Merge with /pub/scm/linux/kernel/git/torvalds/linux-2.6.git
[linux-2.6] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/config.h>
20 #include <linux/module.h>
21 #include <linux/string.h>
22 #include <linux/fs.h>
23 #include <linux/time.h>
24 #include <linux/jbd.h>
25 #include <linux/ext3_fs.h>
26 #include <linux/ext3_jbd.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
29 #include <linux/blkdev.h>
30 #include <linux/parser.h>
31 #include <linux/smp_lock.h>
32 #include <linux/buffer_head.h>
33 #include <linux/vfs.h>
34 #include <linux/random.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/quotaops.h>
38 #include <linux/seq_file.h>
39
40 #include <asm/uaccess.h>
41
42 #include "xattr.h"
43 #include "acl.h"
44 #include "namei.h"
45
46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
47 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
48                                int);
49 static void ext3_commit_super (struct super_block * sb,
50                                struct ext3_super_block * es,
51                                int sync);
52 static void ext3_mark_recovery_complete(struct super_block * sb,
53                                         struct ext3_super_block * es);
54 static void ext3_clear_journal_err(struct super_block * sb,
55                                    struct ext3_super_block * es);
56 static int ext3_sync_fs(struct super_block *sb, int wait);
57 static const char *ext3_decode_error(struct super_block * sb, int errno,
58                                      char nbuf[16]);
59 static int ext3_remount (struct super_block * sb, int * flags, char * data);
60 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
61 static void ext3_unlockfs(struct super_block *sb);
62 static void ext3_write_super (struct super_block * sb);
63 static void ext3_write_super_lockfs(struct super_block *sb);
64
65 /* 
66  * Wrappers for journal_start/end.
67  *
68  * The only special thing we need to do here is to make sure that all
69  * journal_end calls result in the superblock being marked dirty, so
70  * that sync() will call the filesystem's write_super callback if
71  * appropriate. 
72  */
73 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
74 {
75         journal_t *journal;
76
77         if (sb->s_flags & MS_RDONLY)
78                 return ERR_PTR(-EROFS);
79
80         /* Special case here: if the journal has aborted behind our
81          * backs (eg. EIO in the commit thread), then we still need to
82          * take the FS itself readonly cleanly. */
83         journal = EXT3_SB(sb)->s_journal;
84         if (is_journal_aborted(journal)) {
85                 ext3_abort(sb, __FUNCTION__,
86                            "Detected aborted journal");
87                 return ERR_PTR(-EROFS);
88         }
89
90         return journal_start(journal, nblocks);
91 }
92
93 /* 
94  * The only special thing we need to do here is to make sure that all
95  * journal_stop calls result in the superblock being marked dirty, so
96  * that sync() will call the filesystem's write_super callback if
97  * appropriate. 
98  */
99 int __ext3_journal_stop(const char *where, handle_t *handle)
100 {
101         struct super_block *sb;
102         int err;
103         int rc;
104
105         sb = handle->h_transaction->t_journal->j_private;
106         err = handle->h_err;
107         rc = journal_stop(handle);
108
109         if (!err)
110                 err = rc;
111         if (err)
112                 __ext3_std_error(sb, where, err);
113         return err;
114 }
115
116 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
117                 struct buffer_head *bh, handle_t *handle, int err)
118 {
119         char nbuf[16];
120         const char *errstr = ext3_decode_error(NULL, err, nbuf);
121
122         if (bh)
123                 BUFFER_TRACE(bh, "abort");
124
125         if (!handle->h_err)
126                 handle->h_err = err;
127
128         if (is_handle_aborted(handle))
129                 return;
130
131         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
132                caller, errstr, err_fn);
133
134         journal_abort_handle(handle);
135 }
136
137 /* Deal with the reporting of failure conditions on a filesystem such as
138  * inconsistencies detected or read IO failures.
139  *
140  * On ext2, we can store the error state of the filesystem in the
141  * superblock.  That is not possible on ext3, because we may have other
142  * write ordering constraints on the superblock which prevent us from
143  * writing it out straight away; and given that the journal is about to
144  * be aborted, we can't rely on the current, or future, transactions to
145  * write out the superblock safely.
146  *
147  * We'll just use the journal_abort() error code to record an error in
148  * the journal instead.  On recovery, the journal will compain about
149  * that error until we've noted it down and cleared it.
150  */
151
152 static void ext3_handle_error(struct super_block *sb)
153 {
154         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
155
156         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
157         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
158
159         if (sb->s_flags & MS_RDONLY)
160                 return;
161
162         if (test_opt (sb, ERRORS_RO)) {
163                 printk (KERN_CRIT "Remounting filesystem read-only\n");
164                 sb->s_flags |= MS_RDONLY;
165         } else {
166                 journal_t *journal = EXT3_SB(sb)->s_journal;
167
168                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
169                 if (journal)
170                         journal_abort(journal, -EIO);
171         }
172         if (test_opt(sb, ERRORS_PANIC))
173                 panic("EXT3-fs (device %s): panic forced after error\n",
174                         sb->s_id);
175         ext3_commit_super(sb, es, 1);
176 }
177
178 void ext3_error (struct super_block * sb, const char * function,
179                  const char * fmt, ...)
180 {
181         va_list args;
182
183         va_start(args, fmt);
184         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
185         vprintk(fmt, args);
186         printk("\n");
187         va_end(args);
188
189         ext3_handle_error(sb);
190 }
191
192 static const char *ext3_decode_error(struct super_block * sb, int errno,
193                                      char nbuf[16])
194 {
195         char *errstr = NULL;
196
197         switch (errno) {
198         case -EIO:
199                 errstr = "IO failure";
200                 break;
201         case -ENOMEM:
202                 errstr = "Out of memory";
203                 break;
204         case -EROFS:
205                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
206                         errstr = "Journal has aborted";
207                 else
208                         errstr = "Readonly filesystem";
209                 break;
210         default:
211                 /* If the caller passed in an extra buffer for unknown
212                  * errors, textualise them now.  Else we just return
213                  * NULL. */
214                 if (nbuf) {
215                         /* Check for truncated error codes... */
216                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
217                                 errstr = nbuf;
218                 }
219                 break;
220         }
221
222         return errstr;
223 }
224
225 /* __ext3_std_error decodes expected errors from journaling functions
226  * automatically and invokes the appropriate error response.  */
227
228 void __ext3_std_error (struct super_block * sb, const char * function,
229                        int errno)
230 {
231         char nbuf[16];
232         const char *errstr;
233
234         /* Special case: if the error is EROFS, and we're not already
235          * inside a transaction, then there's really no point in logging
236          * an error. */
237         if (errno == -EROFS && journal_current_handle() == NULL &&
238             (sb->s_flags & MS_RDONLY))
239                 return;
240
241         errstr = ext3_decode_error(sb, errno, nbuf);
242         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
243                 sb->s_id, function, errstr);
244
245         ext3_handle_error(sb);
246 }
247
248 /*
249  * ext3_abort is a much stronger failure handler than ext3_error.  The
250  * abort function may be used to deal with unrecoverable failures such
251  * as journal IO errors or ENOMEM at a critical moment in log management.
252  *
253  * We unconditionally force the filesystem into an ABORT|READONLY state,
254  * unless the error response on the fs has been set to panic in which
255  * case we take the easy way out and panic immediately.
256  */
257
258 void ext3_abort (struct super_block * sb, const char * function,
259                  const char * fmt, ...)
260 {
261         va_list args;
262
263         printk (KERN_CRIT "ext3_abort called.\n");
264
265         va_start(args, fmt);
266         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
267         vprintk(fmt, args);
268         printk("\n");
269         va_end(args);
270
271         if (test_opt(sb, ERRORS_PANIC))
272                 panic("EXT3-fs panic from previous error\n");
273
274         if (sb->s_flags & MS_RDONLY)
275                 return;
276
277         printk(KERN_CRIT "Remounting filesystem read-only\n");
278         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
279         sb->s_flags |= MS_RDONLY;
280         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
281         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
282 }
283
284 void ext3_warning (struct super_block * sb, const char * function,
285                    const char * fmt, ...)
286 {
287         va_list args;
288
289         va_start(args, fmt);
290         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
291                sb->s_id, function);
292         vprintk(fmt, args);
293         printk("\n");
294         va_end(args);
295 }
296
297 void ext3_update_dynamic_rev(struct super_block *sb)
298 {
299         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
300
301         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
302                 return;
303
304         ext3_warning(sb, __FUNCTION__,
305                      "updating to rev %d because of new feature flag, "
306                      "running e2fsck is recommended",
307                      EXT3_DYNAMIC_REV);
308
309         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
310         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
311         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
312         /* leave es->s_feature_*compat flags alone */
313         /* es->s_uuid will be set by e2fsck if empty */
314
315         /*
316          * The rest of the superblock fields should be zero, and if not it
317          * means they are likely already in use, so leave them alone.  We
318          * can leave it up to e2fsck to clean up any inconsistencies there.
