Merge master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-misc-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/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/smp_lock.h>
31 #include <linux/buffer_head.h>
32 #include <linux/vfs.h>
33 #include <linux/random.h>
34 #include <linux/mount.h>
35 #include <linux/namei.h>
36 #include <linux/quotaops.h>
37 #include <linux/seq_file.h>
38
39 #include <asm/uaccess.h>
40
41 #include "xattr.h"
42 #include "acl.h"
43 #include "namei.h"
44
45 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
46                              unsigned long journal_devnum);
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 dentry * dentry, 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                 mutex_init(&ei->truncate_mutex);
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                                                 SLAB_MEM_SPREAD),
485                                              init_once, NULL);
486         if (ext3_inode_cachep == NULL)
487                 return -ENOMEM;
488         return 0;
489 }
490
491 static void destroy_inodecache(void)
492 {
493         if (kmem_cache_destroy(ext3_inode_cachep))
494                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
495 }
496
497 static void ext3_clear_inode(struct inode *inode)
498 {
499         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
500 #ifdef CONFIG_EXT3_FS_POSIX_ACL
501         if (EXT3_I(inode)->i_acl &&
502                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
503                 posix_acl_release(EXT3_I(inode)->i_acl);
504                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
505         }
506         if (EXT3_I(inode)->i_default_acl &&
507                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
508                 posix_acl_release(EXT3_I(inode)->i_default_acl);
509                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
510         }
511 #endif
512         ext3_discard_reservation(inode);
513         EXT3_I(inode)->i_block_alloc_info = NULL;
514         if (unlikely(rsv))
515                 kfree(rsv);
516 }
517
518 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
519 {
520 #if defined(CONFIG_QUOTA)
521         struct ext3_sb_info *sbi = EXT3_SB(sb);
522
523         if (sbi->s_jquota_fmt)
524                 seq_printf(seq, ",jqfmt=%s",
525                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
526
527         if (sbi->s_qf_names[USRQUOTA])
528                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
529
530         if (sbi->s_qf_names[GRPQUOTA])
531                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
532
533         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
534                 seq_puts(seq, ",usrquota");
535
536         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
537                 seq_puts(seq, ",grpquota");
538 #endif
539 }
540
541 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
542 {
543         struct super_block *sb = vfs->mnt_sb;
544
545         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
546                 seq_puts(seq, ",data=journal");
547         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
548                 seq_puts(seq, ",data=ordered");
549         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
550                 seq_puts(seq, ",data=writeback");
551
552         ext3_show_quota_options(seq, sb);
553
554         return 0;
555 }
556
557 #ifdef CONFIG_QUOTA
558 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
559 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
560
561 static int ext3_dquot_initialize(struct inode *inode, int type);
562 static int ext3_dquot_drop(struct inode *inode);
563 static int ext3_write_dquot(struct dquot *dquot);
564 static int ext3_acquire_dquot(struct dquot *dquot);
565 static int ext3_release_dquot(struct dquot *dquot);
566 static int ext3_mark_dquot_dirty(struct dquot *dquot);
567 static int ext3_write_info(struct super_block *sb, int type);
568 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
569 static int ext3_quota_on_mount(struct super_block *sb, int type);
570 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
571                                size_t len, loff_t off);
572 static ssize_t ext3_quota_write(struct super_block *sb, int type,
573                                 const char *data, size_t len, loff_t off);
574
575 static struct dquot_operations ext3_quota_operations = {
576         .initialize     = ext3_dquot_initialize,
577         .drop           = ext3_dquot_drop,
578         .alloc_space    = dquot_alloc_space,
579         .alloc_inode    = dquot_alloc_inode,
580         .free_space     = dquot_free_space,
581         .free_inode     = dquot_free_inode,
582         .transfer       = dquot_transfer,
583         .write_dquot    = ext3_write_dquot,
584         .acquire_dquot  = ext3_acquire_dquot,
585         .release_dquot  = ext3_release_dquot,
586         .mark_dirty     = ext3_mark_dquot_dirty,
587         .write_info     = ext3_write_info
588 };
589
590 static struct quotactl_ops ext3_qctl_operations = {
591         .quota_on       = ext3_quota_on,
592         .quota_off      = vfs_quota_off,
593         .quota_sync     = vfs_quota_sync,
594         .get_info       = vfs_get_dqinfo,
595         .set_info       = vfs_set_dqinfo,
596         .get_dqblk      = vfs_get_dqblk,
597         .set_dqblk      = vfs_set_dqblk
598 };
599 #endif
600
601 static struct super_operations ext3_sops = {
602         .alloc_inode    = ext3_alloc_inode,
603         .destroy_inode  = ext3_destroy_inode,
604         .read_inode     = ext3_read_inode,
605         .write_inode    = ext3_write_inode,
606         .dirty_inode    = ext3_dirty_inode,
607         .delete_inode   = ext3_delete_inode,
608         .put_super      = ext3_put_super,
609         .write_super    = ext3_write_super,
610         .sync_fs        = ext3_sync_fs,
611         .write_super_lockfs = ext3_write_super_lockfs,
612         .unlockfs       = ext3_unlockfs,
613         .statfs         = ext3_statfs,
614         .remount_fs     = ext3_remount,
615         .clear_inode    = ext3_clear_inode,
616         .show_options   = ext3_show_options,
617 #ifdef CONFIG_QUOTA
618         .quota_read     = ext3_quota_read,
619         .quota_write    = ext3_quota_write,
620 #endif
621 };
622
623 static struct export_operations ext3_export_ops = {
624         .get_parent = ext3_get_parent,
625 };
626
627 enum {
628         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
629         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
630         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
631         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
632         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
633         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
634         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
635         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
636         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
637         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
638         Opt_grpquota
639 };
640
641 static match_table_t tokens = {
642         {Opt_bsd_df, "bsddf"},
643         {Opt_minix_df, "minixdf"},
644         {Opt_grpid, "grpid"},
645         {Opt_grpid, "bsdgroups"},
646         {Opt_nogrpid, "nogrpid"},
647         {Opt_nogrpid, "sysvgroups"},
648         {Opt_resgid, "resgid=%u"},
649         {Opt_resuid, "resuid=%u"},
650         {Opt_sb, "sb=%u"},
651         {Opt_err_cont, "errors=continue"},
652         {Opt_err_panic, "errors=panic"},
653         {Opt_err_ro, "errors=remount-ro"},
654         {Opt_nouid32, "nouid32"},
655         {Opt_nocheck, "nocheck"},
656         {Opt_nocheck, "check=none"},
657         {Opt_debug, "debug"},
658         {Opt_oldalloc, "oldalloc"},
659         {Opt_orlov, "orlov"},
660         {Opt_user_xattr, "user_xattr"},
661         {Opt_nouser_xattr, "nouser_xattr"},
662         {Opt_acl, "acl"},
663         {Opt_noacl, "noacl"},
664         {Opt_reservation, "reservation"},
665         {Opt_noreservation, "noreservation"},
666         {Opt_noload, "noload"},
667         {Opt_nobh, "nobh"},
668         {Opt_bh, "bh"},
669         {Opt_commit, "commit=%u"},
670         {Opt_journal_update, "journal=update"},
671         {Opt_journal_inum, "journal=%u"},
672         {Opt_journal_dev, "journal_dev=%u"},
673         {Opt_abort, "abort"},
674         {Opt_data_journal, "data=journal"},
675         {Opt_data_ordered, "data=ordered"},
676         {Opt_data_writeback, "data=writeback"},
677         {Opt_offusrjquota, "usrjquota="},
678         {Opt_usrjquota, "usrjquota=%s"},
679         {Opt_offgrpjquota, "grpjquota="},
680         {Opt_grpjquota, "grpjquota=%s"},
681         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
682         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
683         {Opt_grpquota, "grpquota"},
684         {Opt_noquota, "noquota"},
685         {Opt_quota, "quota"},
686         {Opt_usrquota, "usrquota"},
687         {Opt_barrier, "barrier=%u"},
688         {Opt_err, NULL},
689         {Opt_resize, "resize"},
690 };
691
692 static ext3_fsblk_t get_sb_block(void **data)
693 {
694         ext3_fsblk_t    sb_block;
695         char            *options = (char *) *data;
696
697         if (!options || strncmp(options, "sb=", 3) != 0)
698                 return 1;       /* Default location */
699         options += 3;
700         /*todo: use simple_strtoll with >32bit ext3 */
701         sb_block = simple_strtoul(options, &options, 0);
702         if (*options && *options != ',') {
703                 printk("EXT3-fs: Invalid sb specification: %s\n",
704                        (char *) *data);
705                 return 1;
706         }
707         if (*options == ',')
708                 options++;
709         *data = (void *) options;
710         return sb_block;
711 }
712
713 static int parse_options (char *options, struct super_block *sb,
714                           unsigned long *inum, unsigned long *journal_devnum,
715                           ext3_fsblk_t *n_blocks_count, int is_remount)
716 {
717         struct ext3_sb_info *sbi = EXT3_SB(sb);
718         char * p;
719         substring_t args[MAX_OPT_ARGS];
720         int data_opt = 0;
721         int option;
722 #ifdef CONFIG_QUOTA
723         int qtype;
724         char *qname;
725 #endif
726
727         if (!options)
728                 return 1;
729
730         while ((p = strsep (&options, ",")) != NULL) {
731                 int token;
732                 if (!*p)
733                         continue;
734
735                 token = match_token(p, tokens, args);
736                 switch (token) {
737                 case Opt_bsd_df:
738                         clear_opt (sbi->s_mount_opt, MINIX_DF);
739                         break;
740                 case Opt_minix_df:
741                         set_opt (sbi->s_mount_opt, MINIX_DF);
742                         break;
743                 case Opt_grpid:
744                         set_opt (sbi->s_mount_opt, GRPID);
745                         break;
746                 case Opt_nogrpid:
747                         clear_opt (sbi->s_mount_opt, GRPID);
