4bbb5a078bbb8c781ee9f29794af03b1c02a219f
[linux-2.6.git] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/config.h>
20 #include <linux/module.h>
21 #include <linux/string.h>
22 #include <linux/fs.h>
23 #include <linux/time.h>
24 #include <linux/jbd.h>
25 #include <linux/ext3_fs.h>
26 #include <linux/ext3_jbd.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
29 #include <linux/blkdev.h>
30 #include <linux/parser.h>
31 #include <linux/smp_lock.h>
32 #include <linux/buffer_head.h>
33 #include <linux/vfs.h>
34 #include <linux/random.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/quotaops.h>
38 #include <asm/uaccess.h>
39 #include "xattr.h"
40 #include "acl.h"
41
42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
44                                int);
45 static void ext3_commit_super (struct super_block * sb,
46                                struct ext3_super_block * es,
47                                int sync);
48 static void ext3_mark_recovery_complete(struct super_block * sb,
49                                         struct ext3_super_block * es);
50 static void ext3_clear_journal_err(struct super_block * sb,
51                                    struct ext3_super_block * es);
52 static int ext3_sync_fs(struct super_block *sb, int wait);
53
54 /* 
55  * Wrappers for journal_start/end.
56  *
57  * The only special thing we need to do here is to make sure that all
58  * journal_end calls result in the superblock being marked dirty, so
59  * that sync() will call the filesystem's write_super callback if
60  * appropriate. 
61  */
62 handle_t *ext3_journal_start(struct inode *inode, int nblocks)
63 {
64         journal_t *journal;
65
66         if (inode->i_sb->s_flags & MS_RDONLY)
67                 return ERR_PTR(-EROFS);
68
69         /* Special case here: if the journal has aborted behind our
70          * backs (eg. EIO in the commit thread), then we still need to
71          * take the FS itself readonly cleanly. */
72         journal = EXT3_JOURNAL(inode);
73         if (is_journal_aborted(journal)) {
74                 ext3_abort(inode->i_sb, __FUNCTION__,
75                            "Detected aborted journal");
76                 return ERR_PTR(-EROFS);
77         }
78
79         return journal_start(journal, nblocks);
80 }
81
82 /* 
83  * The only special thing we need to do here is to make sure that all
84  * journal_stop calls result in the superblock being marked dirty, so
85  * that sync() will call the filesystem's write_super callback if
86  * appropriate. 
87  */
88 int __ext3_journal_stop(const char *where, handle_t *handle)
89 {
90         struct super_block *sb;
91         int err;
92         int rc;
93
94         sb = handle->h_transaction->t_journal->j_private;
95         err = handle->h_err;
96         rc = journal_stop(handle);
97
98         if (!err)
99                 err = rc;
100         if (err)
101                 __ext3_std_error(sb, where, err);
102         return err;
103 }
104
105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
106                 struct buffer_head *bh, handle_t *handle, int err)
107 {
108         char nbuf[16];
109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
110
111         printk(KERN_ERR "%s: aborting transaction: %s in %s", 
112                caller, errstr, err_fn);
113
114         if (bh)
115                 BUFFER_TRACE(bh, "abort");
116         journal_abort_handle(handle);
117         if (!handle->h_err)
118                 handle->h_err = err;
119 }
120
121 static char error_buf[1024];
122
123 /* Deal with the reporting of failure conditions on a filesystem such as
124  * inconsistencies detected or read IO failures.
125  *
126  * On ext2, we can store the error state of the filesystem in the
127  * superblock.  That is not possible on ext3, because we may have other
128  * write ordering constraints on the superblock which prevent us from
129  * writing it out straight away; and given that the journal is about to
130  * be aborted, we can't rely on the current, or future, transactions to
131  * write out the superblock safely.
132  *
133  * We'll just use the journal_abort() error code to record an error in
134  * the journal instead.  On recovery, the journal will compain about
135  * that error until we've noted it down and cleared it.
136  */
137
138 static void ext3_handle_error(struct super_block *sb)
139 {
140         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
141
142         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
143         es->s_state |= cpu_to_le32(EXT3_ERROR_FS);
144
145         if (sb->s_flags & MS_RDONLY)
146                 return;
147
148         if (test_opt (sb, ERRORS_PANIC))
149                 panic ("EXT3-fs (device %s): panic forced after error\n",
150                        sb->s_id);
151         if (test_opt (sb, ERRORS_RO)) {
152                 printk (KERN_CRIT "Remounting filesystem read-only\n");
153                 sb->s_flags |= MS_RDONLY;
154         } else {
155                 journal_t *journal = EXT3_SB(sb)->s_journal;
156
157                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
158                 if (journal)
159                         journal_abort(journal, -EIO);
160         }
161         ext3_commit_super(sb, es, 1);
162 }
163
164 void ext3_error (struct super_block * sb, const char * function,
165                  const char * fmt, ...)
166 {
167         va_list args;
168
169         va_start (args, fmt);
170         vsprintf (error_buf, fmt, args);
171         va_end (args);
172
173         printk (KERN_CRIT "EXT3-fs error (device %s): %s: %s\n",
174                 sb->s_id, function, error_buf);
175
176         ext3_handle_error(sb);
177 }
178
179 const char *ext3_decode_error(struct super_block * sb, int errno, char nbuf[16])
180 {
181         char *errstr = NULL;
182
183         switch (errno) {
184         case -EIO:
185                 errstr = "IO failure";
186                 break;
187         case -ENOMEM:
188                 errstr = "Out of memory";
189                 break;
190         case -EROFS:
191                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
192                         errstr = "Journal has aborted";
193                 else
194                         errstr = "Readonly filesystem";
195                 break;
196         default:
197                 /* If the caller passed in an extra buffer for unknown
198                  * errors, textualise them now.  Else we just return
199                  * NULL. */
200                 if (nbuf) {
201                         /* Check for truncated error codes... */
202                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
203                                 errstr = nbuf;
204                 }
205                 break;
206         }
207
208         return errstr;
209 }
210
211 /* __ext3_std_error decodes expected errors from journaling functions
212  * automatically and invokes the appropriate error response.  */
213
214 void __ext3_std_error (struct super_block * sb, const char * function,
215                        int errno)
216 {
217         char nbuf[16];
218         const char *errstr = ext3_decode_error(sb, errno, nbuf);
219
220         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
221                 sb->s_id, function, errstr);
222
223         ext3_handle_error(sb);
224 }
225
226 /*
227  * ext3_abort is a much stronger failure handler than ext3_error.  The
228  * abort function may be used to deal with unrecoverable failures such
229  * as journal IO errors or ENOMEM at a critical moment in log management.
230  *
231  * We unconditionally force the filesystem into an ABORT|READONLY state,
232  * unless the error response on the fs has been set to panic in which
233  * case we take the easy way out and panic immediately.
234  */
235
236 void ext3_abort (struct super_block * sb, const char * function,
237                  const char * fmt, ...)
238 {
239         va_list args;
240
241         printk (KERN_CRIT "ext3_abort called.\n");
242
243         va_start (args, fmt);
244         vsprintf (error_buf, fmt, args);
245         va_end (args);
246
247         if (test_opt (sb, ERRORS_PANIC))
248                 panic ("EXT3-fs panic (device %s): %s: %s\n",
249                        sb->s_id, function, error_buf);
250
251         printk (KERN_CRIT "EXT3-fs abort (device %s): %s: %s\n",
252                 sb->s_id, function, error_buf);
253
254         if (sb->s_flags & MS_RDONLY)
255                 return;
256
257         printk (KERN_CRIT "Remounting filesystem read-only\n");
258         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
259         sb->s_flags |= MS_RDONLY;
260         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
261         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
262 }
263
264 /* Deal with the reporting of failure conditions while running, such as
265  * inconsistencies in operation or invalid system states.
266  *
267  * Use ext3_error() for cases of invalid filesystem states, as that will
268  * record an error on disk and force a filesystem check on the next boot.
269  */
270 NORET_TYPE void ext3_panic (struct super_block * sb, const char * function,
271                             const char * fmt, ...)
272 {
273         va_list args;
274
275         va_start (args, fmt);
276         vsprintf (error_buf, fmt, args);
277         va_end (args);
278
279         /* this is to prevent panic from syncing this filesystem */
280         /* AKPM: is this sufficient? */
281         sb->s_flags |= MS_RDONLY;
282         panic ("EXT3-fs panic (device %s): %s: %s\n",
283                sb->s_id, function, error_buf);
284 }
285
286 void ext3_warning (struct super_block * sb, const char * function,
287                    const char * fmt, ...)
288 {
289         va_list args;
290
291         va_start (args, fmt);
292         vsprintf (error_buf, fmt, args);
293         va_end (args);
294         printk (KERN_WARNING "EXT3-fs warning (device %s): %s: %s\n",
295                 sb->s_id, function, error_buf);
296 }
297
298 void ext3_update_dynamic_rev(struct super_block *sb)
299 {
300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
301
302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
303                 return;
304
305         ext3_warning(sb, __FUNCTION__,
306                      "updating to rev %d because of new feature flag, "
307                      "running e2fsck is recommended",
308                      EXT3_DYNAMIC_REV);
309
310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
313         /* leave es->s_feature_*compat flags alone */
314         /* es->s_uuid will be set by e2fsck if empty */
315
316         /*
317          * The rest of the superblock fields should be zero, and if not it
318          * means they are likely already in use, so leave them alone.  We
319          * can leave it up to e2fsck to clean up any inconsistencies there.