319          */
320 }
321
322 /*
323  * Open the external journal device
324  */
325 static struct block_device *ext3_blkdev_get(dev_t dev)
326 {
327         struct block_device *bdev;
328         char b[BDEVNAME_SIZE];
329
330         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
331         if (IS_ERR(bdev))
332                 goto fail;
333         return bdev;
334
335 fail:
336         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
337                         __bdevname(dev, b), PTR_ERR(bdev));
338         return NULL;
339 }
340
341 /*
342  * Release the journal device
343  */
344 static int ext3_blkdev_put(struct block_device *bdev)
345 {
346         bd_release(bdev);
347         return blkdev_put(bdev);
348 }
349
350 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
351 {
352         struct block_device *bdev;
353         int ret = -ENODEV;
354
355         bdev = sbi->journal_bdev;
356         if (bdev) {
357                 ret = ext3_blkdev_put(bdev);
358                 sbi->journal_bdev = NULL;
359         }
360         return ret;
361 }
362
363 static inline struct inode *orphan_list_entry(struct list_head *l)
364 {
365         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
366 }
367
368 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
369 {
370         struct list_head *l;
371
372         printk(KERN_ERR "sb orphan head is %d\n", 
373                le32_to_cpu(sbi->s_es->s_last_orphan));
374
375         printk(KERN_ERR "sb_info orphan list:\n");
376         list_for_each(l, &sbi->s_orphan) {
377                 struct inode *inode = orphan_list_entry(l);
378                 printk(KERN_ERR "  "
379                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
380                        inode->i_sb->s_id, inode->i_ino, inode,
381                        inode->i_mode, inode->i_nlink, 
382                        NEXT_ORPHAN(inode));
383         }
384 }
385
386 static void ext3_put_super (struct super_block * sb)
387 {
388         struct ext3_sb_info *sbi = EXT3_SB(sb);
389         struct ext3_super_block *es = sbi->s_es;
390         int i;
391
392         ext3_xattr_put_super(sb);
393         journal_destroy(sbi->s_journal);
394         if (!(sb->s_flags & MS_RDONLY)) {
395                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
396                 es->s_state = cpu_to_le16(sbi->s_mount_state);
397                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
398                 mark_buffer_dirty(sbi->s_sbh);
399                 ext3_commit_super(sb, es, 1);
400         }
401
402         for (i = 0; i < sbi->s_gdb_count; i++)
403                 brelse(sbi->s_group_desc[i]);
404         kfree(sbi->s_group_desc);
405         percpu_counter_destroy(&sbi->s_freeblocks_counter);
406         percpu_counter_destroy(&sbi->s_freeinodes_counter);
407         percpu_counter_destroy(&sbi->s_dirs_counter);
408         brelse(sbi->s_sbh);
409 #ifdef CONFIG_QUOTA
410         for (i = 0; i < MAXQUOTAS; i++)
411                 kfree(sbi->s_qf_names[i]);
412 #endif
413
414         /* Debugging code just in case the in-memory inode orphan list
415          * isn't empty.  The on-disk one can be non-empty if we've
416          * detected an error and taken the fs readonly, but the
417          * in-memory list had better be clean by this point. */
418         if (!list_empty(&sbi->s_orphan))
419                 dump_orphan_list(sb, sbi);
420         J_ASSERT(list_empty(&sbi->s_orphan));
421
422         invalidate_bdev(sb->s_bdev, 0);
423         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
424                 /*
425                  * Invalidate the journal device's buffers.  We don't want them
426                  * floating about in memory - the physical journal device may
427                  * hotswapped, and it breaks the `ro-after' testing code.
428                  */
429                 sync_blockdev(sbi->journal_bdev);
430                 invalidate_bdev(sbi->journal_bdev, 0);
431                 ext3_blkdev_remove(sbi);
432         }
433         sb->s_fs_info = NULL;
434         kfree(sbi);
435         return;
436 }
437
438 static kmem_cache_t *ext3_inode_cachep;
439
440 /*
441  * Called inside transaction, so use GFP_NOFS
442  */
443 static struct inode *ext3_alloc_inode(struct super_block *sb)
444 {
445         struct ext3_inode_info *ei;
446
447         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
448         if (!ei)
449                 return NULL;
450 #ifdef CONFIG_EXT3_FS_POSIX_ACL
451         ei->i_acl = EXT3_ACL_NOT_CACHED;
452         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
453 #endif
454         ei->i_block_alloc_info = NULL;
455         ei->vfs_inode.i_version = 1;
456         return &ei->vfs_inode;
457 }
458
459 static void ext3_destroy_inode(struct inode *inode)
460 {
461         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
462 }
463
464 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
465 {
466         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
467
468         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
469             SLAB_CTOR_CONSTRUCTOR) {
470                 INIT_LIST_HEAD(&ei->i_orphan);
471 #ifdef CONFIG_EXT3_FS_XATTR
472                 init_rwsem(&ei->xattr_sem);
473 #endif
474                 init_MUTEX(&ei->truncate_sem);
475                 inode_init_once(&ei->vfs_inode);
476         }
477 }
478  
479 static int init_inodecache(void)
480 {
481         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
482                                              sizeof(struct ext3_inode_info),
483                                              0, SLAB_RECLAIM_ACCOUNT,
484                                              init_once, NULL);
485         if (ext3_inode_cachep == NULL)
486                 return -ENOMEM;
487         return 0;
488 }
489
490 static void destroy_inodecache(void)
491 {
492         if (kmem_cache_destroy(ext3_inode_cachep))
493                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
494 }
495
496 static void ext3_clear_inode(struct inode *inode)
497 {
498         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
499 #ifdef CONFIG_EXT3_FS_POSIX_ACL
500        if (EXT3_I(inode)->i_acl &&
501            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
502                posix_acl_release(EXT3_I(inode)->i_acl);
503                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
504        }
505        if (EXT3_I(inode)->i_default_acl &&
506            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
507                posix_acl_release(EXT3_I(inode)->i_default_acl);
508                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
509        }
510 #endif
511         ext3_discard_reservation(inode);
512         EXT3_I(inode)->i_block_alloc_info = NULL;
513         kfree(rsv);
514 }
515
516 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
517 {
518 #if defined(CONFIG_QUOTA)
519         struct ext3_sb_info *sbi = EXT3_SB(sb);
520
521         if (sbi->s_jquota_fmt)
522                 seq_printf(seq, ",jqfmt=%s",
523                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
524
525         if (sbi->s_qf_names[USRQUOTA])
526                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
527
528         if (sbi->s_qf_names[GRPQUOTA])
529                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
530
531         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
532                 seq_puts(seq, ",usrquota");
533
534         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
535                 seq_puts(seq, ",grpquota");
536 #endif
537 }
538
539 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
540 {
541         struct super_block *sb = vfs->mnt_sb;
542
543         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
544                 seq_puts(seq, ",data=journal");
545         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
546                 seq_puts(seq, ",data=ordered");
547         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
548                 seq_puts(seq, ",data=writeback");
549
550         ext3_show_quota_options(seq, sb);
551
552         return 0;
553 }
554
555 #ifdef CONFIG_QUOTA
556 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
557 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
558
559 static int ext3_dquot_initialize(struct inode *inode, int type);
560 static int ext3_dquot_drop(struct inode *inode);
561 static int ext3_write_dquot(struct dquot *dquot);
562 static int ext3_acquire_dquot(struct dquot *dquot);
563 static int ext3_release_dquot(struct dquot *dquot);
564 static int ext3_mark_dquot_dirty(struct dquot *dquot);
565 static int ext3_write_info(struct super_block *sb, int type);
566 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
567 static int ext3_quota_on_mount(struct super_block *sb, int type);
568 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
569                                size_t len, loff_t off);
570 static ssize_t ext3_quota_write(struct super_block *sb, int type,
571                                 const char *data, size_t len, loff_t off);
572
573 static struct dquot_operations ext3_quota_operations = {
574         .initialize     = ext3_dquot_initialize,
575         .drop           = ext3_dquot_drop,
576         .alloc_space    = dquot_alloc_space,
577         .alloc_inode    = dquot_alloc_inode,
578         .free_space     = dquot_free_space,
579         .free_inode     = dquot_free_inode,
580         .transfer       = dquot_transfer,
581         .write_dquot    = ext3_write_dquot,
582         .acquire_dquot  = ext3_acquire_dquot,
583         .release_dquot  = ext3_release_dquot,
584         .mark_dirty     = ext3_mark_dquot_dirty,
585         .write_info     = ext3_write_info
586 };
587
588 static struct quotactl_ops ext3_qctl_operations = {
589         .quota_on       = ext3_quota_on,
590         .quota_off      = vfs_quota_off,
591         .quota_sync     = vfs_quota_sync,
592         .get_info       = vfs_get_dqinfo,
593         .set_info       = vfs_set_dqinfo,
594         .get_dqblk      = vfs_get_dqblk,
595         .set_dqblk      = vfs_set_dqblk
596 };
597 #endif
598
599 static struct super_operations ext3_sops = {
600         .alloc_inode    = ext3_alloc_inode,
601         .destroy_inode  = ext3_destroy_inode,
602         .read_inode     = ext3_read_inode,
603         .write_inode    = ext3_write_inode,
604         .dirty_inode    = ext3_dirty_inode,
605         .delete_inode   = ext3_delete_inode,
606         .put_super      = ext3_put_super,
607         .write_super    = ext3_write_super,
608         .sync_fs        = ext3_sync_fs,
609         .write_super_lockfs = ext3_write_super_lockfs,
610         .unlockfs       = ext3_unlockfs,
611         .statfs         = ext3_statfs,
612         .remount_fs     = ext3_remount,
613         .clear_inode    = ext3_clear_inode,
614         .show_options   = ext3_show_options,
615 #ifdef CONFIG_QUOTA
616         .quota_read     = ext3_quota_read,
617         .quota_write    = ext3_quota_write,
618 #endif
619 };
620
621 static struct export_operations ext3_export_ops = {
622         .get_parent = ext3_get_parent,
623 };
624
625 enum {
626         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
627         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
628         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
629         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
630         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
631         Opt_commit, Opt_journal_update, Opt_journal_inum,
632         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
633         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
634         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
635         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
636         Opt_grpquota
637 };
638
639 static match_table_t tokens = {
640         {Opt_bsd_df, "bsddf"},
641         {Opt_minix_df, "minixdf"},
642         {Opt_grpid, "grpid"},
643         {Opt_grpid, "bsdgroups"},
644         {Opt_nogrpid, "nogrpid"},
645         {Opt_nogrpid, "sysvgroups"},
646         {Opt_resgid, "resgid=%u"},
647         {Opt_resuid, "resuid=%u"},
648         {Opt_sb, "sb=%u"},
649         {Opt_err_cont, "errors=continue"},
650         {Opt_err_panic, "errors=panic"},
651         {Opt_err_ro, "errors=remount-ro"},
652         {Opt_nouid32, "nouid32"},
653         {Opt_nocheck, "nocheck"},
654         {Opt_nocheck, "check=none"},
655         {Opt_debug, "debug"},
656         {Opt_oldalloc, "oldalloc"},
657         {Opt_orlov, "orlov"},
658         {Opt_user_xattr, "user_xattr"},
659         {Opt_nouser_xattr, "nouser_xattr"},
660         {Opt_acl, "acl"},
661         {Opt_noacl, "noacl"},
662         {Opt_reservation, "reservation"},
663         {Opt_noreservation, "noreservation"},
664         {Opt_noload, "noload"},
665         {Opt_nobh, "nobh"},
666         {Opt_commit, "commit=%u"},
667         {Opt_journal_update, "journal=update"},
668         {Opt_journal_inum, "journal=%u"},
669         {Opt_abort, "abort"},
670         {Opt_data_journal, "data=journal"},
671         {Opt_data_ordered, "data=ordered"},
672         {Opt_data_writeback, "data=writeback"},
673         {Opt_offusrjquota, "usrjquota="},
674         {Opt_usrjquota, "usrjquota=%s"},
675         {Opt_offgrpjquota, "grpjquota="},
676         {Opt_grpjquota, "grpjquota=%s"},
677         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
678         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
679         {Opt_grpquota, "grpquota"},
680         {Opt_noquota, "noquota"},
681         {Opt_quota, "quota"},
682         {Opt_usrquota, "usrquota"},
683         {Opt_barrier, "barrier=%u"},
684         {Opt_err, NULL},
685         {Opt_resize, "resize"},
686 };
687
688 static unsigned long get_sb_block(void **data)
689 {
690         unsigned long   sb_block;