748                         break;
749                 case Opt_resuid:
750                         if (match_int(&args[0], &option))
751                                 return 0;
752                         sbi->s_resuid = option;
753                         break;
754                 case Opt_resgid:
755                         if (match_int(&args[0], &option))
756                                 return 0;
757                         sbi->s_resgid = option;
758                         break;
759                 case Opt_sb:
760                         /* handled by get_sb_block() instead of here */
761                         /* *sb_block = match_int(&args[0]); */
762                         break;
763                 case Opt_err_panic:
764                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
765                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
766                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
767                         break;
768                 case Opt_err_ro:
769                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
770                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
771                         set_opt (sbi->s_mount_opt, ERRORS_RO);
772                         break;
773                 case Opt_err_cont:
774                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
775                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
776                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
777                         break;
778                 case Opt_nouid32:
779                         set_opt (sbi->s_mount_opt, NO_UID32);
780                         break;
781                 case Opt_nocheck:
782                         clear_opt (sbi->s_mount_opt, CHECK);
783                         break;
784                 case Opt_debug:
785                         set_opt (sbi->s_mount_opt, DEBUG);
786                         break;
787                 case Opt_oldalloc:
788                         set_opt (sbi->s_mount_opt, OLDALLOC);
789                         break;
790                 case Opt_orlov:
791                         clear_opt (sbi->s_mount_opt, OLDALLOC);
792                         break;
793 #ifdef CONFIG_EXT3_FS_XATTR
794                 case Opt_user_xattr:
795                         set_opt (sbi->s_mount_opt, XATTR_USER);
796                         break;
797                 case Opt_nouser_xattr:
798                         clear_opt (sbi->s_mount_opt, XATTR_USER);
799                         break;
800 #else
801                 case Opt_user_xattr:
802                 case Opt_nouser_xattr:
803                         printk("EXT3 (no)user_xattr options not supported\n");
804                         break;
805 #endif
806 #ifdef CONFIG_EXT3_FS_POSIX_ACL
807                 case Opt_acl:
808                         set_opt(sbi->s_mount_opt, POSIX_ACL);
809                         break;
810                 case Opt_noacl:
811                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
812                         break;
813 #else
814                 case Opt_acl:
815                 case Opt_noacl:
816                         printk("EXT3 (no)acl options not supported\n");
817                         break;
818 #endif
819                 case Opt_reservation:
820                         set_opt(sbi->s_mount_opt, RESERVATION);
821                         break;
822                 case Opt_noreservation:
823                         clear_opt(sbi->s_mount_opt, RESERVATION);
824                         break;
825                 case Opt_journal_update:
826                         /* @@@ FIXME */
827                         /* Eventually we will want to be able to create
828                            a journal file here.  For now, only allow the
829                            user to specify an existing inode to be the
830                            journal file. */
831                         if (is_remount) {
832                                 printk(KERN_ERR "EXT3-fs: cannot specify "
833                                        "journal on remount\n");
834                                 return 0;
835                         }
836                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
837                         break;
838                 case Opt_journal_inum:
839                         if (is_remount) {
840                                 printk(KERN_ERR "EXT3-fs: cannot specify "
841                                        "journal on remount\n");
842                                 return 0;
843                         }
844                         if (match_int(&args[0], &option))
845                                 return 0;
846                         *inum = option;
847                         break;
848                 case Opt_journal_dev:
849                         if (is_remount) {
850                                 printk(KERN_ERR "EXT3-fs: cannot specify "
851                                        "journal on remount\n");
852                                 return 0;
853                         }
854                         if (match_int(&args[0], &option))
855                                 return 0;
856                         *journal_devnum = option;
857                         break;
858                 case Opt_noload:
859                         set_opt (sbi->s_mount_opt, NOLOAD);
860                         break;
861                 case Opt_commit:
862                         if (match_int(&args[0], &option))
863                                 return 0;
864                         if (option < 0)
865                                 return 0;
866                         if (option == 0)
867                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
868                         sbi->s_commit_interval = HZ * option;
869                         break;
870                 case Opt_data_journal:
871                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
872                         goto datacheck;
873                 case Opt_data_ordered:
874                         data_opt = EXT3_MOUNT_ORDERED_DATA;
875                         goto datacheck;
876                 case Opt_data_writeback:
877                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
878                 datacheck:
879                         if (is_remount) {
880                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
881                                                 != data_opt) {
882                                         printk(KERN_ERR
883                                                 "EXT3-fs: cannot change data "
884                                                 "mode on remount\n");
885                                         return 0;
886                                 }
887                         } else {
888                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
889                                 sbi->s_mount_opt |= data_opt;
890                         }
891                         break;
892 #ifdef CONFIG_QUOTA
893                 case Opt_usrjquota:
894                         qtype = USRQUOTA;
895                         goto set_qf_name;
896                 case Opt_grpjquota:
897                         qtype = GRPQUOTA;
898 set_qf_name:
899                         if (sb_any_quota_enabled(sb)) {
900                                 printk(KERN_ERR
901                                         "EXT3-fs: Cannot change journalled "
902                                         "quota options when quota turned on.\n");
903                                 return 0;
904                         }
905                         qname = match_strdup(&args[0]);
906                         if (!qname) {
907                                 printk(KERN_ERR
908                                         "EXT3-fs: not enough memory for "
909                                         "storing quotafile name.\n");
910                                 return 0;
911                         }
912                         if (sbi->s_qf_names[qtype] &&
913                             strcmp(sbi->s_qf_names[qtype], qname)) {
914                                 printk(KERN_ERR
915                                         "EXT3-fs: %s quota file already "
916                                         "specified.\n", QTYPE2NAME(qtype));
917                                 kfree(qname);
918                                 return 0;
919                         }
920                         sbi->s_qf_names[qtype] = qname;
921                         if (strchr(sbi->s_qf_names[qtype], '/')) {
922                                 printk(KERN_ERR
923                                         "EXT3-fs: quotafile must be on "
924                                         "filesystem root.\n");
925                                 kfree(sbi->s_qf_names[qtype]);
926                                 sbi->s_qf_names[qtype] = NULL;
927                                 return 0;
928                         }
929                         set_opt(sbi->s_mount_opt, QUOTA);
930                         break;
931                 case Opt_offusrjquota:
932                         qtype = USRQUOTA;
933                         goto clear_qf_name;
934                 case Opt_offgrpjquota:
935                         qtype = GRPQUOTA;
936 clear_qf_name:
937                         if (sb_any_quota_enabled(sb)) {
938                                 printk(KERN_ERR "EXT3-fs: Cannot change "
939                                         "journalled quota options when "
940                                         "quota turned on.\n");
941                                 return 0;
942                         }
943                         /*
944                          * The space will be released later when all options
945                          * are confirmed to be correct
946                          */
947                         sbi->s_qf_names[qtype] = NULL;
948                         break;
949                 case Opt_jqfmt_vfsold:
950                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
951                         break;
952                 case Opt_jqfmt_vfsv0:
953                         sbi->s_jquota_fmt = QFMT_VFS_V0;
954                         break;
955                 case Opt_quota:
956                 case Opt_usrquota:
957                         set_opt(sbi->s_mount_opt, QUOTA);
958                         set_opt(sbi->s_mount_opt, USRQUOTA);
959                         break;
960                 case Opt_grpquota:
961                         set_opt(sbi->s_mount_opt, QUOTA);
962                         set_opt(sbi->s_mount_opt, GRPQUOTA);
963                         break;
964                 case Opt_noquota:
965                         if (sb_any_quota_enabled(sb)) {
966                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
967                                         "options when quota turned on.\n");
968                                 return 0;
969                         }
970                         clear_opt(sbi->s_mount_opt, QUOTA);
971                         clear_opt(sbi->s_mount_opt, USRQUOTA);
972                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