320          */
321 }
322
323 /*
324  * Open the external journal device
325  */
326 static struct block_device *ext3_blkdev_get(dev_t dev)
327 {
328         struct block_device *bdev;
329         char b[BDEVNAME_SIZE];
330
331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
332         if (IS_ERR(bdev))
333                 goto fail;
334         return bdev;
335
336 fail:
337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
338                         __bdevname(dev, b), PTR_ERR(bdev));
339         return NULL;
340 }
341
342 /*
343  * Release the journal device
344  */
345 static int ext3_blkdev_put(struct block_device *bdev)
346 {
347         bd_release(bdev);
348         return blkdev_put(bdev);
349 }
350
351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
352 {
353         struct block_device *bdev;
354         int ret = -ENODEV;
355
356         bdev = sbi->journal_bdev;
357         if (bdev) {
358                 ret = ext3_blkdev_put(bdev);
359                 sbi->journal_bdev = 0;
360         }
361         return ret;
362 }
363
364 static inline struct inode *orphan_list_entry(struct list_head *l)
365 {
366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
367 }
368
369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
370 {
371         struct list_head *l;
372
373         printk(KERN_ERR "sb orphan head is %d\n", 
374                le32_to_cpu(sbi->s_es->s_last_orphan));
375
376         printk(KERN_ERR "sb_info orphan list:\n");
377         list_for_each(l, &sbi->s_orphan) {
378                 struct inode *inode = orphan_list_entry(l);
379                 printk(KERN_ERR "  "
380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
381                        inode->i_sb->s_id, inode->i_ino, inode,
382                        inode->i_mode, inode->i_nlink, 
383                        le32_to_cpu(NEXT_ORPHAN(inode)));
384         }
385 }
386
387 void ext3_put_super (struct super_block * sb)
388 {
389         struct ext3_sb_info *sbi = EXT3_SB(sb);
390         struct ext3_super_block *es = sbi->s_es;
391         int i;
392
393         ext3_xattr_put_super(sb);
394         journal_destroy(sbi->s_journal);
395         if (!(sb->s_flags & MS_RDONLY)) {
396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
397                 es->s_state = le16_to_cpu(sbi->s_mount_state);
398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
399                 mark_buffer_dirty(sbi->s_sbh);
400                 ext3_commit_super(sb, es, 1);
401         }
402
403         for (i = 0; i < sbi->s_gdb_count; i++)
404                 brelse(sbi->s_group_desc[i]);
405         kfree(sbi->s_group_desc);
406         kfree(sbi->s_debts);
407         brelse(sbi->s_sbh);
408 #ifdef CONFIG_QUOTA
409         for (i = 0; i < MAXQUOTAS; i++) {
410                 if (sbi->s_qf_names[i])
411                         kfree(sbi->s_qf_names[i]);
412         }
413 #endif
414
415         /* Debugging code just in case the in-memory inode orphan list
416          * isn't empty.  The on-disk one can be non-empty if we've
417          * detected an error and taken the fs readonly, but the
418          * in-memory list had better be clean by this point. */
419         if (!list_empty(&sbi->s_orphan))
420                 dump_orphan_list(sb, sbi);
421         J_ASSERT(list_empty(&sbi->s_orphan));
422
423         invalidate_bdev(sb->s_bdev, 0);
424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
425                 /*
426                  * Invalidate the journal device's buffers.  We don't want them
427                  * floating about in memory - the physical journal device may
428                  * hotswapped, and it breaks the `ro-after' testing code.
429                  */
430                 sync_blockdev(sbi->journal_bdev);
431                 invalidate_bdev(sbi->journal_bdev, 0);
432                 ext3_blkdev_remove(sbi);
433         }
434         sb->s_fs_info = NULL;
435         kfree(sbi);
436         return;
437 }
438
439 static kmem_cache_t *ext3_inode_cachep;
440
441 /*
442  * Called inside transaction, so use GFP_NOFS
443  */
444 static struct inode *ext3_alloc_inode(struct super_block *sb)
445 {
446         struct ext3_inode_info *ei;
447
448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
449         if (!ei)
450                 return NULL;
451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
452         ei->i_acl = EXT3_ACL_NOT_CACHED;
453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
454 #endif
455         ei->vfs_inode.i_version = 1;
456         return &ei->vfs_inode;
457 }
458
459 static void ext3_destroy_inode(struct inode *inode)
460 {
461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
462 }
463
464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
465 {
466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
467
468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
469             SLAB_CTOR_CONSTRUCTOR) {
470                 INIT_LIST_HEAD(&ei->i_orphan);
471 #ifdef CONFIG_EXT3_FS_XATTR
472                 init_rwsem(&ei->xattr_sem);
473 #endif
474                 init_MUTEX(&ei->truncate_sem);
475                 inode_init_once(&ei->vfs_inode);
476         }
477 }
478  
479 static int init_inodecache(void)
480 {
481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
482                                              sizeof(struct ext3_inode_info),
483                                              0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
484                                              init_once, NULL);
485         if (ext3_inode_cachep == NULL)
486                 return -ENOMEM;
487         return 0;
488 }
489
490 static void destroy_inodecache(void)
491 {
492         if (kmem_cache_destroy(ext3_inode_cachep))
493                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
494 }
495
496 #ifdef CONFIG_EXT3_FS_POSIX_ACL
497
498 static void ext3_clear_inode(struct inode *inode)
499 {
500        if (EXT3_I(inode)->i_acl &&
501            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
502                posix_acl_release(EXT3_I(inode)->i_acl);
503                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
504        }
505        if (EXT3_I(inode)->i_default_acl &&
506            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
507                posix_acl_release(EXT3_I(inode)->i_default_acl);
508                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
509        }
510 }
511
512 #else
513 # define ext3_clear_inode NULL
514 #endif
515
516 #ifdef CONFIG_QUOTA
517
518 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
519 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
520
521 static int ext3_dquot_initialize(struct inode *inode, int type);
522 static int ext3_dquot_drop(struct inode *inode);
523 static int ext3_write_dquot(struct dquot *dquot);
524 static int ext3_mark_dquot_dirty(struct dquot *dquot);
525 static int ext3_write_info(struct super_block *sb, int type);
526 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
527 static int ext3_quota_on_mount(struct super_block *sb, int type);
528 static int ext3_quota_off_mount(struct super_block *sb, int type);
529
530 static struct dquot_operations ext3_quota_operations = {
531         .initialize     = ext3_dquot_initialize,
532         .drop           = ext3_dquot_drop,
533         .alloc_space    = dquot_alloc_space,
534         .alloc_inode    = dquot_alloc_inode,
535         .free_space     = dquot_free_space,
536         .free_inode     = dquot_free_inode,
537         .transfer       = dquot_transfer,
538         .write_dquot    = ext3_write_dquot,
539         .mark_dirty     = ext3_mark_dquot_dirty,
540         .write_info     = ext3_write_info
541 };
542
543 static struct quotactl_ops ext3_qctl_operations = {
544         .quota_on       = ext3_quota_on,
545         .quota_off      = vfs_quota_off,
546         .quota_sync     = vfs_quota_sync,
547         .get_info       = vfs_get_dqinfo,
548         .set_info       = vfs_set_dqinfo,
549         .get_dqblk      = vfs_get_dqblk,
550         .set_dqblk      = vfs_set_dqblk
551 };
552 #endif
553
554 static struct super_operations ext3_sops = {
555         .alloc_inode    = ext3_alloc_inode,
556         .destroy_inode  = ext3_destroy_inode,
557         .read_inode     = ext3_read_inode,
558         .write_inode    = ext3_write_inode,
559         .dirty_inode    = ext3_dirty_inode,
560         .put_inode      = ext3_put_inode,
561         .delete_inode   = ext3_delete_inode,
562         .put_super      = ext3_put_super,
563         .write_super    = ext3_write_super,
564         .sync_fs        = ext3_sync_fs,
565         .write_super_lockfs = ext3_write_super_lockfs,
566         .unlockfs       = ext3_unlockfs,
567         .statfs         = ext3_statfs,
568         .remount_fs     = ext3_remount,
569         .clear_inode    = ext3_clear_inode,
570 };
571
572 struct dentry *ext3_get_parent(struct dentry *child);
573 static struct export_operations ext3_export_ops = {
574         .get_parent = ext3_get_parent,
575 };
576
577 enum {
578         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
579         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
580         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
581         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl, Opt_noload,
582         Opt_commit, Opt_journal_update, Opt_journal_inum,
583         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
584         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
585         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
586         Opt_ignore, Opt_err,
587 };
588
589 static match_table_t tokens = {
590         {Opt_bsd_df, "bsddf"},
591         {Opt_minix_df, "minixdf"},
592         {Opt_grpid, "grpid"},
593         {Opt_grpid, "bsdgroups"},
594         {Opt_nogrpid, "nogrpid"},
595         {Opt_nogrpid, "sysvgroups"},
596         {Opt_resgid, "resgid=%u"},
597         {Opt_resuid, "resuid=%u"},
598         {Opt_sb, "sb=%u"},
599         {Opt_err_cont, "errors=continue"},
600         {Opt_err_panic, "errors=panic"},
601         {Opt_err_ro, "errors=remount-ro"},
602         {Opt_nouid32, "nouid32"},
603         {Opt_nocheck, "nocheck"},
604         {Opt_nocheck, "check=none"},
605         {Opt_check, "check"},
606         {Opt_debug, "debug"},
607         {Opt_oldalloc, "oldalloc"},
608         {Opt_orlov, "orlov"},
609         {Opt_user_xattr, "user_xattr"},
610         {Opt_nouser_xattr, "nouser_xattr"},
611         {Opt_acl, "acl"},
612         {Opt_noacl, "noacl"},
613         {Opt_noload, "noload"},
614         {Opt_commit, "commit=%u"},
615         {Opt_journal_update, "journal=update"},
616         {Opt_journal_inum, "journal=%u"},
617         {Opt_abort, "abort"},
618         {Opt_data_journal, "data=journal"},
619         {Opt_data_ordered, "data=ordered"},
620         {Opt_data_writeback, "data=writeback"},
621         {Opt_offusrjquota, "usrjquota="},
622         {Opt_usrjquota, "usrjquota=%s"},
623         {Opt_offgrpjquota, "grpjquota="},
624         {Opt_grpjquota, "grpjquota=%s"},
625         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
626         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
627         {Opt_ignore, "grpquota"},
628         {Opt_ignore, "noquota"},
629         {Opt_ignore, "quota"},
630         {Opt_ignore, "usrquota"},
631         {Opt_err, NULL}
632 };
633
634 static unsigned long get_sb_block(void **data)
635 {
636         unsigned long   sb_block;
637         char            *options = (char *) *data;
638
639         if (!options || strncmp(options, "sb=", 3) != 0)
640                 return 1;       /* Default location */
641         options += 3;
642         sb_block = simple_strtoul(options, &options, 0);
643         if (*options && *options != ',') {
644                 printk("EXT3-fs: Invalid sb specification: %s\n",
645                        (char *) *data);
646                 return 1;
