Merge branch 'master' of /usr/src/ntfs-2.6/
[linux-2.6] / 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 <linux/seq_file.h>
39
40 #include <asm/uaccess.h>
41
42 #include "xattr.h"
43 #include "acl.h"
44 #include "namei.h"
45
46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
47 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
48                                int);
49 static void ext3_commit_super (struct super_block * sb,
50                                struct ext3_super_block * es,
51                                int sync);
52 static void ext3_mark_recovery_complete(struct super_block * sb,
53                                         struct ext3_super_block * es);
54 static void ext3_clear_journal_err(struct super_block * sb,
55                                    struct ext3_super_block * es);
56 static int ext3_sync_fs(struct super_block *sb, int wait);
57 static const char *ext3_decode_error(struct super_block * sb, int errno,
58                                      char nbuf[16]);
59 static int ext3_remount (struct super_block * sb, int * flags, char * data);
60 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
61 static void ext3_unlockfs(struct super_block *sb);
62 static void ext3_write_super (struct super_block * sb);
63 static void ext3_write_super_lockfs(struct super_block *sb);
64
65 /* 
66  * Wrappers for journal_start/end.
67  *
68  * The only special thing we need to do here is to make sure that all
69  * journal_end calls result in the superblock being marked dirty, so
70  * that sync() will call the filesystem's write_super callback if
71  * appropriate. 
72  */
73 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
74 {
75         journal_t *journal;
76
77         if (sb->s_flags & MS_RDONLY)
78                 return ERR_PTR(-EROFS);
79
80         /* Special case here: if the journal has aborted behind our
81          * backs (eg. EIO in the commit thread), then we still need to
82          * take the FS itself readonly cleanly. */
83         journal = EXT3_SB(sb)->s_journal;
84         if (is_journal_aborted(journal)) {
85                 ext3_abort(sb, __FUNCTION__,
86                            "Detected aborted journal");
87                 return ERR_PTR(-EROFS);
88         }
89
90         return journal_start(journal, nblocks);
91 }
92
93 /* 
94  * The only special thing we need to do here is to make sure that all
95  * journal_stop calls result in the superblock being marked dirty, so
96  * that sync() will call the filesystem's write_super callback if
97  * appropriate. 
98  */
99 int __ext3_journal_stop(const char *where, handle_t *handle)
100 {
101         struct super_block *sb;
102         int err;
103         int rc;
104
105         sb = handle->h_transaction->t_journal->j_private;
106         err = handle->h_err;
107         rc = journal_stop(handle);
108
109         if (!err)
110                 err = rc;
111         if (err)
112                 __ext3_std_error(sb, where, err);
113         return err;
114 }
115
116 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
117                 struct buffer_head *bh, handle_t *handle, int err)
118 {
119         char nbuf[16];
120         const char *errstr = ext3_decode_error(NULL, err, nbuf);
121
122         if (bh)
123                 BUFFER_TRACE(bh, "abort");
124
125         if (!handle->h_err)
126                 handle->h_err = err;
127
128         if (is_handle_aborted(handle))
129                 return;
130
131         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
132                caller, errstr, err_fn);
133
134         journal_abort_handle(handle);
135 }
136
137 /* Deal with the reporting of failure conditions on a filesystem such as
138  * inconsistencies detected or read IO failures.
139  *
140  * On ext2, we can store the error state of the filesystem in the
141  * superblock.  That is not possible on ext3, because we may have other
142  * write ordering constraints on the superblock which prevent us from
143  * writing it out straight away; and given that the journal is about to
144  * be aborted, we can't rely on the current, or future, transactions to
145  * write out the superblock safely.
146  *
147  * We'll just use the journal_abort() error code to record an error in
148  * the journal instead.  On recovery, the journal will compain about
149  * that error until we've noted it down and cleared it.
150  */
151
152 static void ext3_handle_error(struct super_block *sb)
153 {
154         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
155
156         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
157         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
158
159         if (sb->s_flags & MS_RDONLY)
160                 return;
161
162         if (test_opt (sb, ERRORS_RO)) {
163                 printk (KERN_CRIT "Remounting filesystem read-only\n");
164                 sb->s_flags |= MS_RDONLY;
165         } else {
166                 journal_t *journal = EXT3_SB(sb)->s_journal;
167
168                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
169                 if (journal)
170                         journal_abort(journal, -EIO);
171         }
172         if (test_opt(sb, ERRORS_PANIC))
173                 panic("EXT3-fs (device %s): panic forced after error\n",
174                         sb->s_id);
175         ext3_commit_super(sb, es, 1);
176 }
177
178 void ext3_error (struct super_block * sb, const char * function,
179                  const char * fmt, ...)
180 {
181         va_list args;
182
183         va_start(args, fmt);
184         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
185         vprintk(fmt, args);
186         printk("\n");
187         va_end(args);
188
189         ext3_handle_error(sb);
190 }
191
192 static const char *ext3_decode_error(struct super_block * sb, int errno,
193                                      char nbuf[16])
194 {
195         char *errstr = NULL;
196
197         switch (errno) {
198         case -EIO:
199                 errstr = "IO failure";
200                 break;
201         case -ENOMEM:
202                 errstr = "Out of memory";
203                 break;
204         case -EROFS:
205                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
206                         errstr = "Journal has aborted";
207                 else
208                         errstr = "Readonly filesystem";
209                 break;
210         default:
211                 /* If the caller passed in an extra buffer for unknown
212                  * errors, textualise them now.  Else we just return
213                  * NULL. */
214                 if (nbuf) {
215                         /* Check for truncated error codes... */
216                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
217                                 errstr = nbuf;
218                 }
219                 break;
220         }
221
222         return errstr;
223 }
224
225 /* __ext3_std_error decodes expected errors from journaling functions
226  * automatically and invokes the appropriate error response.  */
227
228 void __ext3_std_error (struct super_block * sb, const char * function,
229                        int errno)
230 {
231         char nbuf[16];
232         const char *errstr;
233
234         /* Special case: if the error is EROFS, and we're not already
235          * inside a transaction, then there's really no point in logging
236          * an error. */
237         if (errno == -EROFS && journal_current_handle() == NULL &&
238             (sb->s_flags & MS_RDONLY))
239                 return;
240
241         errstr = ext3_decode_error(sb, errno, nbuf);
242         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
243                 sb->s_id, function, errstr);
244
245         ext3_handle_error(sb);
246 }
247
248 /*
249  * ext3_abort is a much stronger failure handler than ext3_error.  The
250  * abort function may be used to deal with unrecoverable failures such
251  * as journal IO errors or ENOMEM at a critical moment in log management.
252  *
253  * We unconditionally force the filesystem into an ABORT|READONLY state,
254  * unless the error response on the fs has been set to panic in which
255  * case we take the easy way out and panic immediately.
256  */
257
258 void ext3_abort (struct super_block * sb, const char * function,
259                  const char * fmt, ...)
260 {
261         va_list args;
262
263         printk (KERN_CRIT "ext3_abort called.\n");
264
265         va_start(args, fmt);
266         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
267         vprintk(fmt, args);
268         printk("\n");
269         va_end(args);
270
271         if (test_opt(sb, ERRORS_PANIC))
272                 panic("EXT3-fs panic from previous error\n");
273
274         if (sb->s_flags & MS_RDONLY)
275                 return;
276
277         printk(KERN_CRIT "Remounting filesystem read-only\n");
278         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
279         sb->s_flags |= MS_RDONLY;
280         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
281         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
282 }
283
284 void ext3_warning (struct super_block * sb, const char * function,
285                    const char * fmt, ...)
286 {
287         va_list args;
288
289         va_start(args, fmt);
290         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
291                sb->s_id, function);
292         vprintk(fmt, args);
293         printk("\n");
294         va_end(args);
295 }
296
297 void ext3_update_dynamic_rev(struct super_block *sb)
298 {
299         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
300
301         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
302                 return;
303
304         ext3_warning(sb, __FUNCTION__,
305                      "updating to rev %d because of new feature flag, "
306                      "running e2fsck is recommended",
307                      EXT3_DYNAMIC_REV);
308
309         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
310         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
311         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
312         /* leave es->s_feature_*compat flags alone */
313         /* es->s_uuid will be set by e2fsck if empty */
314
315         /*
316          * The rest of the superblock fields should be zero, and if not it
317          * means they are likely already in use, so leave them alone.  We
318          * can leave it up to e2fsck to clean up any inconsistencies there.
319          */
320 }
321
322 /*
323  * Open the external journal device
324  */
325 static struct block_device *ext3_blkdev_get(dev_t dev)
326 {
327         struct block_device *bdev;
328         char b[BDEVNAME_SIZE];
329
330         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
331         if (IS_ERR(bdev))
332                 goto fail;
333         return bdev;
334
335 fail:
336         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
337                         __bdevname(dev, b), PTR_ERR(bdev));
338         return NULL;
339 }
340
341 /*
342  * Release the journal device
343  */
344 static int ext3_blkdev_put(struct block_device *bdev)
345 {
346         bd_release(bdev);
347         return blkdev_put(bdev);
348 }
349
350 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
351 {
352         struct block_device *bdev;
353         int ret = -ENODEV;
354
355         bdev = sbi->journal_bdev;
356         if (bdev) {
357                 ret = ext3_blkdev_put(bdev);
358                 sbi->journal_bdev = NULL;
359         }
360         return ret;
361 }
362
363 static inline struct inode *orphan_list_entry(struct list_head *l)
364 {
365         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
366 }
367
368 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
369 {
370         struct list_head *l;
371
372         printk(KERN_ERR "sb orphan head is %d\n", 
373                le32_to_cpu(sbi->s_es->s_last_orphan));
374
375         printk(KERN_ERR "sb_info orphan list:\n");
376         list_for_each(l, &sbi->s_orphan) {
377                 struct inode *inode = orphan_list_entry(l);
378                 printk(KERN_ERR "  "
379                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
380                        inode->i_sb->s_id, inode->i_ino, inode,
381                        inode->i_mode, inode->i_nlink, 
382                        NEXT_ORPHAN(inode));
383         }
384 }
385
386 static void ext3_put_super (struct super_block * sb)
387 {
388         struct ext3_sb_info *sbi = EXT3_SB(sb);
389         struct ext3_super_block *es = sbi->s_es;
390         int i;
391
392         ext3_xattr_put_super(sb);
393         journal_destroy(sbi->s_journal);
394         if (!(sb->s_flags & MS_RDONLY)) {
395                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
396                 es->s_state = cpu_to_le16(sbi->s_mount_state);
397                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
398                 mark_buffer_dirty(sbi->s_sbh);
399                 ext3_commit_super(sb, es, 1);
400         }
401
402         for (i = 0; i < sbi->s_gdb_count; i++)
403                 brelse(sbi->s_group_desc[i]);
404         kfree(sbi->s_group_desc);
405         percpu_counter_destroy(&sbi->s_freeblocks_counter);
406         percpu_counter_destroy(&sbi->s_freeinodes_counter);
407         percpu_counter_destroy(&sbi->s_dirs_counter);
408         brelse(sbi->s_sbh);
409 #ifdef CONFIG_QUOTA
410         for (i = 0; i < MAXQUOTAS; i++)
411                 kfree(sbi->s_qf_names[i]);
412 #endif
413
414         /* Debugging code just in case the in-memory inode orphan list
415          * isn't empty.  The on-disk one can be non-empty if we've
416          * detected an error and taken the fs readonly, but the
417          * in-memory list had better be clean by this point. */
418         if (!list_empty(&sbi->s_orphan))
419                 dump_orphan_list(sb, sbi);
420         J_ASSERT(list_empty(&sbi->s_orphan));
421
422         invalidate_bdev(sb->s_bdev, 0);
423         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
424                 /*
425                  * Invalidate the journal device's buffers.  We don't want them
426                  * floating about in memory - the physical journal device may
427                  * hotswapped, and it breaks the `ro-after' testing code.