973                         break;
974 #else
975                 case Opt_quota:
976                 case Opt_usrquota:
977                 case Opt_grpquota:
978                 case Opt_usrjquota:
979                 case Opt_grpjquota:
980                 case Opt_offusrjquota:
981                 case Opt_offgrpjquota:
982                 case Opt_jqfmt_vfsold:
983                 case Opt_jqfmt_vfsv0:
984                         printk(KERN_ERR
985                                 "EXT3-fs: journalled quota options not "
986                                 "supported.\n");
987                         break;
988                 case Opt_noquota:
989                         break;
990 #endif
991                 case Opt_abort:
992                         set_opt(sbi->s_mount_opt, ABORT);
993                         break;
994                 case Opt_barrier:
995                         if (match_int(&args[0], &option))
996                                 return 0;
997                         if (option)
998                                 set_opt(sbi->s_mount_opt, BARRIER);
999                         else
1000                                 clear_opt(sbi->s_mount_opt, BARRIER);
1001                         break;
1002                 case Opt_ignore:
1003                         break;
1004                 case Opt_resize:
1005                         if (!is_remount) {
1006                                 printk("EXT3-fs: resize option only available "
1007                                         "for remount\n");
1008                                 return 0;
1009                         }
1010                         if (match_int(&args[0], &option) != 0)
1011                                 return 0;
1012                         *n_blocks_count = option;
1013                         break;
1014                 case Opt_nobh:
1015                         set_opt(sbi->s_mount_opt, NOBH);
1016                         break;
1017                 case Opt_bh:
1018                         clear_opt(sbi->s_mount_opt, NOBH);
1019                         break;
1020                 default:
1021                         printk (KERN_ERR
1022                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1023                                 "or missing value\n", p);
1024                         return 0;
1025                 }
1026         }
1027 #ifdef CONFIG_QUOTA
1028         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1029                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1030                      sbi->s_qf_names[USRQUOTA])
1031                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1032
1033                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1034                      sbi->s_qf_names[GRPQUOTA])
1035                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1036
1037                 if ((sbi->s_qf_names[USRQUOTA] &&
1038                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1039                     (sbi->s_qf_names[GRPQUOTA] &&
1040                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1041                         printk(KERN_ERR "EXT3-fs: old and new quota "
1042                                         "format mixing.\n");
1043                         return 0;
1044                 }
1045
1046                 if (!sbi->s_jquota_fmt) {
1047                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1048                                         "not specified.\n");
1049                         return 0;
1050                 }
1051         } else {
1052                 if (sbi->s_jquota_fmt) {
1053                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1054                                         "specified with no journalling "
1055                                         "enabled.\n");
1056                         return 0;
1057                 }
1058         }
1059 #endif
1060         return 1;
1061 }
1062
1063 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1064                             int read_only)
1065 {
1066         struct ext3_sb_info *sbi = EXT3_SB(sb);
1067         int res = 0;
1068
1069         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1070                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1071                         "forcing read-only mode\n");
1072                 res = MS_RDONLY;
1073         }
1074         if (read_only)
1075                 return res;
1076         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1077                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1078                         "running e2fsck is recommended\n");
1079         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1080                 printk (KERN_WARNING
1081                         "EXT3-fs warning: mounting fs with errors, "
1082                         "running e2fsck is recommended\n");
1083         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1084                  le16_to_cpu(es->s_mnt_count) >=
1085                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1086                 printk (KERN_WARNING
1087                         "EXT3-fs warning: maximal mount count reached, "
1088                         "running e2fsck is recommended\n");
1089         else if (le32_to_cpu(es->s_checkinterval) &&
1090                 (le32_to_cpu(es->s_lastcheck) +
1091                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1092                 printk (KERN_WARNING
1093                         "EXT3-fs warning: checktime reached, "
1094                         "running e2fsck is recommended\n");
1095 #if 0
1096                 /* @@@ We _will_ want to clear the valid bit if we find
1097                    inconsistencies, to force a fsck at reboot.  But for
1098                    a plain journaled filesystem we can keep it set as
1099                    valid forever! :) */
1100         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1101 #endif
1102         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1103                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1104         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1105         es->s_mtime = cpu_to_le32(get_seconds());
1106         ext3_update_dynamic_rev(sb);
1107         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1108
1109         ext3_commit_super(sb, es, 1);
1110         if (test_opt(sb, DEBUG))
1111                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1112                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1113                         sb->s_blocksize,
1114                         sbi->s_groups_count,
1115                         EXT3_BLOCKS_PER_GROUP(sb),
1116                         EXT3_INODES_PER_GROUP(sb),
1117                         sbi->s_mount_opt);
1118
1119         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1120         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1121                 char b[BDEVNAME_SIZE];
1122
1123                 printk("external journal on %s\n",
1124                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1125         } else {
1126                 printk("internal journal\n");
1127         }
1128         return res;
1129 }
1130
1131 /* Called at mount-time, super-block is locked */
1132 static int ext3_check_descriptors (struct super_block * sb)
1133 {
1134         struct ext3_sb_info *sbi = EXT3_SB(sb);
1135         ext3_fsblk_t block = le32_to_cpu(sbi->s_es->s_first_data_block);
1136         struct ext3_group_desc * gdp = NULL;
1137         int desc_block = 0;
1138         int i;
1139
1140         ext3_debug ("Checking group descriptors");
1141
1142         for (i = 0; i < sbi->s_groups_count; i++)
1143         {
1144                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1145                         gdp = (struct ext3_group_desc *)
1146                                         sbi->s_group_desc[desc_block++]->b_data;
1147                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1148                     le32_to_cpu(gdp->bg_block_bitmap) >=
1149                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1150                 {
1151                         ext3_error (sb, "ext3_check_descriptors",
1152                                     "Block bitmap for group %d"
1153                                     " not in group (block %lu)!",
1154                                     i, (unsigned long)
1155                                         le32_to_cpu(gdp->bg_block_bitmap));
1156                         return 0;
1157                 }
1158                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1159                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1160                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1161                 {
1162                         ext3_error (sb, "ext3_check_descriptors",
1163                                     "Inode bitmap for group %d"
1164                                     " not in group (block %lu)!",
1165                                     i, (unsigned long)
1166                                         le32_to_cpu(gdp->bg_inode_bitmap));
1167                         return 0;
1168                 }
1169                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1170                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1171                     block + EXT3_BLOCKS_PER_GROUP(sb))
1172                 {
1173                         ext3_error (sb, "ext3_check_descriptors",
1174                                     "Inode table for group %d"
1175                                     " not in group (block %lu)!",
1176                                     i, (unsigned long)
1177                                         le32_to_cpu(gdp->bg_inode_table));
1178                         return 0;
1179                 }
1180                 block += EXT3_BLOCKS_PER_GROUP(sb);
1181                 gdp++;
1182         }
1183
1184         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1185         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1186         return 1;
1187 }
1188
1189
1190 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1191  * the superblock) which were deleted from all directories, but held open by
1192  * a process at the time of a crash.  We walk the list and try to delete these
1193  * inodes at recovery time (only with a read-write filesystem).
1194  *
1195  * In order to keep the orphan inode chain consistent during traversal (in
1196  * case of crash during recovery), we link each inode into the superblock
1197  * orphan list_head and handle it the same way as an inode deletion during
1198  * normal operation (which journals the operations for us).
1199  *
1200  * We only do an iget() and an iput() on each inode, which is very safe if we
1201  * accidentally point at an in-use or already deleted inode.  The worst that
1202  * can happen in this case is that we get a "bit already cleared" message from
1203  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1204  * e2fsck was run on this filesystem, and it must have already done the orphan
1205  * inode cleanup for us, so we can safely abort without any further action.