647         }
648         if (*options == ',')
649                 options++;
650         *data = (void *) options;
651         return sb_block;
652 }
653
654 static int parse_options (char * options, struct super_block *sb,
655                           unsigned long * inum, int is_remount)
656 {
657         struct ext3_sb_info *sbi = EXT3_SB(sb);
658         char * p;
659         substring_t args[MAX_OPT_ARGS];
660         int data_opt = 0;
661         int option;
662 #ifdef CONFIG_QUOTA
663         int qtype;
664 #endif
665
666         if (!options)
667                 return 1;
668
669         while ((p = strsep (&options, ",")) != NULL) {
670                 int token;
671                 if (!*p)
672                         continue;
673
674                 token = match_token(p, tokens, args);
675                 switch (token) {
676                 case Opt_bsd_df:
677                         clear_opt (sbi->s_mount_opt, MINIX_DF);
678                         break;
679                 case Opt_minix_df:
680                         set_opt (sbi->s_mount_opt, MINIX_DF);
681                         break;
682                 case Opt_grpid:
683                         set_opt (sbi->s_mount_opt, GRPID);
684                         break;
685                 case Opt_nogrpid:
686                         clear_opt (sbi->s_mount_opt, GRPID);
687                         break;
688                 case Opt_resuid:
689                         if (match_int(&args[0], &option))
690                                 return 0;
691                         sbi->s_resuid = option;
692                         break;
693                 case Opt_resgid:
694                         if (match_int(&args[0], &option))
695                                 return 0;
696                         sbi->s_resgid = option;
697                         break;
698                 case Opt_sb:
699                         /* handled by get_sb_block() instead of here */
700                         /* *sb_block = match_int(&args[0]); */
701                         break;
702                 case Opt_err_panic:
703                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
704                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
705                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
706                         break;
707                 case Opt_err_ro:
708                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
709                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
710                         set_opt (sbi->s_mount_opt, ERRORS_RO);
711                         break;
712                 case Opt_err_cont:
713                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
714                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
715                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
716                         break;
717                 case Opt_nouid32:
718                         set_opt (sbi->s_mount_opt, NO_UID32);
719                         break;
720                 case Opt_check:
721 #ifdef CONFIG_EXT3_CHECK
722                         set_opt (sbi->s_mount_opt, CHECK);
723 #else
724                         printk(KERN_ERR
725                                "EXT3 Check option not supported\n");
726 #endif
727                         break;
728                 case Opt_nocheck:
729                         clear_opt (sbi->s_mount_opt, CHECK);
730                         break;
731                 case Opt_debug:
732                         set_opt (sbi->s_mount_opt, DEBUG);
733                         break;
734                 case Opt_oldalloc:
735                         set_opt (sbi->s_mount_opt, OLDALLOC);
736                         break;
737                 case Opt_orlov:
738                         clear_opt (sbi->s_mount_opt, OLDALLOC);
739                         break;
740 #ifdef CONFIG_EXT3_FS_XATTR
741                 case Opt_user_xattr:
742                         set_opt (sbi->s_mount_opt, XATTR_USER);
743                         break;
744                 case Opt_nouser_xattr:
745                         clear_opt (sbi->s_mount_opt, XATTR_USER);
746                         break;
747 #else
748                 case Opt_user_xattr:
749                 case Opt_nouser_xattr:
750                         printk("EXT3 (no)user_xattr options not supported\n");
751                         break;
752 #endif
753 #ifdef CONFIG_EXT3_FS_POSIX_ACL
754                 case Opt_acl:
755                         set_opt(sbi->s_mount_opt, POSIX_ACL);
756                         break;
757                 case Opt_noacl:
758                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
759                         break;
760 #else
761                 case Opt_acl:
762                 case Opt_noacl:
763                         printk("EXT3 (no)acl options not supported\n");
764                         break;
765 #endif
766                 case Opt_journal_update:
767                         /* @@@ FIXME */
768                         /* Eventually we will want to be able to create
769                            a journal file here.  For now, only allow the
770                            user to specify an existing inode to be the
771                            journal file. */
772                         if (is_remount) {
773                                 printk(KERN_ERR "EXT3-fs: cannot specify "
774                                        "journal on remount\n");
775                                 return 0;
776                         }
777                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
778                         break;
779                 case Opt_journal_inum:
780                         if (is_remount) {
781                                 printk(KERN_ERR "EXT3-fs: cannot specify "
782                                        "journal on remount\n");
783                                 return 0;
784                         }
785                         if (match_int(&args[0], &option))
786                                 return 0;
787                         *inum = option;
788                         break;
789                 case Opt_noload:
790                         set_opt (sbi->s_mount_opt, NOLOAD);
791                         break;
792                 case Opt_commit:
793                         if (match_int(&args[0], &option))
794                                 return 0;
795                         if (option < 0)
796                                 return 0;
797                         if (option == 0)
798                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
799                         sbi->s_commit_interval = HZ * option;
800                         break;
801                 case Opt_data_journal:
802                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
803                         goto datacheck;
804                 case Opt_data_ordered:
805                         data_opt = EXT3_MOUNT_ORDERED_DATA;
806                         goto datacheck;
807                 case Opt_data_writeback:
808                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
809                 datacheck:
810                         if (is_remount) {
811                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
812                                                 != data_opt) {
813                                         printk(KERN_ERR
814                                                 "EXT3-fs: cannot change data "
815                                                 "mode on remount\n");
816                                         return 0;
817                                 }
818                         } else {
819                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
820                                 sbi->s_mount_opt |= data_opt;
821                         }
822                         break;
823 #ifdef CONFIG_QUOTA
824                 case Opt_usrjquota:
825                         qtype = USRQUOTA;
826                         goto set_qf_name;
827                 case Opt_grpjquota:
828                         qtype = GRPQUOTA;
829 set_qf_name:
830                         if (sb_any_quota_enabled(sb)) {
831                                 printk(KERN_ERR
832                                         "EXT3-fs: Cannot change journalled "
833                                         "quota options when quota turned on.\n");
834                                 return 0;
835                         }
836                         if (sbi->s_qf_names[qtype]) {
837                                 printk(KERN_ERR
838                                         "EXT3-fs: %s quota file already "
839                                         "specified.\n", QTYPE2NAME(qtype));
840                                 return 0;
841                         }
842                         sbi->s_qf_names[qtype] = match_strdup(&args[0]);
843                         if (!sbi->s_qf_names[qtype]) {
844                                 printk(KERN_ERR
845                                         "EXT3-fs: not enough memory for "
846                                         "storing quotafile name.\n");
847                                 return 0;
848                         }
849                         if (strchr(sbi->s_qf_names[qtype], '/')) {
850                                 printk(KERN_ERR
851                                         "EXT3-fs: quotafile must be on "
852                                         "filesystem root.\n");
853                                 kfree(sbi->s_qf_names[qtype]);
854                                 sbi->s_qf_names[qtype] = NULL;
855                                 return 0;
856                         }
857                         break;
858                 case Opt_offusrjquota:
859                         qtype = USRQUOTA;
860                         goto clear_qf_name;
861                 case Opt_offgrpjquota:
862                         qtype = GRPQUOTA;
863 clear_qf_name:
864                         if (sb_any_quota_enabled(sb)) {
865                                 printk(KERN_ERR "EXT3-fs: Cannot change "
866                                         "journalled quota options when "
867                                         "quota turned on.\n");
868                                 return 0;
869                         }
870                         if (sbi->s_qf_names[qtype]) {
871                                 kfree(sbi->s_qf_names[qtype]);
872                                 sbi->s_qf_names[qtype] = NULL;
873                         }
874                         break;
875                 case Opt_jqfmt_vfsold:
876                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
877                         break;
878                 case Opt_jqfmt_vfsv0:
879                         sbi->s_jquota_fmt = QFMT_VFS_V0;
880                         break;
881 #else
882                 case Opt_usrjquota:
883                 case Opt_grpjquota:
884                 case Opt_offusrjquota:
885                 case Opt_offgrpjquota:
886                 case Opt_jqfmt_vfsold:
887                 case Opt_jqfmt_vfsv0:
888                         printk(KERN_ERR
889                                 "EXT3-fs: journalled quota options not "
890                                 "supported.\n");
891                         break;
892 #endif
893                 case Opt_abort:
894                         set_opt(sbi->s_mount_opt, ABORT);
895                         break;
896                 case Opt_ignore:
897                         break;
898                 default:
899                         printk (KERN_ERR
900                                 "EXT3-fs: Unrecognized mount option \"%s\" "
901                                 "or missing value\n", p);
902                         return 0;