428                  */
429                 sync_blockdev(sbi->journal_bdev);
430                 invalidate_bdev(sbi->journal_bdev, 0);
431                 ext3_blkdev_remove(sbi);
432         }
433         sb->s_fs_info = NULL;
434         kfree(sbi);
435         return;
436 }
437
438 static kmem_cache_t *ext3_inode_cachep;
439
440 /*
441  * Called inside transaction, so use GFP_NOFS
442  */
443 static struct inode *ext3_alloc_inode(struct super_block *sb)
444 {
445         struct ext3_inode_info *ei;
446
447         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
448         if (!ei)
449                 return NULL;
450 #ifdef CONFIG_EXT3_FS_POSIX_ACL
451         ei->i_acl = EXT3_ACL_NOT_CACHED;
452         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
453 #endif
454         ei->i_block_alloc_info = NULL;
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_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 static void ext3_clear_inode(struct inode *inode)
497 {
498         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
499 #ifdef CONFIG_EXT3_FS_POSIX_ACL
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 #endif
511         ext3_discard_reservation(inode);
512         EXT3_I(inode)->i_block_alloc_info = NULL;
513         kfree(rsv);
514 }
515
516 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
517 {
518 #if defined(CONFIG_QUOTA)
519         struct ext3_sb_info *sbi = EXT3_SB(sb);
520
521         if (sbi->s_jquota_fmt)
522                 seq_printf(seq, ",jqfmt=%s",
523                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
524
525         if (sbi->s_qf_names[USRQUOTA])
526                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
527
528         if (sbi->s_qf_names[GRPQUOTA])
529                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
530
531         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
532                 seq_puts(seq, ",usrquota");
533
534         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
535                 seq_puts(seq, ",grpquota");
536 #endif
537 }
538
539 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
540 {
541         struct super_block *sb = vfs->mnt_sb;
542
543         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
544                 seq_puts(seq, ",data=journal");
545         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
546                 seq_puts(seq, ",data=ordered");
547         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
548                 seq_puts(seq, ",data=writeback");
549
550         ext3_show_quota_options(seq, sb);
551
552         return 0;
553 }
554
555 #ifdef CONFIG_QUOTA
556 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
557 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
558
559 static int ext3_dquot_initialize(struct inode *inode, int type);
560 static int ext3_dquot_drop(struct inode *inode);
561 static int ext3_write_dquot(struct dquot *dquot);
562 static int ext3_acquire_dquot(struct dquot *dquot);
563 static int ext3_release_dquot(struct dquot *dquot);
564 static int ext3_mark_dquot_dirty(struct dquot *dquot);
565 static int ext3_write_info(struct super_block *sb, int type);
566 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
567 static int ext3_quota_on_mount(struct super_block *sb, int type);
568 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
569                                size_t len, loff_t off);
570 static ssize_t ext3_quota_write(struct super_block *sb, int type,
571                                 const char *data, size_t len, loff_t off);
572
573 static struct dquot_operations ext3_quota_operations = {
574         .initialize     = ext3_dquot_initialize,
575         .drop           = ext3_dquot_drop,
576         .alloc_space    = dquot_alloc_space,
577         .alloc_inode    = dquot_alloc_inode,
578         .free_space     = dquot_free_space,
579         .free_inode     = dquot_free_inode,
580         .transfer       = dquot_transfer,
581         .write_dquot    = ext3_write_dquot,
582         .acquire_dquot  = ext3_acquire_dquot,
583         .release_dquot  = ext3_release_dquot,
584         .mark_dirty     = ext3_mark_dquot_dirty,
585         .write_info     = ext3_write_info
586 };
587
588 static struct quotactl_ops ext3_qctl_operations = {
589         .quota_on       = ext3_quota_on,
590         .quota_off      = vfs_quota_off,
591         .quota_sync     = vfs_quota_sync,
592         .get_info       = vfs_get_dqinfo,
593         .set_info       = vfs_set_dqinfo,
594         .get_dqblk      = vfs_get_dqblk,
595         .set_dqblk      = vfs_set_dqblk
596 };
597 #endif
598
599 static struct super_operations ext3_sops = {
600         .alloc_inode    = ext3_alloc_inode,
601         .destroy_inode  = ext3_destroy_inode,
602         .read_inode     = ext3_read_inode,
603         .write_inode    = ext3_write_inode,
604         .dirty_inode    = ext3_dirty_inode,
605         .delete_inode   = ext3_delete_inode,
606         .put_super      = ext3_put_super,
607         .write_super    = ext3_write_super,
608         .sync_fs        = ext3_sync_fs,
609         .write_super_lockfs = ext3_write_super_lockfs,
610         .unlockfs       = ext3_unlockfs,
611         .statfs         = ext3_statfs,
612         .remount_fs     = ext3_remount,
613         .clear_inode    = ext3_clear_inode,
614         .show_options   = ext3_show_options,
615 #ifdef CONFIG_QUOTA
616         .quota_read     = ext3_quota_read,
617         .quota_write    = ext3_quota_write,
618 #endif
619 };
620
621 static struct export_operations ext3_export_ops = {
622         .get_parent = ext3_get_parent,
623 };
624
625 enum {
626         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
627         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
628         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
629         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
630         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
631         Opt_commit, Opt_journal_update, Opt_journal_inum,
632         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
633         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
634         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
635         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
636         Opt_grpquota
637 };
638
639 static match_table_t tokens = {
640         {Opt_bsd_df, "bsddf"},
641         {Opt_minix_df, "minixdf"},
642         {Opt_grpid, "grpid"},
643         {Opt_grpid, "bsdgroups"},
644         {Opt_nogrpid, "nogrpid"},
645         {Opt_nogrpid, "sysvgroups"},
646         {Opt_resgid, "resgid=%u"},
647         {Opt_resuid, "resuid=%u"},
648         {Opt_sb, "sb=%u"},
649         {Opt_err_cont, "errors=continue"},
650         {Opt_err_panic, "errors=panic"},
651         {Opt_err_ro, "errors=remount-ro"},
652         {Opt_nouid32, "nouid32"},
653         {Opt_nocheck, "nocheck"},
654         {Opt_nocheck, "check=none"},
655         {Opt_check, "check"},
656         {Opt_debug, "debug"},
657         {Opt_oldalloc, "oldalloc"},
658         {Opt_orlov, "orlov"},
659         {Opt_user_xattr, "user_xattr"},
660         {Opt_nouser_xattr, "nouser_xattr"},
661         {Opt_acl, "acl"},
662         {Opt_noacl, "noacl"},
663         {Opt_reservation, "reservation"},
664         {Opt_noreservation, "noreservation"},
665         {Opt_noload, "noload"},
666         {Opt_nobh, "nobh"},
667         {Opt_commit, "commit=%u"},
668         {Opt_journal_update, "journal=update"},
669         {Opt_journal_inum, "journal=%u"},
670         {Opt_abort, "abort"},
671         {Opt_data_journal, "data=journal"},
672         {Opt_data_ordered, "data=ordered"},
673         {Opt_data_writeback, "data=writeback"},
674         {Opt_offusrjquota, "usrjquota="},
675         {Opt_usrjquota, "usrjquota=%s"},
676         {Opt_offgrpjquota, "grpjquota="},
677         {Opt_grpjquota, "grpjquota=%s"},
678         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
679         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
680         {Opt_grpquota, "grpquota"},
681         {Opt_noquota, "noquota"},
682         {Opt_quota, "quota"},
683         {Opt_usrquota, "usrquota"},
684         {Opt_barrier, "barrier=%u"},
685         {Opt_err, NULL},
686         {Opt_resize, "resize"},
687 };
688
689 static unsigned long get_sb_block(void **data)
690 {
691         unsigned long   sb_block;
692         char            *options = (char *) *data;
693
694         if (!options || strncmp(options, "sb=", 3) != 0)
695                 return 1;       /* Default location */
696         options += 3;
697         sb_block = simple_strtoul(options, &options, 0);
698         if (*options && *options != ',') {
699                 printk("EXT3-fs: Invalid sb specification: %s\n",
700                        (char *) *data);
701                 return 1;
702         }
703         if (*options == ',')
704                 options++;
705         *data = (void *) options;
706         return sb_block;
707 }
708
709 static int parse_options (char * options, struct super_block *sb,
710                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
711 {
712         struct ext3_sb_info *sbi = EXT3_SB(sb);
713         char * p;
714         substring_t args[MAX_OPT_ARGS];
715         int data_opt = 0;
716         int option;
717 #ifdef CONFIG_QUOTA
718         int qtype;
719         char *qname;
720 #endif
721
722         if (!options)
723                 return 1;
724
725         while ((p = strsep (&options, ",")) != NULL) {
726                 int token;
727                 if (!*p)
728                         continue;
729
730                 token = match_token(p, tokens, args);
731                 switch (token) {
732                 case Opt_bsd_df:
733                         clear_opt (sbi->s_mount_opt, MINIX_DF);
734                         break;
735                 case Opt_minix_df:
736                         set_opt (sbi->s_mount_opt, MINIX_DF);
737                         break;
738                 case Opt_grpid:
739                         set_opt (sbi->s_mount_opt, GRPID);
740                         break;
741                 case Opt_nogrpid:
742                         clear_opt (sbi->s_mount_opt, GRPID);
743                         break;
744                 case Opt_resuid:
745                         if (match_int(&args[0], &option))
746                                 return 0;
747                         sbi->s_resuid = option;
748                         break;
749                 case Opt_resgid:
750                         if (match_int(&args[0], &option))
751                                 return 0;
752                         sbi->s_resgid = option;
753                         break;
754                 case Opt_sb:
755                         /* handled by get_sb_block() instead of here */
756                         /* *sb_block = match_int(&args[0]); */
757                         break;
758                 case Opt_err_panic:
759                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
760                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
761                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
762                         break;
763                 case Opt_err_ro:
764                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
765                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
766                         set_opt (sbi->s_mount_opt, ERRORS_RO);
767                         break;
768                 case Opt_err_cont:
769                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
770                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
771                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
772                         break;
773                 case Opt_nouid32:
774                         set_opt (sbi->s_mount_opt, NO_UID32);
775                         break;
776                 case Opt_check:
777 #ifdef CONFIG_EXT3_CHECK
778                         set_opt (sbi->s_mount_opt, CHECK);
779 #else
780                         printk(KERN_ERR
781                                "EXT3 Check option not supported\n");
782 #endif
783                         break;
