Whamcloud - gitweb
e2fsck: clean up feature test macros with predicate functions
[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 (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                     ext_journal = 0;
293         int                     tried_backup_jnl = 0;
294
295         clear_problem_context(&pctx);
296
297         journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
298         if (!journal) {
299                 return EXT2_ET_NO_MEMORY;
300         }
301
302         dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
303         if (!dev_fs) {
304                 retval = EXT2_ET_NO_MEMORY;
305                 goto errout;
306         }
307         dev_journal = dev_fs+1;
308
309         dev_fs->k_ctx = dev_journal->k_ctx = ctx;
310         dev_fs->k_dev = K_DEV_FS;
311         dev_journal->k_dev = K_DEV_JOURNAL;
312
313         journal->j_dev = dev_journal;
314         journal->j_fs_dev = dev_fs;
315         journal->j_inode = NULL;
316         journal->j_blocksize = ctx->fs->blocksize;
317
318         if (uuid_is_null(sb->s_journal_uuid)) {
319                 if (!sb->s_journal_inum) {
320                         retval = EXT2_ET_BAD_INODE_NUM;
321                         goto errout;
322                 }
323                 j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
324                                                  "journal inode");
325                 if (!j_inode) {
326                         retval = EXT2_ET_NO_MEMORY;
327                         goto errout;
328                 }
329
330                 j_inode->i_ctx = ctx;
331                 j_inode->i_ino = sb->s_journal_inum;
332
333                 if ((retval = ext2fs_read_inode(ctx->fs,
334                                                 sb->s_journal_inum,
335                                                 &j_inode->i_ext2))) {
336                 try_backup_journal:
337                         if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
338                             tried_backup_jnl)
339                                 goto errout;
340                         memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
341                         memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
342                                EXT2_N_BLOCKS*4);
343                         j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
344                         j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
345                         j_inode->i_ext2.i_links_count = 1;
346                         j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
347                         e2fsck_use_inode_shortcuts(ctx, 1);
348                         ctx->stashed_ino = j_inode->i_ino;
349                         ctx->stashed_inode = &j_inode->i_ext2;
350                         tried_backup_jnl++;
351                 }
352                 if (!j_inode->i_ext2.i_links_count ||
353                     !LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
354                         retval = EXT2_ET_NO_JOURNAL;
355                         goto try_backup_journal;
356                 }
357                 if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
358                     JFS_MIN_JOURNAL_BLOCKS) {
359                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
360                         goto try_backup_journal;
361                 }
362                 pb.last_block = -1;
363                 retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
364                                                BLOCK_FLAG_HOLE, 0,
365                                                process_journal_block, &pb);
366                 if ((pb.last_block + 1) * ctx->fs->blocksize <
367                     (int) EXT2_I_SIZE(&j_inode->i_ext2)) {
368                         retval = EXT2_ET_JOURNAL_TOO_SMALL;
369                         goto try_backup_journal;
370                 }
371                 if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
372                         retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
373                                                     &j_inode->i_ext2);
374                         if (retval)
375                                 goto errout;
376                 }
377
378                 journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
379                         journal->j_blocksize;
380
381 #ifdef USE_INODE_IO
382                 retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
383                                                  &j_inode->i_ext2,
384                                                  &journal_name);
385                 if (retval)
386                         goto errout;