903                 }
904         }
905 #ifdef CONFIG_QUOTA
906         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
907             sbi->s_qf_names[GRPQUOTA])) {
908                 printk(KERN_ERR
909                         "EXT3-fs: journalled quota format not specified.\n");
910                 return 0;
911         }
912 #endif
913
914         return 1;
915 }
916
917 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
918                             int read_only)
919 {
920         struct ext3_sb_info *sbi = EXT3_SB(sb);
921         int res = 0;
922
923         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
924                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
925                         "forcing read-only mode\n");
926                 res = MS_RDONLY;
927         }
928         if (read_only)
929                 return res;
930         if (!(sbi->s_mount_state & EXT3_VALID_FS))
931                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
932                         "running e2fsck is recommended\n");
933         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
934                 printk (KERN_WARNING
935                         "EXT3-fs warning: mounting fs with errors, "
936                         "running e2fsck is recommended\n");
937         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
938                  le16_to_cpu(es->s_mnt_count) >=
939                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
940                 printk (KERN_WARNING
941                         "EXT3-fs warning: maximal mount count reached, "
942                         "running e2fsck is recommended\n");
943         else if (le32_to_cpu(es->s_checkinterval) &&
944                 (le32_to_cpu(es->s_lastcheck) +
945                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
946                 printk (KERN_WARNING
947                         "EXT3-fs warning: checktime reached, "
948                         "running e2fsck is recommended\n");
949 #if 0
950                 /* @@@ We _will_ want to clear the valid bit if we find
951                    inconsistencies, to force a fsck at reboot.  But for
952                    a plain journaled filesystem we can keep it set as
953                    valid forever! :) */
954         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
955 #endif
956         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
957                 es->s_max_mnt_count =
958                         (__s16) cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
959         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
960         es->s_mtime = cpu_to_le32(get_seconds());
961         ext3_update_dynamic_rev(sb);
962         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
963
964         ext3_commit_super(sb, es, 1);
965         if (test_opt(sb, DEBUG))
966                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
967                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
968                         sb->s_blocksize,
969                         sbi->s_groups_count,
970                         EXT3_BLOCKS_PER_GROUP(sb),
971                         EXT3_INODES_PER_GROUP(sb),
972                         sbi->s_mount_opt);
973
974         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
975         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
976                 char b[BDEVNAME_SIZE];
977
978                 printk("external journal on %s\n",
979                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
980         } else {
981                 printk("internal journal\n");
982         }
983 #ifdef CONFIG_EXT3_CHECK
984         if (test_opt (sb, CHECK)) {
985                 ext3_check_blocks_bitmap (sb);
986                 ext3_check_inodes_bitmap (sb);
987         }
988 #endif
989         return res;
990 }
991
992 static int ext3_check_descriptors (struct super_block * sb)
993 {
994         struct ext3_sb_info *sbi = EXT3_SB(sb);
995         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
996         struct ext3_group_desc * gdp = NULL;
997         int desc_block = 0;
998         int i;
999
1000         ext3_debug ("Checking group descriptors");
1001
1002         for (i = 0; i < sbi->s_groups_count; i++)
1003         {
1004                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1005                         gdp = (struct ext3_group_desc *)
1006                                         sbi->s_group_desc[desc_block++]->b_data;
1007                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1008                     le32_to_cpu(gdp->bg_block_bitmap) >=
1009                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1010                 {
1011                         ext3_error (sb, "ext3_check_descriptors",
1012                                     "Block bitmap for group %d"
1013                                     " not in group (block %lu)!",
1014                                     i, (unsigned long)
1015                                         le32_to_cpu(gdp->bg_block_bitmap));
1016                         return 0;
1017                 }
1018                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1019                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1020                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1021                 {
1022                         ext3_error (sb, "ext3_check_descriptors",
1023                                     "Inode bitmap for group %d"
1024                                     " not in group (block %lu)!",
1025                                     i, (unsigned long)
1026                                         le32_to_cpu(gdp->bg_inode_bitmap));
1027                         return 0;
1028                 }
1029                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1030                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1031                     block + EXT3_BLOCKS_PER_GROUP(sb))
1032                 {
1033                         ext3_error (sb, "ext3_check_descriptors",
1034                                     "Inode table for group %d"
1035                                     " not in group (block %lu)!",
1036                                     i, (unsigned long)
1037                                         le32_to_cpu(gdp->bg_inode_table));
1038                         return 0;
1039                 }
1040                 block += EXT3_BLOCKS_PER_GROUP(sb);
1041                 gdp++;
1042         }
1043
1044         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1045         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1046         return 1;
1047 }
1048
1049
1050 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1051  * the superblock) which were deleted from all directories, but held open by
1052  * a process at the time of a crash.  We walk the list and try to delete these
1053  * inodes at recovery time (only with a read-write filesystem).
1054  *
1055  * In order to keep the orphan inode chain consistent during traversal (in
1056  * case of crash during recovery), we link each inode into the superblock
1057  * orphan list_head and handle it the same way as an inode deletion during
1058  * normal operation (which journals the operations for us).
1059  *
1060  * We only do an iget() and an iput() on each inode, which is very safe if we
1061  * accidentally point at an in-use or already deleted inode.  The worst that
1062  * can happen in this case is that we get a "bit already cleared" message from
1063  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1064  * e2fsck was run on this filesystem, and it must have already done the orphan
1065  * inode cleanup for us, so we can safely abort without any further action.
1066  */
1067 static void ext3_orphan_cleanup (struct super_block * sb,
1068                                  struct ext3_super_block * es)
1069 {
1070         unsigned int s_flags = sb->s_flags;
1071         int nr_orphans = 0, nr_truncates = 0;
1072 #ifdef CONFIG_QUOTA
1073         int i;
1074 #endif
1075         if (!es->s_last_orphan) {
1076                 jbd_debug(4, "no orphan inodes to clean up\n");
1077                 return;
1078         }
1079
1080         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1081                 if (es->s_last_orphan)
1082                         jbd_debug(1, "Errors on filesystem, "
1083                                   "clearing orphan list.\n");
1084                 es->s_last_orphan = 0;
1085                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1086                 return;
1087         }
1088
1089         if (s_flags & MS_RDONLY) {
1090                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1091                        sb->s_id);
1092                 sb->s_flags &= ~MS_RDONLY;
1093         }
1094 #ifdef CONFIG_QUOTA
1095         /* Needed for iput() to work correctly and not trash data */
1096         sb->s_flags |= MS_ACTIVE;
1097         /* Turn on quotas so that they are updated correctly */
1098         for (i = 0; i < MAXQUOTAS; i++) {
1099                 if (EXT3_SB(sb)->s_qf_names[i]) {
1100                         int ret = ext3_quota_on_mount(sb, i);
1101                         if (ret < 0)
1102                                 printk(KERN_ERR
1103                                         "EXT3-fs: Cannot turn on journalled "
1104                                         "quota: error %d\n", ret);
1105                 }
1106         }
1107 #endif
1108
1109         while (es->s_last_orphan) {
1110                 struct inode *inode;
1111
1112                 if (!(inode =
1113                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1114                         es->s_last_orphan = 0;
1115                         break;
1116                 }
1117
1118                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1119                 DQUOT_INIT(inode);
1120                 if (inode->i_nlink) {
1121                         printk(KERN_DEBUG
1122                                 "%s: truncating inode %ld to %Ld bytes\n",
1123                                 __FUNCTION__, inode->i_ino, inode->i_size);
1124                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1125                                   inode->i_ino, inode->i_size);
1126                         ext3_truncate(inode);
1127                         nr_truncates++;
1128                 } else {
1129                         printk(KERN_DEBUG
1130                                 "%s: deleting unreferenced inode %ld\n",
1131                                 __FUNCTION__, inode->i_ino);
1132                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1133                                   inode->i_ino);
1134                         nr_orphans++;
1135                 }
1136                 iput(inode);  /* The delete magic happens here! */
1137         }
1138
1139 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1140
1141         if (nr_orphans)
1142                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1143                        sb->s_id, PLURAL(nr_orphans));
1144         if (nr_truncates)
1145                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1146                        sb->s_id, PLURAL(nr_truncates));
1147 #ifdef CONFIG_QUOTA
1148         /* Turn quotas off */
1149         for (i = 0; i < MAXQUOTAS; i++) {
1150                 if (sb_dqopt(sb)->files[i])
1151                         ext3_quota_off_mount(sb, i);
1152         }
1153 #endif
1154         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1155 }
1156
1157 #define log2(n) ffz(~(n))
1158
1159 /*
1160  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1161  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1162  * We need to be 1 filesystem block less than the 2^32 sector limit.