1206  */
1207 static void ext3_orphan_cleanup (struct super_block * sb,
1208                                  struct ext3_super_block * es)
1209 {
1210         unsigned int s_flags = sb->s_flags;
1211         int nr_orphans = 0, nr_truncates = 0;
1212 #ifdef CONFIG_QUOTA
1213         int i;
1214 #endif
1215         if (!es->s_last_orphan) {
1216                 jbd_debug(4, "no orphan inodes to clean up\n");
1217                 return;
1218         }
1219
1220         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1221                 if (es->s_last_orphan)
1222                         jbd_debug(1, "Errors on filesystem, "
1223                                   "clearing orphan list.\n");
1224                 es->s_last_orphan = 0;
1225                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1226                 return;
1227         }
1228
1229         if (s_flags & MS_RDONLY) {
1230                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1231                        sb->s_id);
1232                 sb->s_flags &= ~MS_RDONLY;
1233         }
1234 #ifdef CONFIG_QUOTA
1235         /* Needed for iput() to work correctly and not trash data */
1236         sb->s_flags |= MS_ACTIVE;
1237         /* Turn on quotas so that they are updated correctly */
1238         for (i = 0; i < MAXQUOTAS; i++) {
1239                 if (EXT3_SB(sb)->s_qf_names[i]) {
1240                         int ret = ext3_quota_on_mount(sb, i);
1241                         if (ret < 0)
1242                                 printk(KERN_ERR
1243                                         "EXT3-fs: Cannot turn on journalled "
1244                                         "quota: error %d\n", ret);
1245                 }
1246         }
1247 #endif
1248
1249         while (es->s_last_orphan) {
1250                 struct inode *inode;
1251
1252                 if (!(inode =
1253                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1254                         es->s_last_orphan = 0;
1255                         break;
1256                 }
1257
1258                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1259                 DQUOT_INIT(inode);
1260                 if (inode->i_nlink) {
1261                         printk(KERN_DEBUG
1262                                 "%s: truncating inode %ld to %Ld bytes\n",
1263                                 __FUNCTION__, inode->i_ino, inode->i_size);
1264                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1265                                   inode->i_ino, inode->i_size);
1266                         ext3_truncate(inode);
1267                         nr_truncates++;
1268                 } else {
1269                         printk(KERN_DEBUG
1270                                 "%s: deleting unreferenced inode %ld\n",
1271                                 __FUNCTION__, inode->i_ino);
1272                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1273                                   inode->i_ino);
1274                         nr_orphans++;
1275                 }
1276                 iput(inode);  /* The delete magic happens here! */
1277         }
1278
1279 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1280
1281         if (nr_orphans)
1282                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1283                        sb->s_id, PLURAL(nr_orphans));
1284         if (nr_truncates)
1285                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1286                        sb->s_id, PLURAL(nr_truncates));
1287 #ifdef CONFIG_QUOTA
1288         /* Turn quotas off */
1289         for (i = 0; i < MAXQUOTAS; i++) {
1290                 if (sb_dqopt(sb)->files[i])
1291                         vfs_quota_off(sb, i);
1292         }
1293 #endif
1294         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1295 }
1296
1297 #define log2(n) ffz(~(n))
1298
1299 /*
1300  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1301  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1302  * We need to be 1 filesystem block less than the 2^32 sector limit.
1303  */
1304 static loff_t ext3_max_size(int bits)
1305 {
1306         loff_t res = EXT3_NDIR_BLOCKS;
1307         /* This constant is calculated to be the largest file size for a
1308          * dense, 4k-blocksize file such that the total number of
1309          * sectors in the file, including data and all indirect blocks,
1310          * does not exceed 2^32. */
1311         const loff_t upper_limit = 0x1ff7fffd000LL;
1312
1313         res += 1LL << (bits-2);
1314         res += 1LL << (2*(bits-2));
1315         res += 1LL << (3*(bits-2));
1316         res <<= bits;
1317         if (res > upper_limit)
1318                 res = upper_limit;
1319         return res;
1320 }
1321
1322 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1323                                     ext3_fsblk_t logic_sb_block,
1324                                     int nr)
1325 {
1326         struct ext3_sb_info *sbi = EXT3_SB(sb);
1327         unsigned long bg, first_meta_bg;
1328         int has_super = 0;
1329
1330         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1331
1332         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1333             nr < first_meta_bg)
1334                 return (logic_sb_block + nr + 1);
1335         bg = sbi->s_desc_per_block * nr;
1336         if (ext3_bg_has_super(sb, bg))
1337                 has_super = 1;
1338         return (has_super + ext3_group_first_block_no(sb, bg));
1339 }
1340
1341
1342 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1343 {
1344         struct buffer_head * bh;
1345         struct ext3_super_block *es = NULL;
1346         struct ext3_sb_info *sbi;
1347         ext3_fsblk_t block;
1348         ext3_fsblk_t sb_block = get_sb_block(&data);
1349         ext3_fsblk_t logic_sb_block;
1350         unsigned long offset = 0;
1351         unsigned long journal_inum = 0;
1352         unsigned long journal_devnum = 0;
1353         unsigned long def_mount_opts;
1354         struct inode *root;
1355         int blocksize;
1356         int hblock;
1357         int db_count;
1358         int i;
1359         int needs_recovery;
1360         __le32 features;
1361
1362         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1363         if (!sbi)
1364                 return -ENOMEM;
1365         sb->s_fs_info = sbi;
1366         memset(sbi, 0, sizeof(*sbi));
1367         sbi->s_mount_opt = 0;
1368         sbi->s_resuid = EXT3_DEF_RESUID;
1369         sbi->s_resgid = EXT3_DEF_RESGID;
1370
1371         unlock_kernel();
1372
1373         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1374         if (!blocksize) {
1375                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1376                 goto out_fail;
1377         }
1378
1379         /*
1380          * The ext3 superblock will not be buffer aligned for other than 1kB
1381          * block sizes.  We need to calculate the offset from buffer start.
1382          */
1383         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1384                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1385                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1386         } else {
1387                 logic_sb_block = sb_block;
1388         }
1389
1390         if (!(bh = sb_bread(sb, logic_sb_block))) {
1391                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1392                 goto out_fail;
1393         }
1394         /*
1395          * Note: s_es must be initialized as soon as possible because
1396          *       some ext3 macro-instructions depend on its value
1397          */
1398         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1399         sbi->s_es = es;
1400         sb->s_magic = le16_to_cpu(es->s_magic);
1401         if (sb->s_magic != EXT3_SUPER_MAGIC)
1402                 goto cantfind_ext3;
1403
1404         /* Set defaults before we parse the mount options */
1405         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1406         if (def_mount_opts & EXT3_DEFM_DEBUG)
1407                 set_opt(sbi->s_mount_opt, DEBUG);
1408         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1409                 set_opt(sbi->s_mount_opt, GRPID);
1410         if (def_mount_opts & EXT3_DEFM_UID16)
1411                 set_opt(sbi->s_mount_opt, NO_UID32);
1412         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1413                 set_opt(sbi->s_mount_opt, XATTR_USER);
1414         if (def_mount_opts & EXT3_DEFM_ACL)
1415                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1416         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1417                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1418         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1419                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1420         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1421                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1422
1423         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1424                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1425         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1426                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1427
1428         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1429         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1430
1431         set_opt(sbi->s_mount_opt, RESERVATION);
1432
1433         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1434                             NULL, 0))
1435                 goto failed_mount;
1436
1437         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1438                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1439
1440         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1441             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1442              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1443              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1444                 printk(KERN_WARNING 
1445                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1446                        "running e2fsck is recommended\n");
1447         /*
1448          * Check feature flags regardless of the revision level, since we
1449          * previously didn't change the revision level when setting the flags,
1450          * so there is a chance incompat flags are set on a rev 0 filesystem.
1451          */
1452         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1453         if (features) {
1454                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1455                        "unsupported optional features (%x).\n",
1456                        sb->s_id, le32_to_cpu(features));
1457                 goto failed_mount;
1458         }
1459         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1460         if (!(sb->s_flags & MS_RDONLY) && features) {
1461                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1462                        "unsupported optional features (%x).\n",
1463                        sb->s_id, le32_to_cpu(features));
1464                 goto failed_mount;
1465         }
1466         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1467
1468         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1469             blocksize > EXT3_MAX_BLOCK_SIZE) {
1470                 printk(KERN_ERR 
1471                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1472                        blocksize, sb->s_id);
1473                 goto failed_mount;
1474         }
1475
1476         hblock = bdev_hardsect_size(sb->s_bdev);
1477         if (sb->s_blocksize != blocksize) {
1478                 /*
1479                  * Make sure the blocksize for the filesystem is larger
1480                  * than the hardware sectorsize for the machine.