784                 case Opt_nocheck:
785                         clear_opt (sbi->s_mount_opt, CHECK);
786                         break;
787                 case Opt_debug:
788                         set_opt (sbi->s_mount_opt, DEBUG);
789                         break;
790                 case Opt_oldalloc:
791                         set_opt (sbi->s_mount_opt, OLDALLOC);
792                         break;
793                 case Opt_orlov:
794                         clear_opt (sbi->s_mount_opt, OLDALLOC);
795                         break;
796 #ifdef CONFIG_EXT3_FS_XATTR
797                 case Opt_user_xattr:
798                         set_opt (sbi->s_mount_opt, XATTR_USER);
799                         break;
800                 case Opt_nouser_xattr:
801                         clear_opt (sbi->s_mount_opt, XATTR_USER);
802                         break;
803 #else
804                 case Opt_user_xattr:
805                 case Opt_nouser_xattr:
806                         printk("EXT3 (no)user_xattr options not supported\n");
807                         break;
808 #endif
809 #ifdef CONFIG_EXT3_FS_POSIX_ACL
810                 case Opt_acl:
811                         set_opt(sbi->s_mount_opt, POSIX_ACL);
812                         break;
813                 case Opt_noacl:
814                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
815                         break;
816 #else
817                 case Opt_acl:
818                 case Opt_noacl:
819                         printk("EXT3 (no)acl options not supported\n");
820                         break;
821 #endif
822                 case Opt_reservation:
823                         set_opt(sbi->s_mount_opt, RESERVATION);
824                         break;
825                 case Opt_noreservation:
826                         clear_opt(sbi->s_mount_opt, RESERVATION);
827                         break;
828                 case Opt_journal_update:
829                         /* @@@ FIXME */
830                         /* Eventually we will want to be able to create
831                            a journal file here.  For now, only allow the
832                            user to specify an existing inode to be the
833                            journal file. */
834                         if (is_remount) {
835                                 printk(KERN_ERR "EXT3-fs: cannot specify "
836                                        "journal on remount\n");
837                                 return 0;
838                         }
839                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
840                         break;
841                 case Opt_journal_inum:
842                         if (is_remount) {
843                                 printk(KERN_ERR "EXT3-fs: cannot specify "
844                                        "journal on remount\n");
845                                 return 0;
846                         }
847                         if (match_int(&args[0], &option))
848                                 return 0;
849                         *inum = option;
850                         break;
851                 case Opt_noload:
852                         set_opt (sbi->s_mount_opt, NOLOAD);
853                         break;
854                 case Opt_commit:
855                         if (match_int(&args[0], &option))
856                                 return 0;
857                         if (option < 0)
858                                 return 0;
859                         if (option == 0)
860                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
861                         sbi->s_commit_interval = HZ * option;
862                         break;
863                 case Opt_data_journal:
864                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
865                         goto datacheck;
866                 case Opt_data_ordered:
867                         data_opt = EXT3_MOUNT_ORDERED_DATA;
868                         goto datacheck;
869                 case Opt_data_writeback:
870                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
871                 datacheck:
872                         if (is_remount) {
873                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
874                                                 != data_opt) {
875                                         printk(KERN_ERR
876                                                 "EXT3-fs: cannot change data "
877                                                 "mode on remount\n");
878                                         return 0;
879                                 }
880                         } else {
881                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
882                                 sbi->s_mount_opt |= data_opt;
883                         }
884                         break;
885 #ifdef CONFIG_QUOTA
886                 case Opt_usrjquota:
887                         qtype = USRQUOTA;
888                         goto set_qf_name;
889                 case Opt_grpjquota:
890                         qtype = GRPQUOTA;
891 set_qf_name:
892                         if (sb_any_quota_enabled(sb)) {
893                                 printk(KERN_ERR
894                                         "EXT3-fs: Cannot change journalled "
895                                         "quota options when quota turned on.\n");
896                                 return 0;
897                         }
898                         qname = match_strdup(&args[0]);
899                         if (!qname) {
900                                 printk(KERN_ERR
901                                         "EXT3-fs: not enough memory for "
902                                         "storing quotafile name.\n");
903                                 return 0;
904                         }
905                         if (sbi->s_qf_names[qtype] &&
906                             strcmp(sbi->s_qf_names[qtype], qname)) {
907                                 printk(KERN_ERR
908                                         "EXT3-fs: %s quota file already "
909                                         "specified.\n", QTYPE2NAME(qtype));
910                                 kfree(qname);
911                                 return 0;
912                         }
913                         sbi->s_qf_names[qtype] = qname;
914                         if (strchr(sbi->s_qf_names[qtype], '/')) {
915                                 printk(KERN_ERR
916                                         "EXT3-fs: quotafile must be on "
917                                         "filesystem root.\n");
918                                 kfree(sbi->s_qf_names[qtype]);
919                                 sbi->s_qf_names[qtype] = NULL;
920                                 return 0;
921                         }
922                         set_opt(sbi->s_mount_opt, QUOTA);
923                         break;
924                 case Opt_offusrjquota:
925                         qtype = USRQUOTA;
926                         goto clear_qf_name;
927                 case Opt_offgrpjquota:
928                         qtype = GRPQUOTA;
929 clear_qf_name:
930                         if (sb_any_quota_enabled(sb)) {
931                                 printk(KERN_ERR "EXT3-fs: Cannot change "
932                                         "journalled quota options when "
933                                         "quota turned on.\n");
934                                 return 0;
935                         }
936                         /*
937                          * The space will be released later when all options
938                          * are confirmed to be correct
939                          */
940                         sbi->s_qf_names[qtype] = NULL;
941                         break;
942                 case Opt_jqfmt_vfsold:
943                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
944                         break;
945                 case Opt_jqfmt_vfsv0:
946                         sbi->s_jquota_fmt = QFMT_VFS_V0;
947                         break;
948                 case Opt_quota:
949                 case Opt_usrquota:
950                         set_opt(sbi->s_mount_opt, QUOTA);
951                         set_opt(sbi->s_mount_opt, USRQUOTA);
952                         break;
953                 case Opt_grpquota:
954                         set_opt(sbi->s_mount_opt, QUOTA);
955                         set_opt(sbi->s_mount_opt, GRPQUOTA);
956                         break;
957                 case Opt_noquota:
958                         if (sb_any_quota_enabled(sb)) {
959                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
960                                         "options when quota turned on.\n");
961                                 return 0;
962                         }
963                         clear_opt(sbi->s_mount_opt, QUOTA);
964                         clear_opt(sbi->s_mount_opt, USRQUOTA);
965                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
966                         break;
967 #else
968                 case Opt_quota:
969                 case Opt_usrquota:
970                 case Opt_grpquota:
971                 case Opt_usrjquota:
972                 case Opt_grpjquota:
973                 case Opt_offusrjquota:
974                 case Opt_offgrpjquota:
975                 case Opt_jqfmt_vfsold:
976                 case Opt_jqfmt_vfsv0:
977                         printk(KERN_ERR
978                                 "EXT3-fs: journalled quota options not "
979                                 "supported.\n");
980                         break;
981                 case Opt_noquota:
982                         break;
983 #endif
984                 case Opt_abort:
985                         set_opt(sbi->s_mount_opt, ABORT);
986                         break;
987                 case Opt_barrier:
988                         if (match_int(&args[0], &option))
989                                 return 0;
990                         if (option)
991                                 set_opt(sbi->s_mount_opt, BARRIER);
992                         else
993                                 clear_opt(sbi->s_mount_opt, BARRIER);
994                         break;
995                 case Opt_ignore:
996                         break;
997                 case Opt_resize:
998                         if (!is_remount) {
999                                 printk("EXT3-fs: resize option only available "
1000                                         "for remount\n");
1001                                 return 0;
1002                         }
1003                         if (match_int(&args[0], &option) != 0)
1004                                 return 0;
1005                         *n_blocks_count = option;
1006                         break;
1007                 case Opt_nobh:
1008                         set_opt(sbi->s_mount_opt, NOBH);
1009                         break;
1010                 default:
1011                         printk (KERN_ERR
1012                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1013                                 "or missing value\n", p);
1014                         return 0;
1015                 }
1016         }
1017 #ifdef CONFIG_QUOTA
1018         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1019                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1020                      sbi->s_qf_names[USRQUOTA])
1021                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1022
1023                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1024                      sbi->s_qf_names[GRPQUOTA])
1025                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1026
1027                 if ((sbi->s_qf_names[USRQUOTA] &&
1028                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1029                     (sbi->s_qf_names[GRPQUOTA] &&
1030                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1031                         printk(KERN_ERR "EXT3-fs: old and new quota "
1032                                         "format mixing.\n");
1033                         return 0;