1163  */
1164 static loff_t ext3_max_size(int bits)
1165 {
1166         loff_t res = EXT3_NDIR_BLOCKS;
1167         res += 1LL << (bits-2);
1168         res += 1LL << (2*(bits-2));
1169         res += 1LL << (3*(bits-2));
1170         res <<= bits;
1171         if (res > (512LL << 32) - (1 << bits))
1172                 res = (512LL << 32) - (1 << bits);
1173         return res;
1174 }
1175
1176 static unsigned long descriptor_loc(struct super_block *sb,
1177                                     unsigned long logic_sb_block,
1178                                     int nr)
1179 {
1180         struct ext3_sb_info *sbi = EXT3_SB(sb);
1181         unsigned long bg, first_data_block, first_meta_bg;
1182         int has_super = 0;
1183
1184         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1185         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1186
1187         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1188             nr < first_meta_bg)
1189                 return (logic_sb_block + nr + 1);
1190         bg = sbi->s_desc_per_block * nr;
1191         if (ext3_bg_has_super(sb, bg))
1192                 has_super = 1;
1193         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1194 }
1195
1196
1197 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1198 {
1199         struct buffer_head * bh;
1200         struct ext3_super_block *es = 0;
1201         struct ext3_sb_info *sbi;
1202         unsigned long block;
1203         unsigned long sb_block = get_sb_block(&data);
1204         unsigned long logic_sb_block;
1205         unsigned long offset = 0;
1206         unsigned long journal_inum = 0;
1207         unsigned long def_mount_opts;
1208         struct inode *root;
1209         int blocksize;
1210         int hblock;
1211         int db_count;
1212         int i;
1213         int needs_recovery;
1214
1215         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1216         if (!sbi)
1217                 return -ENOMEM;
1218         sb->s_fs_info = sbi;
1219         memset(sbi, 0, sizeof(*sbi));
1220         sbi->s_mount_opt = 0;
1221         sbi->s_resuid = EXT3_DEF_RESUID;
1222         sbi->s_resgid = EXT3_DEF_RESGID;
1223
1224         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1225         if (!blocksize) {
1226                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1227                 goto out_fail;
1228         }
1229
1230         /*
1231          * The ext3 superblock will not be buffer aligned for other than 1kB
1232          * block sizes.  We need to calculate the offset from buffer start.
1233          */
1234         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1235                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1236                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1237         } else {
1238                 logic_sb_block = sb_block;
1239         }
1240
1241         if (!(bh = sb_bread(sb, logic_sb_block))) {
1242                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1243                 goto out_fail;
1244         }
1245         /*
1246          * Note: s_es must be initialized as soon as possible because
1247          *       some ext3 macro-instructions depend on its value
1248          */
1249         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1250         sbi->s_es = es;
1251         sb->s_magic = le16_to_cpu(es->s_magic);
1252         if (sb->s_magic != EXT3_SUPER_MAGIC) {
1253                 if (!silent)
1254                         printk(KERN_ERR 
1255                                "VFS: Can't find ext3 filesystem on dev %s.\n",
1256                                sb->s_id);
1257                 goto failed_mount;
1258         }
1259
1260         /* Set defaults before we parse the mount options */
1261         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1262         if (def_mount_opts & EXT3_DEFM_DEBUG)
1263                 set_opt(sbi->s_mount_opt, DEBUG);
1264         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1265                 set_opt(sbi->s_mount_opt, GRPID);
1266         if (def_mount_opts & EXT3_DEFM_UID16)
1267                 set_opt(sbi->s_mount_opt, NO_UID32);
1268         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1269                 set_opt(sbi->s_mount_opt, XATTR_USER);
1270         if (def_mount_opts & EXT3_DEFM_ACL)
1271                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1272         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1273                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1274         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1275                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1276         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1277                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1278
1279         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1280                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1281         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1282                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1283
1284         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1285         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1286
1287         if (!parse_options ((char *) data, sb, &journal_inum, 0))
1288                 goto failed_mount;
1289
1290         sb->s_flags |= MS_ONE_SECOND;
1291         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1292                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1293
1294         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1295             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1296              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1297              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1298                 printk(KERN_WARNING 
1299                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1300                        "running e2fsck is recommended\n");
1301         /*
1302          * Check feature flags regardless of the revision level, since we
1303          * previously didn't change the revision level when setting the flags,
1304          * so there is a chance incompat flags are set on a rev 0 filesystem.
1305          */
1306         if ((i = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))) {
1307                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1308                        "unsupported optional features (%x).\n",
1309                        sb->s_id, i);
1310                 goto failed_mount;
1311         }
1312         if (!(sb->s_flags & MS_RDONLY) &&
1313             (i = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))){
1314                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1315                        "unsupported optional features (%x).\n",
1316                        sb->s_id, i);
1317                 goto failed_mount;
1318         }
1319         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1320
1321         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1322             blocksize > EXT3_MAX_BLOCK_SIZE) {
1323                 printk(KERN_ERR 
1324                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1325                        blocksize, sb->s_id);
1326                 goto failed_mount;
1327         }
1328
1329         hblock = bdev_hardsect_size(sb->s_bdev);
1330         if (sb->s_blocksize != blocksize) {
1331                 /*
1332                  * Make sure the blocksize for the filesystem is larger
1333                  * than the hardware sectorsize for the machine.
1334                  */
1335                 if (blocksize < hblock) {
1336                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1337                                "device blocksize %d.\n", blocksize, hblock);
1338                         goto failed_mount;
1339                 }
1340
1341                 brelse (bh);
1342                 sb_set_blocksize(sb, blocksize);
1343                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1344                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1345                 bh = sb_bread(sb, logic_sb_block);
1346                 if (!bh) {
1347                         printk(KERN_ERR 
1348                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1349                         goto failed_mount;
1350                 }
1351                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1352                 sbi->s_es = es;
1353                 if (es->s_magic != le16_to_cpu(EXT3_SUPER_MAGIC)) {
1354                         printk (KERN_ERR 
1355                                 "EXT3-fs: Magic mismatch, very weird !\n");
1356                         goto failed_mount;
1357                 }
1358         }
1359
1360         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1361
1362         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1363                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1364                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1365         } else {
1366                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1367                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1368                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1369                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1370                     (sbi->s_inode_size > blocksize)) {
1371                         printk (KERN_ERR
1372                                 "EXT3-fs: unsupported inode size: %d\n",
1373                                 sbi->s_inode_size);
1374                         goto failed_mount;
1375                 }
1376         }
1377         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1378                                    le32_to_cpu(es->s_log_frag_size);
1379         if (blocksize != sbi->s_frag_size) {
1380                 printk(KERN_ERR
1381                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1382                        sbi->s_frag_size, blocksize);
1383                 goto failed_mount;
1384         }
1385         sbi->s_frags_per_block = 1;
1386         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1387         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1388         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1389         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1390         sbi->s_itb_per_group = sbi->s_inodes_per_group /sbi->s_inodes_per_block;
1391         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1392         sbi->s_sbh = bh;
1393         sbi->s_mount_state = le16_to_cpu(es->s_state);
1394         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1395         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1396         for (i=0; i < 4; i++)
1397                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1398         sbi->s_def_hash_version = es->s_def_hash_version;
1399
1400         if (sbi->s_blocks_per_group > blocksize * 8) {
1401                 printk (KERN_ERR
1402                         "EXT3-fs: #blocks per group too big: %lu\n",
1403                         sbi->s_blocks_per_group);
1404                 goto failed_mount;
1405         }
1406         if (sbi->s_frags_per_group > blocksize * 8) {
1407                 printk (KERN_ERR
1408                         "EXT3-fs: #fragments per group too big: %lu\n",
1409                         sbi->s_frags_per_group);
1410                 goto failed_mount;
1411         }
1412         if (sbi->s_inodes_per_group > blocksize * 8) {
1413                 printk (KERN_ERR
1414                         "EXT3-fs: #inodes per group too big: %lu\n",
1415                         sbi->s_inodes_per_group);
1416                 goto failed_mount;
1417         }
1418
1419         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1420                                le32_to_cpu(es->s_first_data_block) +
1421                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1422                               EXT3_BLOCKS_PER_GROUP(sb);
1423         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1424                    EXT3_DESC_PER_BLOCK(sb);
1425         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1426                                     GFP_KERNEL);
1427         if (sbi->s_group_desc == NULL) {
1428                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1429                 goto failed_mount;
1430         }
1431         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(u8),
1432                         GFP_KERNEL);
1433         if (!sbi->s_debts) {
1434                 printk("EXT3-fs: not enough memory to allocate s_bgi\n");
1435                 goto failed_mount2;
1436         }
1437         memset(sbi->s_debts, 0,  sbi->s_groups_count * sizeof(u8));
1438
1439         percpu_counter_init(&sbi->s_freeblocks_counter);
1440         percpu_counter_init(&sbi->s_freeinodes_counter);
1441         percpu_counter_init(&sbi->s_dirs_counter);
1442         bgl_lock_init(&sbi->s_blockgroup_lock);
1443
1444         for (i = 0; i < db_count; i++) {
1445                 block = descriptor_loc(sb, logic_sb_block, i);
1446                 sbi->s_group_desc[i] = sb_bread(sb, block);
1447                 if (!sbi->s_group_desc[i]) {
1448                         printk (KERN_ERR "EXT3-fs: "
1449                                 "can't read group descriptor %d\n", i);
1450                         db_count = i;
1451                         goto failed_mount2;
1452                 }
1453         }
1454         if (!ext3_check_descriptors (sb)) {
1455                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1456                 goto failed_mount2;
1457         }
1458         sbi->s_gdb_count = db_count;
1459         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1460         spin_lock_init(&sbi->s_next_gen_lock);
1461         /*
1462          * set up enough so that it can read an inode
1463          */
1464         sb->s_op = &ext3_sops;
1465         sb->s_export_op = &ext3_export_ops;
1466 #ifdef CONFIG_QUOTA
1467         sb->s_qcop = &ext3_qctl_operations;
1468         sb->dq_op = &ext3_quota_operations;
1469 #endif
1470         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1471
1472         sb->s_root = 0;
1473
1474         needs_recovery = (es->s_last_orphan != 0 ||
1475                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1476                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1477
1478         /*
1479          * The first inode we look at is the journal inode.  Don't try
1480          * root first: it may be modified in the journal!