1481                  */
1482                 if (blocksize < hblock) {
1483                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1484                                "device blocksize %d.\n", blocksize, hblock);
1485                         goto failed_mount;
1486                 }
1487
1488                 brelse (bh);
1489                 sb_set_blocksize(sb, blocksize);
1490                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1491                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1492                 bh = sb_bread(sb, logic_sb_block);
1493                 if (!bh) {
1494                         printk(KERN_ERR 
1495                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1496                         goto failed_mount;
1497                 }
1498                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1499                 sbi->s_es = es;
1500                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1501                         printk (KERN_ERR 
1502                                 "EXT3-fs: Magic mismatch, very weird !\n");
1503                         goto failed_mount;
1504                 }
1505         }
1506
1507         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1508
1509         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1510                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1511                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1512         } else {
1513                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1514                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1515                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1516                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1517                     (sbi->s_inode_size > blocksize)) {
1518                         printk (KERN_ERR
1519                                 "EXT3-fs: unsupported inode size: %d\n",
1520                                 sbi->s_inode_size);
1521                         goto failed_mount;
1522                 }
1523         }
1524         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1525                                    le32_to_cpu(es->s_log_frag_size);
1526         if (blocksize != sbi->s_frag_size) {
1527                 printk(KERN_ERR
1528                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1529                        sbi->s_frag_size, blocksize);
1530                 goto failed_mount;
1531         }
1532         sbi->s_frags_per_block = 1;
1533         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1534         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1535         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1536         if (EXT3_INODE_SIZE(sb) == 0)
1537                 goto cantfind_ext3;
1538         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1539         if (sbi->s_inodes_per_block == 0)
1540                 goto cantfind_ext3;
1541         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1542                                         sbi->s_inodes_per_block;
1543         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1544         sbi->s_sbh = bh;
1545         sbi->s_mount_state = le16_to_cpu(es->s_state);
1546         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1547         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1548         for (i=0; i < 4; i++)
1549                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1550         sbi->s_def_hash_version = es->s_def_hash_version;
1551
1552         if (sbi->s_blocks_per_group > blocksize * 8) {
1553                 printk (KERN_ERR
1554                         "EXT3-fs: #blocks per group too big: %lu\n",
1555                         sbi->s_blocks_per_group);
1556                 goto failed_mount;
1557         }
1558         if (sbi->s_frags_per_group > blocksize * 8) {
1559                 printk (KERN_ERR
1560                         "EXT3-fs: #fragments per group too big: %lu\n",
1561                         sbi->s_frags_per_group);
1562                 goto failed_mount;
1563         }
1564         if (sbi->s_inodes_per_group > blocksize * 8) {
1565                 printk (KERN_ERR
1566                         "EXT3-fs: #inodes per group too big: %lu\n",
1567                         sbi->s_inodes_per_group);
1568                 goto failed_mount;
1569         }
1570
1571         if (le32_to_cpu(es->s_blocks_count) >
1572                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1573                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1574                         " too large to mount safely\n", sb->s_id);
1575                 if (sizeof(sector_t) < 8)
1576                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1577                                         "enabled\n");
1578                 goto failed_mount;
1579         }
1580
1581         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1582                 goto cantfind_ext3;
1583         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1584                                le32_to_cpu(es->s_first_data_block) +
1585                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1586                               EXT3_BLOCKS_PER_GROUP(sb);
1587         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1588                    EXT3_DESC_PER_BLOCK(sb);
1589         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1590                                     GFP_KERNEL);
1591         if (sbi->s_group_desc == NULL) {
1592                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1593                 goto failed_mount;
1594         }
1595
1596         bgl_lock_init(&sbi->s_blockgroup_lock);
1597
1598         for (i = 0; i < db_count; i++) {
1599                 block = descriptor_loc(sb, logic_sb_block, i);
1600                 sbi->s_group_desc[i] = sb_bread(sb, block);
1601                 if (!sbi->s_group_desc[i]) {
1602                         printk (KERN_ERR "EXT3-fs: "
1603                                 "can't read group descriptor %d\n", i);
1604                         db_count = i;
1605                         goto failed_mount2;
1606                 }
1607         }
1608         if (!ext3_check_descriptors (sb)) {
1609                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1610                 goto failed_mount2;
1611         }
1612         sbi->s_gdb_count = db_count;
1613         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1614         spin_lock_init(&sbi->s_next_gen_lock);
1615
1616         percpu_counter_init(&sbi->s_freeblocks_counter,
1617                 ext3_count_free_blocks(sb));
1618         percpu_counter_init(&sbi->s_freeinodes_counter,
1619                 ext3_count_free_inodes(sb));
1620         percpu_counter_init(&sbi->s_dirs_counter,
1621                 ext3_count_dirs(sb));
1622
1623         /* per fileystem reservation list head & lock */
1624         spin_lock_init(&sbi->s_rsv_window_lock);
1625         sbi->s_rsv_window_root = RB_ROOT;
1626         /* Add a single, static dummy reservation to the start of the
1627          * reservation window list --- it gives us a placeholder for
1628          * append-at-start-of-list which makes the allocation logic
1629          * _much_ simpler. */
1630         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1631         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1632         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1633         sbi->s_rsv_window_head.rsv_goal_size = 0;
1634         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1635
1636         /*
1637          * set up enough so that it can read an inode
1638          */
1639         sb->s_op = &ext3_sops;
1640         sb->s_export_op = &ext3_export_ops;
1641         sb->s_xattr = ext3_xattr_handlers;
1642 #ifdef CONFIG_QUOTA
1643         sb->s_qcop = &ext3_qctl_operations;
1644         sb->dq_op = &ext3_quota_operations;
1645 #endif
1646         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1647
1648         sb->s_root = NULL;
1649
1650         needs_recovery = (es->s_last_orphan != 0 ||
1651                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1652                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1653
1654         /*
1655          * The first inode we look at is the journal inode.  Don't try
1656          * root first: it may be modified in the journal!
1657          */
1658         if (!test_opt(sb, NOLOAD) &&
1659             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1660                 if (ext3_load_journal(sb, es, journal_devnum))
1661                         goto failed_mount3;
1662         } else if (journal_inum) {
1663                 if (ext3_create_journal(sb, es, journal_inum))
1664                         goto failed_mount3;
1665         } else {
1666                 if (!silent)
1667                         printk (KERN_ERR
1668                                 "ext3: No journal on filesystem on %s\n",
1669                                 sb->s_id);
1670                 goto failed_mount3;
1671         }
1672
1673         /* We have now updated the journal if required, so we can
1674          * validate the data journaling mode. */
1675         switch (test_opt(sb, DATA_FLAGS)) {
1676         case 0:
1677                 /* No mode set, assume a default based on the journal
1678                    capabilities: ORDERED_DATA if the journal can
1679                    cope, else JOURNAL_DATA */
1680                 if (journal_check_available_features
1681                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1682                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1683                 else
1684                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1685                 break;
1686
1687         case EXT3_MOUNT_ORDERED_DATA:
1688         case EXT3_MOUNT_WRITEBACK_DATA:
1689                 if (!journal_check_available_features
1690                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1691                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1692                                "requested data journaling mode\n");
1693                         goto failed_mount4;
1694                 }
1695         default:
1696                 break;
1697         }
1698
1699         if (test_opt(sb, NOBH)) {
1700                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1701                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1702                                 "its supported only with writeback mode\n");
1703                         clear_opt(sbi->s_mount_opt, NOBH);
1704                 }
1705         }
1706         /*
1707          * The journal_load will have done any necessary log recovery,
1708          * so we can safely mount the rest of the filesystem now.
1709          */
1710
1711         root = iget(sb, EXT3_ROOT_INO);
1712         sb->s_root = d_alloc_root(root);
1713         if (!sb->s_root) {
1714                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1715                 iput(root);
1716                 goto failed_mount4;
1717         }
1718         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1719                 dput(sb->s_root);
1720                 sb->s_root = NULL;
1721                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1722                 goto failed_mount4;
1723         }
1724
1725         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1726         /*
1727          * akpm: core read_super() calls in here with the superblock locked.
1728          * That deadlocks, because orphan cleanup needs to lock the superblock
1729          * in numerous places.  Here we just pop the lock - it's relatively
1730          * harmless, because we are now ready to accept write_super() requests,
1731          * and aviro says that's the only reason for hanging onto the
1732          * superblock lock.
1733          */
1734         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1735         ext3_orphan_cleanup(sb, es);
1736         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1737         if (needs_recovery)
1738                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1739         ext3_mark_recovery_complete(sb, es);
1740         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1741                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1742                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1743                 "writeback");
1744
1745         lock_kernel();
1746         return 0;
1747
1748 cantfind_ext3:
1749         if (!silent)
1750                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1751                        sb->s_id);
1752         goto failed_mount;
1753
1754 failed_mount4:
1755         journal_destroy(sbi->s_journal);
1756 failed_mount3:
1757         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1758         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1759         percpu_counter_destroy(&sbi->s_dirs_counter);
1760 failed_mount2:
1761         for (i = 0; i < db_count; i++)
1762                 brelse(sbi->s_group_desc[i]);
1763         kfree(sbi->s_group_desc);
1764 failed_mount:
1765 #ifdef CONFIG_QUOTA
1766         for (i = 0; i < MAXQUOTAS; i++)
1767                 kfree(sbi->s_qf_names[i]);
1768 #endif
1769         ext3_blkdev_remove(sbi);
1770         brelse(bh);
1771 out_fail:
1772         sb->s_fs_info = NULL;
1773         kfree(sbi);
1774         lock_kernel();
1775         return -EINVAL;
1776 }
1777
1778 /*
1779  * Setup any per-fs journal parameters now.  We'll do this both on
1780  * initial mount, once the journal has been initialised but before we've
1781  * done any recovery; and again on any subsequent remount. 