691         char            *options = (char *) *data;
692
693         if (!options || strncmp(options, "sb=", 3) != 0)
694                 return 1;       /* Default location */
695         options += 3;
696         sb_block = simple_strtoul(options, &options, 0);
697         if (*options && *options != ',') {
698                 printk("EXT3-fs: Invalid sb specification: %s\n",
699                        (char *) *data);
700                 return 1;
701         }
702         if (*options == ',')
703                 options++;
704         *data = (void *) options;
705         return sb_block;
706 }
707
708 static int parse_options (char * options, struct super_block *sb,
709                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
710 {
711         struct ext3_sb_info *sbi = EXT3_SB(sb);
712         char * p;
713         substring_t args[MAX_OPT_ARGS];
714         int data_opt = 0;
715         int option;
716 #ifdef CONFIG_QUOTA
717         int qtype;
718         char *qname;
719 #endif
720
721         if (!options)
722                 return 1;
723
724         while ((p = strsep (&options, ",")) != NULL) {
725                 int token;
726                 if (!*p)
727                         continue;
728
729                 token = match_token(p, tokens, args);
730                 switch (token) {
731                 case Opt_bsd_df:
732                         clear_opt (sbi->s_mount_opt, MINIX_DF);
733                         break;
734                 case Opt_minix_df:
735                         set_opt (sbi->s_mount_opt, MINIX_DF);
736                         break;
737                 case Opt_grpid:
738                         set_opt (sbi->s_mount_opt, GRPID);
739                         break;
740                 case Opt_nogrpid:
741                         clear_opt (sbi->s_mount_opt, GRPID);
742                         break;
743                 case Opt_resuid:
744                         if (match_int(&args[0], &option))
745                                 return 0;
746                         sbi->s_resuid = option;
747                         break;
748                 case Opt_resgid:
749                         if (match_int(&args[0], &option))
750                                 return 0;
751                         sbi->s_resgid = option;
752                         break;
753                 case Opt_sb:
754                         /* handled by get_sb_block() instead of here */
755                         /* *sb_block = match_int(&args[0]); */
756                         break;
757                 case Opt_err_panic:
758                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
759                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
760                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
761                         break;
762                 case Opt_err_ro:
763                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
764                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
765                         set_opt (sbi->s_mount_opt, ERRORS_RO);
766                         break;
767                 case Opt_err_cont:
768                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
769                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
770                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
771                         break;
772                 case Opt_nouid32:
773                         set_opt (sbi->s_mount_opt, NO_UID32);
774                         break;
775                 case Opt_nocheck:
776                         clear_opt (sbi->s_mount_opt, CHECK);
777                         break;
778                 case Opt_debug:
779                         set_opt (sbi->s_mount_opt, DEBUG);
780                         break;
781                 case Opt_oldalloc:
782                         set_opt (sbi->s_mount_opt, OLDALLOC);
783                         break;
784                 case Opt_orlov:
785                         clear_opt (sbi->s_mount_opt, OLDALLOC);
786                         break;
787 #ifdef CONFIG_EXT3_FS_XATTR
788                 case Opt_user_xattr:
789                         set_opt (sbi->s_mount_opt, XATTR_USER);
790                         break;
791                 case Opt_nouser_xattr:
792                         clear_opt (sbi->s_mount_opt, XATTR_USER);
793                         break;
794 #else
795                 case Opt_user_xattr:
796                 case Opt_nouser_xattr:
797                         printk("EXT3 (no)user_xattr options not supported\n");
798                         break;
799 #endif
800 #ifdef CONFIG_EXT3_FS_POSIX_ACL
801                 case Opt_acl:
802                         set_opt(sbi->s_mount_opt, POSIX_ACL);
803                         break;
804                 case Opt_noacl:
805                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
806                         break;
807 #else
808                 case Opt_acl:
809                 case Opt_noacl:
810                         printk("EXT3 (no)acl options not supported\n");
811                         break;
812 #endif
813                 case Opt_reservation:
814                         set_opt(sbi->s_mount_opt, RESERVATION);
815                         break;
816                 case Opt_noreservation:
817                         clear_opt(sbi->s_mount_opt, RESERVATION);
818                         break;
819                 case Opt_journal_update:
820                         /* @@@ FIXME */
821                         /* Eventually we will want to be able to create
822                            a journal file here.  For now, only allow the
823                            user to specify an existing inode to be the
824                            journal file. */
825                         if (is_remount) {
826                                 printk(KERN_ERR "EXT3-fs: cannot specify "
827                                        "journal on remount\n");
828                                 return 0;
829                         }
830                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
831                         break;
832                 case Opt_journal_inum:
833                         if (is_remount) {
834                                 printk(KERN_ERR "EXT3-fs: cannot specify "
835                                        "journal on remount\n");
836                                 return 0;
837                         }
838                         if (match_int(&args[0], &option))
839                                 return 0;
840                         *inum = option;
841                         break;
842                 case Opt_noload:
843                         set_opt (sbi->s_mount_opt, NOLOAD);
844                         break;
845                 case Opt_commit:
846                         if (match_int(&args[0], &option))
847                                 return 0;
848                         if (option < 0)
849                                 return 0;
850                         if (option == 0)
851                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
852                         sbi->s_commit_interval = HZ * option;
853                         break;
854                 case Opt_data_journal:
855                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
856                         goto datacheck;
857                 case Opt_data_ordered:
858                         data_opt = EXT3_MOUNT_ORDERED_DATA;
859                         goto datacheck;
860                 case Opt_data_writeback:
861                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
862                 datacheck:
863                         if (is_remount) {
864                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
865                                                 != data_opt) {
866                                         printk(KERN_ERR
867                                                 "EXT3-fs: cannot change data "
868                                                 "mode on remount\n");
869                                         return 0;
870                                 }
871                         } else {
872                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
873                                 sbi->s_mount_opt |= data_opt;
874                         }
875                         break;
876 #ifdef CONFIG_QUOTA
877                 case Opt_usrjquota:
878                         qtype = USRQUOTA;
879                         goto set_qf_name;
880                 case Opt_grpjquota:
881                         qtype = GRPQUOTA;
882 set_qf_name:
883                         if (sb_any_quota_enabled(sb)) {
884                                 printk(KERN_ERR
885                                         "EXT3-fs: Cannot change journalled "
886                                         "quota options when quota turned on.\n");
887                                 return 0;
888                         }
889                         qname = match_strdup(&args[0]);
890                         if (!qname) {
891                                 printk(KERN_ERR
892                                         "EXT3-fs: not enough memory for "
893                                         "storing quotafile name.\n");
894                                 return 0;
895                         }
896                         if (sbi->s_qf_names[qtype] &&
897                             strcmp(sbi->s_qf_names[qtype], qname)) {
898                                 printk(KERN_ERR
899                                         "EXT3-fs: %s quota file already "
900                                         "specified.\n", QTYPE2NAME(qtype));
901                                 kfree(qname);
902                                 return 0;
903                         }
904                         sbi->s_qf_names[qtype] = qname;
905                         if (strchr(sbi->s_qf_names[qtype], '/')) {
906                                 printk(KERN_ERR
907                                         "EXT3-fs: quotafile must be on "
908                                         "filesystem root.\n");
909                                 kfree(sbi->s_qf_names[qtype]);
910                                 sbi->s_qf_names[qtype] = NULL;
911                                 return 0;
912                         }
913                         set_opt(sbi->s_mount_opt, QUOTA);
914                         break;
915                 case Opt_offusrjquota:
916                         qtype = USRQUOTA;
917                         goto clear_qf_name;
918                 case Opt_offgrpjquota:
919                         qtype = GRPQUOTA;
920 clear_qf_name:
921                         if (sb_any_quota_enabled(sb)) {
922                                 printk(KERN_ERR "EXT3-fs: Cannot change "
923                                         "journalled quota options when "
924                                         "quota turned on.\n");
925                                 return 0;
926                         }
927                         /*
928                          * The space will be released later when all options
929                          * are confirmed to be correct
930                          */
931                         sbi->s_qf_names[qtype] = NULL;
932                         break;
933                 case Opt_jqfmt_vfsold:
934                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
935                         break;
936                 case Opt_jqfmt_vfsv0:
937                         sbi->s_jquota_fmt = QFMT_VFS_V0;
938                         break;
939                 case Opt_quota:
940                 case Opt_usrquota:
941                         set_opt(sbi->s_mount_opt, QUOTA);
942                         set_opt(sbi->s_mount_opt, USRQUOTA);
943                         break;
944                 case Opt_grpquota:
945                         set_opt(sbi->s_mount_opt, QUOTA);
946                         set_opt(sbi->s_mount_opt, GRPQUOTA);
947                         break;
948                 case Opt_noquota:
949                         if (sb_any_quota_enabled(sb)) {
950                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
951                                         "options when quota turned on.\n");
952                                 return 0;
953                         }
954                         clear_opt(sbi->s_mount_opt, QUOTA);
955                         clear_opt(sbi->s_mount_opt, USRQUOTA);
956                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
957                         break;
958 #else
959                 case Opt_quota:
960                 case Opt_usrquota:
961                 case Opt_grpquota:
962                 case Opt_usrjquota:
963                 case Opt_grpjquota:
964                 case Opt_offusrjquota:
965                 case Opt_offgrpjquota:
966                 case Opt_jqfmt_vfsold:
967                 case Opt_jqfmt_vfsv0:
968                         printk(KERN_ERR
969                                 "EXT3-fs: journalled quota options not "
970                                 "supported.\n");
971                         break;
972                 case Opt_noquota:
973                         break;
974 #endif
975                 case Opt_abort:
976                         set_opt(sbi->s_mount_opt, ABORT);
977                         break;
978                 case Opt_barrier:
979                         if (match_int(&args[0], &option))
980                                 return 0;
981                         if (option)
982                                 set_opt(sbi->s_mount_opt, BARRIER);
983                         else
984                                 clear_opt(sbi->s_mount_opt, BARRIER);
985                         break;
986                 case Opt_ignore:
987                         break;
988                 case Opt_resize:
989                         if (!is_remount) {
990                                 printk("EXT3-fs: resize option only available "
991                                         "for remount\n");
992                                 return 0;
993                         }
994                         if (match_int(&args[0], &option) != 0)
995                                 return 0;
996                         *n_blocks_count = option;
997                         break;
998                 case Opt_nobh:
999                         set_opt(sbi->s_mount_opt, NOBH);
1000                         break;
1001                 default:
1002                         printk (KERN_ERR
1003                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1004                                 "or missing value\n", p);
1005                         return 0;
1006                 }
1007         }
1008 #ifdef CONFIG_QUOTA
1009         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1010                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1011                      sbi->s_qf_names[USRQUOTA])
1012                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1013
1014                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1015                      sbi->s_qf_names[GRPQUOTA])
1016                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1017
1018                 if ((sbi->s_qf_names[USRQUOTA] &&
1019                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1020                     (sbi->s_qf_names[GRPQUOTA] &&
1021                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1022                         printk(KERN_ERR "EXT3-fs: old and new quota "
1023                                         "format mixing.\n");
1024                         return 0;
1025                 }
1026
1027                 if (!sbi->s_jquota_fmt) {
1028                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1029                                         "not specified.\n");
1030                         return 0;
1031                 }
1032         } else {
1033                 if (sbi->s_jquota_fmt) {
1034                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1035                                         "specified with no journalling "
1036                                         "enabled.\n");
1037                         return 0;
1038                 }
1039         }
1040 #endif
1041         return 1;
1042 }
1043
1044 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1045                             int read_only)
1046 {
1047         struct ext3_sb_info *sbi = EXT3_SB(sb);
1048         int res = 0;
1049
1050         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1051                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1052                         "forcing read-only mode\n");
1053                 res = MS_RDONLY;
1054         }
1055         if (read_only)
1056                 return res;
1057         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1058                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1059                         "running e2fsck is recommended\n");
1060         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1061                 printk (KERN_WARNING
1062                         "EXT3-fs warning: mounting fs with errors, "
1063                         "running e2fsck is recommended\n");
1064         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1065                  le16_to_cpu(es->s_mnt_count) >=
1066                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1067                 printk (KERN_WARNING
1068                         "EXT3-fs warning: maximal mount count reached, "
1069                         "running e2fsck is recommended\n");
1070         else if (le32_to_cpu(es->s_checkinterval) &&
1071                 (le32_to_cpu(es->s_lastcheck) +
1072                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1073                 printk (KERN_WARNING
1074                         "EXT3-fs warning: checktime reached, "
1075                         "running e2fsck is recommended\n");
1076 #if 0
1077                 /* @@@ We _will_ want to clear the valid bit if we find
1078                    inconsistencies, to force a fsck at reboot.  But for
1079                    a plain journaled filesystem we can keep it set as
1080                    valid forever! :) */
1081         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1082 #endif
1083         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1084                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1085         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1086         es->s_mtime = cpu_to_le32(get_seconds());
1087         ext3_update_dynamic_rev(sb);
1088         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1089
1090         ext3_commit_super(sb, es, 1);
1091         if (test_opt(sb, DEBUG))
1092                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1093                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1094                         sb->s_blocksize,
1095                         sbi->s_groups_count,
1096                         EXT3_BLOCKS_PER_GROUP(sb),
1097                         EXT3_INODES_PER_GROUP(sb),
1098                         sbi->s_mount_opt);
1099
1100         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1101         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1102                 char b[BDEVNAME_SIZE];
1103
1104                 printk("external journal on %s\n",
1105                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1106         } else {
1107                 printk("internal journal\n");
1108         }
1109         return res;
1110 }
1111
1112 /* Called at mount-time, super-block is locked */
1113 static int ext3_check_descriptors (struct super_block * sb)
1114 {
1115         struct ext3_sb_info *sbi = EXT3_SB(sb);
1116         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1117         struct ext3_group_desc * gdp = NULL;
1118         int desc_block = 0;
1119         int i;
1120
1121         ext3_debug ("Checking group descriptors");
1122
1123         for (i = 0; i < sbi->s_groups_count; i++)
1124         {
1125                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1126                         gdp = (struct ext3_group_desc *)
1127                                         sbi->s_group_desc[desc_block++]->b_data;
1128                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1129                     le32_to_cpu(gdp->bg_block_bitmap) >=
1130                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1131                 {
1132                         ext3_error (sb, "ext3_check_descriptors",
1133                                     "Block bitmap for group %d"
1134                                     " not in group (block %lu)!",
1135                                     i, (unsigned long)
1136                                         le32_to_cpu(gdp->bg_block_bitmap));
1137                         return 0;
1138                 }
1139                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1140                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1141                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1142                 {
1143                         ext3_error (sb, "ext3_check_descriptors",
1144                                     "Inode bitmap for group %d"
1145                                     " not in group (block %lu)!",
1146                                     i, (unsigned long)
1147                                         le32_to_cpu(gdp->bg_inode_bitmap));
1148                         return 0;
1149                 }
1150                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1151                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1152                     block + EXT3_BLOCKS_PER_GROUP(sb))
1153                 {
1154                         ext3_error (sb, "ext3_check_descriptors",
1155                                     "Inode table for group %d"
1156                                     " not in group (block %lu)!",
1157                                     i, (unsigned long)
1158                                         le32_to_cpu(gdp->bg_inode_table));
1159                         return 0;
1160                 }
1161                 block += EXT3_BLOCKS_PER_GROUP(sb);
1162                 gdp++;
1163         }
1164
1165         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1166         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1167         return 1;
1168 }
1169
1170
1171 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1172  * the superblock) which were deleted from all directories, but held open by
1173  * a process at the time of a crash.  We walk the list and try to delete these
1174  * inodes at recovery time (only with a read-write filesystem).
1175  *
1176  * In order to keep the orphan inode chain consistent during traversal (in
1177  * case of crash during recovery), we link each inode into the superblock
1178  * orphan list_head and handle it the same way as an inode deletion during
1179  * normal operation (which journals the operations for us).
1180  *
1181  * We only do an iget() and an iput() on each inode, which is very safe if we
1182  * accidentally point at an in-use or already deleted inode.  The worst that
1183  * can happen in this case is that we get a "bit already cleared" message from
1184  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1185  * e2fsck was run on this filesystem, and it must have already done the orphan
1186  * inode cleanup for us, so we can safely abort without any further action.
1187  */
1188 static void ext3_orphan_cleanup (struct super_block * sb,
1189                                  struct ext3_super_block * es)
1190 {
1191         unsigned int s_flags = sb->s_flags;
1192         int nr_orphans = 0, nr_truncates = 0;
1193 #ifdef CONFIG_QUOTA
1194         int i;
1195 #endif
1196         if (!es->s_last_orphan) {
1197                 jbd_debug(4, "no orphan inodes to clean up\n");
1198                 return;
1199         }
1200
1201         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1202                 if (es->s_last_orphan)
1203                         jbd_debug(1, "Errors on filesystem, "
1204                                   "clearing orphan list.\n");
1205                 es->s_last_orphan = 0;
1206                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1207                 return;
1208         }
1209
1210         if (s_flags & MS_RDONLY) {
1211                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1212                        sb->s_id);
1213                 sb->s_flags &= ~MS_RDONLY;
1214         }
1215 #ifdef CONFIG_QUOTA
1216         /* Needed for iput() to work correctly and not trash data */
1217         sb->s_flags |= MS_ACTIVE;
1218         /* Turn on quotas so that they are updated correctly */
1219         for (i = 0; i < MAXQUOTAS; i++) {
1220                 if (EXT3_SB(sb)->s_qf_names[i]) {
1221                         int ret = ext3_quota_on_mount(sb, i);
1222                         if (ret < 0)
1223                                 printk(KERN_ERR
1224                                         "EXT3-fs: Cannot turn on journalled "
1225                                         "quota: error %d\n", ret);
1226                 }
1227         }
1228 #endif
1229
1230         while (es->s_last_orphan) {
1231                 struct inode *inode;
1232
1233                 if (!(inode =
1234                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1235                         es->s_last_orphan = 0;
1236                         break;
1237                 }
1238
1239                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1240                 DQUOT_INIT(inode);
1241                 if (inode->i_nlink) {
1242                         printk(KERN_DEBUG
1243                                 "%s: truncating inode %ld to %Ld bytes\n",
1244                                 __FUNCTION__, inode->i_ino, inode->i_size);
1245                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1246                                   inode->i_ino, inode->i_size);
1247                         ext3_truncate(inode);
1248                         nr_truncates++;
1249                 } else {
1250                         printk(KERN_DEBUG
1251                                 "%s: deleting unreferenced inode %ld\n",
1252                                 __FUNCTION__, inode->i_ino);
1253                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1254                                   inode->i_ino);
1255                         nr_orphans++;
1256                 }
1257                 iput(inode);  /* The delete magic happens here! */
1258         }
1259
1260 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1261
1262         if (nr_orphans)
1263                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1264                        sb->s_id, PLURAL(nr_orphans));
1265         if (nr_truncates)
1266                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1267                        sb->s_id, PLURAL(nr_truncates));
1268 #ifdef CONFIG_QUOTA
1269         /* Turn quotas off */
1270         for (i = 0; i < MAXQUOTAS; i++) {
1271                 if (sb_dqopt(sb)->files[i])
1272                         vfs_quota_off(sb, i);
1273         }
1274 #endif
1275         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1276 }
1277
1278 #define log2(n) ffz(~(n))
1279
1280 /*
1281  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1282  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1283  * We need to be 1 filesystem block less than the 2^32 sector limit.