387
388                 io_ptr = inode_io_manager;
389 #else
390                 journal->j_inode = j_inode;
391                 ctx->journal_io = ctx->fs->io;
392                 if ((retval = (errcode_t) journal_bmap(journal, 0, &start)) != 0)
393                         goto errout;
394 #endif
395         } else {
396                 ext_journal = 1;
397                 if (!ctx->journal_name) {
398                         char uuid[37];
399
400                         uuid_unparse(sb->s_journal_uuid, uuid);
401                         ctx->journal_name = blkid_get_devname(ctx->blkid,
402                                                               "UUID", uuid);
403                         if (!ctx->journal_name)
404                                 ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
405                 }
406                 journal_name = ctx->journal_name;
407
408                 if (!journal_name) {
409                         fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
410                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
411                         goto errout;
412                 }
413
414                 jfs_debug(1, "Using journal file %s\n", journal_name);
415                 io_ptr = unix_io_manager;
416         }
417
418 #if 0
419         test_io_backing_manager = io_ptr;
420         io_ptr = test_io_manager;
421 #endif
422 #ifndef USE_INODE_IO
423         if (ext_journal)
424 #endif
425         {
426                 int flags = IO_FLAG_RW;
427                 if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
428                       ctx->mount_flags & EXT2_MF_READONLY))
429                         flags |= IO_FLAG_EXCLUSIVE;
430                 if ((ctx->mount_flags & EXT2_MF_READONLY) &&
431                     (ctx->options & E2F_OPT_FORCE))
432                         flags &= ~IO_FLAG_EXCLUSIVE;
433
434
435                 retval = io_ptr->open(journal_name, flags,
436                                       &ctx->journal_io);
437         }
438         if (retval)
439                 goto errout;
440
441         io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
442
443         if (ext_journal) {
444                 blk64_t maxlen;
445
446                 start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
447                 bh = getblk(dev_journal, start, ctx->fs->blocksize);
448                 if (!bh) {
449                         retval = EXT2_ET_NO_MEMORY;
450                         goto errout;
451                 }
452                 ll_rw_block(READ, 1, &bh);
453                 if ((retval = bh->b_err) != 0) {
454                         brelse(bh);
455                         goto errout;
456                 }
457                 memcpy(&jsuper, start ? bh->b_data :  bh->b_data + SUPERBLOCK_OFFSET,
458                        sizeof(jsuper));
459 #ifdef WORDS_BIGENDIAN
460                 if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
461                         ext2fs_swap_super(&jsuper);
462 #endif
463                 if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
464                     !ext2fs_has_feature_journal_dev(&jsuper)) {
465                         fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
466                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
467                         brelse(bh);
468                         goto errout;
469                 }
470                 /* Make sure the journal UUID is correct */
471                 if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
472                            sizeof(jsuper.s_uuid))) {
473                         fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
474                         retval = EXT2_ET_LOAD_EXT_JOURNAL;
475                         brelse(bh);
476                         goto errout;
477                 }
478
479                 /* Check the superblock checksum */
480                 if (ext2fs_has_feature_metadata_csum(&jsuper)) {
481                         struct struct_ext2_filsys fsx;
482                         struct ext2_super_block superx;
483                         void *p;
484
485                         p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
486                         memcpy(&fsx, ctx->fs, sizeof(fsx));
487                         memcpy(&superx, ctx->fs->super, sizeof(superx));
488                         fsx.super = &superx;
489                         ext2fs_set_feature_metadata_csum(fsx.super);
490                         if (!ext2fs_superblock_csum_verify(&fsx, p) &&
491                             fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
492                                         &pctx)) {