1782  */
1783 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1784 {
1785         struct ext3_sb_info *sbi = EXT3_SB(sb);
1786
1787         if (sbi->s_commit_interval)
1788                 journal->j_commit_interval = sbi->s_commit_interval;
1789         /* We could also set up an ext3-specific default for the commit
1790          * interval here, but for now we'll just fall back to the jbd
1791          * default. */
1792
1793         spin_lock(&journal->j_state_lock);
1794         if (test_opt(sb, BARRIER))
1795                 journal->j_flags |= JFS_BARRIER;
1796         else
1797                 journal->j_flags &= ~JFS_BARRIER;
1798         spin_unlock(&journal->j_state_lock);
1799 }
1800
1801 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1802 {
1803         struct inode *journal_inode;
1804         journal_t *journal;
1805
1806         /* First, test for the existence of a valid inode on disk.  Bad
1807          * things happen if we iget() an unused inode, as the subsequent
1808          * iput() will try to delete it. */
1809
1810         journal_inode = iget(sb, journal_inum);
1811         if (!journal_inode) {
1812                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1813                 return NULL;
1814         }
1815         if (!journal_inode->i_nlink) {
1816                 make_bad_inode(journal_inode);
1817                 iput(journal_inode);
1818                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1819                 return NULL;
1820         }
1821
1822         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1823                   journal_inode, journal_inode->i_size);
1824         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1825                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1826                 iput(journal_inode);
1827                 return NULL;
1828         }
1829
1830         journal = journal_init_inode(journal_inode);
1831         if (!journal) {
1832                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1833                 iput(journal_inode);
1834                 return NULL;
1835         }
1836         journal->j_private = sb;
1837         ext3_init_journal_params(sb, journal);
1838         return journal;
1839 }
1840
1841 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1842                                        dev_t j_dev)
1843 {
1844         struct buffer_head * bh;
1845         journal_t *journal;
1846         ext3_fsblk_t start;
1847         ext3_fsblk_t len;
1848         int hblock, blocksize;
1849         ext3_fsblk_t sb_block;
1850         unsigned long offset;
1851         struct ext3_super_block * es;
1852         struct block_device *bdev;
1853
1854         bdev = ext3_blkdev_get(j_dev);
1855         if (bdev == NULL)
1856                 return NULL;
1857
1858         if (bd_claim(bdev, sb)) {
1859                 printk(KERN_ERR
1860                         "EXT3: failed to claim external journal device.\n");
1861                 blkdev_put(bdev);
1862                 return NULL;
1863         }
1864
1865         blocksize = sb->s_blocksize;
1866         hblock = bdev_hardsect_size(bdev);
1867         if (blocksize < hblock) {
1868                 printk(KERN_ERR
1869                         "EXT3-fs: blocksize too small for journal device.\n");
1870                 goto out_bdev;
1871         }
1872
1873         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1874         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1875         set_blocksize(bdev, blocksize);
1876         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1877                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1878                        "external journal\n");
1879                 goto out_bdev;
1880         }
1881
1882         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1883         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1884             !(le32_to_cpu(es->s_feature_incompat) &
1885               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1886                 printk(KERN_ERR "EXT3-fs: external journal has "
1887                                         "bad superblock\n");
1888                 brelse(bh);
1889                 goto out_bdev;
1890         }
1891
1892         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1893                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1894                 brelse(bh);
1895                 goto out_bdev;
1896         }
1897
1898         len = le32_to_cpu(es->s_blocks_count);
1899         start = sb_block + 1;
1900         brelse(bh);     /* we're done with the superblock */
1901
1902         journal = journal_init_dev(bdev, sb->s_bdev,
1903                                         start, len, blocksize);
1904         if (!journal) {
1905                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1906                 goto out_bdev;
1907         }
1908         journal->j_private = sb;
1909         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1910         wait_on_buffer(journal->j_sb_buffer);
1911         if (!buffer_uptodate(journal->j_sb_buffer)) {
1912                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1913                 goto out_journal;
1914         }
1915         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1916                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1917                                         "user (unsupported) - %d\n",
1918                         be32_to_cpu(journal->j_superblock->s_nr_users));
1919                 goto out_journal;
1920         }
1921         EXT3_SB(sb)->journal_bdev = bdev;
1922         ext3_init_journal_params(sb, journal);
1923         return journal;
1924 out_journal:
1925         journal_destroy(journal);
1926 out_bdev:
1927         ext3_blkdev_put(bdev);
1928         return NULL;
1929 }
1930
1931 static int ext3_load_journal(struct super_block *sb,
1932                              struct ext3_super_block *es,
1933                              unsigned long journal_devnum)
1934 {
1935         journal_t *journal;
1936         int journal_inum = le32_to_cpu(es->s_journal_inum);
1937         dev_t journal_dev;
1938         int err = 0;
1939         int really_read_only;
1940
1941         if (journal_devnum &&
1942             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1943                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1944                         "numbers have changed\n");
1945                 journal_dev = new_decode_dev(journal_devnum);
1946         } else
1947                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1948
1949         really_read_only = bdev_read_only(sb->s_bdev);
1950
1951         /*
1952          * Are we loading a blank journal or performing recovery after a
1953          * crash?  For recovery, we need to check in advance whether we
1954          * can get read-write access to the device.
1955          */
1956
1957         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1958                 if (sb->s_flags & MS_RDONLY) {
1959                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1960                                         "required on readonly filesystem.\n");
1961                         if (really_read_only) {
1962                                 printk(KERN_ERR "EXT3-fs: write access "
1963                                         "unavailable, cannot proceed.\n");
1964                                 return -EROFS;
1965                         }
1966                         printk (KERN_INFO "EXT3-fs: write access will "
1967                                         "be enabled during recovery.\n");
1968                 }
1969         }
1970
1971         if (journal_inum && journal_dev) {
1972                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1973                        "and inode journals!\n");
1974                 return -EINVAL;
1975         }
1976
1977         if (journal_inum) {
1978                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1979                         return -EINVAL;
1980         } else {
1981                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1982                         return -EINVAL;
1983         }
1984
1985         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1986                 err = journal_update_format(journal);
1987                 if (err)  {
1988                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1989                         journal_destroy(journal);
1990                         return err;
1991                 }
1992         }
1993
1994         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1995                 err = journal_wipe(journal, !really_read_only);
1996         if (!err)
1997                 err = journal_load(journal);
1998
1999         if (err) {
2000                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2001                 journal_destroy(journal);
2002                 return err;
2003         }
2004
2005         EXT3_SB(sb)->s_journal = journal;
2006         ext3_clear_journal_err(sb, es);
2007
2008         if (journal_devnum &&
2009             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2010                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2011                 sb->s_dirt = 1;
2012
2013                 /* Make sure we flush the recovery flag to disk. */
2014                 ext3_commit_super(sb, es, 1);
2015         }
2016
2017         return 0;
2018 }
2019
2020 static int ext3_create_journal(struct super_block * sb,
2021                                struct ext3_super_block * es,
2022                                int journal_inum)
2023 {
2024         journal_t *journal;
2025
2026         if (sb->s_flags & MS_RDONLY) {
2027                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2028                                 "create journal.\n");
2029                 return -EROFS;
2030         }
2031
2032         if (!(journal = ext3_get_journal(sb, journal_inum)))
2033                 return -EINVAL;
2034
2035         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2036                journal_inum);
2037
2038         if (journal_create(journal)) {
2039                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2040                 journal_destroy(journal);
2041                 return -EIO;
2042         }
2043
2044         EXT3_SB(sb)->s_journal = journal;
2045
2046         ext3_update_dynamic_rev(sb);
2047         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2048         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2049
2050         es->s_journal_inum = cpu_to_le32(journal_inum);
2051         sb->s_dirt = 1;
2052
2053         /* Make sure we flush the recovery flag to disk. */
2054         ext3_commit_super(sb, es, 1);
2055
2056         return 0;
2057 }
2058
2059 static void ext3_commit_super (struct super_block * sb,
2060                                struct ext3_super_block * es,
2061                                int sync)
2062 {
2063         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2064
2065         if (!sbh)
2066                 return;
2067         es->s_wtime = cpu_to_le32(get_seconds());
2068         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2069         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2070         BUFFER_TRACE(sbh, "marking dirty");
2071         mark_buffer_dirty(sbh);
2072         if (sync)
2073                 sync_dirty_buffer(sbh);
2074 }
2075
2076
2077 /*
2078  * Have we just finished recovery?  If so, and if we are mounting (or
2079  * remounting) the filesystem readonly, then we will end up with a
2080  * consistent fs on disk.  Record that fact.
2081  */
2082 static void ext3_mark_recovery_complete(struct super_block * sb,
2083                                         struct ext3_super_block * es)
2084 {
2085         journal_t *journal = EXT3_SB(sb)->s_journal;
2086
2087         journal_lock_updates(journal);
2088         journal_flush(journal);
2089         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2090             sb->s_flags & MS_RDONLY) {
2091                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2092                 sb->s_dirt = 0;
2093                 ext3_commit_super(sb, es, 1);
2094         }
2095         journal_unlock_updates(journal);
2096 }
2097
2098 /*
2099  * If we are mounting (or read-write remounting) a filesystem whose journal
2100  * has recorded an error from a previous lifetime, move that error to the
2101  * main filesystem now.
2102  */
2103 static void ext3_clear_journal_err(struct super_block * sb,
2104                                    struct ext3_super_block * es)
2105 {
2106         journal_t *journal;
2107         int j_errno;
2108         const char *errstr;
2109
2110         journal = EXT3_SB(sb)->s_journal;
2111
2112         /*
2113          * Now check for any error status which may have been recorded in the
2114          * journal by a prior ext3_error() or ext3_abort()
2115          */
2116
2117         j_errno = journal_errno(journal);
2118         if (j_errno) {
2119                 char nbuf[16];
2120
2121                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2122                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2123                              "from previous mount: %s", errstr);
2124                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2125                              "filesystem check.");
2126
2127                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2128                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2129                 ext3_commit_super (sb, es, 1);
2130
2131                 journal_clear_err(journal);
2132         }
2133 }
2134
2135 /*
2136  * Force the running and committing transactions to commit,
2137  * and wait on the commit.