1034                 }
1035
1036                 if (!sbi->s_jquota_fmt) {
1037                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1038                                         "not specified.\n");
1039                         return 0;
1040                 }
1041         } else {
1042                 if (sbi->s_jquota_fmt) {
1043                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1044                                         "specified with no journalling "
1045                                         "enabled.\n");
1046                         return 0;
1047                 }
1048         }
1049 #endif
1050         return 1;
1051 }
1052
1053 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1054                             int read_only)
1055 {
1056         struct ext3_sb_info *sbi = EXT3_SB(sb);
1057         int res = 0;
1058
1059         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1060                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1061                         "forcing read-only mode\n");
1062                 res = MS_RDONLY;
1063         }
1064         if (read_only)
1065                 return res;
1066         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1067                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1068                         "running e2fsck is recommended\n");
1069         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1070                 printk (KERN_WARNING
1071                         "EXT3-fs warning: mounting fs with errors, "
1072                         "running e2fsck is recommended\n");
1073         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1074                  le16_to_cpu(es->s_mnt_count) >=
1075                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1076                 printk (KERN_WARNING
1077                         "EXT3-fs warning: maximal mount count reached, "
1078                         "running e2fsck is recommended\n");
1079         else if (le32_to_cpu(es->s_checkinterval) &&
1080                 (le32_to_cpu(es->s_lastcheck) +
1081                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1082                 printk (KERN_WARNING
1083                         "EXT3-fs warning: checktime reached, "
1084                         "running e2fsck is recommended\n");
1085 #if 0
1086                 /* @@@ We _will_ want to clear the valid bit if we find
1087                    inconsistencies, to force a fsck at reboot.  But for
1088                    a plain journaled filesystem we can keep it set as
1089                    valid forever! :) */
1090         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1091 #endif
1092         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1093                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1094         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1095         es->s_mtime = cpu_to_le32(get_seconds());
1096         ext3_update_dynamic_rev(sb);
1097         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1098
1099         ext3_commit_super(sb, es, 1);
1100         if (test_opt(sb, DEBUG))
1101                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1102                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1103                         sb->s_blocksize,
1104                         sbi->s_groups_count,
1105                         EXT3_BLOCKS_PER_GROUP(sb),
1106                         EXT3_INODES_PER_GROUP(sb),
1107                         sbi->s_mount_opt);
1108
1109         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1110         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1111                 char b[BDEVNAME_SIZE];
1112
1113                 printk("external journal on %s\n",
1114                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1115         } else {
1116                 printk("internal journal\n");
1117         }
1118 #ifdef CONFIG_EXT3_CHECK
1119         if (test_opt (sb, CHECK)) {
1120                 ext3_check_blocks_bitmap (sb);
1121                 ext3_check_inodes_bitmap (sb);
1122         }
1123 #endif
1124         return res;
1125 }
1126
1127 /* Called at mount-time, super-block is locked */
1128 static int ext3_check_descriptors (struct super_block * sb)
1129 {
1130         struct ext3_sb_info *sbi = EXT3_SB(sb);
1131         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1132         struct ext3_group_desc * gdp = NULL;
1133         int desc_block = 0;
1134         int i;
1135
1136         ext3_debug ("Checking group descriptors");
1137
1138         for (i = 0; i < sbi->s_groups_count; i++)
1139         {
1140                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1141                         gdp = (struct ext3_group_desc *)
1142                                         sbi->s_group_desc[desc_block++]->b_data;
1143                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1144                     le32_to_cpu(gdp->bg_block_bitmap) >=
1145                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1146                 {
1147                         ext3_error (sb, "ext3_check_descriptors",
1148                                     "Block bitmap for group %d"
1149                                     " not in group (block %lu)!",
1150                                     i, (unsigned long)
1151                                         le32_to_cpu(gdp->bg_block_bitmap));
1152                         return 0;
1153                 }
1154                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1155                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1156                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1157                 {
1158                         ext3_error (sb, "ext3_check_descriptors",
1159                                     "Inode bitmap for group %d"
1160                                     " not in group (block %lu)!",
1161                                     i, (unsigned long)
1162                                         le32_to_cpu(gdp->bg_inode_bitmap));
1163                         return 0;
1164                 }
1165                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1166                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1167                     block + EXT3_BLOCKS_PER_GROUP(sb))
1168                 {
1169                         ext3_error (sb, "ext3_check_descriptors",
1170                                     "Inode table for group %d"
1171                                     " not in group (block %lu)!",
1172                                     i, (unsigned long)
1173                                         le32_to_cpu(gdp->bg_inode_table));
1174                         return 0;
1175                 }
1176                 block += EXT3_BLOCKS_PER_GROUP(sb);
1177                 gdp++;
1178         }
1179
1180         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1181         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1182         return 1;
1183 }
1184
1185
1186 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1187  * the superblock) which were deleted from all directories, but held open by
1188  * a process at the time of a crash.  We walk the list and try to delete these
1189  * inodes at recovery time (only with a read-write filesystem).
1190  *
1191  * In order to keep the orphan inode chain consistent during traversal (in
1192  * case of crash during recovery), we link each inode into the superblock
1193  * orphan list_head and handle it the same way as an inode deletion during
1194  * normal operation (which journals the operations for us).
1195  *
1196  * We only do an iget() and an iput() on each inode, which is very safe if we
1197  * accidentally point at an in-use or already deleted inode.  The worst that
1198  * can happen in this case is that we get a "bit already cleared" message from
1199  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1200  * e2fsck was run on this filesystem, and it must have already done the orphan
1201  * inode cleanup for us, so we can safely abort without any further action.
1202  */
1203 static void ext3_orphan_cleanup (struct super_block * sb,
1204                                  struct ext3_super_block * es)
1205 {
1206         unsigned int s_flags = sb->s_flags;
1207         int nr_orphans = 0, nr_truncates = 0;
1208 #ifdef CONFIG_QUOTA
1209         int i;
1210 #endif
1211         if (!es->s_last_orphan) {
1212                 jbd_debug(4, "no orphan inodes to clean up\n");
1213                 return;
1214         }
1215
1216         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1217                 if (es->s_last_orphan)
1218                         jbd_debug(1, "Errors on filesystem, "
1219                                   "clearing orphan list.\n");
1220                 es->s_last_orphan = 0;
1221                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1222                 return;
1223         }
1224
1225         if (s_flags & MS_RDONLY) {
1226                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1227                        sb->s_id);
1228                 sb->s_flags &= ~MS_RDONLY;
1229         }
1230 #ifdef CONFIG_QUOTA
1231         /* Needed for iput() to work correctly and not trash data */
1232         sb->s_flags |= MS_ACTIVE;
1233         /* Turn on quotas so that they are updated correctly */
1234         for (i = 0; i < MAXQUOTAS; i++) {
1235                 if (EXT3_SB(sb)->s_qf_names[i]) {
1236                         int ret = ext3_quota_on_mount(sb, i);
1237                         if (ret < 0)
1238                                 printk(KERN_ERR
1239                                         "EXT3-fs: Cannot turn on journalled "
1240                                         "quota: error %d\n", ret);
1241                 }
1242         }
1243 #endif
1244
1245         while (es->s_last_orphan) {
1246                 struct inode *inode;
1247
1248                 if (!(inode =
1249                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1250                         es->s_last_orphan = 0;
1251                         break;
1252                 }
1253
1254                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1255                 DQUOT_INIT(inode);
1256                 if (inode->i_nlink) {
1257                         printk(KERN_DEBUG
1258                                 "%s: truncating inode %ld to %Ld bytes\n",
1259                                 __FUNCTION__, inode->i_ino, inode->i_size);
1260                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1261                                   inode->i_ino, inode->i_size);
1262                         ext3_truncate(inode);
1263                         nr_truncates++;
1264                 } else {
1265                         printk(KERN_DEBUG
1266                                 "%s: deleting unreferenced inode %ld\n",
1267                                 __FUNCTION__, inode->i_ino);
1268                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1269                                   inode->i_ino);
1270                         nr_orphans++;
1271                 }
1272                 iput(inode);  /* The delete magic happens here! */
1273         }
1274
1275 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1276
1277         if (nr_orphans)
1278                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1279                        sb->s_id, PLURAL(nr_orphans));
1280         if (nr_truncates)
1281                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1282                        sb->s_id, PLURAL(nr_truncates));
1283 #ifdef CONFIG_QUOTA
1284         /* Turn quotas off */
1285         for (i = 0; i < MAXQUOTAS; i++) {
1286                 if (sb_dqopt(sb)->files[i])
1287                         vfs_quota_off(sb, i);
1288         }
1289 #endif
1290         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1291 }
1292
1293 #define log2(n) ffz(~(n))
1294
1295 /*
1296  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1297  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1298  * We need to be 1 filesystem block less than the 2^32 sector limit.