493                                 ext2fs_superblock_csum_set(&fsx, p);
494                                 mark_buffer_dirty(bh);
495                         }
496                 }
497                 brelse(bh);
498
499                 maxlen = ext2fs_blocks_count(&jsuper);
500                 journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
501                 start++;
502         }
503
504         if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
505                 retval = EXT2_ET_NO_MEMORY;
506                 goto errout;
507         }
508
509         journal->j_sb_buffer = bh;
510         journal->j_superblock = (journal_superblock_t *)bh->b_data;
511
512 #ifdef USE_INODE_IO
513         if (j_inode)
514                 ext2fs_free_mem(&j_inode);
515 #endif
516
517         *ret_journal = journal;
518         e2fsck_use_inode_shortcuts(ctx, 0);
519         return 0;
520
521 errout:
522         e2fsck_use_inode_shortcuts(ctx, 0);
523         if (dev_fs)
524                 ext2fs_free_mem(&dev_fs);
525         if (j_inode)
526                 ext2fs_free_mem(&j_inode);
527         if (journal)
528                 ext2fs_free_mem(&journal);
529         return retval;
530 }
531
532 static errcode_t e2fsck_journal_fix_bad_inode(e2fsck_t ctx,
533                                               struct problem_context *pctx)
534 {
535         struct ext2_super_block *sb = ctx->fs->super;
536         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
537         int has_journal = ext2fs_has_feature_journal(ctx->fs->super);
538
539         if (has_journal || sb->s_journal_inum) {
540                 /* The journal inode is bogus, remove and force full fsck */
541                 pctx->ino = sb->s_journal_inum;
542                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_INODE, pctx)) {
543                         if (has_journal && sb->s_journal_inum)
544                                 printf("*** journal has been deleted ***\n\n");
545                         ext2fs_clear_feature_journal(sb);
546                         sb->s_journal_inum = 0;
547                         memset(sb->s_jnl_blocks, 0, sizeof(sb->s_jnl_blocks));
548                         ctx->flags |= E2F_FLAG_JOURNAL_INODE;
549                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
550                         e2fsck_clear_recover(ctx, 1);
551                         return 0;
552                 }
553                 return EXT2_ET_BAD_INODE_NUM;
554         } else if (recover) {
555                 if (fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, pctx)) {
556                         e2fsck_clear_recover(ctx, 1);
557                         return 0;
558                 }
559                 return EXT2_ET_UNSUPP_FEATURE;
560         }
561         return 0;
562 }
563
564 #define V1_SB_SIZE      0x0024
565 static void clear_v2_journal_fields(journal_t *journal)
566 {
567         e2fsck_t ctx = journal->j_dev->k_ctx;
568         struct problem_context pctx;
569
570         clear_problem_context(&pctx);
571
572         if (!fix_problem(ctx, PR_0_CLEAR_V2_JOURNAL, &pctx))
573                 return;
574
575         memset(((char *) journal->j_superblock) + V1_SB_SIZE, 0,
576                ctx->fs->blocksize-V1_SB_SIZE);
577         mark_buffer_dirty(journal->j_sb_buffer);
578 }
579
580
581 static errcode_t e2fsck_journal_load(journal_t *journal)
582 {
583         e2fsck_t ctx = journal->j_dev->k_ctx;
584         journal_superblock_t *jsb;
585         struct buffer_head *jbh = journal->j_sb_buffer;
586         struct problem_context pctx;
587
588         clear_problem_context(&pctx);
589
590         ll_rw_block(READ, 1, &jbh);
591         if (jbh->b_err) {
592                 com_err(ctx->device_name, jbh->b_err, "%s",
593                         _("reading journal superblock\n"));
594                 return jbh->b_err;
595         }
596
597         jsb = journal->j_superblock;
598         /* If we don't even have JFS_MAGIC, we probably have a wrong inode */
599         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER))
600                 return e2fsck_journal_fix_bad_inode(ctx, &pctx);
601
602         switch (ntohl(jsb->s_header.h_blocktype)) {
603         case JFS_SUPERBLOCK_V1:
604                 journal->j_format_version = 1;
605                 if (jsb->s_feature_compat ||
606                     jsb->s_feature_incompat ||
607                     jsb->s_feature_ro_compat ||
608                     jsb->s_nr_users)
609                         clear_v2_journal_fields(journal);
610                 break;
611
612         case JFS_SUPERBLOCK_V2:
613                 journal->j_format_version = 2;
614                 if (ntohl(jsb->s_nr_users) > 1 &&
615                     uuid_is_null(ctx->fs->super->s_journal_uuid))