1284  */
1285 static loff_t ext3_max_size(int bits)
1286 {
1287         loff_t res = EXT3_NDIR_BLOCKS;
1288         /* This constant is calculated to be the largest file size for a
1289          * dense, 4k-blocksize file such that the total number of
1290          * sectors in the file, including data and all indirect blocks,
1291          * does not exceed 2^32. */
1292         const loff_t upper_limit = 0x1ff7fffd000LL;
1293
1294         res += 1LL << (bits-2);
1295         res += 1LL << (2*(bits-2));
1296         res += 1LL << (3*(bits-2));
1297         res <<= bits;
1298         if (res > upper_limit)
1299                 res = upper_limit;
1300         return res;
1301 }
1302
1303 static unsigned long descriptor_loc(struct super_block *sb,
1304                                     unsigned long logic_sb_block,
1305                                     int nr)
1306 {
1307         struct ext3_sb_info *sbi = EXT3_SB(sb);
1308         unsigned long bg, first_data_block, first_meta_bg;
1309         int has_super = 0;
1310
1311         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1312         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1313
1314         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1315             nr < first_meta_bg)
1316                 return (logic_sb_block + nr + 1);
1317         bg = sbi->s_desc_per_block * nr;
1318         if (ext3_bg_has_super(sb, bg))
1319                 has_super = 1;
1320         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1321 }
1322
1323
1324 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1325 {
1326         struct buffer_head * bh;
1327         struct ext3_super_block *es = NULL;
1328         struct ext3_sb_info *sbi;
1329         unsigned long block;
1330         unsigned long sb_block = get_sb_block(&data);
1331         unsigned long logic_sb_block;
1332         unsigned long offset = 0;
1333         unsigned long journal_inum = 0;
1334         unsigned long def_mount_opts;
1335         struct inode *root;
1336         int blocksize;
1337         int hblock;
1338         int db_count;
1339         int i;
1340         int needs_recovery;
1341         __le32 features;
1342
1343         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1344         if (!sbi)
1345                 return -ENOMEM;
1346         sb->s_fs_info = sbi;
1347         memset(sbi, 0, sizeof(*sbi));
1348         sbi->s_mount_opt = 0;
1349         sbi->s_resuid = EXT3_DEF_RESUID;
1350         sbi->s_resgid = EXT3_DEF_RESGID;
1351
1352         unlock_kernel();
1353
1354         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1355         if (!blocksize) {
1356                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1357                 goto out_fail;
1358         }
1359
1360         /*
1361          * The ext3 superblock will not be buffer aligned for other than 1kB
1362          * block sizes.  We need to calculate the offset from buffer start.
1363          */
1364         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1365                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1366                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1367         } else {
1368                 logic_sb_block = sb_block;
1369         }
1370
1371         if (!(bh = sb_bread(sb, logic_sb_block))) {
1372                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1373                 goto out_fail;
1374         }
1375         /*
1376          * Note: s_es must be initialized as soon as possible because
1377          *       some ext3 macro-instructions depend on its value
1378          */
1379         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1380         sbi->s_es = es;
1381         sb->s_magic = le16_to_cpu(es->s_magic);
1382         if (sb->s_magic != EXT3_SUPER_MAGIC)
1383                 goto cantfind_ext3;
1384
1385         /* Set defaults before we parse the mount options */
1386         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1387         if (def_mount_opts & EXT3_DEFM_DEBUG)
1388                 set_opt(sbi->s_mount_opt, DEBUG);
1389         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1390                 set_opt(sbi->s_mount_opt, GRPID);
1391         if (def_mount_opts & EXT3_DEFM_UID16)
1392                 set_opt(sbi->s_mount_opt, NO_UID32);
1393         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1394                 set_opt(sbi->s_mount_opt, XATTR_USER);
1395         if (def_mount_opts & EXT3_DEFM_ACL)
1396                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1397         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1398                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1399         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1400                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1401         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1402                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1403
1404         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1405                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1406         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1407                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1408
1409         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1410         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1411
1412         set_opt(sbi->s_mount_opt, RESERVATION);
1413
1414         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1415                 goto failed_mount;
1416
1417         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1418                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1419
1420         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1421             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1422              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1423              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1424                 printk(KERN_WARNING 
1425                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1426                        "running e2fsck is recommended\n");
1427         /*
1428          * Check feature flags regardless of the revision level, since we
1429          * previously didn't change the revision level when setting the flags,
1430          * so there is a chance incompat flags are set on a rev 0 filesystem.
1431          */
1432         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1433         if (features) {
1434                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1435                        "unsupported optional features (%x).\n",
1436                        sb->s_id, le32_to_cpu(features));
1437                 goto failed_mount;
1438         }
1439         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1440         if (!(sb->s_flags & MS_RDONLY) && features) {
1441                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1442                        "unsupported optional features (%x).\n",
1443                        sb->s_id, le32_to_cpu(features));
1444                 goto failed_mount;
1445         }
1446         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1447
1448         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1449             blocksize > EXT3_MAX_BLOCK_SIZE) {
1450                 printk(KERN_ERR 
1451                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1452                        blocksize, sb->s_id);
1453                 goto failed_mount;
1454         }
1455
1456         hblock = bdev_hardsect_size(sb->s_bdev);
1457         if (sb->s_blocksize != blocksize) {
1458                 /*
1459                  * Make sure the blocksize for the filesystem is larger
1460                  * than the hardware sectorsize for the machine.
1461                  */
1462                 if (blocksize < hblock) {
1463                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1464                                "device blocksize %d.\n", blocksize, hblock);
1465                         goto failed_mount;
1466                 }
1467
1468                 brelse (bh);
1469                 sb_set_blocksize(sb, blocksize);
1470                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1471                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1472                 bh = sb_bread(sb, logic_sb_block);
1473                 if (!bh) {
1474                         printk(KERN_ERR 
1475                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1476                         goto failed_mount;
1477                 }
1478                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1479                 sbi->s_es = es;
1480                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1481                         printk (KERN_ERR 
1482                                 "EXT3-fs: Magic mismatch, very weird !\n");
1483                         goto failed_mount;
1484                 }
1485         }
1486
1487         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1488
1489         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1490                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1491                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1492         } else {
1493                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1494                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1495                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1496                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1497                     (sbi->s_inode_size > blocksize)) {
1498                         printk (KERN_ERR
1499                                 "EXT3-fs: unsupported inode size: %d\n",
1500                                 sbi->s_inode_size);
1501                         goto failed_mount;
1502                 }
1503         }
1504         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1505                                    le32_to_cpu(es->s_log_frag_size);
1506         if (blocksize != sbi->s_frag_size) {
1507                 printk(KERN_ERR
1508                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1509                        sbi->s_frag_size, blocksize);
1510                 goto failed_mount;
1511         }
1512         sbi->s_frags_per_block = 1;
1513         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1514         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1515         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1516         if (EXT3_INODE_SIZE(sb) == 0)
1517                 goto cantfind_ext3;
1518         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1519         if (sbi->s_inodes_per_block == 0)
1520                 goto cantfind_ext3;
1521         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1522                                         sbi->s_inodes_per_block;
1523         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1524         sbi->s_sbh = bh;
1525         sbi->s_mount_state = le16_to_cpu(es->s_state);
1526         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1527         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1528         for (i=0; i < 4; i++)
1529                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1530         sbi->s_def_hash_version = es->s_def_hash_version;
1531
1532         if (sbi->s_blocks_per_group > blocksize * 8) {
1533                 printk (KERN_ERR
1534                         "EXT3-fs: #blocks per group too big: %lu\n",
1535                         sbi->s_blocks_per_group);
1536                 goto failed_mount;
1537         }
1538         if (sbi->s_frags_per_group > blocksize * 8) {
1539                 printk (KERN_ERR
1540                         "EXT3-fs: #fragments per group too big: %lu\n",
1541                         sbi->s_frags_per_group);
1542                 goto failed_mount;
1543         }
1544         if (sbi->s_inodes_per_group > blocksize * 8) {
1545                 printk (KERN_ERR
1546                         "EXT3-fs: #inodes per group too big: %lu\n",
1547                         sbi->s_inodes_per_group);
1548                 goto failed_mount;
1549         }
1550
1551         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1552                 goto cantfind_ext3;
1553         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1554                                le32_to_cpu(es->s_first_data_block) +
1555                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1556                               EXT3_BLOCKS_PER_GROUP(sb);
1557         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1558                    EXT3_DESC_PER_BLOCK(sb);
1559         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1560                                     GFP_KERNEL);
1561         if (sbi->s_group_desc == NULL) {
1562                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1563                 goto failed_mount;
1564         }
1565
1566         percpu_counter_init(&sbi->s_freeblocks_counter);
1567         percpu_counter_init(&sbi->s_freeinodes_counter);
1568         percpu_counter_init(&sbi->s_dirs_counter);
1569         bgl_lock_init(&sbi->s_blockgroup_lock);
1570
1571         for (i = 0; i < db_count; i++) {
1572                 block = descriptor_loc(sb, logic_sb_block, i);
1573                 sbi->s_group_desc[i] = sb_bread(sb, block);
1574                 if (!sbi->s_group_desc[i]) {
1575                         printk (KERN_ERR "EXT3-fs: "
1576                                 "can't read group descriptor %d\n", i);
1577                         db_count = i;
1578                         goto failed_mount2;
1579                 }
1580         }
1581         if (!ext3_check_descriptors (sb)) {
1582                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1583                 goto failed_mount2;
1584         }
1585         sbi->s_gdb_count = db_count;
1586         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1587         spin_lock_init(&sbi->s_next_gen_lock);
1588         /* per fileystem reservation list head & lock */
1589         spin_lock_init(&sbi->s_rsv_window_lock);
1590         sbi->s_rsv_window_root = RB_ROOT;
1591         /* Add a single, static dummy reservation to the start of the
1592          * reservation window list --- it gives us a placeholder for
1593          * append-at-start-of-list which makes the allocation logic
1594          * _much_ simpler. */
1595         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1596         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1597         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1598         sbi->s_rsv_window_head.rsv_goal_size = 0;
1599         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1600
1601         /*
1602          * set up enough so that it can read an inode
1603          */
1604         sb->s_op = &ext3_sops;
1605         sb->s_export_op = &ext3_export_ops;
1606         sb->s_xattr = ext3_xattr_handlers;
1607 #ifdef CONFIG_QUOTA
1608         sb->s_qcop = &ext3_qctl_operations;
1609         sb->dq_op = &ext3_quota_operations;
1610 #endif
1611         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1612
1613         sb->s_root = NULL;
1614
1615         needs_recovery = (es->s_last_orphan != 0 ||
1616                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1617                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1618
1619         /*
1620          * The first inode we look at is the journal inode.  Don't try
1621          * root first: it may be modified in the journal!