1299  */
1300 static loff_t ext3_max_size(int bits)
1301 {
1302         loff_t res = EXT3_NDIR_BLOCKS;
1303         /* This constant is calculated to be the largest file size for a
1304          * dense, 4k-blocksize file such that the total number of
1305          * sectors in the file, including data and all indirect blocks,
1306          * does not exceed 2^32. */
1307         const loff_t upper_limit = 0x1ff7fffd000LL;
1308
1309         res += 1LL << (bits-2);
1310         res += 1LL << (2*(bits-2));
1311         res += 1LL << (3*(bits-2));
1312         res <<= bits;
1313         if (res > upper_limit)
1314                 res = upper_limit;
1315         return res;
1316 }
1317
1318 static unsigned long descriptor_loc(struct super_block *sb,
1319                                     unsigned long logic_sb_block,
1320                                     int nr)
1321 {
1322         struct ext3_sb_info *sbi = EXT3_SB(sb);
1323         unsigned long bg, first_data_block, first_meta_bg;
1324         int has_super = 0;
1325
1326         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1327         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1328
1329         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1330             nr < first_meta_bg)
1331                 return (logic_sb_block + nr + 1);
1332         bg = sbi->s_desc_per_block * nr;
1333         if (ext3_bg_has_super(sb, bg))
1334                 has_super = 1;
1335         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1336 }
1337
1338
1339 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1340 {
1341         struct buffer_head * bh;
1342         struct ext3_super_block *es = NULL;
1343         struct ext3_sb_info *sbi;
1344         unsigned long block;
1345         unsigned long sb_block = get_sb_block(&data);
1346         unsigned long logic_sb_block;
1347         unsigned long offset = 0;
1348         unsigned long journal_inum = 0;
1349         unsigned long def_mount_opts;
1350         struct inode *root;
1351         int blocksize;
1352         int hblock;
1353         int db_count;
1354         int i;
1355         int needs_recovery;
1356         __le32 features;
1357
1358         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1359         if (!sbi)
1360                 return -ENOMEM;
1361         sb->s_fs_info = sbi;
1362         memset(sbi, 0, sizeof(*sbi));
1363         sbi->s_mount_opt = 0;
1364         sbi->s_resuid = EXT3_DEF_RESUID;
1365         sbi->s_resgid = EXT3_DEF_RESGID;
1366
1367         unlock_kernel();
1368
1369         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1370         if (!blocksize) {
1371                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1372                 goto out_fail;
1373         }
1374
1375         /*
1376          * The ext3 superblock will not be buffer aligned for other than 1kB
1377          * block sizes.  We need to calculate the offset from buffer start.
1378          */
1379         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1380                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1381                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1382         } else {
1383                 logic_sb_block = sb_block;
1384         }
1385
1386         if (!(bh = sb_bread(sb, logic_sb_block))) {
1387                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1388                 goto out_fail;
1389         }
1390         /*
1391          * Note: s_es must be initialized as soon as possible because
1392          *       some ext3 macro-instructions depend on its value
1393          */
1394         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1395         sbi->s_es = es;
1396         sb->s_magic = le16_to_cpu(es->s_magic);
1397         if (sb->s_magic != EXT3_SUPER_MAGIC)
1398                 goto cantfind_ext3;
1399
1400         /* Set defaults before we parse the mount options */
1401         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1402         if (def_mount_opts & EXT3_DEFM_DEBUG)
1403                 set_opt(sbi->s_mount_opt, DEBUG);
1404         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1405                 set_opt(sbi->s_mount_opt, GRPID);
1406         if (def_mount_opts & EXT3_DEFM_UID16)
1407                 set_opt(sbi->s_mount_opt, NO_UID32);
1408         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1409                 set_opt(sbi->s_mount_opt, XATTR_USER);
1410         if (def_mount_opts & EXT3_DEFM_ACL)
1411                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1412         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1413                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1414         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1415                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1416         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1417                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1418
1419         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1420                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1421         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1422                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1423
1424         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1425         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1426
1427         set_opt(sbi->s_mount_opt, RESERVATION);
1428
1429         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1430                 goto failed_mount;
1431
1432         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1433                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1434
1435         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1436             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1437              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1438              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1439                 printk(KERN_WARNING 
1440                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1441                        "running e2fsck is recommended\n");
1442         /*
1443          * Check feature flags regardless of the revision level, since we
1444          * previously didn't change the revision level when setting the flags,
1445          * so there is a chance incompat flags are set on a rev 0 filesystem.
1446          */
1447         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1448         if (features) {
1449                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1450                        "unsupported optional features (%x).\n",
1451                        sb->s_id, le32_to_cpu(features));
1452                 goto failed_mount;
1453         }
1454         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1455         if (!(sb->s_flags & MS_RDONLY) && features) {
1456                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1457                        "unsupported optional features (%x).\n",
1458                        sb->s_id, le32_to_cpu(features));
1459                 goto failed_mount;
1460         }
1461         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1462
1463         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1464             blocksize > EXT3_MAX_BLOCK_SIZE) {
1465                 printk(KERN_ERR 
1466                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1467                        blocksize, sb->s_id);
1468                 goto failed_mount;
1469         }
1470
1471         hblock = bdev_hardsect_size(sb->s_bdev);
1472         if (sb->s_blocksize != blocksize) {
1473                 /*
1474                  * Make sure the blocksize for the filesystem is larger
1475                  * than the hardware sectorsize for the machine.
1476                  */
1477                 if (blocksize < hblock) {
1478                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1479                                "device blocksize %d.\n", blocksize, hblock);
1480                         goto failed_mount;
1481                 }
1482
1483                 brelse (bh);
1484                 sb_set_blocksize(sb, blocksize);
1485                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1486                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1487                 bh = sb_bread(sb, logic_sb_block);
1488                 if (!bh) {
1489                         printk(KERN_ERR 
1490                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1491                         goto failed_mount;
1492                 }
1493                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1494                 sbi->s_es = es;
1495                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1496                         printk (KERN_ERR 
1497                                 "EXT3-fs: Magic mismatch, very weird !\n");
1498                         goto failed_mount;
1499                 }
1500         }
1501
1502         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1503
1504         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1505                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1506                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1507         } else {
1508                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1509                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1510                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1511                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1512                     (sbi->s_inode_size > blocksize)) {
1513                         printk (KERN_ERR
1514                                 "EXT3-fs: unsupported inode size: %d\n",
1515                                 sbi->s_inode_size);
1516                         goto failed_mount;
1517                 }
1518         }
1519         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1520                                    le32_to_cpu(es->s_log_frag_size);
1521         if (blocksize != sbi->s_frag_size) {
1522                 printk(KERN_ERR
1523                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1524                        sbi->s_frag_size, blocksize);
1525                 goto failed_mount;
1526         }
1527         sbi->s_frags_per_block = 1;
1528         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1529         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1530         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1531         if (EXT3_INODE_SIZE(sb) == 0)
1532                 goto cantfind_ext3;
1533         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1534         if (sbi->s_inodes_per_block == 0)
1535                 goto cantfind_ext3;
1536         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1537                                         sbi->s_inodes_per_block;
1538         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1539         sbi->s_sbh = bh;
1540         sbi->s_mount_state = le16_to_cpu(es->s_state);
1541         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1542         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1543         for (i=0; i < 4; i++)
1544                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1545         sbi->s_def_hash_version = es->s_def_hash_version;
1546
1547         if (sbi->s_blocks_per_group > blocksize * 8) {
1548                 printk (KERN_ERR
1549                         "EXT3-fs: #blocks per group too big: %lu\n",
1550                         sbi->s_blocks_per_group);
1551                 goto failed_mount;
1552         }
1553         if (sbi->s_frags_per_group > blocksize * 8) {
1554                 printk (KERN_ERR
1555                         "EXT3-fs: #fragments per group too big: %lu\n",
1556                         sbi->s_frags_per_group);
1557                 goto failed_mount;
1558         }
1559         if (sbi->s_inodes_per_group > blocksize * 8) {
1560                 printk (KERN_ERR
1561                         "EXT3-fs: #inodes per group too big: %lu\n",
1562                         sbi->s_inodes_per_group);
1563                 goto failed_mount;
1564         }
1565
1566         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1567                 goto cantfind_ext3;
1568         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1569                                le32_to_cpu(es->s_first_data_block) +
1570                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1571                               EXT3_BLOCKS_PER_GROUP(sb);
1572         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1573                    EXT3_DESC_PER_BLOCK(sb);
1574         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1575                                     GFP_KERNEL);
1576         if (sbi->s_group_desc == NULL) {
1577                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1578                 goto failed_mount;
1579         }
1580
1581         percpu_counter_init(&sbi->s_freeblocks_counter);
1582         percpu_counter_init(&sbi->s_freeinodes_counter);
1583         percpu_counter_init(&sbi->s_dirs_counter);
1584         bgl_lock_init(&sbi->s_blockgroup_lock);
1585
1586         for (i = 0; i < db_count; i++) {
1587                 block = descriptor_loc(sb, logic_sb_block, i);
1588                 sbi->s_group_desc[i] = sb_bread(sb, block);
1589                 if (!sbi->s_group_desc[i]) {
1590                         printk (KERN_ERR "EXT3-fs: "
1591                                 "can't read group descriptor %d\n", i);
1592                         db_count = i;
1593                         goto failed_mount2;
1594                 }
1595         }
1596         if (!ext3_check_descriptors (sb)) {
1597                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1598                 goto failed_mount2;
1599         }
1600         sbi->s_gdb_count = db_count;
1601         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1602         spin_lock_init(&sbi->s_next_gen_lock);
1603         /* per fileystem reservation list head & lock */
1604         spin_lock_init(&sbi->s_rsv_window_lock);
1605         sbi->s_rsv_window_root = RB_ROOT;
1606         /* Add a single, static dummy reservation to the start of the
1607          * reservation window list --- it gives us a placeholder for
1608          * append-at-start-of-list which makes the allocation logic
1609          * _much_ simpler. */
1610         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1611         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1612         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1613         sbi->s_rsv_window_head.rsv_goal_size = 0;
1614         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1615
1616         /*
1617          * set up enough so that it can read an inode
1618          */
1619         sb->s_op = &ext3_sops;
1620         sb->s_export_op = &ext3_export_ops;
1621         sb->s_xattr = ext3_xattr_handlers;
1622 #ifdef CONFIG_QUOTA
1623         sb->s_qcop = &ext3_qctl_operations;
1624         sb->dq_op = &ext3_quota_operations;
1625 #endif
1626         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1627
1628         sb->s_root = NULL;
1629
1630         needs_recovery = (es->s_last_orphan != 0 ||
1631                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1632                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1633
1634         /*
1635          * The first inode we look at is the journal inode.  Don't try
1636          * root first: it may be modified in the journal!