616                         clear_v2_journal_fields(journal);
617                 if (ntohl(jsb->s_nr_users) > 1) {
618                         fix_problem(ctx, PR_0_JOURNAL_UNSUPP_MULTIFS, &pctx);
619                         return EXT2_ET_JOURNAL_UNSUPP_VERSION;
620                 }
621                 break;
622
623         /*
624          * These should never appear in a journal super block, so if
625          * they do, the journal is badly corrupted.
626          */
627         case JFS_DESCRIPTOR_BLOCK:
628         case JFS_COMMIT_BLOCK:
629         case JFS_REVOKE_BLOCK:
630                 return EXT2_ET_CORRUPT_SUPERBLOCK;
631
632         /* If we don't understand the superblock major type, but there
633          * is a magic number, then it is likely to be a new format we
634          * just don't understand, so leave it alone. */
635         default:
636                 return EXT2_ET_JOURNAL_UNSUPP_VERSION;
637         }
638
639         if (JFS_HAS_INCOMPAT_FEATURE(journal, ~JFS_KNOWN_INCOMPAT_FEATURES))
640                 return EXT2_ET_UNSUPP_FEATURE;
641
642         if (JFS_HAS_RO_COMPAT_FEATURE(journal, ~JFS_KNOWN_ROCOMPAT_FEATURES))
643                 return EXT2_ET_RO_UNSUPP_FEATURE;
644
645         /* Checksum v1-3 are mutually exclusive features. */
646         if (jfs_has_feature_csum2(journal) && jfs_has_feature_csum3(journal))
647                 return EXT2_ET_CORRUPT_SUPERBLOCK;
648
649         if (journal_has_csum_v2or3(journal) &&
650             jfs_has_feature_checksum(journal))
651                 return EXT2_ET_CORRUPT_SUPERBLOCK;
652
653         if (!e2fsck_journal_verify_csum_type(journal, jsb) ||
654             !e2fsck_journal_sb_csum_verify(journal, jsb))
655                 return EXT2_ET_CORRUPT_SUPERBLOCK;
656
657         if (journal_has_csum_v2or3(journal))
658                 journal->j_csum_seed = jbd2_chksum(journal, ~0, jsb->s_uuid,
659                                                    sizeof(jsb->s_uuid));
660
661         /* We have now checked whether we know enough about the journal
662          * format to be able to proceed safely, so any other checks that
663          * fail we should attempt to recover from. */
664         if (jsb->s_blocksize != htonl(journal->j_blocksize)) {
665                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
666                         _("%s: no valid journal superblock found\n"),
667                         ctx->device_name);
668                 return EXT2_ET_CORRUPT_SUPERBLOCK;
669         }
670
671         if (ntohl(jsb->s_maxlen) < journal->j_maxlen)
672                 journal->j_maxlen = ntohl(jsb->s_maxlen);
673         else if (ntohl(jsb->s_maxlen) > journal->j_maxlen) {
674                 com_err(ctx->program_name, EXT2_ET_CORRUPT_SUPERBLOCK,
675                         _("%s: journal too short\n"),
676                         ctx->device_name);
677                 return EXT2_ET_CORRUPT_SUPERBLOCK;
678         }
679
680         journal->j_tail_sequence = ntohl(jsb->s_sequence);
681         journal->j_transaction_sequence = journal->j_tail_sequence;
682         journal->j_tail = ntohl(jsb->s_start);
683         journal->j_first = ntohl(jsb->s_first);
684         journal->j_last = ntohl(jsb->s_maxlen);
685
686         return 0;
687 }
688
689 static void e2fsck_journal_reset_super(e2fsck_t ctx, journal_superblock_t *jsb,
690                                        journal_t *journal)
691 {
692         char *p;
693         union {
694                 uuid_t uuid;
695                 __u32 val[4];
696         } u;
697         __u32 new_seq = 0;
698         int i;
699
700         /* Leave a valid existing V1 superblock signature alone.
701          * Anything unrecognisable we overwrite with a new V2
702          * signature. */
703
704         if (jsb->s_header.h_magic != htonl(JFS_MAGIC_NUMBER) ||
705             jsb->s_header.h_blocktype != htonl(JFS_SUPERBLOCK_V1)) {
706                 jsb->s_header.h_magic = htonl(JFS_MAGIC_NUMBER);
707                 jsb->s_header.h_blocktype = htonl(JFS_SUPERBLOCK_V2);
708         }
709
710         /* Zero out everything else beyond the superblock header */
711
712         p = ((char *) jsb) + sizeof(journal_header_t);
713         memset (p, 0, ctx->fs->blocksize-sizeof(journal_header_t));
714
715         jsb->s_blocksize = htonl(ctx->fs->blocksize);
716         jsb->s_maxlen = htonl(journal->j_maxlen);
717         jsb->s_first = htonl(1);
718
719         /* Initialize the journal sequence number so that there is "no"
720          * chance we will find old "valid" transactions in the journal.
721          * This avoids the need to zero the whole journal (slow to do,
722          * and risky when we are just recovering the filesystem).