1622          */
1623         if (!test_opt(sb, NOLOAD) &&
1624             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1625                 if (ext3_load_journal(sb, es))
1626                         goto failed_mount2;
1627         } else if (journal_inum) {
1628                 if (ext3_create_journal(sb, es, journal_inum))
1629                         goto failed_mount2;
1630         } else {
1631                 if (!silent)
1632                         printk (KERN_ERR
1633                                 "ext3: No journal on filesystem on %s\n",
1634                                 sb->s_id);
1635                 goto failed_mount2;
1636         }
1637
1638         /* We have now updated the journal if required, so we can
1639          * validate the data journaling mode. */
1640         switch (test_opt(sb, DATA_FLAGS)) {
1641         case 0:
1642                 /* No mode set, assume a default based on the journal
1643                    capabilities: ORDERED_DATA if the journal can
1644                    cope, else JOURNAL_DATA */
1645                 if (journal_check_available_features
1646                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1647                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1648                 else
1649                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1650                 break;
1651
1652         case EXT3_MOUNT_ORDERED_DATA:
1653         case EXT3_MOUNT_WRITEBACK_DATA:
1654                 if (!journal_check_available_features
1655                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1656                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1657                                "requested data journaling mode\n");
1658                         goto failed_mount3;
1659                 }
1660         default:
1661                 break;
1662         }
1663
1664         if (test_opt(sb, NOBH)) {
1665                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1666                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1667                                 "since filesystem blocksize doesn't match "
1668                                 "pagesize\n");
1669                         clear_opt(sbi->s_mount_opt, NOBH);
1670                 }
1671                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1672                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1673                                 "its supported only with writeback mode\n");
1674                         clear_opt(sbi->s_mount_opt, NOBH);
1675                 }
1676         }
1677         /*
1678          * The journal_load will have done any necessary log recovery,
1679          * so we can safely mount the rest of the filesystem now.
1680          */
1681
1682         root = iget(sb, EXT3_ROOT_INO);
1683         sb->s_root = d_alloc_root(root);
1684         if (!sb->s_root) {
1685                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1686                 iput(root);
1687                 goto failed_mount3;
1688         }
1689         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1690                 dput(sb->s_root);
1691                 sb->s_root = NULL;
1692                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1693                 goto failed_mount3;
1694         }
1695
1696         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1697         /*
1698          * akpm: core read_super() calls in here with the superblock locked.
1699          * That deadlocks, because orphan cleanup needs to lock the superblock
1700          * in numerous places.  Here we just pop the lock - it's relatively
1701          * harmless, because we are now ready to accept write_super() requests,
1702          * and aviro says that's the only reason for hanging onto the
1703          * superblock lock.
1704          */
1705         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1706         ext3_orphan_cleanup(sb, es);
1707         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1708         if (needs_recovery)
1709                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1710         ext3_mark_recovery_complete(sb, es);
1711         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1712                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1713                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1714                 "writeback");
1715
1716         percpu_counter_mod(&sbi->s_freeblocks_counter,
1717                 ext3_count_free_blocks(sb));
1718         percpu_counter_mod(&sbi->s_freeinodes_counter,
1719                 ext3_count_free_inodes(sb));
1720         percpu_counter_mod(&sbi->s_dirs_counter,
1721                 ext3_count_dirs(sb));
1722
1723         lock_kernel();
1724         return 0;
1725
1726 cantfind_ext3:
1727         if (!silent)
1728                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1729                        sb->s_id);
1730         goto failed_mount;
1731
1732 failed_mount3:
1733         journal_destroy(sbi->s_journal);
1734 failed_mount2:
1735         for (i = 0; i < db_count; i++)
1736                 brelse(sbi->s_group_desc[i]);
1737         kfree(sbi->s_group_desc);
1738 failed_mount:
1739 #ifdef CONFIG_QUOTA
1740         for (i = 0; i < MAXQUOTAS; i++)
1741                 kfree(sbi->s_qf_names[i]);
1742 #endif
1743         ext3_blkdev_remove(sbi);
1744         brelse(bh);
1745 out_fail:
1746         sb->s_fs_info = NULL;
1747         kfree(sbi);
1748         lock_kernel();
1749         return -EINVAL;
1750 }
1751
1752 /*
1753  * Setup any per-fs journal parameters now.  We'll do this both on
1754  * initial mount, once the journal has been initialised but before we've
1755  * done any recovery; and again on any subsequent remount. 
1756  */
1757 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1758 {
1759         struct ext3_sb_info *sbi = EXT3_SB(sb);
1760
1761         if (sbi->s_commit_interval)
1762                 journal->j_commit_interval = sbi->s_commit_interval;
1763         /* We could also set up an ext3-specific default for the commit
1764          * interval here, but for now we'll just fall back to the jbd
1765          * default. */
1766
1767         spin_lock(&journal->j_state_lock);
1768         if (test_opt(sb, BARRIER))
1769                 journal->j_flags |= JFS_BARRIER;
1770         else
1771                 journal->j_flags &= ~JFS_BARRIER;
1772         spin_unlock(&journal->j_state_lock);
1773 }
1774
1775 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1776 {
1777         struct inode *journal_inode;
1778         journal_t *journal;
1779
1780         /* First, test for the existence of a valid inode on disk.  Bad
1781          * things happen if we iget() an unused inode, as the subsequent
1782          * iput() will try to delete it. */
1783
1784         journal_inode = iget(sb, journal_inum);
1785         if (!journal_inode) {
1786                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1787                 return NULL;
1788         }
1789         if (!journal_inode->i_nlink) {
1790                 make_bad_inode(journal_inode);
1791                 iput(journal_inode);
1792                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1793                 return NULL;
1794         }
1795
1796         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1797                   journal_inode, journal_inode->i_size);
1798         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1799                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1800                 iput(journal_inode);
1801                 return NULL;
1802         }
1803
1804         journal = journal_init_inode(journal_inode);
1805         if (!journal) {
1806                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1807                 iput(journal_inode);
1808                 return NULL;
1809         }
1810         journal->j_private = sb;
1811         ext3_init_journal_params(sb, journal);
1812         return journal;
1813 }
1814
1815 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1816                                        dev_t j_dev)
1817 {
1818         struct buffer_head * bh;
1819         journal_t *journal;
1820         int start;
1821         int len;
1822         int hblock, blocksize;
1823         unsigned long sb_block;
1824         unsigned long offset;
1825         struct ext3_super_block * es;
1826         struct block_device *bdev;
1827
1828         bdev = ext3_blkdev_get(j_dev);
1829         if (bdev == NULL)
1830                 return NULL;
1831
1832         if (bd_claim(bdev, sb)) {
1833                 printk(KERN_ERR
1834                         "EXT3: failed to claim external journal device.\n");
1835                 blkdev_put(bdev);
1836                 return NULL;
1837         }
1838
1839         blocksize = sb->s_blocksize;
1840         hblock = bdev_hardsect_size(bdev);
1841         if (blocksize < hblock) {
1842                 printk(KERN_ERR
1843                         "EXT3-fs: blocksize too small for journal device.\n");
1844                 goto out_bdev;
1845         }
1846
1847         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1848         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1849         set_blocksize(bdev, blocksize);
1850         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1851                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1852                        "external journal\n");
1853                 goto out_bdev;
1854         }
1855
1856         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1857         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1858             !(le32_to_cpu(es->s_feature_incompat) &
1859               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1860                 printk(KERN_ERR "EXT3-fs: external journal has "
1861                                         "bad superblock\n");
1862                 brelse(bh);
1863                 goto out_bdev;
1864         }
1865
1866         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1867                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1868                 brelse(bh);
1869                 goto out_bdev;
1870         }
1871
1872         len = le32_to_cpu(es->s_blocks_count);
1873         start = sb_block + 1;
1874         brelse(bh);     /* we're done with the superblock */
1875
1876         journal = journal_init_dev(bdev, sb->s_bdev,
1877                                         start, len, blocksize);
1878         if (!journal) {
1879                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1880                 goto out_bdev;
1881         }
1882         journal->j_private = sb;
1883         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1884         wait_on_buffer(journal->j_sb_buffer);
1885         if (!buffer_uptodate(journal->j_sb_buffer)) {
1886                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1887                 goto out_journal;
1888         }
1889         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1890                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1891                                         "user (unsupported) - %d\n",
1892                         be32_to_cpu(journal->j_superblock->s_nr_users));
1893                 goto out_journal;
1894         }
1895         EXT3_SB(sb)->journal_bdev = bdev;
1896         ext3_init_journal_params(sb, journal);
1897         return journal;
1898 out_journal:
1899         journal_destroy(journal);
1900 out_bdev:
1901         ext3_blkdev_put(bdev);
1902         return NULL;
1903 }
1904
1905 static int ext3_load_journal(struct super_block * sb,
1906                              struct ext3_super_block * es)
1907 {
1908         journal_t *journal;
1909         int journal_inum = le32_to_cpu(es->s_journal_inum);
1910         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1911         int err = 0;
1912         int really_read_only;
1913
1914         really_read_only = bdev_read_only(sb->s_bdev);
1915
1916         /*
1917          * Are we loading a blank journal or performing recovery after a
1918          * crash?  For recovery, we need to check in advance whether we
1919          * can get read-write access to the device.