1637          */
1638         if (!test_opt(sb, NOLOAD) &&
1639             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1640                 if (ext3_load_journal(sb, es))
1641                         goto failed_mount2;
1642         } else if (journal_inum) {
1643                 if (ext3_create_journal(sb, es, journal_inum))
1644                         goto failed_mount2;
1645         } else {
1646                 if (!silent)
1647                         printk (KERN_ERR
1648                                 "ext3: No journal on filesystem on %s\n",
1649                                 sb->s_id);
1650                 goto failed_mount2;
1651         }
1652
1653         /* We have now updated the journal if required, so we can
1654          * validate the data journaling mode. */
1655         switch (test_opt(sb, DATA_FLAGS)) {
1656         case 0:
1657                 /* No mode set, assume a default based on the journal
1658                    capabilities: ORDERED_DATA if the journal can
1659                    cope, else JOURNAL_DATA */
1660                 if (journal_check_available_features
1661                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1662                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1663                 else
1664                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1665                 break;
1666
1667         case EXT3_MOUNT_ORDERED_DATA:
1668         case EXT3_MOUNT_WRITEBACK_DATA:
1669                 if (!journal_check_available_features
1670                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1671                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1672                                "requested data journaling mode\n");
1673                         goto failed_mount3;
1674                 }
1675         default:
1676                 break;
1677         }
1678
1679         if (test_opt(sb, NOBH)) {
1680                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1681                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1682                                 "since filesystem blocksize doesn't match "
1683                                 "pagesize\n");
1684                         clear_opt(sbi->s_mount_opt, NOBH);
1685                 }
1686                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1687                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1688                                 "its supported only with writeback mode\n");
1689                         clear_opt(sbi->s_mount_opt, NOBH);
1690                 }
1691         }
1692         /*
1693          * The journal_load will have done any necessary log recovery,
1694          * so we can safely mount the rest of the filesystem now.
1695          */
1696
1697         root = iget(sb, EXT3_ROOT_INO);
1698         sb->s_root = d_alloc_root(root);
1699         if (!sb->s_root) {
1700                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1701                 iput(root);
1702                 goto failed_mount3;
1703         }
1704         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1705                 dput(sb->s_root);
1706                 sb->s_root = NULL;
1707                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1708                 goto failed_mount3;
1709         }
1710
1711         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1712         /*
1713          * akpm: core read_super() calls in here with the superblock locked.
1714          * That deadlocks, because orphan cleanup needs to lock the superblock
1715          * in numerous places.  Here we just pop the lock - it's relatively
1716          * harmless, because we are now ready to accept write_super() requests,
1717          * and aviro says that's the only reason for hanging onto the
1718          * superblock lock.
1719          */
1720         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1721         ext3_orphan_cleanup(sb, es);
1722         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1723         if (needs_recovery)
1724                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1725         ext3_mark_recovery_complete(sb, es);
1726         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1727                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1728                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1729                 "writeback");
1730
1731         percpu_counter_mod(&sbi->s_freeblocks_counter,
1732                 ext3_count_free_blocks(sb));
1733         percpu_counter_mod(&sbi->s_freeinodes_counter,
1734                 ext3_count_free_inodes(sb));
1735         percpu_counter_mod(&sbi->s_dirs_counter,
1736                 ext3_count_dirs(sb));
1737
1738         lock_kernel();
1739         return 0;
1740
1741 cantfind_ext3:
1742         if (!silent)
1743                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1744                        sb->s_id);
1745         goto failed_mount;
1746
1747 failed_mount3:
1748         journal_destroy(sbi->s_journal);
1749 failed_mount2:
1750         for (i = 0; i < db_count; i++)
1751                 brelse(sbi->s_group_desc[i]);
1752         kfree(sbi->s_group_desc);
1753 failed_mount:
1754 #ifdef CONFIG_QUOTA
1755         for (i = 0; i < MAXQUOTAS; i++)
1756                 kfree(sbi->s_qf_names[i]);
1757 #endif
1758         ext3_blkdev_remove(sbi);
1759         brelse(bh);
1760 out_fail:
1761         sb->s_fs_info = NULL;
1762         kfree(sbi);
1763         lock_kernel();
1764         return -EINVAL;
1765 }
1766
1767 /*
1768  * Setup any per-fs journal parameters now.  We'll do this both on
1769  * initial mount, once the journal has been initialised but before we've
1770  * done any recovery; and again on any subsequent remount. 
1771  */
1772 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1773 {
1774         struct ext3_sb_info *sbi = EXT3_SB(sb);
1775
1776         if (sbi->s_commit_interval)
1777                 journal->j_commit_interval = sbi->s_commit_interval;
1778         /* We could also set up an ext3-specific default for the commit
1779          * interval here, but for now we'll just fall back to the jbd
1780          * default. */
1781
1782         spin_lock(&journal->j_state_lock);
1783         if (test_opt(sb, BARRIER))
1784                 journal->j_flags |= JFS_BARRIER;
1785         else
1786                 journal->j_flags &= ~JFS_BARRIER;
1787         spin_unlock(&journal->j_state_lock);
1788 }
1789
1790 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1791 {
1792         struct inode *journal_inode;
1793         journal_t *journal;
1794
1795         /* First, test for the existence of a valid inode on disk.  Bad
1796          * things happen if we iget() an unused inode, as the subsequent
1797          * iput() will try to delete it. */
1798
1799         journal_inode = iget(sb, journal_inum);
1800         if (!journal_inode) {
1801                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1802                 return NULL;
1803         }
1804         if (!journal_inode->i_nlink) {
1805                 make_bad_inode(journal_inode);
1806                 iput(journal_inode);
1807                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1808                 return NULL;
1809         }
1810
1811         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1812                   journal_inode, journal_inode->i_size);
1813         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1814                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1815                 iput(journal_inode);
1816                 return NULL;
1817         }
1818
1819         journal = journal_init_inode(journal_inode);
1820         if (!journal) {
1821                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1822                 iput(journal_inode);
1823                 return NULL;
1824         }
1825         journal->j_private = sb;
1826         ext3_init_journal_params(sb, journal);
1827         return journal;
1828 }
1829
1830 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1831                                        dev_t j_dev)
1832 {
1833         struct buffer_head * bh;
1834         journal_t *journal;
1835         int start;
1836         int len;
1837         int hblock, blocksize;
1838         unsigned long sb_block;
1839         unsigned long offset;
1840         struct ext3_super_block * es;
1841         struct block_device *bdev;
1842
1843         bdev = ext3_blkdev_get(j_dev);
1844         if (bdev == NULL)
1845                 return NULL;
1846
1847         if (bd_claim(bdev, sb)) {
1848                 printk(KERN_ERR
1849                         "EXT3: failed to claim external journal device.\n");
1850                 blkdev_put(bdev);
1851                 return NULL;
1852         }
1853
1854         blocksize = sb->s_blocksize;
1855         hblock = bdev_hardsect_size(bdev);
1856         if (blocksize < hblock) {
1857                 printk(KERN_ERR
1858                         "EXT3-fs: blocksize too small for journal device.\n");
1859                 goto out_bdev;
1860         }
1861
1862         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1863         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1864         set_blocksize(bdev, blocksize);
1865         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1866                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1867                        "external journal\n");
1868                 goto out_bdev;
1869         }
1870
1871         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1872         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1873             !(le32_to_cpu(es->s_feature_incompat) &
1874               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1875                 printk(KERN_ERR "EXT3-fs: external journal has "
1876                                         "bad superblock\n");
1877                 brelse(bh);
1878                 goto out_bdev;
1879         }
1880
1881         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1882                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1883                 brelse(bh);
1884                 goto out_bdev;
1885         }
1886
1887         len = le32_to_cpu(es->s_blocks_count);
1888         start = sb_block + 1;
1889         brelse(bh);     /* we're done with the superblock */
1890
1891         journal = journal_init_dev(bdev, sb->s_bdev,
1892                                         start, len, blocksize);
1893         if (!journal) {
1894                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1895                 goto out_bdev;
1896         }
1897         journal->j_private = sb;
1898         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1899         wait_on_buffer(journal->j_sb_buffer);
1900         if (!buffer_uptodate(journal->j_sb_buffer)) {
1901                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1902                 goto out_journal;
1903         }
1904         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1905                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1906                                         "user (unsupported) - %d\n",
1907                         be32_to_cpu(journal->j_superblock->s_nr_users));
1908                 goto out_journal;
1909         }
1910         EXT3_SB(sb)->journal_bdev = bdev;
1911         ext3_init_journal_params(sb, journal);
1912         return journal;
1913 out_journal:
1914         journal_destroy(journal);
1915 out_bdev:
1916         ext3_blkdev_put(bdev);
1917         return NULL;
1918 }
1919
1920 static int ext3_load_journal(struct super_block * sb,
1921                              struct ext3_super_block * es)
1922 {
1923         journal_t *journal;
1924         int journal_inum = le32_to_cpu(es->s_journal_inum);
1925         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1926         int err = 0;
1927         int really_read_only;
1928
1929         really_read_only = bdev_read_only(sb->s_bdev);
1930
1931         /*
1932          * Are we loading a blank journal or performing recovery after a
1933          * crash?  For recovery, we need to check in advance whether we
1934          * can get read-write access to the device.
1935          */
1936
1937         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1938                 if (sb->s_flags & MS_RDONLY) {
1939                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1940                                         "required on readonly filesystem.\n");
1941                         if (really_read_only) {
1942                                 printk(KERN_ERR "EXT3-fs: write access "
1943                                         "unavailable, cannot proceed.\n");
1944                                 return -EROFS;
1945                         }
1946                         printk (KERN_INFO "EXT3-fs: write access will "
1947                                         "be enabled during recovery.\n");
1948                 }
1949         }
1950
1951         if (journal_inum && journal_dev) {
1952                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1953                        "and inode journals!\n");
1954                 return -EINVAL;
1955         }
1956
1957         if (journal_inum) {
1958                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1959                         return -EINVAL;
1960         } else {
1961                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1962                         return -EINVAL;
1963         }
1964
1965         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1966                 err = journal_update_format(journal);
1967                 if (err)  {
1968                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1969                         journal_destroy(journal);
1970                         return err;
1971                 }
1972         }
1973
1974         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1975                 err = journal_wipe(journal, !really_read_only);
1976         if (!err)
1977                 err = journal_load(journal);
1978
1979         if (err) {
1980                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1981                 journal_destroy(journal);
1982                 return err;
1983         }
1984
1985         EXT3_SB(sb)->s_journal = journal;
1986         ext3_clear_journal_err(sb, es);
1987         return 0;
1988 }
1989
1990 static int ext3_create_journal(struct super_block * sb,
1991                                struct ext3_super_block * es,
1992                                int journal_inum)
1993 {
1994         journal_t *journal;
1995
1996         if (sb->s_flags & MS_RDONLY) {
1997                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1998                                 "create journal.\n");
1999                 return -EROFS;
2000         }
2001
2002         if (!(journal = ext3_get_journal(sb, journal_inum)))
2003                 return -EINVAL;
2004
2005         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2006                journal_inum);
2007
2008         if (journal_create(journal)) {
2009                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2010                 journal_destroy(journal);
2011                 return -EIO;
2012         }
2013
2014         EXT3_SB(sb)->s_journal = journal;
2015
2016         ext3_update_dynamic_rev(sb);
2017         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2018         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2019
2020         es->s_journal_inum = cpu_to_le32(journal_inum);
2021         sb->s_dirt = 1;
2022
2023         /* Make sure we flush the recovery flag to disk. */
2024         ext3_commit_super(sb, es, 1);
2025
2026         return 0;
2027 }
2028
2029 static void ext3_commit_super (struct super_block * sb,
2030                                struct ext3_super_block * es,
2031                                int sync)
2032 {
2033         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2034
2035         if (!sbh)
2036                 return;
2037         es->s_wtime = cpu_to_le32(get_seconds());
2038         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2039         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2040         BUFFER_TRACE(sbh, "marking dirty");
2041         mark_buffer_dirty(sbh);
2042         if (sync)
2043                 sync_dirty_buffer(sbh);
2044 }
2045
2046
2047 /*
2048  * Have we just finished recovery?  If so, and if we are mounting (or
2049  * remounting) the filesystem readonly, then we will end up with a
2050  * consistent fs on disk.  Record that fact.