723          */
724         uuid_generate(u.uuid);
725         for (i = 0; i < 4; i ++)
726                 new_seq ^= u.val[i];
727         jsb->s_sequence = htonl(new_seq);
728         e2fsck_journal_sb_csum_set(journal, jsb);
729
730         mark_buffer_dirty(journal->j_sb_buffer);
731         ll_rw_block(WRITE, 1, &journal->j_sb_buffer);
732 }
733
734 static errcode_t e2fsck_journal_fix_corrupt_super(e2fsck_t ctx,
735                                                   journal_t *journal,
736                                                   struct problem_context *pctx)
737 {
738         struct ext2_super_block *sb = ctx->fs->super;
739         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
740
741         if (ext2fs_has_feature_journal(sb)) {
742                 if (fix_problem(ctx, PR_0_JOURNAL_BAD_SUPER, pctx)) {
743                         e2fsck_journal_reset_super(ctx, journal->j_superblock,
744                                                    journal);
745                         journal->j_transaction_sequence = 1;
746                         e2fsck_clear_recover(ctx, recover);
747                         return 0;
748                 }
749                 return EXT2_ET_CORRUPT_SUPERBLOCK;
750         } else if (e2fsck_journal_fix_bad_inode(ctx, pctx))
751                 return EXT2_ET_CORRUPT_SUPERBLOCK;
752
753         return 0;
754 }
755
756 static void e2fsck_journal_release(e2fsck_t ctx, journal_t *journal,
757                                    int reset, int drop)
758 {
759         journal_superblock_t *jsb;
760
761         if (drop)
762                 mark_buffer_clean(journal->j_sb_buffer);
763         else if (!(ctx->options & E2F_OPT_READONLY)) {
764                 jsb = journal->j_superblock;
765                 jsb->s_sequence = htonl(journal->j_tail_sequence);
766                 if (reset)
767                         jsb->s_start = 0; /* this marks the journal as empty */
768                 e2fsck_journal_sb_csum_set(journal, jsb);
769                 mark_buffer_dirty(journal->j_sb_buffer);
770         }
771         brelse(journal->j_sb_buffer);
772
773         if (ctx->journal_io) {
774                 if (ctx->fs && ctx->fs->io != ctx->journal_io)
775                         io_channel_close(ctx->journal_io);
776                 ctx->journal_io = 0;
777         }
778
779 #ifndef USE_INODE_IO
780         if (journal->j_inode)
781                 ext2fs_free_mem(&journal->j_inode);
782 #endif
783         if (journal->j_fs_dev)
784                 ext2fs_free_mem(&journal->j_fs_dev);
785         ext2fs_free_mem(&journal);
786 }
787
788 /*
789  * This function makes sure that the superblock fields regarding the
790  * journal are consistent.
791  */
792 errcode_t e2fsck_check_ext3_journal(e2fsck_t ctx)
793 {
794         struct ext2_super_block *sb = ctx->fs->super;
795         journal_t *journal;
796         int recover = ext2fs_has_feature_journal_needs_recovery(ctx->fs->super);
797         struct problem_context pctx;
798         problem_t problem;
799         int reset = 0, force_fsck = 0;
800         errcode_t retval;
801
802         /* If we don't have any journal features, don't do anything more */
803         if (!ext2fs_has_feature_journal(sb) &&
804             !recover && sb->s_journal_inum == 0 && sb->s_journal_dev == 0 &&
805             uuid_is_null(sb->s_journal_uuid))
806                 return 0;
807
808         clear_problem_context(&pctx);
809         pctx.num = sb->s_journal_inum;
810
811         retval = e2fsck_get_journal(ctx, &journal);
812         if (retval) {
813                 if ((retval == EXT2_ET_BAD_INODE_NUM) ||
814                     (retval == EXT2_ET_BAD_BLOCK_NUM) ||
815                     (retval == EXT2_ET_JOURNAL_TOO_SMALL) ||
816                     (retval == EXT2_ET_NO_JOURNAL))
817                         return e2fsck_journal_fix_bad_inode(ctx, &pctx);
818                 return retval;
819         }
820
821         retval = e2fsck_journal_load(journal);
822         if (retval) {
823                 if ((retval == EXT2_ET_CORRUPT_SUPERBLOCK) ||
824                     ((retval == EXT2_ET_UNSUPP_FEATURE) &&
825                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_INCOMPAT,
826                                   &pctx))) ||
827                     ((retval == EXT2_ET_RO_UNSUPP_FEATURE) &&
828                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_ROCOMPAT,
829                                   &pctx))) ||
830                     ((retval == EXT2_ET_JOURNAL_UNSUPP_VERSION) &&
831                     (!fix_problem(ctx, PR_0_JOURNAL_UNSUPP_VERSION, &pctx))))
832                         retval = e2fsck_journal_fix_corrupt_super(ctx, journal,
833                                                                   &pctx);
834                 e2fsck_journal_release(ctx, journal, 0, 1);
835                 return retval;
836         }
837
838         /*
839          * We want to make the flags consistent here.  We will not leave with
840          * needs_recovery set but has_journal clear.  We can't get in a loop
841          * with -y, -n, or -p, only if a user isn't making up their mind.
842          */
843 no_has_journal:
844         if (!ext2fs_has_feature_journal(sb)) {
845                 recover = ext2fs_has_feature_journal_needs_recovery(sb);
846                 if (fix_problem(ctx, PR_0_JOURNAL_HAS_JOURNAL, &pctx)) {
847                         if (recover &&
848                             !fix_problem(ctx, PR_0_JOURNAL_RECOVER_SET, &pctx))
849                                 goto no_has_journal;
850                         /*
851                          * Need a full fsck if we are releasing a
852                          * journal stored on a reserved inode.