1920          */
1921
1922         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1923                 if (sb->s_flags & MS_RDONLY) {
1924                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1925                                         "required on readonly filesystem.\n");
1926                         if (really_read_only) {
1927                                 printk(KERN_ERR "EXT3-fs: write access "
1928                                         "unavailable, cannot proceed.\n");
1929                                 return -EROFS;
1930                         }
1931                         printk (KERN_INFO "EXT3-fs: write access will "
1932                                         "be enabled during recovery.\n");
1933                 }
1934         }
1935
1936         if (journal_inum && journal_dev) {
1937                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1938                        "and inode journals!\n");
1939                 return -EINVAL;
1940         }
1941
1942         if (journal_inum) {
1943                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1944                         return -EINVAL;
1945         } else {
1946                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1947                         return -EINVAL;
1948         }
1949
1950         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1951                 err = journal_update_format(journal);
1952                 if (err)  {
1953                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1954                         journal_destroy(journal);
1955                         return err;
1956                 }
1957         }
1958
1959         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1960                 err = journal_wipe(journal, !really_read_only);
1961         if (!err)
1962                 err = journal_load(journal);
1963
1964         if (err) {
1965                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1966                 journal_destroy(journal);
1967                 return err;
1968         }
1969
1970         EXT3_SB(sb)->s_journal = journal;
1971         ext3_clear_journal_err(sb, es);
1972         return 0;
1973 }
1974
1975 static int ext3_create_journal(struct super_block * sb,
1976                                struct ext3_super_block * es,
1977                                int journal_inum)
1978 {
1979         journal_t *journal;
1980
1981         if (sb->s_flags & MS_RDONLY) {
1982                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1983                                 "create journal.\n");
1984                 return -EROFS;
1985         }
1986
1987         if (!(journal = ext3_get_journal(sb, journal_inum)))
1988                 return -EINVAL;
1989
1990         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1991                journal_inum);
1992
1993         if (journal_create(journal)) {
1994                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1995                 journal_destroy(journal);
1996                 return -EIO;
1997         }
1998
1999         EXT3_SB(sb)->s_journal = journal;
2000
2001         ext3_update_dynamic_rev(sb);
2002         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2003         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2004
2005         es->s_journal_inum = cpu_to_le32(journal_inum);
2006         sb->s_dirt = 1;
2007
2008         /* Make sure we flush the recovery flag to disk. */
2009         ext3_commit_super(sb, es, 1);
2010
2011         return 0;
2012 }
2013
2014 static void ext3_commit_super (struct super_block * sb,
2015                                struct ext3_super_block * es,
2016                                int sync)
2017 {
2018         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2019
2020         if (!sbh)
2021                 return;
2022         es->s_wtime = cpu_to_le32(get_seconds());
2023         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2024         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2025         BUFFER_TRACE(sbh, "marking dirty");
2026         mark_buffer_dirty(sbh);
2027         if (sync)
2028                 sync_dirty_buffer(sbh);
2029 }
2030
2031
2032 /*
2033  * Have we just finished recovery?  If so, and if we are mounting (or
2034  * remounting) the filesystem readonly, then we will end up with a
2035  * consistent fs on disk.  Record that fact.
2036  */
2037 static void ext3_mark_recovery_complete(struct super_block * sb,
2038                                         struct ext3_super_block * es)
2039 {
2040         journal_t *journal = EXT3_SB(sb)->s_journal;
2041
2042         journal_lock_updates(journal);
2043         journal_flush(journal);
2044         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2045             sb->s_flags & MS_RDONLY) {
2046                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2047                 sb->s_dirt = 0;
2048                 ext3_commit_super(sb, es, 1);
2049         }
2050         journal_unlock_updates(journal);
2051 }
2052
2053 /*
2054  * If we are mounting (or read-write remounting) a filesystem whose journal
2055  * has recorded an error from a previous lifetime, move that error to the
2056  * main filesystem now.
2057  */
2058 static void ext3_clear_journal_err(struct super_block * sb,
2059                                    struct ext3_super_block * es)
2060 {
2061         journal_t *journal;
2062         int j_errno;
2063         const char *errstr;
2064
2065         journal = EXT3_SB(sb)->s_journal;
2066
2067         /*
2068          * Now check for any error status which may have been recorded in the
2069          * journal by a prior ext3_error() or ext3_abort()
2070          */
2071
2072         j_errno = journal_errno(journal);
2073         if (j_errno) {
2074                 char nbuf[16];
2075
2076                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2077                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2078                              "from previous mount: %s", errstr);
2079                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2080                              "filesystem check.");
2081
2082                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2083                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2084                 ext3_commit_super (sb, es, 1);
2085
2086                 journal_clear_err(journal);
2087         }
2088 }
2089
2090 /*
2091  * Force the running and committing transactions to commit,
2092  * and wait on the commit.
2093  */
2094 int ext3_force_commit(struct super_block *sb)
2095 {
2096         journal_t *journal;
2097         int ret;
2098
2099         if (sb->s_flags & MS_RDONLY)
2100                 return 0;
2101
2102         journal = EXT3_SB(sb)->s_journal;
2103         sb->s_dirt = 0;
2104         ret = ext3_journal_force_commit(journal);
2105         return ret;
2106 }
2107
2108 /*
2109  * Ext3 always journals updates to the superblock itself, so we don't
2110  * have to propagate any other updates to the superblock on disk at this
2111  * point.  Just start an async writeback to get the buffers on their way
2112  * to the disk.
2113  *
2114  * This implicitly triggers the writebehind on sync().
2115  */
2116
2117 static void ext3_write_super (struct super_block * sb)
2118 {
2119         if (down_trylock(&sb->s_lock) == 0)
2120                 BUG();
2121         sb->s_dirt = 0;
2122 }
2123
2124 static int ext3_sync_fs(struct super_block *sb, int wait)
2125 {
2126         tid_t target;
2127
2128         sb->s_dirt = 0;
2129         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2130                 if (wait)
2131                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2132         }
2133         return 0;
2134 }
2135
2136 /*
2137  * LVM calls this function before a (read-only) snapshot is created.  This
2138  * gives us a chance to flush the journal completely and mark the fs clean.
2139  */
2140 static void ext3_write_super_lockfs(struct super_block *sb)
2141 {
2142         sb->s_dirt = 0;
2143
2144         if (!(sb->s_flags & MS_RDONLY)) {
2145                 journal_t *journal = EXT3_SB(sb)->s_journal;
2146
2147                 /* Now we set up the journal barrier. */
2148                 journal_lock_updates(journal);
2149                 journal_flush(journal);
2150
2151                 /* Journal blocked and flushed, clear needs_recovery flag. */
2152                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2153                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2154         }
2155 }
2156
2157 /*
2158  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2159  * flag here, even though the filesystem is not technically dirty yet.
2160  */
2161 static void ext3_unlockfs(struct super_block *sb)
2162 {
2163         if (!(sb->s_flags & MS_RDONLY)) {
2164                 lock_super(sb);
2165                 /* Reser the needs_recovery flag before the fs is unlocked. */
2166                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2167                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2168                 unlock_super(sb);
2169                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2170         }
2171 }
2172
2173 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2174 {
2175         struct ext3_super_block * es;
2176         struct ext3_sb_info *sbi = EXT3_SB(sb);
2177         unsigned long n_blocks_count = 0;
2178         unsigned long old_sb_flags;
2179         struct ext3_mount_options old_opts;
2180         int err;
2181 #ifdef CONFIG_QUOTA
2182         int i;
2183 #endif
2184
2185         /* Store the original options */
2186         old_sb_flags = sb->s_flags;
2187         old_opts.s_mount_opt = sbi->s_mount_opt;
2188         old_opts.s_resuid = sbi->s_resuid;
2189         old_opts.s_resgid = sbi->s_resgid;
2190         old_opts.s_commit_interval = sbi->s_commit_interval;
2191 #ifdef CONFIG_QUOTA
2192         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2193         for (i = 0; i < MAXQUOTAS; i++)
2194                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2195 #endif
2196
2197         /*
2198          * Allow the "check" option to be passed as a remount option.
2199          */
2200         if (!parse_options(data, sb, NULL, &n_blocks_count, 1)) {
2201                 err = -EINVAL;
2202                 goto restore_opts;
2203         }
2204
2205         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2206                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2207
2208         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2209                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2210
2211         es = sbi->s_es;
2212
2213         ext3_init_journal_params(sb, sbi->s_journal);
2214
2215         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2216                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2217                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2218                         err = -EROFS;
2219                         goto restore_opts;
2220                 }
2221
2222                 if (*flags & MS_RDONLY) {
2223                         /*
2224                          * First of all, the unconditional stuff we have to do
2225                          * to disable replay of the journal when we next remount
2226                          */
2227                         sb->s_flags |= MS_RDONLY;
2228
2229                         /*
2230                          * OK, test if we are remounting a valid rw partition
2231                          * readonly, and if so set the rdonly flag and then
2232                          * mark the partition as valid again.
2233                          */
2234                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2235                             (sbi->s_mount_state & EXT3_VALID_FS))
2236                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2237
2238                         ext3_mark_recovery_complete(sb, es);
2239                 } else {
2240                         __le32 ret;
2241                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2242                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2243                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2244                                        "remount RDWR because of unsupported "
2245                                        "optional features (%x).\n",
2246                                        sb->s_id, le32_to_cpu(ret));
2247                                 err = -EROFS;
2248                                 goto restore_opts;
2249                         }
2250                         /*
2251                          * Mounting a RDONLY partition read-write, so reread
2252                          * and store the current valid flag.  (It may have
2253                          * been changed by e2fsck since we originally mounted
2254                          * the partition.)
2255                          */
2256                         ext3_clear_journal_err(sb, es);
2257                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2258                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2259                                 err = ret;
2260                                 goto restore_opts;
2261                         }
2262                         if (!ext3_setup_super (sb, es, 0))
2263                                 sb->s_flags &= ~MS_RDONLY;
2264                 }
2265         }
2266 #ifdef CONFIG_QUOTA
2267         /* Release old quota file names */
2268         for (i = 0; i < MAXQUOTAS; i++)
2269                 if (old_opts.s_qf_names[i] &&
2270                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2271                         kfree(old_opts.s_qf_names[i]);
2272 #endif
2273         return 0;
2274 restore_opts:
2275         sb->s_flags = old_sb_flags;
2276         sbi->s_mount_opt = old_opts.s_mount_opt;
2277         sbi->s_resuid = old_opts.s_resuid;
2278         sbi->s_resgid = old_opts.s_resgid;
2279         sbi->s_commit_interval = old_opts.s_commit_interval;
2280 #ifdef CONFIG_QUOTA
2281         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2282         for (i = 0; i < MAXQUOTAS; i++) {
2283                 if (sbi->s_qf_names[i] &&
2284                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2285                         kfree(sbi->s_qf_names[i]);
2286                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2287         }
2288 #endif
2289         return err;
2290 }
2291
2292 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2293 {
2294         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2295         unsigned long overhead;
2296         int i;
2297
2298         if (test_opt (sb, MINIX_DF))
2299                 overhead = 0;
2300         else {
2301                 unsigned long ngroups;
2302                 ngroups = EXT3_SB(sb)->s_groups_count;
2303                 smp_rmb();
2304
2305                 /*
2306                  * Compute the overhead (FS structures)
2307                  */
2308
2309                 /*
2310                  * All of the blocks before first_data_block are
2311                  * overhead
2312                  */
2313                 overhead = le32_to_cpu(es->s_first_data_block);
2314
2315                 /*
2316                  * Add the overhead attributed to the superblock and
2317                  * block group descriptors.  If the sparse superblocks
2318                  * feature is turned on, then not all groups have this.