1481          */
1482         if (!test_opt(sb, NOLOAD) &&
1483             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1484                 if (ext3_load_journal(sb, es))
1485                         goto failed_mount2;
1486         } else if (journal_inum) {
1487                 if (ext3_create_journal(sb, es, journal_inum))
1488                         goto failed_mount2;
1489         } else {
1490                 if (!silent)
1491                         printk (KERN_ERR
1492                                 "ext3: No journal on filesystem on %s\n",
1493                                 sb->s_id);
1494                 goto failed_mount2;
1495         }
1496
1497         /* We have now updated the journal if required, so we can
1498          * validate the data journaling mode. */
1499         switch (test_opt(sb, DATA_FLAGS)) {
1500         case 0:
1501                 /* No mode set, assume a default based on the journal
1502                    capabilities: ORDERED_DATA if the journal can
1503                    cope, else JOURNAL_DATA */
1504                 if (journal_check_available_features
1505                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1506                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1507                 else
1508                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1509                 break;
1510
1511         case EXT3_MOUNT_ORDERED_DATA:
1512         case EXT3_MOUNT_WRITEBACK_DATA:
1513                 if (!journal_check_available_features
1514                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1515                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1516                                "requested data journaling mode\n");
1517                         goto failed_mount3;
1518                 }
1519         default:
1520                 break;
1521         }
1522
1523         /*
1524          * The journal_load will have done any necessary log recovery,
1525          * so we can safely mount the rest of the filesystem now.
1526          */
1527
1528         root = iget(sb, EXT3_ROOT_INO);
1529         sb->s_root = d_alloc_root(root);
1530         if (!sb->s_root) {
1531                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1532                 iput(root);
1533                 goto failed_mount3;
1534         }
1535         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1536                 dput(sb->s_root);
1537                 sb->s_root = NULL;
1538                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1539                 goto failed_mount3;
1540         }
1541
1542         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1543         /*
1544          * akpm: core read_super() calls in here with the superblock locked.
1545          * That deadlocks, because orphan cleanup needs to lock the superblock
1546          * in numerous places.  Here we just pop the lock - it's relatively
1547          * harmless, because we are now ready to accept write_super() requests,
1548          * and aviro says that's the only reason for hanging onto the
1549          * superblock lock.
1550          */
1551         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1552         ext3_orphan_cleanup(sb, es);
1553         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1554         if (needs_recovery)
1555                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1556         ext3_mark_recovery_complete(sb, es);
1557         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1558                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1559                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1560                 "writeback");
1561
1562         percpu_counter_mod(&sbi->s_freeblocks_counter,
1563                 ext3_count_free_blocks(sb));
1564         percpu_counter_mod(&sbi->s_freeinodes_counter,
1565                 ext3_count_free_inodes(sb));
1566         percpu_counter_mod(&sbi->s_dirs_counter,
1567                 ext3_count_dirs(sb));
1568
1569         return 0;
1570
1571 failed_mount3:
1572         journal_destroy(sbi->s_journal);
1573 failed_mount2:
1574         kfree(sbi->s_debts);
1575         for (i = 0; i < db_count; i++)
1576                 brelse(sbi->s_group_desc[i]);
1577         kfree(sbi->s_group_desc);
1578 failed_mount:
1579 #ifdef CONFIG_QUOTA
1580         for (i = 0; i < MAXQUOTAS; i++) {
1581                 if (sbi->s_qf_names[i])
1582                         kfree(sbi->s_qf_names[i]);
1583         }
1584 #endif
1585         ext3_blkdev_remove(sbi);
1586         brelse(bh);
1587 out_fail:
1588         sb->s_fs_info = NULL;
1589         kfree(sbi);
1590         return -EINVAL;
1591 }
1592
1593 /*
1594  * Setup any per-fs journal parameters now.  We'll do this both on
1595  * initial mount, once the journal has been initialised but before we've
1596  * done any recovery; and again on any subsequent remount. 
1597  */
1598 static void ext3_init_journal_params(struct ext3_sb_info *sbi, 
1599                                      journal_t *journal)
1600 {
1601         if (sbi->s_commit_interval)
1602                 journal->j_commit_interval = sbi->s_commit_interval;
1603         /* We could also set up an ext3-specific default for the commit
1604          * interval here, but for now we'll just fall back to the jbd
1605          * default. */
1606 }
1607
1608
1609 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1610 {
1611         struct inode *journal_inode;
1612         journal_t *journal;
1613
1614         /* First, test for the existence of a valid inode on disk.  Bad
1615          * things happen if we iget() an unused inode, as the subsequent
1616          * iput() will try to delete it. */
1617
1618         journal_inode = iget(sb, journal_inum);
1619         if (!journal_inode) {
1620                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1621                 return NULL;
1622         }
1623         if (!journal_inode->i_nlink) {
1624                 make_bad_inode(journal_inode);
1625                 iput(journal_inode);
1626                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1627                 return NULL;
1628         }
1629
1630         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1631                   journal_inode, journal_inode->i_size);
1632         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1633                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1634                 iput(journal_inode);
1635                 return NULL;
1636         }
1637
1638         journal = journal_init_inode(journal_inode);
1639         if (!journal) {
1640                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1641                 iput(journal_inode);
1642         }
1643         journal->j_private = sb;
1644         ext3_init_journal_params(EXT3_SB(sb), journal);
1645         return journal;
1646 }
1647
1648 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1649                                        dev_t j_dev)
1650 {
1651         struct buffer_head * bh;
1652         journal_t *journal;
1653         int start;
1654         int len;
1655         int hblock, blocksize;
1656         unsigned long sb_block;
1657         unsigned long offset;
1658         struct ext3_super_block * es;
1659         struct block_device *bdev;
1660
1661         bdev = ext3_blkdev_get(j_dev);
1662         if (bdev == NULL)
1663                 return NULL;
1664
1665         if (bd_claim(bdev, sb)) {
1666                 printk(KERN_ERR
1667                         "EXT3: failed to claim external journal device.\n");
1668                 blkdev_put(bdev);
1669                 return NULL;
1670         }
1671
1672         blocksize = sb->s_blocksize;
1673         hblock = bdev_hardsect_size(bdev);
1674         if (blocksize < hblock) {
1675                 printk(KERN_ERR
1676                         "EXT3-fs: blocksize too small for journal device.\n");
1677                 goto out_bdev;
1678         }
1679
1680         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1681         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1682         set_blocksize(bdev, blocksize);
1683         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1684                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1685                        "external journal\n");
1686                 goto out_bdev;
1687         }
1688
1689         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1690         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1691             !(le32_to_cpu(es->s_feature_incompat) &
1692               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1693                 printk(KERN_ERR "EXT3-fs: external journal has "
1694                                         "bad superblock\n");
1695                 brelse(bh);
1696                 goto out_bdev;
1697         }
1698
1699         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1700                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1701                 brelse(bh);
1702                 goto out_bdev;
1703         }
1704
1705         len = le32_to_cpu(es->s_blocks_count);
1706         start = sb_block + 1;
1707         brelse(bh);     /* we're done with the superblock */
1708
1709         journal = journal_init_dev(bdev, sb->s_bdev,
1710                                         start, len, blocksize);
1711         if (!journal) {
1712                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1713                 goto out_bdev;
1714         }
1715         journal->j_private = sb;
1716         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1717         wait_on_buffer(journal->j_sb_buffer);
1718         if (!buffer_uptodate(journal->j_sb_buffer)) {
1719                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1720                 goto out_journal;
1721         }
1722         if (ntohl(journal->j_superblock->s_nr_users) != 1) {
1723                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1724                                         "user (unsupported) - %d\n",
1725                         ntohl(journal->j_superblock->s_nr_users));
1726                 goto out_journal;
1727         }
1728         EXT3_SB(sb)->journal_bdev = bdev;
1729         ext3_init_journal_params(EXT3_SB(sb), journal);
1730         return journal;
1731 out_journal:
1732         journal_destroy(journal);
1733 out_bdev:
1734         ext3_blkdev_put(bdev);
1735         return NULL;
1736 }
1737
1738 static int ext3_load_journal(struct super_block * sb,
1739                              struct ext3_super_block * es)
1740 {
1741         journal_t *journal;
1742         int journal_inum = le32_to_cpu(es->s_journal_inum);
1743         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1744         int err = 0;
1745         int really_read_only;
1746
1747         really_read_only = bdev_read_only(sb->s_bdev);
1748
1749         /*
1750          * Are we loading a blank journal or performing recovery after a
1751          * crash?  For recovery, we need to check in advance whether we
1752          * can get read-write access to the device.