853                          */
854                         force_fsck = recover ||
855                                 (sb->s_journal_inum < EXT2_FIRST_INODE(sb));
856                         /* Clear all of the journal fields */
857                         sb->s_journal_inum = 0;
858                         sb->s_journal_dev = 0;
859                         memset(sb->s_journal_uuid, 0,
860                                sizeof(sb->s_journal_uuid));
861                         e2fsck_clear_recover(ctx, force_fsck);
862                 } else if (!(ctx->options & E2F_OPT_READONLY)) {
863                         ext2fs_set_feature_journal(sb);
864                         ctx->fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
865                         ext2fs_mark_super_dirty(ctx->fs);
866                 }
867         }
868
869         if (ext2fs_has_feature_journal(sb) &&
870             !ext2fs_has_feature_journal_needs_recovery(sb) &&
871             journal->j_superblock->s_start != 0) {
872                 /* Print status information */
873                 fix_problem(ctx, PR_0_JOURNAL_RECOVERY_CLEAR, &pctx);
874                 if (ctx->superblock)
875                         problem = PR_0_JOURNAL_RUN_DEFAULT;
876                 else
877                         problem = PR_0_JOURNAL_RUN;
878                 if (fix_problem(ctx, problem, &pctx)) {
879                         ctx->options |= E2F_OPT_FORCE;
880                         ext2fs_set_feature_journal_needs_recovery(sb);
881                         ext2fs_mark_super_dirty(ctx->fs);
882                 } else if (fix_problem(ctx,
883                                        PR_0_JOURNAL_RESET_JOURNAL, &pctx)) {
884                         reset = 1;
885                         sb->s_state &= ~EXT2_VALID_FS;
886                         ext2fs_mark_super_dirty(ctx->fs);
887                 }
888                 /*
889                  * If the user answers no to the above question, we
890                  * ignore the fact that journal apparently has data;
891                  * accidentally replaying over valid data would be far
892                  * worse than skipping a questionable recovery.
893                  *
894                  * XXX should we abort with a fatal error here?  What
895                  * will the ext3 kernel code do if a filesystem with
896                  * !NEEDS_RECOVERY but with a non-zero
897                  * journal->j_superblock->s_start is mounted?
898                  */
899         }
900
901         /*
902          * If we don't need to do replay the journal, check to see if
903          * the journal's errno is set; if so, we need to mark the file
904          * system as being corrupt and clear the journal's s_errno.
905          */
906         if (!ext2fs_has_feature_journal_needs_recovery(sb) &&
907             journal->j_superblock->s_errno) {
908                 ctx->fs->super->s_state |= EXT2_ERROR_FS;
909                 ext2fs_mark_super_dirty(ctx->fs);
910                 journal->j_superblock->s_errno = 0;
911                 e2fsck_journal_sb_csum_set(journal, journal->j_superblock);
912                 mark_buffer_dirty(journal->j_sb_buffer);
913         }
914
915         e2fsck_journal_release(ctx, journal, reset, 0);
916         return retval;
917 }
918
919 static errcode_t recover_ext3_journal(e2fsck_t ctx)
920 {
921         struct problem_context  pctx;
922         journal_t *journal;
923         errcode_t retval;
924
925         clear_problem_context(&pctx);
926
927         journal_init_revoke_caches();
928         retval = e2fsck_get_journal(ctx, &journal);
929         if (retval)
930                 return retval;
931
932         retval = e2fsck_journal_load(journal);
933         if (retval)
934                 goto errout;
935
936         retval = journal_init_revoke(journal, 1024);
937         if (retval)
938                 goto errout;
939
940         retval = -journal_recover(journal);
941         if (retval)
942                 goto errout;
943
944         if (journal->j_failed_commit) {
945                 pctx.ino = journal->j_failed_commit;
946                 fix_problem(ctx, PR_0_JNL_TXN_CORRUPT, &pctx);
947                 journal->j_superblock->s_errno = -EINVAL;
948                 mark_buffer_dirty(journal->j_sb_buffer);
949         }
950
951 errout:
952         journal_destroy_revoke(journal);
953         journal_destroy_revoke_caches();
954         e2fsck_journal_release(ctx, journal, 1, 0);
955         return retval;
956 }
957
958 errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
959 {
960         io_manager io_ptr = ctx->fs->io->manager;
961         int blocksize = ctx->fs->blocksize;
962         errcode_t       retval, recover_retval;
963         io_stats        stats = 0;
964         unsigned long long kbytes_written = 0;
965
966         printf(_("%s: recovering journal\n"), ctx->device_name);
967         if (ctx->options & E2F_OPT_READONLY) {
968                 printf(_("%s: won't do journal recovery while read-only\n"),
969                        ctx->device_name);
970                 return EXT2_ET_FILE_RO;
971         }
972
973         if (ctx->fs->flags & EXT2_FLAG_DIRTY)
974                 ext2fs_flush(ctx->fs);  /* Force out any modifications */
975
976         recover_retval = recover_ext3_journal(ctx);
977
978         /*
979          * Reload the filesystem context to get up-to-date data from disk
980          * because journal recovery will change the filesystem under us.