2319                  */
2320                 for (i = 0; i < ngroups; i++) {
2321                         overhead += ext3_bg_has_super(sb, i) +
2322                                 ext3_bg_num_gdb(sb, i);
2323                         cond_resched();
2324                 }
2325
2326                 /*
2327                  * Every block group has an inode bitmap, a block
2328                  * bitmap, and an inode table.
2329                  */
2330                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2331         }
2332
2333         buf->f_type = EXT3_SUPER_MAGIC;
2334         buf->f_bsize = sb->s_blocksize;
2335         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2336         buf->f_bfree = ext3_count_free_blocks (sb);
2337         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2338         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2339                 buf->f_bavail = 0;
2340         buf->f_files = le32_to_cpu(es->s_inodes_count);
2341         buf->f_ffree = ext3_count_free_inodes (sb);
2342         buf->f_namelen = EXT3_NAME_LEN;
2343         return 0;
2344 }
2345
2346 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2347  * is locked for write. Otherwise the are possible deadlocks:
2348  * Process 1                         Process 2
2349  * ext3_create()                     quota_sync()
2350  *   journal_start()                   write_dquot()
2351  *   DQUOT_INIT()                        down(dqio_sem)
2352  *     down(dqio_sem)                    journal_start()
2353  *
2354  */
2355
2356 #ifdef CONFIG_QUOTA
2357
2358 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2359 {
2360         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2361 }
2362
2363 static int ext3_dquot_initialize(struct inode *inode, int type)
2364 {
2365         handle_t *handle;
2366         int ret, err;
2367
2368         /* We may create quota structure so we need to reserve enough blocks */
2369         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2370         if (IS_ERR(handle))
2371                 return PTR_ERR(handle);
2372         ret = dquot_initialize(inode, type);
2373         err = ext3_journal_stop(handle);
2374         if (!ret)
2375                 ret = err;
2376         return ret;
2377 }
2378
2379 static int ext3_dquot_drop(struct inode *inode)
2380 {
2381         handle_t *handle;
2382         int ret, err;
2383
2384         /* We may delete quota structure so we need to reserve enough blocks */
2385         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2386         if (IS_ERR(handle))
2387                 return PTR_ERR(handle);
2388         ret = dquot_drop(inode);
2389         err = ext3_journal_stop(handle);
2390         if (!ret)
2391                 ret = err;
2392         return ret;
2393 }
2394
2395 static int ext3_write_dquot(struct dquot *dquot)
2396 {
2397         int ret, err;
2398         handle_t *handle;
2399         struct inode *inode;
2400
2401         inode = dquot_to_inode(dquot);
2402         handle = ext3_journal_start(inode,
2403                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2404         if (IS_ERR(handle))
2405                 return PTR_ERR(handle);
2406         ret = dquot_commit(dquot);
2407         err = ext3_journal_stop(handle);
2408         if (!ret)
2409                 ret = err;
2410         return ret;
2411 }
2412
2413 static int ext3_acquire_dquot(struct dquot *dquot)
2414 {
2415         int ret, err;
2416         handle_t *handle;
2417
2418         handle = ext3_journal_start(dquot_to_inode(dquot),
2419                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2420         if (IS_ERR(handle))
2421                 return PTR_ERR(handle);
2422         ret = dquot_acquire(dquot);
2423         err = ext3_journal_stop(handle);
2424         if (!ret)
2425                 ret = err;
2426         return ret;
2427 }
2428
2429 static int ext3_release_dquot(struct dquot *dquot)
2430 {
2431         int ret, err;
2432         handle_t *handle;
2433
2434         handle = ext3_journal_start(dquot_to_inode(dquot),
2435                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2436         if (IS_ERR(handle))
2437                 return PTR_ERR(handle);
2438         ret = dquot_release(dquot);
2439         err = ext3_journal_stop(handle);
2440         if (!ret)
2441                 ret = err;
2442         return ret;
2443 }
2444
2445 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2446 {
2447         /* Are we journalling quotas? */
2448         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2449             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2450                 dquot_mark_dquot_dirty(dquot);
2451                 return ext3_write_dquot(dquot);
2452         } else {
2453                 return dquot_mark_dquot_dirty(dquot);
2454         }
2455 }
2456
2457 static int ext3_write_info(struct super_block *sb, int type)
2458 {
2459         int ret, err;
2460         handle_t *handle;
2461
2462         /* Data block + inode block */
2463         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2464         if (IS_ERR(handle))
2465                 return PTR_ERR(handle);
2466         ret = dquot_commit_info(sb, type);
2467         err = ext3_journal_stop(handle);
2468         if (!ret)
2469                 ret = err;
2470         return ret;
2471 }
2472
2473 /*
2474  * Turn on quotas during mount time - we need to find
2475  * the quota file and such...
2476  */
2477 static int ext3_quota_on_mount(struct super_block *sb, int type)
2478 {
2479         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2480                         EXT3_SB(sb)->s_jquota_fmt, type);
2481 }
2482
2483 /*
2484  * Standard function to be called on quota_on
2485  */
2486 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2487                          char *path)
2488 {
2489         int err;
2490         struct nameidata nd;
2491
2492         if (!test_opt(sb, QUOTA))
2493                 return -EINVAL;
2494         /* Not journalling quota? */
2495         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2496             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2497                 return vfs_quota_on(sb, type, format_id, path);
2498         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2499         if (err)
2500                 return err;
2501         /* Quotafile not on the same filesystem? */
2502         if (nd.mnt->mnt_sb != sb) {
2503                 path_release(&nd);
2504                 return -EXDEV;
2505         }
2506         /* Quotafile not of fs root? */
2507         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2508                 printk(KERN_WARNING
2509                         "EXT3-fs: Quota file not on filesystem root. "
2510                         "Journalled quota will not work.\n");
2511         path_release(&nd);
2512         return vfs_quota_on(sb, type, format_id, path);
2513 }
2514
2515 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2516  * acquiring the locks... As quota files are never truncated and quota code
2517  * itself serializes the operations (and noone else should touch the files)
2518  * we don't have to be afraid of races */
2519 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2520                                size_t len, loff_t off)
2521 {
2522         struct inode *inode = sb_dqopt(sb)->files[type];
2523         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2524         int err = 0;
2525         int offset = off & (sb->s_blocksize - 1);
2526         int tocopy;
2527         size_t toread;
2528         struct buffer_head *bh;
2529         loff_t i_size = i_size_read(inode);
2530
2531         if (off > i_size)
2532                 return 0;
2533         if (off+len > i_size)
2534                 len = i_size-off;
2535         toread = len;
2536         while (toread > 0) {
2537                 tocopy = sb->s_blocksize - offset < toread ?
2538                                 sb->s_blocksize - offset : toread;
2539                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2540                 if (err)
2541                         return err;
2542                 if (!bh)        /* A hole? */
2543                         memset(data, 0, tocopy);
2544                 else
2545                         memcpy(data, bh->b_data+offset, tocopy);
2546                 brelse(bh);
2547                 offset = 0;
2548                 toread -= tocopy;
2549                 data += tocopy;
2550                 blk++;
2551         }
2552         return len;
2553 }
2554
2555 /* Write to quotafile (we know the transaction is already started and has
2556  * enough credits) */
2557 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2558                                 const char *data, size_t len, loff_t off)
2559 {
2560         struct inode *inode = sb_dqopt(sb)->files[type];
2561         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2562         int err = 0;
2563         int offset = off & (sb->s_blocksize - 1);
2564         int tocopy;
2565         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2566         size_t towrite = len;
2567         struct buffer_head *bh;
2568         handle_t *handle = journal_current_handle();
2569
2570         down(&inode->i_sem);
2571         while (towrite > 0) {
2572                 tocopy = sb->s_blocksize - offset < towrite ?
2573                                 sb->s_blocksize - offset : towrite;
2574                 bh = ext3_bread(handle, inode, blk, 1, &err);
2575                 if (!bh)
2576                         goto out;
2577                 if (journal_quota) {
2578                         err = ext3_journal_get_write_access(handle, bh);
2579                         if (err) {
2580                                 brelse(bh);
2581                                 goto out;
2582                         }
2583                 }
2584                 lock_buffer(bh);
2585                 memcpy(bh->b_data+offset, data, tocopy);
2586                 flush_dcache_page(bh->b_page);
2587                 unlock_buffer(bh);
2588                 if (journal_quota)
2589                         err = ext3_journal_dirty_metadata(handle, bh);
2590                 else {
2591                         /* Always do at least ordered writes for quotas */
2592                         err = ext3_journal_dirty_data(handle, bh);
2593                         mark_buffer_dirty(bh);
2594                 }
2595                 brelse(bh);
2596                 if (err)
2597                         goto out;
2598                 offset = 0;
2599                 towrite -= tocopy;
2600                 data += tocopy;
2601                 blk++;
2602         }
2603 out:
2604         if (len == towrite)
2605                 return err;
2606         if (inode->i_size < off+len-towrite) {
2607                 i_size_write(inode, off+len-towrite);
2608                 EXT3_I(inode)->i_disksize = inode->i_size;
2609         }
2610         inode->i_version++;
2611         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2612         ext3_mark_inode_dirty(handle, inode);
2613         up(&inode->i_sem);
2614         return len - towrite;
2615 }
2616
2617 #endif
2618
2619 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2620         int flags, const char *dev_name, void *data)
2621 {
2622         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2623 }
2624
2625 static struct file_system_type ext3_fs_type = {
2626         .owner          = THIS_MODULE,
2627         .name           = "ext3",
2628         .get_sb         = ext3_get_sb,
2629         .kill_sb        = kill_block_super,
2630         .fs_flags       = FS_REQUIRES_DEV,
2631 };
2632
2633 static int __init init_ext3_fs(void)
2634 {
2635         int err = init_ext3_xattr();
2636         if (err)
2637                 return err;
2638         err = init_inodecache();
2639         if (err)
2640                 goto out1;
2641         err = register_filesystem(&ext3_fs_type);
2642         if (err)
2643                 goto out;
2644         return 0;
2645 out:
2646         destroy_inodecache();
2647 out1:
2648         exit_ext3_xattr();
2649         return err;
2650 }
2651
2652 static void __exit exit_ext3_fs(void)
2653 {
2654         unregister_filesystem(&ext3_fs_type);
2655         destroy_inodecache();
2656         exit_ext3_xattr();
2657 }
2658
2659 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2660 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2661 MODULE_LICENSE("GPL");
2662 module_init(init_ext3_fs)
2663 module_exit(exit_ext3_fs)