2051  */
2052 static void ext3_mark_recovery_complete(struct super_block * sb,
2053                                         struct ext3_super_block * es)
2054 {
2055         journal_t *journal = EXT3_SB(sb)->s_journal;
2056
2057         journal_lock_updates(journal);
2058         journal_flush(journal);
2059         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2060             sb->s_flags & MS_RDONLY) {
2061                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2062                 sb->s_dirt = 0;
2063                 ext3_commit_super(sb, es, 1);
2064         }
2065         journal_unlock_updates(journal);
2066 }
2067
2068 /*
2069  * If we are mounting (or read-write remounting) a filesystem whose journal
2070  * has recorded an error from a previous lifetime, move that error to the
2071  * main filesystem now.
2072  */
2073 static void ext3_clear_journal_err(struct super_block * sb,
2074                                    struct ext3_super_block * es)
2075 {
2076         journal_t *journal;
2077         int j_errno;
2078         const char *errstr;
2079
2080         journal = EXT3_SB(sb)->s_journal;
2081
2082         /*
2083          * Now check for any error status which may have been recorded in the
2084          * journal by a prior ext3_error() or ext3_abort()
2085          */
2086
2087         j_errno = journal_errno(journal);
2088         if (j_errno) {
2089                 char nbuf[16];
2090
2091                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2092                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2093                              "from previous mount: %s", errstr);
2094                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2095                              "filesystem check.");
2096
2097                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2098                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2099                 ext3_commit_super (sb, es, 1);
2100
2101                 journal_clear_err(journal);
2102         }
2103 }
2104
2105 /*
2106  * Force the running and committing transactions to commit,
2107  * and wait on the commit.
2108  */
2109 int ext3_force_commit(struct super_block *sb)
2110 {
2111         journal_t *journal;
2112         int ret;
2113
2114         if (sb->s_flags & MS_RDONLY)
2115                 return 0;
2116
2117         journal = EXT3_SB(sb)->s_journal;
2118         sb->s_dirt = 0;
2119         ret = ext3_journal_force_commit(journal);
2120         return ret;
2121 }
2122
2123 /*
2124  * Ext3 always journals updates to the superblock itself, so we don't
2125  * have to propagate any other updates to the superblock on disk at this
2126  * point.  Just start an async writeback to get the buffers on their way
2127  * to the disk.
2128  *
2129  * This implicitly triggers the writebehind on sync().
2130  */
2131
2132 static void ext3_write_super (struct super_block * sb)
2133 {
2134         if (down_trylock(&sb->s_lock) == 0)
2135                 BUG();
2136         sb->s_dirt = 0;
2137 }
2138
2139 static int ext3_sync_fs(struct super_block *sb, int wait)
2140 {
2141         tid_t target;
2142
2143         sb->s_dirt = 0;
2144         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2145                 if (wait)
2146                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2147         }
2148         return 0;
2149 }
2150
2151 /*
2152  * LVM calls this function before a (read-only) snapshot is created.  This
2153  * gives us a chance to flush the journal completely and mark the fs clean.
2154  */
2155 static void ext3_write_super_lockfs(struct super_block *sb)
2156 {
2157         sb->s_dirt = 0;
2158
2159         if (!(sb->s_flags & MS_RDONLY)) {
2160                 journal_t *journal = EXT3_SB(sb)->s_journal;
2161
2162                 /* Now we set up the journal barrier. */
2163                 journal_lock_updates(journal);
2164                 journal_flush(journal);
2165
2166                 /* Journal blocked and flushed, clear needs_recovery flag. */
2167                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2168                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2169         }
2170 }
2171
2172 /*
2173  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2174  * flag here, even though the filesystem is not technically dirty yet.
2175  */
2176 static void ext3_unlockfs(struct super_block *sb)
2177 {
2178         if (!(sb->s_flags & MS_RDONLY)) {
2179                 lock_super(sb);
2180                 /* Reser the needs_recovery flag before the fs is unlocked. */
2181                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2182                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2183                 unlock_super(sb);
2184                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2185         }
2186 }
2187
2188 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2189 {
2190         struct ext3_super_block * es;
2191         struct ext3_sb_info *sbi = EXT3_SB(sb);
2192         unsigned long n_blocks_count = 0;
2193         unsigned long old_sb_flags;
2194         struct ext3_mount_options old_opts;
2195         int err;
2196 #ifdef CONFIG_QUOTA
2197         int i;
2198 #endif
2199
2200         /* Store the original options */
2201         old_sb_flags = sb->s_flags;
2202         old_opts.s_mount_opt = sbi->s_mount_opt;
2203         old_opts.s_resuid = sbi->s_resuid;
2204         old_opts.s_resgid = sbi->s_resgid;
2205         old_opts.s_commit_interval = sbi->s_commit_interval;
2206 #ifdef CONFIG_QUOTA
2207         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2208         for (i = 0; i < MAXQUOTAS; i++)
2209                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2210 #endif
2211
2212         /*
2213          * Allow the "check" option to be passed as a remount option.
2214          */
2215         if (!parse_options(data, sb, NULL, &n_blocks_count, 1)) {
2216                 err = -EINVAL;
2217                 goto restore_opts;
2218         }
2219
2220         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2221                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2222
2223         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2224                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2225
2226         es = sbi->s_es;
2227
2228         ext3_init_journal_params(sb, sbi->s_journal);
2229
2230         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2231                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2232                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2233                         err = -EROFS;
2234                         goto restore_opts;
2235                 }
2236
2237                 if (*flags & MS_RDONLY) {
2238                         /*
2239                          * First of all, the unconditional stuff we have to do
2240                          * to disable replay of the journal when we next remount
2241                          */
2242                         sb->s_flags |= MS_RDONLY;
2243
2244                         /*
2245                          * OK, test if we are remounting a valid rw partition
2246                          * readonly, and if so set the rdonly flag and then
2247                          * mark the partition as valid again.
2248                          */
2249                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2250                             (sbi->s_mount_state & EXT3_VALID_FS))
2251                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2252
2253                         ext3_mark_recovery_complete(sb, es);
2254                 } else {
2255                         __le32 ret;
2256                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2257                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2258                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2259                                        "remount RDWR because of unsupported "
2260                                        "optional features (%x).\n",
2261                                        sb->s_id, le32_to_cpu(ret));
2262                                 err = -EROFS;
2263                                 goto restore_opts;
2264                         }
2265                         /*
2266                          * Mounting a RDONLY partition read-write, so reread
2267                          * and store the current valid flag.  (It may have
2268                          * been changed by e2fsck since we originally mounted
2269                          * the partition.)
2270                          */
2271                         ext3_clear_journal_err(sb, es);
2272                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2273                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2274                                 err = ret;
2275                                 goto restore_opts;
2276                         }
2277                         if (!ext3_setup_super (sb, es, 0))
2278                                 sb->s_flags &= ~MS_RDONLY;
2279                 }
2280         }
2281 #ifdef CONFIG_QUOTA
2282         /* Release old quota file names */
2283         for (i = 0; i < MAXQUOTAS; i++)
2284                 if (old_opts.s_qf_names[i] &&
2285                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2286                         kfree(old_opts.s_qf_names[i]);
2287 #endif
2288         return 0;
2289 restore_opts:
2290         sb->s_flags = old_sb_flags;
2291         sbi->s_mount_opt = old_opts.s_mount_opt;
2292         sbi->s_resuid = old_opts.s_resuid;
2293         sbi->s_resgid = old_opts.s_resgid;
2294         sbi->s_commit_interval = old_opts.s_commit_interval;
2295 #ifdef CONFIG_QUOTA
2296         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2297         for (i = 0; i < MAXQUOTAS; i++) {
2298                 if (sbi->s_qf_names[i] &&
2299                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2300                         kfree(sbi->s_qf_names[i]);
2301                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2302         }
2303 #endif
2304         return err;
2305 }
2306
2307 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2308 {
2309         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2310         unsigned long overhead;
2311         int i;
2312
2313         if (test_opt (sb, MINIX_DF))
2314                 overhead = 0;
2315         else {
2316                 unsigned long ngroups;
2317                 ngroups = EXT3_SB(sb)->s_groups_count;
2318                 smp_rmb();
2319
2320                 /*
2321                  * Compute the overhead (FS structures)
2322                  */
2323
2324                 /*
2325                  * All of the blocks before first_data_block are
2326                  * overhead
2327                  */
2328                 overhead = le32_to_cpu(es->s_first_data_block);
2329
2330                 /*
2331                  * Add the overhead attributed to the superblock and
2332                  * block group descriptors.  If the sparse superblocks
2333                  * feature is turned on, then not all groups have this.
2334                  */
2335                 for (i = 0; i < ngroups; i++) {
2336                         overhead += ext3_bg_has_super(sb, i) +
2337                                 ext3_bg_num_gdb(sb, i);
2338                         cond_resched();
2339                 }
2340
2341                 /*
2342                  * Every block group has an inode bitmap, a block
2343                  * bitmap, and an inode table.