981          */
982         if (ctx->fs->super->s_kbytes_written &&
983             ctx->fs->io->manager->get_stats)
984                 ctx->fs->io->manager->get_stats(ctx->fs->io, &stats);
985         if (stats && stats->bytes_written)
986                 kbytes_written = stats->bytes_written >> 10;
987
988         ext2fs_mmp_stop(ctx->fs);
989         ext2fs_free(ctx->fs);
990         retval = ext2fs_open(ctx->filesystem_name, ctx->openfs_flags,
991                              ctx->superblock, blocksize, io_ptr,
992                              &ctx->fs);
993         if (retval) {
994                 com_err(ctx->program_name, retval,
995                         _("while trying to re-open %s"),
996                         ctx->device_name);
997                 fatal_error(ctx, 0);
998         }
999         ctx->fs->priv_data = ctx;
1000         ctx->fs->now = ctx->now;
1001         ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1002         ctx->fs->super->s_kbytes_written += kbytes_written;
1003
1004         /* Set the superblock flags */
1005         e2fsck_clear_recover(ctx, recover_retval != 0);
1006
1007         /*
1008          * Do one last sanity check, and propagate journal->s_errno to
1009          * the EXT2_ERROR_FS flag in the fs superblock if needed.
1010          */
1011         retval = e2fsck_check_ext3_journal(ctx);
1012         return retval ? retval : recover_retval;
1013 }
1014
1015 /*
1016  * This function will move the journal inode from a visible file in
1017  * the filesystem directory hierarchy to the reserved inode if necessary.
1018  */
1019 static const char * const journal_names[] = {
1020         ".journal", "journal", ".journal.dat", "journal.dat", 0 };
1021
1022 void e2fsck_move_ext3_journal(e2fsck_t ctx)
1023 {
1024         struct ext2_super_block *sb = ctx->fs->super;
1025         struct problem_context  pctx;
1026         struct ext2_inode       inode;
1027         ext2_filsys             fs = ctx->fs;
1028         ext2_ino_t              ino;
1029         errcode_t               retval;
1030         const char * const *    cpp;
1031         dgrp_t                  group;
1032         int                     mount_flags;
1033
1034         clear_problem_context(&pctx);
1035
1036         /*
1037          * If the filesystem is opened read-only, or there is no
1038          * journal, then do nothing.
1039          */
1040         if ((ctx->options & E2F_OPT_READONLY) ||
1041             (sb->s_journal_inum == 0) ||
1042             !ext2fs_has_feature_journal(sb))
1043                 return;
1044
1045         /*
1046          * Read in the journal inode
1047          */
1048         if (ext2fs_read_inode(fs, sb->s_journal_inum, &inode) != 0)
1049                 return;
1050
1051         /*
1052          * If it's necessary to backup the journal inode, do so.
1053          */
1054         if ((sb->s_jnl_backup_type == 0) ||
1055             ((sb->s_jnl_backup_type == EXT3_JNL_BACKUP_BLOCKS) &&
1056              memcmp(inode.i_block, sb->s_jnl_blocks, EXT2_N_BLOCKS*4))) {
1057                 if (fix_problem(ctx, PR_0_BACKUP_JNL, &pctx)) {
1058                         memcpy(sb->s_jnl_blocks, inode.i_block,
1059                                EXT2_N_BLOCKS*4);
1060                         sb->s_jnl_blocks[15] = inode.i_size_high;
1061                         sb->s_jnl_blocks[16] = inode.i_size;
1062                         sb->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
1063                         ext2fs_mark_super_dirty(fs);
1064                         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1065                 }
1066         }
1067
1068         /*
1069          * If the journal is already the hidden inode, then do nothing
1070          */
1071         if (sb->s_journal_inum == EXT2_JOURNAL_INO)
1072                 return;
1073
1074         /*
1075          * The journal inode had better have only one link and not be readable.