1753          */
1754
1755         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1756                 if (sb->s_flags & MS_RDONLY) {
1757                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1758                                         "required on readonly filesystem.\n");
1759                         if (really_read_only) {
1760                                 printk(KERN_ERR "EXT3-fs: write access "
1761                                         "unavailable, cannot proceed.\n");
1762                                 return -EROFS;
1763                         }
1764                         printk (KERN_INFO "EXT3-fs: write access will "
1765                                         "be enabled during recovery.\n");
1766                 }
1767         }
1768
1769         if (journal_inum && journal_dev) {
1770                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1771                        "and inode journals!\n");
1772                 return -EINVAL;
1773         }
1774
1775         if (journal_inum) {
1776                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1777                         return -EINVAL;
1778         } else {
1779                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1780                         return -EINVAL;
1781         }
1782
1783         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1784                 err = journal_update_format(journal);
1785                 if (err)  {
1786                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1787                         journal_destroy(journal);
1788                         return err;
1789                 }
1790         }
1791
1792         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1793                 err = journal_wipe(journal, !really_read_only);
1794         if (!err)
1795                 err = journal_load(journal);
1796
1797         if (err) {
1798                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1799                 journal_destroy(journal);
1800                 return err;
1801         }
1802
1803         EXT3_SB(sb)->s_journal = journal;
1804         ext3_clear_journal_err(sb, es);
1805         return 0;
1806 }
1807
1808 static int ext3_create_journal(struct super_block * sb,
1809                                struct ext3_super_block * es,
1810                                int journal_inum)
1811 {
1812         journal_t *journal;
1813
1814         if (sb->s_flags & MS_RDONLY) {
1815                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1816                                 "create journal.\n");
1817                 return -EROFS;
1818         }
1819
1820         if (!(journal = ext3_get_journal(sb, journal_inum)))
1821                 return -EINVAL;
1822
1823         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1824                journal_inum);
1825
1826         if (journal_create(journal)) {
1827                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1828                 journal_destroy(journal);
1829                 return -EIO;
1830         }
1831
1832         EXT3_SB(sb)->s_journal = journal;
1833
1834         ext3_update_dynamic_rev(sb);
1835         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1836         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1837
1838         es->s_journal_inum = cpu_to_le32(journal_inum);
1839         sb->s_dirt = 1;
1840
1841         /* Make sure we flush the recovery flag to disk. */
1842         ext3_commit_super(sb, es, 1);
1843
1844         return 0;
1845 }
1846
1847 static void ext3_commit_super (struct super_block * sb,
1848                                struct ext3_super_block * es,
1849                                int sync)
1850 {
1851         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1852
1853         if (!sbh)
1854                 return;
1855         es->s_wtime = cpu_to_le32(get_seconds());
1856         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1857         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1858         BUFFER_TRACE(sbh, "marking dirty");
1859         mark_buffer_dirty(sbh);
1860         if (sync)
1861                 sync_dirty_buffer(sbh);
1862 }
1863
1864
1865 /*
1866  * Have we just finished recovery?  If so, and if we are mounting (or
1867  * remounting) the filesystem readonly, then we will end up with a
1868  * consistent fs on disk.  Record that fact.
1869  */
1870 static void ext3_mark_recovery_complete(struct super_block * sb,
1871                                         struct ext3_super_block * es)
1872 {
1873         journal_flush(EXT3_SB(sb)->s_journal);
1874         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1875             sb->s_flags & MS_RDONLY) {
1876                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1877                 sb->s_dirt = 0;
1878                 ext3_commit_super(sb, es, 1);
1879         }
1880 }
1881
1882 /*
1883  * If we are mounting (or read-write remounting) a filesystem whose journal
1884  * has recorded an error from a previous lifetime, move that error to the
1885  * main filesystem now.
1886  */
1887 static void ext3_clear_journal_err(struct super_block * sb,
1888                                    struct ext3_super_block * es)
1889 {
1890         journal_t *journal;
1891         int j_errno;
1892         const char *errstr;
1893
1894         journal = EXT3_SB(sb)->s_journal;
1895
1896         /*
1897          * Now check for any error status which may have been recorded in the
1898          * journal by a prior ext3_error() or ext3_abort()
1899          */
1900
1901         j_errno = journal_errno(journal);
1902         if (j_errno) {
1903                 char nbuf[16];
1904
1905                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1906                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1907                              "from previous mount: %s", errstr);
1908                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1909                              "filesystem check.");
1910
1911                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1912                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1913                 ext3_commit_super (sb, es, 1);
1914
1915                 journal_clear_err(journal);
1916         }
1917 }
1918
1919 /*
1920  * Force the running and committing transactions to commit,
1921  * and wait on the commit.
1922  */
1923 int ext3_force_commit(struct super_block *sb)
1924 {
1925         journal_t *journal;
1926         int ret;
1927
1928         if (sb->s_flags & MS_RDONLY)
1929                 return 0;
1930
1931         journal = EXT3_SB(sb)->s_journal;
1932         sb->s_dirt = 0;
1933         ret = ext3_journal_force_commit(journal);
1934         return ret;
1935 }
1936
1937 /*
1938  * Ext3 always journals updates to the superblock itself, so we don't
1939  * have to propagate any other updates to the superblock on disk at this
1940  * point.  Just start an async writeback to get the buffers on their way
1941  * to the disk.
1942  *
1943  * This implicitly triggers the writebehind on sync().
1944  */
1945
1946 void ext3_write_super (struct super_block * sb)
1947 {
1948         if (down_trylock(&sb->s_lock) == 0)
1949                 BUG();
1950         sb->s_dirt = 0;
1951 }
1952
1953 static int ext3_sync_fs(struct super_block *sb, int wait)
1954 {
1955         tid_t target;
1956
1957         sb->s_dirt = 0;
1958         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
1959                 if (wait)
1960                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
1961         }
1962         return 0;
1963 }
1964
1965 /*
1966  * LVM calls this function before a (read-only) snapshot is created.  This
1967  * gives us a chance to flush the journal completely and mark the fs clean.
1968  */
1969 void ext3_write_super_lockfs(struct super_block *sb)
1970 {
1971         sb->s_dirt = 0;
1972
1973         if (!(sb->s_flags & MS_RDONLY)) {
1974                 journal_t *journal = EXT3_SB(sb)->s_journal;
1975
1976                 /* Now we set up the journal barrier. */
1977                 journal_lock_updates(journal);
1978                 journal_flush(journal);
1979
1980                 /* Journal blocked and flushed, clear needs_recovery flag. */
1981                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1982                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
1983         }
1984 }
1985
1986 /*
1987  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
1988  * flag here, even though the filesystem is not technically dirty yet.
1989  */
1990 void ext3_unlockfs(struct super_block *sb)
1991 {
1992         if (!(sb->s_flags & MS_RDONLY)) {
1993                 lock_super(sb);
1994                 /* Reser the needs_recovery flag before the fs is unlocked. */
1995                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1996                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
1997                 unlock_super(sb);
1998                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
1999         }
2000 }
2001
2002 int ext3_remount (struct super_block * sb, int * flags, char * data)
2003 {
2004         struct ext3_super_block * es;
2005         struct ext3_sb_info *sbi = EXT3_SB(sb);
2006         unsigned long tmp;
2007
2008         /*
2009          * Allow the "check" option to be passed as a remount option.
