023  * the filesystem directory hierarchy to the reserved inode if necessary.
1024  */
1025 static const char * const journal_names[] = {
1026         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
1027
1028 void e2fsck_move_ext3_journal(e2fsck_t ctx)
1029 {
1030         struct ext2_super_block *sb = ctx->fs->super;
1031         struct problem_context  pctx;
1032         struct ext2_inode       inode;
1033         ext2_filsys             fs = ctx->fs;
1034         ext2_ino_t              ino;
1035         errcode_t               retval;
1036         const char * const *    cpp;
1037         dgrp_t                  group;
1038         int                     mount_flags;
1039
1040         clear_problem_context(&pctx);
1041
1042         /*
1043          * If the filesystem is opened read-only, or there is no
1044          * journal, then do nothing.
1045          */
1046         if ((ctx->options & E2F_OPT_READONLY) ||
1047             (sb->s_journal_inum == 0) ||
1048             !ext2fs_has_feature_journal(sb))
1049                 return;
1050
1051         /*
1052          * Read in the journal inode
1053          */
1054         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
1055                 return;
1056
1057         /*
1058          * If it's necessary to backup the journal inode, do so.
1059          */
1060         if ((sb->s_jnl_backup_type == 0) ||
1061             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
1062              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
1063                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
1064                         memcpy(sb->s_jnl_blocks, inode.i_block,
1065                                EXT2_N_BLOCKS*4);
1066                         sb->s_jnl_blocks[15] = inode.i_size_high;
1067                         sb->s_jnl_blocks[16] = inode.i_size;
1068                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
1069                         ext2fs_mark_super_dirty(fs);
1070                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1071                 }
1072         }
1073
1074         /*
1075          * If the journal is already the hidden inode, then do nothing
1076          */
1077         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
1078                 return;
1079
1080         /*
1081          * The journal inode had better have only one link and not be readable.
1082          */
1083         if (inode.i_links_count != 1)
1084                 return;
1085
1086         /*
1087          * If the filesystem is mounted, or we can't tell whether
1088          * or not it's mounted, do nothing.
1089          */
1090         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1091         if (retval || (mount_flags & EXT2_MF_MOUNTED))
1092                 return;
1093
1094         /*
1095          * If we can't find the name of the journal inode, then do
1096          * nothing.
1097          */
1098         for (cpp = journal_names; *cpp; cpp++) {
1099                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1100                                        strlen(*cpp), 0, &ino);
1101                 if ((retval == 0) && (ino == sb->s_journal_inum))
1102                         break;
1103         }
1104         if (*cpp == 0)
1105                 return;
1106
1107         /* We need the inode bitmap to be loaded */
1108         retval = ext2fs_read_bitmaps(fs);
1109         if (retval)
1110                 return;
1111
1112         pctx.str = *cpp;
1113         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1114                 return;
1115
1116         /*
1117          * OK, we've done all the checks, let's actually move the
1118          * journal inode.  Errors at this point mean we need to force
1119          * an ext2 filesystem check.
1120          */
1121         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1122                 goto err_out;
1123         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1124                 goto err_out;
1125         sb->s_journal_inum = EXT2_JOURNAL_INO;
1126         ext2fs_mark_super_dirty(fs);
1127         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1128         inode.i_links_count = 0;
1129         inode.i_dtime = ctx->now;
1130         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1131                 goto err_out;
1132
1133         group = ext2fs_group_of_ino(fs, ino);
1134         ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1135         ext2fs_mark_ib_dirty(fs);
1136         ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1137         ext2fs_group_desc_csum_set(fs, group);
1138         fs->super->s_free_inodes_count++;
1139         return;
1140
1141 err_out:
1142         pctx.errcode = retval;
1143         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1144         fs->super->s_state &= ~EXT2_VALID_FS;
1145         ext2fs_mark_super_dirty(fs);
1146         return;
1147 }
1148
1149 /*
1150  * This function makes sure the superblock hint for the external
1151  * journal is correct.
1152  */
1153 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1154 {
1155         struct ext2_super_block *sb = ctx->fs->super;
1156         struct problem_context pctx;
1157         char uuid[37], *journal_name;
1158         struct stat st;
1159
1160         if (!ext2fs_has_feature_journal(sb) ||
1161             uuid_is_null(sb->s_journal_uuid))
1162                 return 0;
1163
1164         uuid_unparse(sb->s_journal_uuid, uuid);
1165         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1166         if (!journal_name)
1167                 return 0;
1168
1169         if (stat(journal_name, &st) < 0) {
1170                 free(journal_name);
1171                 return 0;
1172         }
1173
1174         if (st.st_rdev != sb->s_journal_dev) {
1175                 clear_problem_context(&pctx);
1176                 pctx.num = st.st_rdev;
1177                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1178                         sb->s_journal_dev = st.st_rdev;
1179                         ext2fs_mark_super_dirty(ctx->fs);
1180                 }
1181         }
1182
1183         free(journal_name);
1184         return 0;
1185 }