1076          */
1077         if (inode.i_links_count != 1)
1078                 return;
1079
1080         /*
1081          * If the filesystem is mounted, or we can't tell whether
1082          * or not it's mounted, do nothing.
1083          */
1084         retval = ext2fs_check_if_mounted(ctx->filesystem_name, &mount_flags);
1085         if (retval || (mount_flags & EXT2_MF_MOUNTED))
1086                 return;
1087
1088         /*
1089          * If we can't find the name of the journal inode, then do
1090          * nothing.
1091          */
1092         for (cpp = journal_names; *cpp; cpp++) {
1093                 retval = ext2fs_lookup(fs, EXT2_ROOT_INO, *cpp,
1094                                        strlen(*cpp), 0, &ino);
1095                 if ((retval == 0) && (ino == sb->s_journal_inum))
1096                         break;
1097         }
1098         if (*cpp == 0)
1099                 return;
1100
1101         /* We need the inode bitmap to be loaded */
1102         retval = ext2fs_read_bitmaps(fs);
1103         if (retval)
1104                 return;
1105
1106         pctx.str = *cpp;
1107         if (!fix_problem(ctx, PR_0_MOVE_JOURNAL, &pctx))
1108                 return;
1109
1110         /*
1111          * OK, we've done all the checks, let's actually move the
1112          * journal inode.  Errors at this point mean we need to force
1113          * an ext2 filesystem check.
1114          */
1115         if ((retval = ext2fs_unlink(fs, EXT2_ROOT_INO, *cpp, ino, 0)) != 0)
1116                 goto err_out;
1117         if ((retval = ext2fs_write_inode(fs, EXT2_JOURNAL_INO, &inode)) != 0)
1118                 goto err_out;
1119         sb->s_journal_inum = EXT2_JOURNAL_INO;
1120         ext2fs_mark_super_dirty(fs);
1121         fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
1122         inode.i_links_count = 0;
1123         inode.i_dtime = ctx->now;
1124         if ((retval = ext2fs_write_inode(fs, ino, &inode)) != 0)
1125                 goto err_out;
1126
1127         group = ext2fs_group_of_ino(fs, ino);
1128         ext2fs_unmark_inode_bitmap2(fs->inode_map, ino);
1129         ext2fs_mark_ib_dirty(fs);
1130         ext2fs_bg_free_inodes_count_set(fs, group, ext2fs_bg_free_inodes_count(fs, group) + 1);
1131         ext2fs_group_desc_csum_set(fs, group);
1132         fs->super->s_free_inodes_count++;
1133         return;
1134
1135 err_out:
1136         pctx.errcode = retval;
1137         fix_problem(ctx, PR_0_ERR_MOVE_JOURNAL, &pctx);
1138         fs->super->s_state &= ~EXT2_VALID_FS;
1139         ext2fs_mark_super_dirty(fs);
1140         return;
1141 }
1142
1143 /*
1144  * This function makes sure the superblock hint for the external
1145  * journal is correct.
1146  */
1147 int e2fsck_fix_ext3_journal_hint(e2fsck_t ctx)
1148 {
1149         struct ext2_super_block *sb = ctx->fs->super;
1150         struct problem_context pctx;
1151         char uuid[37], *journal_name;
1152         struct stat st;
1153
1154         if (!ext2fs_has_feature_journal(sb) ||
1155             uuid_is_null(sb->s_journal_uuid))
1156                 return 0;
1157
1158         uuid_unparse(sb->s_journal_uuid, uuid);
1159         journal_name = blkid_get_devname(ctx->blkid, "UUID", uuid);
1160         if (!journal_name)
1161                 return 0;
1162
1163         if (stat(journal_name, &st) < 0) {
1164                 free(journal_name);
1165                 return 0;
1166         }
1167
1168         if (st.st_rdev != sb->s_journal_dev) {
1169                 clear_problem_context(&pctx);
1170                 pctx.num = st.st_rdev;
1171                 if (fix_problem(ctx, PR_0_EXTERNAL_JOURNAL_HINT, &pctx)) {
1172                         sb->s_journal_dev = st.st_rdev;
1173                         ext2fs_mark_super_dirty(ctx->fs);
1174                 }
1175         }
1176
1177         free(journal_name);
1178         return 0;
1179 }