2344                  */
2345                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2346         }
2347
2348         buf->f_type = EXT3_SUPER_MAGIC;
2349         buf->f_bsize = sb->s_blocksize;
2350         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2351         buf->f_bfree = ext3_count_free_blocks (sb);
2352         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2353         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2354                 buf->f_bavail = 0;
2355         buf->f_files = le32_to_cpu(es->s_inodes_count);
2356         buf->f_ffree = ext3_count_free_inodes (sb);
2357         buf->f_namelen = EXT3_NAME_LEN;
2358         return 0;
2359 }
2360
2361 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2362  * is locked for write. Otherwise the are possible deadlocks:
2363  * Process 1                         Process 2
2364  * ext3_create()                     quota_sync()
2365  *   journal_start()                   write_dquot()
2366  *   DQUOT_INIT()                        down(dqio_sem)
2367  *     down(dqio_sem)                    journal_start()
2368  *
2369  */
2370
2371 #ifdef CONFIG_QUOTA
2372
2373 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2374 {
2375         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2376 }
2377
2378 static int ext3_dquot_initialize(struct inode *inode, int type)
2379 {
2380         handle_t *handle;
2381         int ret, err;
2382
2383         /* We may create quota structure so we need to reserve enough blocks */
2384         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2385         if (IS_ERR(handle))
2386                 return PTR_ERR(handle);
2387         ret = dquot_initialize(inode, type);
2388         err = ext3_journal_stop(handle);
2389         if (!ret)
2390                 ret = err;
2391         return ret;
2392 }
2393
2394 static int ext3_dquot_drop(struct inode *inode)
2395 {
2396         handle_t *handle;
2397         int ret, err;
2398
2399         /* We may delete quota structure so we need to reserve enough blocks */
2400         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2401         if (IS_ERR(handle))
2402                 return PTR_ERR(handle);
2403         ret = dquot_drop(inode);
2404         err = ext3_journal_stop(handle);
2405         if (!ret)
2406                 ret = err;
2407         return ret;
2408 }
2409
2410 static int ext3_write_dquot(struct dquot *dquot)
2411 {
2412         int ret, err;
2413         handle_t *handle;
2414         struct inode *inode;
2415
2416         inode = dquot_to_inode(dquot);
2417         handle = ext3_journal_start(inode,
2418                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2419         if (IS_ERR(handle))
2420                 return PTR_ERR(handle);
2421         ret = dquot_commit(dquot);
2422         err = ext3_journal_stop(handle);
2423         if (!ret)
2424                 ret = err;
2425         return ret;
2426 }
2427
2428 static int ext3_acquire_dquot(struct dquot *dquot)
2429 {
2430         int ret, err;
2431         handle_t *handle;
2432
2433         handle = ext3_journal_start(dquot_to_inode(dquot),
2434                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2435         if (IS_ERR(handle))
2436                 return PTR_ERR(handle);
2437         ret = dquot_acquire(dquot);
2438         err = ext3_journal_stop(handle);
2439         if (!ret)
2440                 ret = err;
2441         return ret;
2442 }
2443
2444 static int ext3_release_dquot(struct dquot *dquot)
2445 {
2446         int ret, err;
2447         handle_t *handle;
2448
2449         handle = ext3_journal_start(dquot_to_inode(dquot),
2450                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2451         if (IS_ERR(handle))
2452                 return PTR_ERR(handle);
2453         ret = dquot_release(dquot);
2454         err = ext3_journal_stop(handle);
2455         if (!ret)
2456                 ret = err;
2457         return ret;
2458 }
2459
2460 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2461 {
2462         /* Are we journalling quotas? */
2463         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2464             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2465                 dquot_mark_dquot_dirty(dquot);
2466                 return ext3_write_dquot(dquot);
2467         } else {
2468                 return dquot_mark_dquot_dirty(dquot);
2469         }
2470 }
2471
2472 static int ext3_write_info(struct super_block *sb, int type)
2473 {
2474         int ret, err;
2475         handle_t *handle;
2476
2477         /* Data block + inode block */
2478         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2479         if (IS_ERR(handle))
2480                 return PTR_ERR(handle);
2481         ret = dquot_commit_info(sb, type);
2482         err = ext3_journal_stop(handle);
2483         if (!ret)
2484                 ret = err;
2485         return ret;
2486 }
2487
2488 /*
2489  * Turn on quotas during mount time - we need to find
2490  * the quota file and such...
2491  */
2492 static int ext3_quota_on_mount(struct super_block *sb, int type)
2493 {
2494         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2495                         EXT3_SB(sb)->s_jquota_fmt, type);
2496 }
2497
2498 /*
2499  * Standard function to be called on quota_on
2500  */
2501 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2502                          char *path)
2503 {
2504         int err;
2505         struct nameidata nd;
2506
2507         if (!test_opt(sb, QUOTA))
2508                 return -EINVAL;
2509         /* Not journalling quota? */
2510         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2511             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2512                 return vfs_quota_on(sb, type, format_id, path);
2513         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2514         if (err)
2515                 return err;
2516         /* Quotafile not on the same filesystem? */
2517         if (nd.mnt->mnt_sb != sb) {
2518                 path_release(&nd);
2519                 return -EXDEV;
2520         }
2521         /* Quotafile not of fs root? */
2522         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2523                 printk(KERN_WARNING
2524                         "EXT3-fs: Quota file not on filesystem root. "
2525                         "Journalled quota will not work.\n");
2526         path_release(&nd);
2527         return vfs_quota_on(sb, type, format_id, path);
2528 }
2529
2530 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2531  * acquiring the locks... As quota files are never truncated and quota code
2532  * itself serializes the operations (and noone else should touch the files)
2533  * we don't have to be afraid of races */
2534 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2535                                size_t len, loff_t off)
2536 {
2537         struct inode *inode = sb_dqopt(sb)->files[type];
2538         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2539         int err = 0;
2540         int offset = off & (sb->s_blocksize - 1);
2541         int tocopy;
2542         size_t toread;
2543         struct buffer_head *bh;
2544         loff_t i_size = i_size_read(inode);
2545
2546         if (off > i_size)
2547                 return 0;
2548         if (off+len > i_size)
2549                 len = i_size-off;
2550         toread = len;
2551         while (toread > 0) {
2552                 tocopy = sb->s_blocksize - offset < toread ?
2553                                 sb->s_blocksize - offset : toread;
2554                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2555                 if (err)
2556                         return err;
2557                 if (!bh)        /* A hole? */
2558                         memset(data, 0, tocopy);
2559                 else
2560                         memcpy(data, bh->b_data+offset, tocopy);
2561                 brelse(bh);
2562                 offset = 0;
2563                 toread -= tocopy;
2564                 data += tocopy;
2565                 blk++;
2566         }
2567         return len;
2568 }
2569
2570 /* Write to quotafile (we know the transaction is already started and has
2571  * enough credits) */
2572 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2573                                 const char *data, size_t len, loff_t off)
2574 {
2575         struct inode *inode = sb_dqopt(sb)->files[type];
2576         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2577         int err = 0;
2578         int offset = off & (sb->s_blocksize - 1);
2579         int tocopy;
2580         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2581         size_t towrite = len;
2582         struct buffer_head *bh;
2583         handle_t *handle = journal_current_handle();
2584
2585         down(&inode->i_sem);
2586         while (towrite > 0) {
2587                 tocopy = sb->s_blocksize - offset < towrite ?
2588                                 sb->s_blocksize - offset : towrite;
2589                 bh = ext3_bread(handle, inode, blk, 1, &err);
2590                 if (!bh)
2591                         goto out;
2592                 if (journal_quota) {
2593                         err = ext3_journal_get_write_access(handle, bh);
2594                         if (err) {
2595                                 brelse(bh);
2596                                 goto out;
2597                         }
2598                 }
2599                 lock_buffer(bh);
2600                 memcpy(bh->b_data+offset, data, tocopy);
2601                 flush_dcache_page(bh->b_page);
2602                 unlock_buffer(bh);
2603                 if (journal_quota)
2604                         err = ext3_journal_dirty_metadata(handle, bh);
2605                 else {
2606                         /* Always do at least ordered writes for quotas */
2607                         err = ext3_journal_dirty_data(handle, bh);
2608                         mark_buffer_dirty(bh);
2609                 }
2610                 brelse(bh);
2611                 if (err)
2612                         goto out;
2613                 offset = 0;
2614                 towrite -= tocopy;
2615                 data += tocopy;
2616                 blk++;
2617         }
2618 out:
2619         if (len == towrite)
2620                 return err;
2621         if (inode->i_size < off+len-towrite) {
2622                 i_size_write(inode, off+len-towrite);
2623                 EXT3_I(inode)->i_disksize = inode->i_size;
2624         }
2625         inode->i_version++;
2626         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2627         ext3_mark_inode_dirty(handle, inode);
2628         up(&inode->i_sem);
2629         return len - towrite;
2630 }
2631
2632 #endif
2633
2634 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2635         int flags, const char *dev_name, void *data)
2636 {
2637         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2638 }
2639
2640 static struct file_system_type ext3_fs_type = {
2641         .owner          = THIS_MODULE,
2642         .name           = "ext3",
2643         .get_sb         = ext3_get_sb,
2644         .kill_sb        = kill_block_super,
2645         .fs_flags       = FS_REQUIRES_DEV,
2646 };
2647
2648 static int __init init_ext3_fs(void)
2649 {
2650         int err = init_ext3_xattr();
2651         if (err)
2652                 return err;
2653         err = init_inodecache();
2654         if (err)
2655                 goto out1;
2656         err = register_filesystem(&ext3_fs_type);
2657         if (err)
2658                 goto out;
2659         return 0;
2660 out:
2661         destroy_inodecache();
2662 out1:
2663         exit_ext3_xattr();
2664         return err;
2665 }
2666
2667 static void __exit exit_ext3_fs(void)
2668 {
2669         unregister_filesystem(&ext3_fs_type);
2670         destroy_inodecache();
2671         exit_ext3_xattr();
2672 }
2673
2674 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2675 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2676 MODULE_LICENSE("GPL");
2677 module_init(init_ext3_fs)
2678 module_exit(exit_ext3_fs)