2138  */
2139 int ext3_force_commit(struct super_block *sb)
2140 {
2141         journal_t *journal;
2142         int ret;
2143
2144         if (sb->s_flags & MS_RDONLY)
2145                 return 0;
2146
2147         journal = EXT3_SB(sb)->s_journal;
2148         sb->s_dirt = 0;
2149         ret = ext3_journal_force_commit(journal);
2150         return ret;
2151 }
2152
2153 /*
2154  * Ext3 always journals updates to the superblock itself, so we don't
2155  * have to propagate any other updates to the superblock on disk at this
2156  * point.  Just start an async writeback to get the buffers on their way
2157  * to the disk.
2158  *
2159  * This implicitly triggers the writebehind on sync().
2160  */
2161
2162 static void ext3_write_super (struct super_block * sb)
2163 {
2164         if (mutex_trylock(&sb->s_lock) != 0)
2165                 BUG();
2166         sb->s_dirt = 0;
2167 }
2168
2169 static int ext3_sync_fs(struct super_block *sb, int wait)
2170 {
2171         tid_t target;
2172
2173         sb->s_dirt = 0;
2174         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2175                 if (wait)
2176                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2177         }
2178         return 0;
2179 }
2180
2181 /*
2182  * LVM calls this function before a (read-only) snapshot is created.  This
2183  * gives us a chance to flush the journal completely and mark the fs clean.
2184  */
2185 static void ext3_write_super_lockfs(struct super_block *sb)
2186 {
2187         sb->s_dirt = 0;
2188
2189         if (!(sb->s_flags & MS_RDONLY)) {
2190                 journal_t *journal = EXT3_SB(sb)->s_journal;
2191
2192                 /* Now we set up the journal barrier. */
2193                 journal_lock_updates(journal);
2194                 journal_flush(journal);
2195
2196                 /* Journal blocked and flushed, clear needs_recovery flag. */
2197                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2198                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2199         }
2200 }
2201
2202 /*
2203  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2204  * flag here, even though the filesystem is not technically dirty yet.
2205  */
2206 static void ext3_unlockfs(struct super_block *sb)
2207 {
2208         if (!(sb->s_flags & MS_RDONLY)) {
2209                 lock_super(sb);
2210                 /* Reser the needs_recovery flag before the fs is unlocked. */
2211                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2212                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2213                 unlock_super(sb);
2214                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2215         }
2216 }
2217
2218 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2219 {
2220         struct ext3_super_block * es;
2221         struct ext3_sb_info *sbi = EXT3_SB(sb);
2222         ext3_fsblk_t n_blocks_count = 0;
2223         unsigned long old_sb_flags;
2224         struct ext3_mount_options old_opts;
2225         int err;
2226 #ifdef CONFIG_QUOTA
2227         int i;
2228 #endif
2229
2230         /* Store the original options */
2231         old_sb_flags = sb->s_flags;
2232         old_opts.s_mount_opt = sbi->s_mount_opt;
2233         old_opts.s_resuid = sbi->s_resuid;
2234         old_opts.s_resgid = sbi->s_resgid;
2235         old_opts.s_commit_interval = sbi->s_commit_interval;
2236 #ifdef CONFIG_QUOTA
2237         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2238         for (i = 0; i < MAXQUOTAS; i++)
2239                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2240 #endif
2241
2242         /*
2243          * Allow the "check" option to be passed as a remount option.
2244          */
2245         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2246                 err = -EINVAL;
2247                 goto restore_opts;
2248         }
2249
2250         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2251                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2252
2253         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2254                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2255
2256         es = sbi->s_es;
2257
2258         ext3_init_journal_params(sb, sbi->s_journal);
2259
2260         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2261                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2262                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2263                         err = -EROFS;
2264                         goto restore_opts;
2265                 }
2266
2267                 if (*flags & MS_RDONLY) {
2268                         /*
2269                          * First of all, the unconditional stuff we have to do
2270                          * to disable replay of the journal when we next remount
2271                          */
2272                         sb->s_flags |= MS_RDONLY;
2273
2274                         /*
2275                          * OK, test if we are remounting a valid rw partition
2276                          * readonly, and if so set the rdonly flag and then
2277                          * mark the partition as valid again.
2278                          */
2279                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2280                             (sbi->s_mount_state & EXT3_VALID_FS))
2281                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2282
2283                         ext3_mark_recovery_complete(sb, es);
2284                 } else {
2285                         __le32 ret;
2286                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2287                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2288                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2289                                        "remount RDWR because of unsupported "
2290                                        "optional features (%x).\n",
2291                                        sb->s_id, le32_to_cpu(ret));
2292                                 err = -EROFS;
2293                                 goto restore_opts;
2294                         }
2295                         /*
2296                          * Mounting a RDONLY partition read-write, so reread
2297                          * and store the current valid flag.  (It may have
2298                          * been changed by e2fsck since we originally mounted
2299                          * the partition.)
2300                          */
2301                         ext3_clear_journal_err(sb, es);
2302                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2303                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2304                                 err = ret;
2305                                 goto restore_opts;
2306                         }
2307                         if (!ext3_setup_super (sb, es, 0))
2308                                 sb->s_flags &= ~MS_RDONLY;
2309                 }
2310         }
2311 #ifdef CONFIG_QUOTA
2312         /* Release old quota file names */
2313         for (i = 0; i < MAXQUOTAS; i++)
2314                 if (old_opts.s_qf_names[i] &&
2315                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2316                         kfree(old_opts.s_qf_names[i]);
2317 #endif
2318         return 0;
2319 restore_opts:
2320         sb->s_flags = old_sb_flags;
2321         sbi->s_mount_opt = old_opts.s_mount_opt;
2322         sbi->s_resuid = old_opts.s_resuid;
2323         sbi->s_resgid = old_opts.s_resgid;
2324         sbi->s_commit_interval = old_opts.s_commit_interval;
2325 #ifdef CONFIG_QUOTA
2326         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2327         for (i = 0; i < MAXQUOTAS; i++) {
2328                 if (sbi->s_qf_names[i] &&
2329                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2330                         kfree(sbi->s_qf_names[i]);
2331                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2332         }
2333 #endif
2334         return err;
2335 }
2336
2337 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2338 {
2339         struct super_block *sb = dentry->d_sb;
2340         struct ext3_sb_info *sbi = EXT3_SB(sb);
2341         struct ext3_super_block *es = sbi->s_es;
2342         ext3_fsblk_t overhead;
2343         int i;
2344
2345         if (test_opt (sb, MINIX_DF))
2346                 overhead = 0;
2347         else {
2348                 unsigned long ngroups;
2349                 ngroups = EXT3_SB(sb)->s_groups_count;
2350                 smp_rmb();
2351
2352                 /*
2353                  * Compute the overhead (FS structures)
2354                  */
2355
2356                 /*
2357                  * All of the blocks before first_data_block are
2358                  * overhead
2359                  */
2360                 overhead = le32_to_cpu(es->s_first_data_block);
2361
2362                 /*
2363                  * Add the overhead attributed to the superblock and
2364                  * block group descriptors.  If the sparse superblocks
2365                  * feature is turned on, then not all groups have this.
2366                  */
2367                 for (i = 0; i < ngroups; i++) {
2368                         overhead += ext3_bg_has_super(sb, i) +
2369                                 ext3_bg_num_gdb(sb, i);
2370                         cond_resched();
2371                 }
2372
2373                 /*
2374                  * Every block group has an inode bitmap, a block
2375                  * bitmap, and an inode table.