2010          */
2011         if (!parse_options(data, sb, &tmp, 1))
2012                 return -EINVAL;
2013
2014         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2015                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2016
2017         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2018                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2019
2020         es = sbi->s_es;
2021
2022         ext3_init_journal_params(sbi, sbi->s_journal);
2023
2024         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2025                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2026                         return -EROFS;
2027
2028                 if (*flags & MS_RDONLY) {
2029                         /*
2030                          * First of all, the unconditional stuff we have to do
2031                          * to disable replay of the journal when we next remount
2032                          */
2033                         sb->s_flags |= MS_RDONLY;
2034
2035                         /*
2036                          * OK, test if we are remounting a valid rw partition
2037                          * readonly, and if so set the rdonly flag and then
2038                          * mark the partition as valid again.
2039                          */
2040                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2041                             (sbi->s_mount_state & EXT3_VALID_FS))
2042                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2043
2044                         ext3_mark_recovery_complete(sb, es);
2045                 } else {
2046                         int ret;
2047                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2048                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2049                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2050                                        "remount RDWR because of unsupported "
2051                                        "optional features (%x).\n",
2052                                        sb->s_id, ret);
2053                                 return -EROFS;
2054                         }
2055                         /*
2056                          * Mounting a RDONLY partition read-write, so reread
2057                          * and store the current valid flag.  (It may have
2058                          * been changed by e2fsck since we originally mounted
2059                          * the partition.)
2060                          */
2061                         ext3_clear_journal_err(sb, es);
2062                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2063                         if (!ext3_setup_super (sb, es, 0))
2064                                 sb->s_flags &= ~MS_RDONLY;
2065                 }
2066         }
2067         return 0;
2068 }
2069
2070 int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2071 {
2072         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2073         unsigned long overhead;
2074         int i;
2075
2076         if (test_opt (sb, MINIX_DF))
2077                 overhead = 0;
2078         else {
2079                 /*
2080                  * Compute the overhead (FS structures)
2081                  */
2082
2083                 /*
2084                  * All of the blocks before first_data_block are
2085                  * overhead
2086                  */
2087                 overhead = le32_to_cpu(es->s_first_data_block);
2088
2089                 /*
2090                  * Add the overhead attributed to the superblock and
2091                  * block group descriptors.  If the sparse superblocks
2092                  * feature is turned on, then not all groups have this.
2093                  */
2094                 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++)
2095                         overhead += ext3_bg_has_super(sb, i) +
2096                                 ext3_bg_num_gdb(sb, i);
2097
2098                 /*
2099                  * Every block group has an inode bitmap, a block
2100                  * bitmap, and an inode table.
2101                  */
2102                 overhead += (EXT3_SB(sb)->s_groups_count *
2103                              (2 + EXT3_SB(sb)->s_itb_per_group));
2104         }
2105
2106         buf->f_type = EXT3_SUPER_MAGIC;
2107         buf->f_bsize = sb->s_blocksize;
2108         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2109         buf->f_bfree = ext3_count_free_blocks (sb);
2110         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2111         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2112                 buf->f_bavail = 0;
2113         buf->f_files = le32_to_cpu(es->s_inodes_count);
2114         buf->f_ffree = ext3_count_free_inodes (sb);
2115         buf->f_namelen = EXT3_NAME_LEN;
2116         return 0;
2117 }
2118
2119 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2120  * is locked for write. Otherwise the are possible deadlocks:
2121  * Process 1                         Process 2
2122  * ext3_create()                     quota_sync()
2123  *   journal_start()                   write_dquot()
2124  *   DQUOT_INIT()                        down(dqio_sem)
2125  *     down(dqio_sem)                    journal_start()
2126  *
2127  */
2128
2129 #ifdef CONFIG_QUOTA
2130
2131 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2132 {
2133         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type]->f_dentry->d_inode;
2134 }
2135
2136 static int ext3_dquot_initialize(struct inode *inode, int type)
2137 {
2138         handle_t *handle;
2139         int ret, err;
2140
2141         /* We may create quota structure so we need to reserve enough blocks */
2142         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2143         if (IS_ERR(handle))
2144                 return PTR_ERR(handle);
2145         ret = dquot_initialize(inode, type);
2146         err = ext3_journal_stop(handle);
2147         if (!ret)
2148                 ret = err;
2149         return ret;
2150 }
2151
2152 static int ext3_dquot_drop(struct inode *inode)
2153 {
2154         handle_t *handle;
2155         int ret, err;
2156
2157         /* We may delete quota structure so we need to reserve enough blocks */
2158         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2159         if (IS_ERR(handle))
2160                 return PTR_ERR(handle);
2161         ret = dquot_drop(inode);
2162         err = ext3_journal_stop(handle);
2163         if (!ret)
2164                 ret = err;
2165         return ret;
2166 }
2167
2168 static int ext3_write_dquot(struct dquot *dquot)
2169 {
2170         int ret, err;
2171         handle_t *handle;
2172
2173         handle = ext3_journal_start(dquot_to_inode(dquot),
2174                                         EXT3_QUOTA_TRANS_BLOCKS);
2175         if (IS_ERR(handle))
2176                 return PTR_ERR(handle);
2177         ret = dquot_commit(dquot);
2178         err = ext3_journal_stop(handle);
2179         if (!ret)
2180                 ret = err;
2181         return ret;
2182 }
2183
2184 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2185 {
2186         /* Are we journalling quotas? */
2187         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2188             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2189                 dquot_mark_dquot_dirty(dquot);
2190                 return ext3_write_dquot(dquot);
2191         } else {
2192                 return dquot_mark_dquot_dirty(dquot);
2193         }
2194 }
2195
2196 static int ext3_write_info(struct super_block *sb, int type)
2197 {
2198         int ret, err;
2199         handle_t *handle;
2200
2201         /* Data block + inode block */
2202         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2203         if (IS_ERR(handle))
2204                 return PTR_ERR(handle);
2205         ret = dquot_commit_info(sb, type);
2206         err = ext3_journal_stop(handle);
2207         if (!ret)
2208                 ret = err;
2209         return ret;
2210 }
2211
2212 /*
2213  * Turn on quotas during mount time - we need to find
2214  * the quota file and such...
2215  */
2216 static int ext3_quota_on_mount(struct super_block *sb, int type)
2217 {
2218         int err;
2219         struct dentry *dentry;
2220         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2221                              .hash = 0,
2222                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2223
2224         dentry = lookup_hash(&name, sb->s_root);
2225         if (IS_ERR(dentry))
2226                 return PTR_ERR(dentry);
2227         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2228         if (err)
2229                 dput(dentry);
2230         /* We keep the dentry reference if everything went ok - we drop it
2231          * on quota_off time */
2232         return err;
2233 }
2234
2235 /* Turn quotas off during mount time */
2236 static int ext3_quota_off_mount(struct super_block *sb, int type)
2237 {
2238         int err;
2239         struct dentry *dentry;
2240
2241         dentry = sb_dqopt(sb)->files[type]->f_dentry;
2242         err = vfs_quota_off_mount(sb, type);
2243         /* We invalidate dentry - it has at least wrong hash... */
2244         d_invalidate(dentry);
2245         dput(dentry);
2246         return err;
2247 }
2248
2249 /*
2250  * Standard function to be called on quota_on
2251  */
2252 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2253                          char *path)
2254 {
2255         int err;
2256         struct nameidata nd;
2257
2258         /* Not journalling quota? */
2259         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2260             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2261                 return vfs_quota_on(sb, type, format_id, path);
2262         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2263         if (err)
2264                 return err;
2265         /* Quotafile not on the same filesystem? */
2266         if (nd.mnt->mnt_sb != sb)
2267                 return -EXDEV;
2268         /* Quotafile not of fs root? */
2269         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2270                 printk(KERN_WARNING
2271                         "EXT3-fs: Quota file not on filesystem root. "
2272                         "Journalled quota will not work.\n");
2273         if (!ext3_should_journal_data(nd.dentry->d_inode))
2274                 printk(KERN_WARNING "EXT3-fs: Quota file does not have "
2275                         "data-journalling. Journalled quota will not work.\n");
2276         path_release(&nd);
2277         return vfs_quota_on(sb, type, format_id, path);
2278 }
2279
2280 #endif
2281
2282 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2283         int flags, const char *dev_name, void *data)
2284 {
2285         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2286 }
2287
2288 static struct file_system_type ext3_fs_type = {
2289         .owner          = THIS_MODULE,
2290         .name           = "ext3",
2291         .get_sb         = ext3_get_sb,
2292         .kill_sb        = kill_block_super,
2293         .fs_flags       = FS_REQUIRES_DEV,
2294 };
2295
2296 static int __init init_ext3_fs(void)
2297 {
2298         int err = init_ext3_xattr();
2299         if (err)
2300                 return err;
2301         err = init_inodecache();
2302         if (err)
2303                 goto out1;
2304         err = register_filesystem(&ext3_fs_type);
2305         if (err)
2306                 goto out;
2307         return 0;
2308 out:
2309         destroy_inodecache();
2310 out1:
2311         exit_ext3_xattr();
2312         return err;
2313 }
2314
2315 static void __exit exit_ext3_fs(void)
2316 {
2317         unregister_filesystem(&ext3_fs_type);
2318         destroy_inodecache();
2319         exit_ext3_xattr();
2320 }
2321
2322 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2323 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2324 MODULE_LICENSE("GPL");
2325 module_init(init_ext3_fs)
2326 module_exit(exit_ext3_fs)