2376                  */
2377                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2378         }
2379
2380         buf->f_type = EXT3_SUPER_MAGIC;
2381         buf->f_bsize = sb->s_blocksize;
2382         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2383         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2384         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2385         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2386                 buf->f_bavail = 0;
2387         buf->f_files = le32_to_cpu(es->s_inodes_count);
2388         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2389         buf->f_namelen = EXT3_NAME_LEN;
2390         return 0;
2391 }
2392
2393 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2394  * is locked for write. Otherwise the are possible deadlocks:
2395  * Process 1                         Process 2
2396  * ext3_create()                     quota_sync()
2397  *   journal_start()                   write_dquot()
2398  *   DQUOT_INIT()                        down(dqio_mutex)
2399  *     down(dqio_mutex)                    journal_start()
2400  *
2401  */
2402
2403 #ifdef CONFIG_QUOTA
2404
2405 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2406 {
2407         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2408 }
2409
2410 static int ext3_dquot_initialize(struct inode *inode, int type)
2411 {
2412         handle_t *handle;
2413         int ret, err;
2414
2415         /* We may create quota structure so we need to reserve enough blocks */
2416         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2417         if (IS_ERR(handle))
2418                 return PTR_ERR(handle);
2419         ret = dquot_initialize(inode, type);
2420         err = ext3_journal_stop(handle);
2421         if (!ret)
2422                 ret = err;
2423         return ret;
2424 }
2425
2426 static int ext3_dquot_drop(struct inode *inode)
2427 {
2428         handle_t *handle;
2429         int ret, err;
2430
2431         /* We may delete quota structure so we need to reserve enough blocks */
2432         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2433         if (IS_ERR(handle))
2434                 return PTR_ERR(handle);
2435         ret = dquot_drop(inode);
2436         err = ext3_journal_stop(handle);
2437         if (!ret)
2438                 ret = err;
2439         return ret;
2440 }
2441
2442 static int ext3_write_dquot(struct dquot *dquot)
2443 {
2444         int ret, err;
2445         handle_t *handle;
2446         struct inode *inode;
2447
2448         inode = dquot_to_inode(dquot);
2449         handle = ext3_journal_start(inode,
2450                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2451         if (IS_ERR(handle))
2452                 return PTR_ERR(handle);
2453         ret = dquot_commit(dquot);
2454         err = ext3_journal_stop(handle);
2455         if (!ret)
2456                 ret = err;
2457         return ret;
2458 }
2459
2460 static int ext3_acquire_dquot(struct dquot *dquot)
2461 {
2462         int ret, err;
2463         handle_t *handle;
2464
2465         handle = ext3_journal_start(dquot_to_inode(dquot),
2466                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2467         if (IS_ERR(handle))
2468                 return PTR_ERR(handle);
2469         ret = dquot_acquire(dquot);
2470         err = ext3_journal_stop(handle);
2471         if (!ret)
2472                 ret = err;
2473         return ret;
2474 }
2475
2476 static int ext3_release_dquot(struct dquot *dquot)
2477 {
2478         int ret, err;
2479         handle_t *handle;
2480
2481         handle = ext3_journal_start(dquot_to_inode(dquot),
2482                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2483         if (IS_ERR(handle))
2484                 return PTR_ERR(handle);
2485         ret = dquot_release(dquot);
2486         err = ext3_journal_stop(handle);
2487         if (!ret)
2488                 ret = err;
2489         return ret;
2490 }
2491
2492 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2493 {
2494         /* Are we journalling quotas? */
2495         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2496             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2497                 dquot_mark_dquot_dirty(dquot);
2498                 return ext3_write_dquot(dquot);
2499         } else {
2500                 return dquot_mark_dquot_dirty(dquot);
2501         }
2502 }
2503
2504 static int ext3_write_info(struct super_block *sb, int type)
2505 {
2506         int ret, err;
2507         handle_t *handle;
2508
2509         /* Data block + inode block */
2510         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2511         if (IS_ERR(handle))
2512                 return PTR_ERR(handle);
2513         ret = dquot_commit_info(sb, type);
2514         err = ext3_journal_stop(handle);
2515         if (!ret)
2516                 ret = err;
2517         return ret;
2518 }
2519
2520 /*
2521  * Turn on quotas during mount time - we need to find
2522  * the quota file and such...
2523  */
2524 static int ext3_quota_on_mount(struct super_block *sb, int type)
2525 {
2526         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2527                         EXT3_SB(sb)->s_jquota_fmt, type);
2528 }
2529
2530 /*
2531  * Standard function to be called on quota_on
2532  */
2533 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2534                          char *path)
2535 {
2536         int err;
2537         struct nameidata nd;
2538
2539         if (!test_opt(sb, QUOTA))
2540                 return -EINVAL;
2541         /* Not journalling quota? */
2542         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2543             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2544                 return vfs_quota_on(sb, type, format_id, path);
2545         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2546         if (err)
2547                 return err;
2548         /* Quotafile not on the same filesystem? */
2549         if (nd.mnt->mnt_sb != sb) {
2550                 path_release(&nd);
2551                 return -EXDEV;
2552         }
2553         /* Quotafile not of fs root? */
2554         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2555                 printk(KERN_WARNING
2556                         "EXT3-fs: Quota file not on filesystem root. "
2557                         "Journalled quota will not work.\n");
2558         path_release(&nd);
2559         return vfs_quota_on(sb, type, format_id, path);
2560 }
2561
2562 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2563  * acquiring the locks... As quota files are never truncated and quota code
2564  * itself serializes the operations (and noone else should touch the files)
2565  * we don't have to be afraid of races */
2566 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2567                                size_t len, loff_t off)
2568 {
2569         struct inode *inode = sb_dqopt(sb)->files[type];
2570         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2571         int err = 0;
2572         int offset = off & (sb->s_blocksize - 1);
2573         int tocopy;
2574         size_t toread;
2575         struct buffer_head *bh;
2576         loff_t i_size = i_size_read(inode);
2577
2578         if (off > i_size)
2579                 return 0;
2580         if (off+len > i_size)
2581                 len = i_size-off;
2582         toread = len;
2583         while (toread > 0) {
2584                 tocopy = sb->s_blocksize - offset < toread ?
2585                                 sb->s_blocksize - offset : toread;
2586                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2587                 if (err)
2588                         return err;
2589                 if (!bh)        /* A hole? */
2590                         memset(data, 0, tocopy);
2591                 else
2592                         memcpy(data, bh->b_data+offset, tocopy);
2593                 brelse(bh);
2594                 offset = 0;
2595                 toread -= tocopy;
2596                 data += tocopy;
2597                 blk++;
2598         }
2599         return len;
2600 }
2601
2602 /* Write to quotafile (we know the transaction is already started and has
2603  * enough credits) */
2604 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2605                                 const char *data, size_t len, loff_t off)
2606 {
2607         struct inode *inode = sb_dqopt(sb)->files[type];
2608         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2609         int err = 0;
2610         int offset = off & (sb->s_blocksize - 1);
2611         int tocopy;
2612         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2613         size_t towrite = len;
2614         struct buffer_head *bh;
2615         handle_t *handle = journal_current_handle();
2616
2617         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2618         while (towrite > 0) {
2619                 tocopy = sb->s_blocksize - offset < towrite ?
2620                                 sb->s_blocksize - offset : towrite;
2621                 bh = ext3_bread(handle, inode, blk, 1, &err);
2622                 if (!bh)
2623                         goto out;
2624                 if (journal_quota) {
2625                         err = ext3_journal_get_write_access(handle, bh);
2626                         if (err) {
2627                                 brelse(bh);
2628                                 goto out;
2629                         }
2630                 }
2631                 lock_buffer(bh);
2632                 memcpy(bh->b_data+offset, data, tocopy);
2633                 flush_dcache_page(bh->b_page);
2634                 unlock_buffer(bh);
2635                 if (journal_quota)
2636                         err = ext3_journal_dirty_metadata(handle, bh);
2637                 else {
2638                         /* Always do at least ordered writes for quotas */
2639                         err = ext3_journal_dirty_data(handle, bh);
2640                         mark_buffer_dirty(bh);
2641                 }
2642                 brelse(bh);
2643                 if (err)
2644                         goto out;
2645                 offset = 0;
2646                 towrite -= tocopy;
2647                 data += tocopy;
2648                 blk++;
2649         }
2650 out:
2651         if (len == towrite)
2652                 return err;
2653         if (inode->i_size < off+len-towrite) {
2654                 i_size_write(inode, off+len-towrite);
2655                 EXT3_I(inode)->i_disksize = inode->i_size;
2656         }
2657         inode->i_version++;
2658         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2659         ext3_mark_inode_dirty(handle, inode);
2660         mutex_unlock(&inode->i_mutex);
2661         return len - towrite;
2662 }
2663
2664 #endif
2665
2666 static int ext3_get_sb(struct file_system_type *fs_type,
2667         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2668 {
2669         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2670 }
2671
2672 static struct file_system_type ext3_fs_type = {
2673         .owner          = THIS_MODULE,
2674         .name           = "ext3",
2675         .get_sb         = ext3_get_sb,
2676         .kill_sb        = kill_block_super,
2677         .fs_flags       = FS_REQUIRES_DEV,
2678 };
2679
2680 static int __init init_ext3_fs(void)
2681 {
2682         int err = init_ext3_xattr();
2683         if (err)
2684                 return err;
2685         err = init_inodecache();
2686         if (err)
2687                 goto out1;
2688         err = register_filesystem(&ext3_fs_type);
2689         if (err)
2690                 goto out;
2691         return 0;
2692 out:
2693         destroy_inodecache();
2694 out1:
2695         exit_ext3_xattr();
2696         return err;
2697 }
2698
2699 static void __exit exit_ext3_fs(void)
2700 {
2701         unregister_filesystem(&ext3_fs_type);
2702         destroy_inodecache();
2703         exit_ext3_xattr();
2704 }
2705
2706 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2707 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2708 MODULE_LICENSE("GPL");
2709 module_init(init_ext3_fs)
2710 module_exit(exit_ext3_fs)